[go: up one dir, main page]

CN120541419B - Quality assessment method for bottle nitrogen filling process by using artificial intelligence - Google Patents

Quality assessment method for bottle nitrogen filling process by using artificial intelligence

Info

Publication number
CN120541419B
CN120541419B CN202510628010.8A CN202510628010A CN120541419B CN 120541419 B CN120541419 B CN 120541419B CN 202510628010 A CN202510628010 A CN 202510628010A CN 120541419 B CN120541419 B CN 120541419B
Authority
CN
China
Prior art keywords
airflow
characteristic
data set
air flow
vortex
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202510628010.8A
Other languages
Chinese (zh)
Other versions
CN120541419A (en
Inventor
张建明
张建国
徐锦泉
朱月华
谢卓航
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Xitai Packaging Equipment Co ltd
Original Assignee
Guangdong Xitai Packaging Equipment Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Xitai Packaging Equipment Co ltd filed Critical Guangdong Xitai Packaging Equipment Co ltd
Priority to CN202510628010.8A priority Critical patent/CN120541419B/en
Publication of CN120541419A publication Critical patent/CN120541419A/en
Application granted granted Critical
Publication of CN120541419B publication Critical patent/CN120541419B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/27—Regression, e.g. linear or logistic regression
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/08—Fluids
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Biomedical Technology (AREA)
  • Algebra (AREA)
  • Fluid Mechanics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请提供一种应用人工智能的瓶体充氮过程质量评估方法，包括：获取充氮设备瓶口处气流速度分布、涡流强度及流线形态，进行特征提取和时间序列分析，生成气流特征数据集；若气流特征稳定性低于预设阈值，则通过长短时记忆网络算法预测气流速度分布、涡流强度及流线形态的短期变化趋势，得到气流特征的未来演化状态；根据气流特征的未来演化状态调整充氮设备的氮气注入速率与注入角度，生成优化的充氮控制参数；根据优化后的充氮控制参数调整氮气注入参数，实时更新充氮设备的运行状态，生成更新后的气流特征数据集。

This application provides a quality assessment method for the nitrogen filling process of a bottle using artificial intelligence, comprising: acquiring the airflow velocity distribution, vortex intensity, and streamline morphology at the bottle opening of the nitrogen filling equipment, performing feature extraction and time series analysis to generate an airflow feature dataset; if the stability of the airflow features is lower than a preset threshold, predicting the short-term change trend of the airflow velocity distribution, vortex intensity, and streamline morphology using a long short-term memory network algorithm to obtain the future evolution state of the airflow features; adjusting the nitrogen injection rate and injection angle of the nitrogen filling equipment according to the future evolution state of the airflow features to generate optimized nitrogen filling control parameters; adjusting the nitrogen injection parameters according to the optimized nitrogen filling control parameters, updating the operating status of the nitrogen filling equipment in real time, and generating an updated airflow feature dataset.

Description

Quality assessment method for bottle nitrogen filling process by using artificial intelligence
Technical Field
The invention relates to the technical field of information, in particular to a quality evaluation method for a bottle nitrogen filling process by using artificial intelligence.
Background
The quality evaluation of nitrogen filling is a key link in the packaging process of the industries of food, medicine and the like, and directly relates to the quality guarantee period and the safety of the product. The nitrogen filling process is to inject nitrogen into the sealed container to replace oxygen so as to inhibit oxidation reaction, and the quality evaluation is critical to the production efficiency and the product quality. The traditional nitrogen filling quality assessment methods are mostly dependent on pressure detection or gas concentration analysis, and although the methods are simple to operate, when the pressure in the bottle is close to the external pressure, the method is difficult to accurately capture the fine change of the gas flow at the bottle mouth, so that the judgment of the nitrogen filling completion state is not accurate enough. In addition, the existing method has insufficient analysis on the dynamic flow field characteristics such as airflow speed distribution, vortex intensity, streamline morphology and the like, and limits the optimal control of the nitrogen charging process. The key challenges lie in the complex changes of the gas flow characteristics at the bottle mouth during pressure balance, including the change of the uniformity of the gas flow velocity distribution, the attenuation characteristics of the vortex intensity and the stability fluctuation of the streamline form, which are difficult to quantify in real time by the traditional detection means, and directly affect the accurate judgment of the nitrogen charging completion state. The prior art lacks a dynamic analysis means for the flow field characteristics, and is difficult to reduce energy consumption and improve production efficiency while guaranteeing quality. Therefore, how to accurately judge the completion state of the nitrogen charging process based on the bottleneck airflow characteristics and combining with an artificial intelligent flow field analysis algorithm becomes a key problem for optimizing the energy consumption and the production efficiency of the nitrogen charging system.
Disclosure of Invention
The invention provides a quality evaluation method for a bottle nitrogen filling process by applying artificial intelligence, which mainly comprises the following steps:
Acquiring airflow speed distribution, vortex intensity and streamline form at the bottleneck of nitrogen charging equipment, and carrying out feature extraction and time sequence analysis to generate an airflow feature data set;
Extracting uniformity characteristics of airflow speed distribution, attenuation characteristics of vortex intensity and stability characteristics of streamline morphology in the airflow characteristic data set, and performing characteristic fusion processing to obtain multidimensional vector representation of bottleneck airflow characteristics;
Carrying out multidimensional vector representation on the airflow velocity distribution, the vortex intensity and the streamline morphology, constructing a pre-trained flow field dynamic analysis model, predicting the dynamic change rule of the airflow characteristics in a pressure balance state, and obtaining the stability index of the airflow characteristics of the bottle mouth through stability analysis;
if the stability of the airflow characteristics is lower than a preset threshold, predicting short-term change trends of airflow speed distribution, vortex strength and streamline form through a long-short-term memory network algorithm to obtain future evolution states of the airflow characteristics;
adjusting the nitrogen injection rate and the injection angle of the nitrogen charging equipment according to the future evolution state of the airflow characteristics to generate optimized nitrogen charging control parameters;
According to the optimized nitrogen filling control parameters, adjusting nitrogen filling parameters, updating the running state of nitrogen filling equipment in real time, and generating an updated airflow characteristic data set;
inputting the updated airflow characteristic data set into a flow field dynamic analysis model, recalculating characteristic vectors of airflow speed distribution, vortex strength and streamline form, and outputting an updated airflow characteristic stability index so as to determine the completion state of the nitrogen charging process;
if the air flow characteristic stability index reaches the threshold value, classifying and updating the index through a logistic regression algorithm, outputting the nitrogen charging completion probability, generating a termination signal according to the probability, controlling the nitrogen charging equipment to stop, recording an air flow characteristic data set, and determining the production efficiency improvement amplitude.
Further, the method comprises the steps of obtaining the airflow speed distribution, the vortex intensity and the streamline form at the bottleneck of the nitrogen charging equipment, carrying out feature extraction and time sequence analysis to generate an airflow feature data set, wherein an air sensor array is arranged at grid points divided by the diameter of the bottleneck, measuring the airflow speed data of each point in the array by adopting an ultrasonic air flowmeter, acquiring air inlet pressure data by adopting an air pressure sensor, and carrying out noise reduction and standardization treatment on the sensor data to obtain a first airflow data set. And constructing an airflow velocity field distribution diagram according to the first airflow data set, scanning a bottleneck plane by adopting an airflow probe to obtain streamline track data, calculating the vorticity value of the airflow field, and superposing a vorticity contour line on the basis of the airflow velocity field distribution diagram to obtain a first vortex distribution diagram. And calculating a vortex intensity gradient according to the first vortex distribution diagram, performing hierarchical clustering on the vortex intensity gradient data to obtain a vortex intensity interval division result, and extracting curvature characteristic parameters aiming at the streamline track data. And acquiring gas concentration data in the nitrogen filling process by adopting a sensor array, constructing a Bayesian probability network by combining the first airflow data set, and extracting a gas concentration distribution characteristic vector from an airflow velocity field distribution map. And performing db4 wavelet transformation decomposition on the airflow pulsation frequency signal sequence, extracting a detail coefficient sequence, and constructing airflow pulsation characteristic data according to the coefficient sequence. And superposing a streamline track curve on the basis of the first vortex distribution diagram, calculating a gas diffusion coefficient according to the gas concentration distribution characteristic vector, and quantitatively describing the structure form of the flow field by adopting the diffusion coefficient to obtain a gas flow characteristic time sequence data set so as to form a gas flow characteristic data set.
Further, the method comprises the steps of extracting uniformity characteristics of airflow speed distribution, attenuation characteristics of vortex intensity and stability characteristics of streamline morphology in airflow characteristic data, carrying out characteristic fusion processing to obtain multidimensional vector representation of airflow characteristics of a bottle mouth, cleaning the airflow characteristic data set, removing abnormal values by adopting a median filter, and segmenting time sequence data by a sliding window method to obtain a preprocessed airflow characteristic data set. Calculating an airflow velocity field variance distribution diagram of the bottleneck area according to the preprocessed airflow characteristic data set, calculating a first velocity field fluctuation value by adopting gridding variance, and carrying out maximum and minimum normalization processing on the fluctuation value to obtain a first airflow uniformity index. And extracting an eddy current intensity time sequence according to the preprocessed air current characteristic data set, calculating the eddy current intensity difference value at adjacent moments to obtain a first eddy current intensity change curve, and fitting the change curve by adopting an exponential decay function to obtain eddy current decay parameters. And extracting a streamline track curve according to the preprocessed airflow characteristic data set, calculating a curvature change rate to obtain a first curvature sequence, and performing dimension reduction treatment on the curvature sequence by adopting a principal component analysis method to obtain a three-dimensional principal characteristic vector. Constructing an eight-layer recurrent neural network, setting the number of nodes of an input layer to be three, setting the number of nodes of a hidden layer to be sixteen, training the network by adopting a first curvature sequence, and extracting network output layer data to obtain a streamline form stability feature vector. And constructing a feature correlation matrix, wherein the dimension of the matrix is the same as the sum of the first airflow uniformity index, the vortex attenuation parameter and the streamline form stability feature vector dimension, and calculating the pairwise relation number between the features to obtain a feature weight vector. And carrying out weighted summation on the first airflow uniformity index, the vortex attenuation parameter and the streamline form stability characteristic vector according to the characteristic weight vector to obtain a multidimensional vector representation of the airflow characteristics of the bottle mouth.
Further, multidimensional vector representation is carried out on airflow speed distribution, vortex intensity and streamline morphology, a pre-trained flow field dynamic analysis model is constructed, dynamic change rules of airflow characteristics in a pressure balance state are predicted, and stability indexes of bottleneck airflow characteristics are obtained through stability analysis. And constructing a speed distribution feature vector according to the first airflow data set, extracting airflow pulsation frequency spectrum features by adopting discrete Fourier transform, superposing vortex intensity data on the basis of the frequency spectrum features, and extracting curvature change features from streamline form parameters to obtain a first multidimensional feature vector. And (3) pre-training the depth self-encoder by adopting the marking data, wherein the number of nodes of the input layer is the same as the dimension of the first multidimensional feature vector, the number of nodes of the middle layer is halved, and the feature dimension reduction is carried out on the self-encoder after the pre-training to obtain a second multidimensional feature vector. And constructing a residual neural network predictor, wherein an input layer adopts a second multidimensional feature vector, sequential features are extracted by stacking residual blocks, the predictor is trained by adopting historical data, a flow field dynamic analysis model is obtained, and an airflow dynamic prediction result is output. Calculating a bottleneck pressure change sequence according to the airflow dynamic prediction result, smoothing the pressure change sequence by adopting a sliding average method, and extracting a third multidimensional feature vector in a pressure balance state. And calculating the fluctuation amplitude and the fluctuation frequency aiming at the third multidimensional feature vector, and calculating the maximum Lyapunov exponent of the feature sequence by adopting a phase space reconstruction method. Dividing a stability interval according to the maximum Lyapunov exponent, determining a stability threshold through a feature sequence variance, and extracting an airflow feature stability index from a third multi-dimensional feature vector.
Further, if the stability of the air flow characteristic is lower than a preset threshold, the short-term change trend of the air flow speed distribution, the vortex strength and the streamline form is predicted through a long-short-term memory network algorithm, and the future evolution state of the air flow characteristic is obtained. Four layers of decomposition are carried out on the first preprocessing data set by wavelet transformation, airflow speed distribution characteristics are extracted from wavelet coefficients, vortex intensity time sequence characteristics are calculated through an autocorrelation function, and a first characteristic vector is constructed by combining streamline curvature change data. And constructing a long-short-time memory network predictor, wherein the number of nodes of an input layer is the same as the dimension of the first feature vector, the hidden layer comprises three layers of memory units, the number of each layer of memory units is 64, and the number of nodes of an output layer is the same as the dimension of a prediction target. And carrying out sequence division on the first feature vector by adopting a sliding window method, wherein the window length is set to be 50 sampling points, and the sliding step length is set to be 10 sampling points, so as to generate a training data sequence. And training the long-short time memory network predictor according to the training data sequence, optimizing network parameters by adopting a back propagation algorithm, and calculating a prediction error on the verification data set to obtain a first prediction result. And extracting airflow speed distribution trend, vortex intensity change rule and streamline morphology evolution characteristics of 100 future sampling points according to the first prediction result, and fusing the characteristics by a weighted average method to obtain a second characteristic vector. And constructing an airflow characteristic evolution sequence according to the second characteristic vector, processing the evolution sequence by adopting an exponential smoothing method, and acquiring the future evolution state of the airflow characteristic from the smoothed sequence.
Further, adjusting the nitrogen injection rate and the injection angle of the nitrogen filling equipment according to the future evolution state of the airflow characteristics to generate optimized nitrogen filling control parameters, wherein the method comprises the steps of normalizing the evolution state data of the airflow characteristics, removing abnormal values by adopting a median filter, and segmenting the processed data by a sliding window method to obtain a first evolution data set. Extracting an airflow speed distribution sequence according to the first evolution data set, calculating speed distribution frequency spectrum characteristics by adopting Fourier transformation, acquiring airflow fluctuation period data from the frequency spectrum characteristics, and constructing a nitrogen injection rate range aiming at the fluctuation period. Reconstructing the vortex intensity sequence by adopting a phase space reconstruction method, setting the embedding dimension as 3, setting the time delay as 5, and calculating the radius of curvature of the reconstruction track to obtain a first vortex evolution rule. And constructing an injection angle range according to the first vortex evolution rule, generating an angle adjustment sequence by adopting a uniform division method, and constructing a multi-objective optimization function by combining the injection rate range. Setting parameters of a particle swarm optimizer, wherein the number of particles is 50, the maximum iteration number is 100, the inertia weight is 0.7, the learning factor is 2, and solving the multi-objective optimization function to obtain a first control parameter combination. And (3) carrying out air flow stability prediction on the first control parameter combination by adopting a recurrent neural network, and generating a parameter adjustment sequence through a control parameter updating rule. And calculating an air flow stability index according to the parameter adjustment sequence, dynamically correcting the stability index by adopting a Kalman filter, and extracting the optimized nitrogen charging control parameter from corrected data.
Further, adjusting nitrogen injection parameters according to the optimized nitrogen filling control parameters, updating the running state of the nitrogen filling equipment in real time, and generating an updated air flow characteristic data set, including: generating an injection parameter adjustment instruction according to the optimized nitrogen filling control parameter, adjusting the nitrogen injection rate by adopting a proportional integral regulator, setting an integral time constant to be one fourth of a response period, acquiring bottleneck airflow data in real time through a sensor array, and performing Gaussian filtering processing on the acquired data to obtain a first airflow data set. And calculating gridding variance distribution aiming at the first airflow data set, acquiring a first speed field uniformity index by adopting a regional block statistical method, and extracting the difference value of adjacent sampling points from the vortex intensity sequence to obtain a first attenuation rate curve. And calculating a curve slope sequence according to the first attenuation rate curve, fitting a curve slope change trend by adopting a least square method, and constructing a first response characteristic vector by combining the streamline curvature parameter. And carrying out feature extraction on the first response characteristic vector by adopting a support vector regressor, wherein the input feature dimension is the same as the vector dimension, the kernel function is a radial basis function, and the second response characteristic vector is obtained from a regression result. And setting a dynamic response judgment rule aiming at the second response characteristic vector, judging that the speed distribution is uniform if the speed field uniformity index is larger than a preset threshold value, and judging that the eddy current attenuation is stable if the eddy current intensity attenuation rate is in a preset interval. And updating the flow form parameters in real time by adopting a sliding window method, setting the window length to be twice of the sampling period, and calculating the fluctuation amplitude of the parameters in the window to obtain a first stability index. And constructing a Kalman filter according to the first stability index, obtaining a measurement noise covariance matrix through statistics of historical data, and obtaining an updated airflow characteristic data set through filter output.
Further, the updated airflow characteristic data set is input into a flow field dynamic analysis model, the characteristic vectors of airflow speed distribution, vortex intensity and streamline form are recalculated, and the updated airflow characteristic stability index is output, so that the completion state of the nitrogen charging process is determined. Four-layer decomposition is carried out on the first processing data set by wavelet transformation, a low-frequency component and a high-frequency component of airflow velocity distribution data are extracted, and a 16-dimensional velocity distribution characteristic vector is constructed from decomposition coefficients. And calculating the vortex intensity change rate according to the 16-dimensional speed distribution feature vector, smoothing the change rate sequence by adopting an exponential weighting method, and setting an upper limit value and a lower limit value of the self-adaptive threshold interval to obtain the 32-dimensional vortex feature vector. And constructing a long-short memory network, wherein the number of nodes of an input layer is 32, the number of units of each layer is 64, the number of nodes of an output layer is 16, and dynamically predicting the eddy current characteristic vector. And extracting streamline form parameters according to the prediction result, adopting a principal component analysis method to reduce the dimension of the parameters, and reserving principal components with contribution rate exceeding a preset threshold value to obtain 8-dimensional streamline feature vectors. And fusing the speed distribution feature vector, the vortex feature vector and the streamline feature vector by adopting a weighted summation method, and obtaining a weight coefficient through statistics of historical data to obtain an airflow feature stability index. And constructing a five-layer recurrent neural network to track the change of the stability index, and if the stability index is higher than a preset completion threshold value in ten continuous sampling periods, judging that the nitrogen charging process is completed.
Further, if the air flow characteristic stability index reaches a threshold value, classifying and updating the index through a logistic regression algorithm, outputting a nitrogen charging completion probability, generating a termination signal according to the probability, controlling nitrogen charging equipment to stop, recording an air flow characteristic data set, and determining the production efficiency improvement amplitude, wherein the method comprises the steps of comparing the air flow characteristic stability index with a preset threshold value, if the stability index reaches the preset threshold value, segmenting the stability index by adopting a sliding window with the length of 100 sampling points, and setting the window overlapping rate to be 50% to obtain a first index sequence. And extracting features of the first index sequence, calculating four statistical features of mean, variance, skewness and kurtosis, extracting three time sequence features of a maximum value, a minimum value and a peak Gu Chazhi of the sequence, and constructing a feature vector. And classifying the seven-dimensional feature vectors by adopting a logistic regression algorithm, mapping the classification result into a probability value by using a sigmoid function, and generating a termination signal if five continuous sampling periods of the probability value exceed a preset probability threshold. And constructing a device control instruction according to the termination signal, executing by adopting a proportional-integral controller, setting an integral time constant to be one fourth of a sampling period, and obtaining a proportional coefficient by statistics of historical data. And recording the air flow characteristic data through a data acquisition unit, setting the sampling frequency to be 200 Hz, and covering the complete nitrogen charging process for the recording duration to generate a process parameter data set. And extracting an airflow speed uniformity index, a vortex intensity stability index and a streamline form continuity index from the process parameter data set, and constructing an eight-dimensional feature vector. And classifying the eight-dimensional feature vectors by adopting a support vector machine, selecting a radial basis function as a kernel function, optimizing kernel parameters by a grid search method, and extracting an efficiency improvement index from a classification result.
The technical scheme provided by the embodiment of the invention can have the following beneficial effects:
The invention discloses a quality evaluation method for a bottle nitrogen filling process by applying artificial intelligence, which comprises the steps of obtaining air flow speed distribution, vortex strength and streamline form data at a bottle mouth, and extracting features and carrying out multidimensional vector representation, and constructing a flow field dynamic analysis model to predict the change rule of the airflow features. When the characteristic stability of the air flow is lower than a threshold value, predicting short-term variation trend by using a long-short-term memory network algorithm, and adjusting the nitrogen injection parameter according to the short-term variation trend. According to the invention, the airflow characteristic data set is updated in real time, the characteristic vector is recalculated, and the stability index is output until the preset threshold value is reached. And finally judging the nitrogen charging completion probability through a logistic regression algorithm, and generating a termination signal. The method can realize the dynamic optimization control of the nitrogen charging process, improve the production efficiency and ensure the stability of the nitrogen charging quality.
Drawings
FIG. 1 is a flow chart of a method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence.
FIG. 2 is a schematic diagram of a method for evaluating the quality of a bottle nitrogen filling process using artificial intelligence according to the present invention.
Detailed Description
For a further understanding of the present application, the present application will be described in detail with reference to the drawings and examples. The application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not limiting of the application. It should be noted that, for convenience of description, only the portions related to the application are shown in the drawings.
In the embodiment of the invention, the nitrogen filling equipment is used for filling nitrogen before the bottle body is sealed, and replacing oxygen so as to prolong the shelf life of the product. The airflow characteristics at the bottle mouth include airflow speed distribution, vortex intensity and streamline form, and the nitrogen charging effect is directly affected. While the traditional method is difficult to quantify the characteristics in real time, the invention realizes dynamic analysis through an artificial intelligence algorithm and improves the evaluation precision. The following embodiments do not limit implementation details of the technical solution too much, and can be determined by a skilled person according to actual scenes.
Fig. 1 shows a flow chart of a quality evaluation method for a bottle nitrogen filling process using artificial intelligence according to an embodiment of the present invention, which is described in detail below:
s101, acquiring airflow speed distribution, vortex intensity and streamline form data at the bottleneck, and carrying out feature extraction and time sequence analysis to generate an airflow feature data set containing time sequences.
In the embodiment of the invention, after the nitrogen charging equipment is started, the dynamic characteristics of the air flow at the bottleneck are required to be monitored in real time so as to evaluate the quality of the nitrogen charging process. The generation of the airflow characteristic data set is the basis of subsequent analysis, and relates to multi-dimensional sensor data acquisition and intelligent processing. The specific implementation of the step comprises the following sub-steps:
as shown in fig. 2, s1011 deploys a sensor array to collect airflow speed and pressure data at the bottleneck, and performs noise reduction and standardization processing to generate a first airflow data set.
In the embodiment of the invention, an ultrasonic gas flow sensor and a gas pressure sensor are arranged at 4X 4 grid points with the diameter of the bottleneck being divided, the sampling frequency is set to be 200 Hz, and the data of the gas flow speed and the gas inlet pressure are acquired. The original data is affected by environmental noise and equipment vibration, a Butterworth low-pass filter is adopted for noise reduction, the cut-off frequency is 80 Hz, and high-frequency interference is filtered. And after noise reduction, the data is subjected to zero mean value standardization processing to generate a first airflow data set, so that the consistency and reliability of the data are ensured. The sensor array covers all grid points of the bottleneck, so that the comprehensiveness of data acquisition is ensured.
S1012, an airflow velocity field distribution diagram is constructed based on the first airflow data set, a first vortex distribution diagram is generated by combining streamline tracks and vorticity data, and flow field characteristics are extracted.
In the embodiment of the invention, based on the first airflow data set, the air flow speed data is spatially interpolated by adopting a kriging interpolation method, the interpolation grid density is 0.5 millimeter, an air flow speed field distribution map is generated, and the speed distribution uniformity at the bottle mouth is reflected. And scanning the bottle mouth in a plane 5mm away from the bottle mouth by using a gas flow probe, wherein the distance is 1mm, and recording the gas flow track to generate streamline data. And calculating the vorticity value of the airflow field, wherein the range is-50 to 50, 10 contour lines are arranged, the interval is 10, and the contour lines are overlapped on the velocity field distribution diagram to form a first vortex distribution diagram. Aiming at the vortex intensity data, an improved K-means clustering algorithm is adopted for layering, the clustering center number is 5, and a classification result of the vortex intensity interval is obtained. Curvature features are extracted from the streamline trajectories, with a radius of curvature ranging from 2 to 20 millimeters, reflecting the degree of deflection of the airflow.
S1013, collecting gas concentration data, constructing a Bayesian probability network, extracting concentration distribution characteristic vectors, and generating an airflow characteristic time sequence data set by combining diffusion coefficients.
In the embodiment of the invention, the gas concentration data in the nitrogen filling process is acquired through the sensor array, the concentration of the center of the bottle mouth is up to 95%, and the concentration of the edge area is reduced to 75%. And (3) constructing a Bayesian probability network by combining the first airflow data set, extracting a gas concentration distribution feature vector, wherein the dimension is 8, and the statistical features comprise mean, variance, skewness and the like. And calculating a gas diffusion coefficient based on the concentration distribution characteristic vector, wherein the diffusion coefficient reaches 2.5 when the vortex strength is 30. Performing db4 wavelet transformation on the airflow pulsation frequency signal, decomposing 4 layers, extracting detail coefficient sequences of 0-25 Hz, 25-50 Hz, 50-75 Hz and 75-100 Hz, and constructing an airflow pulsation characteristic data set. And superposing a streamline track curve on the first vortex distribution diagram, and carrying out quantitative description on a flow field structure by combining a diffusion coefficient to generate a time sequence data set containing characteristics such as a speed field, vortex intensity, concentration distribution and the like.
In the embodiment of the invention, the generated airflow characteristic data set comprehensively reflects the airflow dynamic characteristics of the bottle mouth through the data acquisition and the intelligent processing of the multidimensional sensor. Compared with the traditional method, the method can capture the uniformity of the airflow speed distribution, the spatial layering of the vortex intensity and the curvature change of the streamline form in real time, and provides high-quality data support for subsequent analysis. The finally generated time sequence data set lays a foundation for dynamic optimization and completion state judgment of the nitrogen charging process, and the evaluation precision and the production efficiency are obviously improved.
S102, extracting uniformity characteristics of airflow speed distribution, attenuation characteristics of vortex intensity and stability characteristics of streamline morphology in airflow characteristic data, and performing characteristic fusion processing to generate multidimensional vector representation of airflow characteristics of a bottle mouth.
In the embodiment of the invention, the airflow characteristic data set comprises time series data of airflow speed, vortex intensity and streamline form at the bottleneck, and key characteristics are required to be extracted and fused to generate a multidimensional vector representation so as to reflect the airflow dynamic characteristics. The method comprehensively quantifies the airflow characteristics through data cleaning, characteristic extraction and weighted fusion, and provides high-quality input for subsequent analysis. The specific implementation of the step comprises the following sub-steps:
S1021, cleaning and segmenting the airflow characteristic data set to generate a preprocessed airflow characteristic data set.
In the embodiment of the invention, the airflow characteristic data set is influenced by equipment vibration and environmental noise, and data cleaning is needed to ensure analysis accuracy. And a 5-point median filter is adopted to remove abnormal values, the length of a filtering window is 5 sampling points, and short-term fluctuation is effectively smoothed. Based on the time series characteristics, the data set is segmented by a sliding window method, the window length is set to be 100 milliseconds, the overlapping rate is set to be 50 percent, and the preprocessed air flow characteristic data set is generated. The segmentation processing reserves the time continuity of the data, reduces the interference of noise on feature extraction, and provides a reliable data basis for subsequent analysis.
S1022, calculating uniformity characteristics of the airflow velocity field based on the preprocessed airflow characteristic data set, and generating a first airflow uniformity index.
In the embodiment of the invention, the uniformity of the airflow velocity distribution is a key index for evaluating the nitrogen charging stability. Dividing the bottleneck area into 8 multiplied by 8 grids, calculating the variance of the airflow velocity in each grid, generating a variance distribution map, and reflecting the local velocity fluctuation characteristic. And the variance value ranges from 0.2 to 2.5, and the maximum and minimum value normalization processing is adopted to map the variance value to a range from 0 to 1, so as to obtain a first airflow uniformity index. The closer the uniformity index value is to 0, the more uniform the gas flow distribution, and the higher the stability of the nitrogen charging process. The step quantifies the spatial distribution characteristic of the speed field through gridding calculation and normalization processing, and provides important input for feature fusion.
S1023, extracting an eddy current intensity time sequence, fitting attenuation characteristics, and generating eddy current attenuation parameters.
In the embodiment of the invention, the attenuation characteristic of the vortex strength reflects the dynamic evolution of the nitrogen charging process. A time series of vortex strengths is extracted from the pre-processed airflow characterization dataset, gradually decreasing from an initial value of 85 to 25 over a 200 millisecond time window. And calculating the eddy current intensity difference between adjacent sampling points to generate a first eddy current intensity change curve. The curve is fitted using an exponential decay function of the form I (t) =i 0*e(-kt), where I 0 is the initial intensity and k is the decay coefficient. And (3) optimizing fitting by a least square method to obtain a damping coefficient k=0.015, wherein the goodness of fit is 0.92, which shows that the eddy current strength shows a remarkable exponential damping trend. The eddy current attenuation parameter k is used as characteristic input, and the eddy current dynamic change rule is quantized.
S1024, extracting streamline track curvature characteristics, and generating streamline form stability characteristic vectors by combining a recurrent neural network.
In the embodiment of the invention, the stability of the streamline form reflects the complexity of the airflow movement and directly influences the nitrogen charging effect. And extracting a streamline track curve from the preprocessed airflow characteristic data set, calculating a curvature change rate, and generating a first curvature sequence. The curvature change rate of the central area of the bottle mouth is small, the maximum value is not more than 0.05, the curvature change of the edge area is severe, and the maximum value reaches 0.15. And performing dimension reduction treatment on the curvature sequence by adopting a principal component analysis method, wherein the cumulative contribution rate of the first three principal components reaches 85%, and generating a three-dimensional principal feature vector. An eight-layer recurrent neural network is constructed, the number of nodes of an input layer is 3, the number of nodes of each hidden layer is 16, and a long-short-time memory unit structure is adopted to capture the time dependency relationship. Training the network by using 250 groups of labeling data, setting the learning rate to be 0.001, training for 1000 times, and extracting 8-dimensional streamline form stability feature vectors from an output layer after network convergence. The dynamic stability of the streamline form is accurately described through neural network deep learning.
S1025, constructing a feature correlation matrix, calculating a feature weight vector, and carrying out weighted fusion to generate a multidimensional vector representation of the bottleneck airflow feature.
In the embodiment of the invention, for integrating the airflow characteristics, the uniformity index, the vortex attenuation parameter and the streamline stability characteristics are required to be fused. And constructing a 12 multiplied by 12 characteristic correlation matrix, wherein matrix elements are the pearson correlation coefficients among the first airflow uniformity index, the vortex attenuation parameter and the streamline form stability characteristic vector. Feature pairs with absolute values of correlation coefficients greater than 0.7 have significant correlation, indicating strong interactions between features. And calculating a characteristic weight vector based on the correlation coefficient, wherein the uniformity index weight is 0.35, the vortex decay parameter weight is 0.25, and the streamline stability characteristic weight is 0.4. And (3) adopting a weighted summation method to fuse the three types of feature vectors according to weights to generate a 12-dimensional airflow feature vector. The vector comprehensively describes the uniformity of the airflow speed distribution, the attenuation characteristic of the vortex intensity and the stability of the streamline form, and provides high-quality characteristic representation for subsequent nitrogen charging state evaluation.
In the embodiment of the invention, the generated bottleneck airflow characteristic multidimensional vector can accurately reflect the dynamic characteristic of the nitrogen charging process through multidimensional characteristic extraction and fusion. Compared with the traditional method, the method has the advantages that the accuracy and the robustness of feature extraction are remarkably improved through median filtering, gridding calculation and neural network analysis. The application of the feature correlation matrix further optimizes the feature weight distribution, ensures the representativeness of fusion features, lays a solid foundation for the subsequent dynamic analysis and equipment optimization, and finally improves the reliability and the production efficiency of nitrogen filling quality evaluation.
S103, constructing a pre-trained flow field dynamic analysis model based on multidimensional vector representation of bottleneck airflow characteristics, predicting dynamic change rules of the airflow characteristics in a pressure balance state, and generating an airflow characteristic stability index through stability analysis.
In the embodiment of the invention, based on multidimensional vector representation of bottleneck airflow characteristics, a flow field dynamic analysis model is further required to be constructed so as to predict dynamic evolution trend of airflow speed distribution, vortex strength and streamline morphology in a pressure balance state, and stability analysis is adopted to quantify the stability of the airflow characteristics. The method realizes accurate evaluation of the dynamic characteristics of the nitrogen charging process through feature pretreatment, deep learning model construction and stability index calculation, and provides scientific basis for follow-up optimization control. The specific implementation of the step comprises the following sub-steps:
s1031, preprocessing the air flow characteristic data, extracting pulsation spectrum characteristics, fusing vortex and streamline characteristics and generating a first multidimensional characteristic vector.
In the embodiment of the invention, the air flow speed distribution data contains noise and abnormal fluctuation, and preprocessing is needed to ensure the input quality of the model. The data are mapped to a 0-1 interval by adopting a normalization method, short-term fluctuation is removed by combining a 5-point median filter with a window length of 200 milliseconds, a sliding window interval is set to be 50 milliseconds, and a first airflow data set containing 2000 groups of sampling points is generated in a segmented mode. For the data set, 1024-point fast Fourier transform is adopted to extract airflow pulsation spectrum characteristics, three main frequency peaks of 15 Hz, 35 Hz and 65 Hz are identified in the range of 0 to 100 Hz, and a 16-dimensional spectrum component vector is generated. In order to integrate flow field characteristics, vortex intensity data and curvature change characteristics of streamline morphology are fused, a vortex intensity sequence reflects a dynamic attenuation trend, the curvature change rate of the streamline is smaller than 0.05 in a central area, and the curvature change rate of the streamline is as high as 0.15 in an edge area. And finally generating a 32-dimensional first multidimensional feature vector which comprises frequency spectrum components, vortex strength and curvature features and comprehensively describing the dynamic characteristics of the air flow.
S1032 performs a dimension reduction process on the first multi-dimensional feature vector using the depth self-encoder to generate a second multi-dimensional feature vector.
In the embodiment of the invention, in order to reduce the feature dimension and keep the core information, a depth self-encoder is adopted for feature compression. The depth self-encoder is designed into a symmetrical structure, 32 nodes of the input layer correspond to the dimension of the first multidimensional feature vector, 16 nodes of the intermediate encoding layer realize feature compression, and 32 nodes of the decoding layer are used for feature reconstruction. The 1500 groups of marking data are used for pre-training, the training round is set to be 500, the optimization target is to minimize the mean square error, and the error after training is reduced to 0.015, so that the feature extraction effect is good. After the pre-training is completed, the first multidimensional feature vector is input into the self-encoder, and the 16-dimensional second multidimensional feature vector is extracted through the intermediate encoding layer. The vector keeps main information of airflow characteristics, reduces calculation complexity, and provides high-efficiency input for a subsequent prediction model.
S1033, constructing a residual neural network predictor, predicting airflow dynamic change trend, and generating an airflow dynamic prediction result.
In the embodiment of the invention, a residual neural network predictor is constructed for capturing the time dependence of the airflow characteristics. The network comprises 8 residual blocks, each residual block consists of two layers of convolution layers and one shortcut connection, the input layer receives 16-dimensional second multidimensional feature vectors, and the output layer predicts airflow change trend within 100 milliseconds. 1000 sets of historical data covering different working conditions are adopted for training, the learning rate is set to be 0.001, the training is iterated for 2000 times, and the prediction mean square error is controlled within 0.025. After training, the network can accurately predict the short-term evolution trend of the airflow velocity distribution, the vortex intensity and the streamline form, and generate an airflow dynamic prediction result. The result reflects the dynamic change rule of the airflow characteristics in the pressure balance state, and provides data support for stability analysis.
S1034 calculates a pressure change sequence and a stability index based on the airflow dynamic prediction result, and generates an airflow characteristic stability index.
In the embodiment of the invention, the pressure balance state is a key for evaluating the nitrogen charging stability. Based on the air flow dynamic prediction result, calculating a bottleneck pressure change sequence, and smoothing by adopting a 15-point moving average method, wherein the window length is 150 milliseconds. When the pressure fluctuation amplitude is smaller than 0.05 bar and the duration exceeds 300 milliseconds, judging that the pressure balance state is reached, extracting 24-dimensional characteristic vectors under the state, and describing the characteristics of the airflow speed, the vortex and the streamline. The dynamics characteristic of the feature vector is analyzed by adopting a phase space reconstruction method, the embedding dimension is set to be 3, the time delay is set to be 5 sampling points, the maximum Lyapunov exponent is calculated, and the value is set to be 0.15, so that the system has certain instability. To quantify stability, the variance of the feature sequence over different time scales was calculated, setting the stability threshold to 0.08. When the characteristic sequence variance is less than the threshold and the maximum lyapunov exponent is less than 0.1, the airflow is determined to be in a steady state. And finally generating an airflow characteristic stability index, wherein the value range is 0 to 1, the airflow characteristic stability index is distributed in 0.75 to 0.95 in practical application, and the closer the value is 1, the more stable the airflow characteristic is.
In the embodiment of the invention, the accurate prediction of the dynamic change of the airflow characteristics is realized by combining the depth self-encoder and the residual neural network. The phase space reconstruction and Lyapunov exponent analysis further quantify the stability of the airflow system, and compared with the traditional method, the method has the advantage that the evaluation accuracy is remarkably improved. The generated stability index provides a reliable basis for judging the completion state of the nitrogen filling process, and lays a foundation for optimizing nitrogen injection parameters, so that the boosting production efficiency is improved, and the energy consumption is reduced.
S104, if the airflow characteristic stability index is lower than a preset threshold value, predicting short-term change trend of airflow speed distribution, vortex strength and streamline form by adopting a long-short-term memory network algorithm, and generating future evolution state of the airflow characteristic.
In the embodiment of the invention, when the airflow characteristic stability index is lower than the preset threshold value, the airflow state in the nitrogen charging process is indicated to have fluctuation, and the short-term evolution trend of the airflow state needs to be predicted to optimize the equipment control. The method comprises the steps of generating future evolution states of airflow characteristics through data preprocessing, characteristic extraction, long-short-time memory network prediction and characteristic fusion, and providing basis for dynamically adjusting nitrogen injection parameters. The specific implementation of the step comprises the following sub-steps:
S1041, preprocessing the airflow characteristic data, extracting characteristics by wavelet transformation, and generating a first characteristic vector.
In the embodiment of the invention, the air flow characteristic data is influenced by measurement noise and environmental interference, and preprocessing is needed to improve the data quality. Mapping the data to a range from-1 to 1 by adopting a standardized method, and removing abnormal fluctuation by using a 15-point median filter to generate a first preprocessing data set. For the data set, four layers of wavelet transformation are carried out by adopting db4 wavelet basis functions, and detail coefficients of four frequency scales of 0-25 Hz, 25-50 Hz, 50-75 Hz and 75-100 Hz and an approximation coefficient are obtained through decomposition. The detail coefficient captures the dynamic variation characteristics of the airflow velocity distribution in different frequency bands. And calculating a correlation coefficient of the vortex intensity time sequence through an autocorrelation function, setting the correlation length as 50 sampling points, and quantifying the time sequence correlation of the vortex intensity. And combining the curvature change rate of the streamline track, wherein the curvature change of the central area is less than 0.05, the curvature change of the edge area is as high as 0.12, and constructing a 48-dimensional first characteristic vector which comprises airflow speed distribution, vortex strength and streamline morphology characteristics and provides comprehensive input for subsequent prediction.
S1042 generates a training data sequence based on the first feature vector, trains the long-short-term memory network predictor, and generates a first prediction result.
In the embodiment of the invention, a long-and-short-term memory network predictor is constructed for capturing time sequence dependency of air flow characteristics. The network comprises an input layer, three hidden layers and an output layer, the number of nodes of the input layer is 48 and is consistent with the dimension of a first feature vector, each hidden layer comprises 64 memory units, a forgetting gate, an input gate and an output gate are adopted to control information flow, the threshold value of the forgetting gate is set to be 0.5, an activation function is sigmoid, a tanh function is used for state updating, and 32 nodes of the output layer correspond to the dimension of a prediction target. And carrying out sequence division on the first feature vector by adopting a sliding window method, wherein the window length is 50 sampling points, the step length is 10 sampling points, and adjacent windows overlap 40 sampling points to generate 2000 groups of training samples. The random gradient descent method is used for training the network, the batch size is 32, the initial learning rate is 0.01, the attenuation is 0.9 times of each 50 rounds, and the mean square error of the verification set is reduced to 0.015 after 1000 rounds of training, so that the network convergence is shown. After training is completed, the network predicts the airflow characteristic change trend of 100 sampling points in the future, generates a first prediction result, and controls the prediction standard deviation within 0.08 and the relative error of the eddy current strength is less than 5%.
S1043, extracting trend features of the first prediction result, fusing and generating a second feature vector, and constructing and smoothing an airflow feature evolution sequence.
In the embodiment of the invention, the first prediction result comprises airflow velocity distribution trend, vortex intensity change rule and streamline morphology evolution characteristics. And extracting the standard deviation of speed distribution, the periodical change amplitude of vortex intensity and the change rate of streamline curvature of 100 sampling points in the future, wherein the accuracy rate of speed distribution prediction reaches 90%, and the consistency of streamline morphology prediction is 92%. And fusing the characteristics by a weighted average method, and generating a 32-dimensional second characteristic vector by distributing weights of 0.4 of speed distribution, 0.3 of vortex strength and 0.3 of streamline form. And constructing an airflow characteristic evolution sequence based on the second characteristic vector, adopting an exponential smoothing method to process, setting a smoothing coefficient to be 0.85, and enhancing the continuity of the sequence trend. The smoothed sequence shows that the airflow speed fluctuation is gradually increased within 300 milliseconds, the vortex intensity shows periodic oscillation, the streamline form is distorted in the edge area, the predicted result and the actual observation fit degree reach 88%, and a reliable basis is provided for evaluating the future evolution state of the airflow.
In the embodiment of the invention, the stability threshold is set to 0.75, and based on experimental data statistics, when the index is lower than the value, a prediction process is triggered to cope with airflow fluctuation. The wavelet transformation and the application of the long-short-time memory network obviously improve the capturing capacity of the dynamic characteristics of the airflow, and compared with the traditional method, the predicted result reflects the short-term change trend of the airflow speed, the vortex strength and the streamline morphology more accurately. The generated evolution state provides data support for optimizing the operation parameters of the nitrogen charging equipment, and is beneficial to reducing energy consumption and improving production efficiency.
S105, adjusting the nitrogen injection rate and the injection angle of the nitrogen charging equipment according to the future evolution state dynamic characteristics of the airflow characteristics, and generating optimized nitrogen charging control parameters.
In the embodiment of the invention, the future evolution state of the airflow characteristics reflects the dynamic trend of the airflow speed, the vortex strength and the streamline form in the nitrogen charging process, and the nitrogen injection parameters are required to be optimized according to the dynamic trend to ensure the process stability. The specific implementation of the step comprises the following sub-steps:
s1051, preprocessing the airflow characteristic evolution state data, extracting an airflow fluctuation period and an eddy evolution rule, and generating a first evolution data set and related characteristics.
In the embodiment of the invention, the airflow characteristic evolution state data comprises airflow speed, vortex intensity and streamline morphology characteristics, is influenced by measurement noise, and needs to be preprocessed to improve data quality. Mapping data to a 0-1 interval by adopting a normalization method, smoothing abnormal fluctuation by using an 11-point median filter, segmenting by a sliding window method, wherein the window length is 100 sampling points, and the overlapping rate is 50%, so as to generate a first evolution data set. Extracting an airflow speed distribution sequence from the data set, calculating spectrum characteristics by adopting 1024-point fast Fourier transform, identifying main frequency components at 15 Hz, 35 Hz and 55 Hz, respectively having amplitudes of 0.8, 0.6 and 0.4, determining an airflow fluctuation period, and setting a nitrogen injection rate range to be 2-8 meters per second. In order to analyze the dynamic characteristics of the vortex, a phase space reconstruction method is adopted to process a vortex intensity sequence, an embedding dimension is set to be 3, time delay is set to be 5 sampling points, a reconstruction track is in a spiral shape, the curvature radius range is 5-20 mm, the periodic change of the vortex intensity is reflected, and a first vortex evolution rule is generated. The method accurately captures the dynamic characteristics of the airflow through spectrum analysis and phase space reconstruction, and provides a data basis for parameter optimization.
S1052, constructing a multi-objective optimization function, optimizing the injection rate and angle by adopting a particle swarm algorithm, and generating a first control parameter combination.
In the embodiment of the invention, based on the first vortex evolution rule and fluctuation period data, the nitrogen injection angle range is 15-45 degrees, the step length is 5 degrees, and the injection rate range is 2-8 meters per second. A multi-objective optimization function is designed, wherein the multi-objective optimization function comprises three sub-objectives of airflow uniformity, vortex stability and streamline continuity, and weights are respectively 0.4, 0.3 and 0.3. Solving by adopting a particle swarm optimization algorithm, setting the number of particles as 50, setting the maximum iteration number as 100, setting the inertia weight as 0.7 and setting the learning factor as 2. The algorithm obtains an optimal injection rate of 5.5 meters per second and an injection angle of 32 degrees by iteratively updating particle positions, optimizing an objective function value and converging to 0.92, and forms a first control parameter combination. The step balances the mutual restriction relation of the airflow characteristics through multi-objective optimization, and ensures the overall performance of control parameters.
S1053, predicting the air flow stability of the control parameters by using the recurrent neural network, and generating optimized nitrogen charging control parameters by combining the Kalman filtering correction stability index.
In the embodiment of the invention, in order to verify the effect of the first control parameter combination, a three-layer recurrent neural network is constructed to predict the air flow stability. The network input layer 16 nodes receive control parameters and airflow characteristics, the hidden layer 32 nodes capture time sequence relations, and the output layer 8 nodes predict stability trends. Training by using 1000 groups of labeling data, the learning rate is set to be 0.001, and the prediction accuracy rate reaches 90%. When the prediction stability is lower than 0.8, triggering self-adaptive adjustment, wherein the injection rate adjustment step length is 0.5 m/s, the angle adjustment step length is 3 degrees, and generating a parameter adjustment sequence. Based on the sequence, an airflow stability index is calculated, dynamic correction is carried out by adopting a Kalman filter, the measurement noise variance is set to be 0.01, and the process noise variance is set to be 0.005. After correction, the stability index shows that the fluctuation amplitude of the air flow speed is reduced by 45%, the variation amplitude of the vortex intensity is reduced by 38%, the streamline form is kept continuous, and the index value is maintained above 0.85. The optimized nitrogen charging control parameters are extracted from the correction data, and the method is suitable for various working conditions.
In the embodiment of the invention, the optimized control parameters ensure the stability of the nitrogen charging process by precisely capturing the dynamic change of the air flow characteristics. Compared with the traditional method relying on experience adjustment, the method combines Fourier transformation, particle swarm optimization and neural network prediction, and improves scientificity and adaptability of parameter optimization. The optimization result effectively reduces system fluctuation, improves production efficiency, reduces nitrogen consumption, and provides a high-efficiency solution for nitrogen charging technology in the industries of food, medicine and the like.
S106, adjusting nitrogen injection parameters according to the optimized nitrogen filling control parameters, updating the running state of the nitrogen filling equipment in real time, generating an updated airflow characteristic data set, and analyzing the dynamic response characteristics of the nitrogen filling process, including airflow speed distribution uniformity, vortex intensity attenuation rate and streamline morphology stability.
In the embodiment of the invention, the optimized nitrogen charging control parameters are used for dynamically adjusting the nitrogen injection rate and angle so as to ensure the stability of the nitrogen charging process. And generating an updated airflow characteristic data set by acquiring the airflow data of the bottle mouth, carrying out characteristic analysis and dynamic response evaluation in real time, quantifying the airflow dynamic characteristics, and providing a basis for continuously optimizing the operation of equipment. The specific implementation of the step comprises the following sub-steps:
S1061, acquiring bottleneck airflow data and preprocessing to generate a first airflow data set, and extracting speed field uniformity indexes and attenuation rate characteristics.
In the embodiment of the invention, when the nitrogen charging equipment operates, the air flow speed, pressure and streamline data of the bottle mouth are collected in real time through the sensor array at the sampling frequency of 200 Hz. The original data is affected by random noise, the original data is processed by a Gaussian filter, the standard deviation is set to be 1.5, the window length is 15 sampling points, short-term fluctuation is effectively smoothed, and a first airflow data set is generated. Dividing the bottleneck area into 8 multiplied by 8 grids, wherein the grid size is 5 mm multiplied by 5 mm, the overlapping rate is 25%, and calculating the standard deviation of the air flow velocity in each grid to generate grid variance distribution. When the standard deviation of more than 90% of grids is smaller than 0.2, the speed field is judged to be uniformly distributed, and a first speed field uniformity index is generated. And extracting the difference value of adjacent sampling points from the eddy current intensity sequence, wherein the sampling interval is 5 milliseconds, and generating a first attenuation rate curve to reflect the dynamic change rate of the eddy current intensity. The method provides a reliable basis for dynamic response characteristic evaluation through high-precision data acquisition and gridding analysis.
S1062, constructing a response characteristic vector, extracting features by using a support vector regressor, generating a second response characteristic vector, and judging dynamic response characteristics.
In the embodiment of the invention, based on a first attenuation rate curve, a 5-order polynomial fitting slope change trend is adopted, the fitting goodness reaches 0.95, and a 16-dimensional first response characteristic vector is constructed by combining streamline curvature parameters, wherein the first response characteristic vector comprises speed field uniformity, vortex attenuation and streamline morphology characteristics, and each dimension is normalized to 0 to 1. And extracting features by adopting a support vector regressor, setting a radial basis function, setting a gamma parameter to be 0.1, setting a relaxation factor C to be 10, using 1000 groups of training samples, and generating an 8-dimensional second response characteristic vector, wherein the cross validation error is smaller than 0.05. The dynamic response judgment rule is set, wherein the speed field uniformity index is larger than 0.85, the speed distribution is judged to be uniform, the vortex intensity attenuation rate is 0.2-0.4, the vortex attenuation stability is judged, the streamline form stability index is larger than 0.8, and the streamline form stability is judged. The step accurately quantifies the dynamic response characteristic of the airflow through regression analysis and rule judgment.
And S1063, updating the airflow characteristic data set in real time, and generating the airflow characteristic data set by combining the Kalman filtering optimization data.
In the embodiment of the invention, a sliding window method is adopted to update the airflow speed distribution and the streamline morphology parameters in real time, the window length is 400 milliseconds, the sampling period is about twice, the fluctuation amplitude of the parameters in the window is calculated, and the streamline morphology is judged to be stable when the amplitude is smaller than 0.15, so that a first stability index is generated. And constructing Kalman filter optimization data, deriving a state transition matrix based on a physical model, and determining a measurement noise covariance matrix through statistics of 1000 groups of historical data, wherein the process noise covariance matrix is a diagonal matrix, and the diagonal element is 0.01. The filter tracks the change of the airflow characteristics in real time, and outputs an updated airflow characteristic data set to accurately reflect the dynamic response characteristic of the nitrogen charging process. The step ensures the high quality and real-time performance of the data set through dynamic updating and filtering processing.
In the embodiment of the invention, the proportional integral regulator is used for adjusting the nitrogen injection parameter in real time, the proportional coefficient is 0.8, the integral time constant is 50 milliseconds, and the response speed is high. When the injection rate is adjusted from 5 meters per second to 7 meters per second, the speed field uniformity index is increased from 0.75 to 0.88 in 150 milliseconds, the eddy current intensity decay rate is stabilized at 0.35 in 200 milliseconds, and the streamline form stability is maintained above 0.85, which indicates that the system is rapidly in a stable state. Compared with the traditional fixed parameter control, the method has the advantages that the response performance and stability of the nitrogen charging process are remarkably improved through real-time monitoring and dynamic adjustment, the method is suitable for various working conditions, the nitrogen waste is reduced, and the production efficiency is improved.
S107, inputting the updated airflow characteristic data set into a flow field dynamic analysis model, recalculating the characteristic vectors of airflow speed distribution, vortex strength and streamline morphology, and generating an updated airflow characteristic stability index to judge the completion state of the nitrogen charging process.
In the embodiment of the invention, the updated airflow characteristic data set reflects the airflow dynamic characteristics adjusted by the nitrogen charging equipment, and the characteristics are re-extracted and the stability is evaluated through a flow field dynamic analysis model to determine whether the nitrogen charging process is completed or not. The specific implementation of the step comprises the following sub-steps:
S1071, outlier detection and preprocessing are carried out on the airflow characteristic data set, wavelet transformation is adopted to extract speed distribution characteristics, and a first processing data set and characteristic vectors are generated.
In the embodiment of the invention, the airflow characteristic data set may contain measurement noise or abnormal fluctuation, and cleaning is needed to improve the data quality. And detecting abnormal values by adopting a three-time standard deviation principle, calculating a data mean value and a standard deviation, setting a boundary to be the mean value plus or minus three times the standard deviation, wherein the range of the standard deviation of the air flow speed is 0.5 to 1.2, and removing data points exceeding the boundary. The data is smoothed using a 15-point median filter to eliminate short term fluctuations and generate a first processed data set. For the data set, four layers of wavelet transformation are carried out by adopting db4 wavelet basis functions, and detail coefficients of 0-25 Hz, 25-50 Hz, 50-75 Hz and 75-100 Hz and an approximate coefficient are obtained through decomposition. The low frequency component reflects the overall trend of the airflow velocity, the high frequency component captures local disturbance, and the 16-dimensional velocity distribution characteristic vector is extracted from the decomposition coefficient and comprises the multi-scale velocity distribution characteristic. The method ensures the accuracy of feature extraction through outlier detection and wavelet transformation, and lays a foundation for subsequent analysis.
S1072, calculating the change rate of the eddy current strength, performing smoothing treatment, generating an eddy current characteristic vector, and performing dynamic prediction by combining a long-time memory network and a short-time memory network.
In the embodiment of the invention, the dynamic change of the vortex strength is the key for evaluating the nitrogen charging stability. Based on the 16-dimensional speed distribution feature vector, calculating the vortex strength change rate by adopting a 5-point difference method, and generating a change rate sequence. And smoothing the sequence by using an exponential weighted average method, wherein the weight attenuation coefficient is 0.85, the self-adaptive threshold interval is set according to historical data, the upper limit is the mean value plus 1.5 times of standard deviation, and the lower limit is the mean value minus 1.5 times of standard deviation, so that a 32-dimensional eddy current characteristic vector is generated. The method comprises the steps of constructing a long and short-time memory network to conduct dynamic prediction, wherein the network comprises an input layer, three layers of hidden layers and an output layer, 32 nodes of the input layer receive eddy characteristic vectors, 64 memory units of each hidden layer are reserved by adopting forgetting gate control information, a threshold value is set to be 0.5, a sigmoid activation function is used for input gates and output gates, and 16 nodes of the output layer predict future trends. 1500 groups of training samples are used, the learning rate is 0.001, the training is 1000 rounds, and the prediction accuracy rate reaches 92%. The step accurately captures the evolution rule of the vortex strength through smoothing and time sequence prediction.
S1073, extracting streamline morphology features, reducing dimension, fusing multidimensional feature vectors and generating an airflow feature stability index.
In the embodiment of the invention, the stability of the streamline form directly influences the nitrogen charging effect. And extracting streamline form parameters from the long-short-term memory network prediction result, and calculating curvature change rate, wherein the curvature change of a central area is less than 0.05, and the curvature change of an edge area is as high as 0.15. And reducing the dimension by adopting a principal component analysis method, reserving the first 8 principal components with the contribution rate exceeding 85%, and generating an 8-dimension streamline eigenvector. And fusing the speed distribution feature vector, the vortex feature vector and the streamline feature vector, adopting a weighted summation method, and determining weights of 0.4, 0.35 and 0.25 respectively through 1000 groups of historical data statistics. The generated air flow characteristic stability index value is between 0 and 1, and reflects the comprehensive stability of the air flow. In the step, the dynamic characteristics of the air flow are comprehensively quantized through dimension reduction and feature fusion.
S1074, constructing a recurrent neural network to track the stability index change, and judging the nitrogen charging process completion state.
In the embodiment of the invention, in order to monitor the stability trend in real time, a five-layer recurrent neural network is constructed, the number of layer nodes is 32, 64, 32 and 16 in sequence, and a tanh activation function is used. The network inputs the stability index at the current moment, outputs the predicted value at the next moment, and the training samples cover 1000 groups of different working conditions, and the prediction error is smaller than 0.05. And setting the finishing threshold to be 0.85, and judging that the nitrogen charging process is finished when the stability index is higher than the value for 10 continuous sampling periods. In practical application, the initial stage index fluctuates from 0.6 to 0.7, and stabilizes at 0.88 to 0.92 within 800 milliseconds after adjustment, indicating that the system reaches a steady state. The accuracy of the completion state judgment is ensured through dynamic tracking.
Compared with the traditional method relying on pressure detection, the method provided by the embodiment of the invention has the advantages that the accuracy of airflow characteristic analysis is obviously improved through multi-scale characteristic extraction and neural network prediction. The generated stability index accurately reflects the dynamic evolution of the nitrogen charging process, judges the completion state in real time, avoids the resource waste caused by early or late termination, optimizes the production efficiency, and is suitable for the high-standard nitrogen charging requirements of the industries such as food, medicine and the like.
S108, if the air flow characteristic stability index reaches a preset threshold, classifying the updated index through a logistic regression algorithm, calculating the nitrogen charging completion probability, generating a termination signal to control the nitrogen charging equipment to stop, and recording an air flow characteristic data set to evaluate the production efficiency improvement amplitude.
In the embodiment of the invention, when the air flow characteristic stability index reaches the preset threshold value, the nitrogen charging process is indicated to enter a stable state, the completion state is confirmed through a classification algorithm, a control instruction is generated, and meanwhile, the process data is recorded to improve the quantization efficiency. The specific implementation of the step comprises the following sub-steps:
S1081, carrying out sectional processing on the airflow characteristic stability index, extracting statistics and time sequence characteristics, and constructing a seven-dimensional characteristic vector.
In the embodiment of the invention, the airflow characteristic stability index reflects the dynamic stability of the nitrogen charging process, and the characteristics of the airflow characteristic stability index need to be quantified through characteristic extraction. The index sequence is segmented by adopting a sliding window method, the window length is 100 sampling points, 500 milliseconds are covered, the overlapping rate is 50%, and a first index sequence is generated. For the sequence, statistical features are calculated, including mean, variance, skewness, and kurtosis, where the mean ranges from 0.85 to 0.95, variance is less than 0.02, skewness is from-0.5 to 0.5, kurtosis is near 3, indicating that the data distribution is stable and nearly normal. And extracting time sequence characteristics, including a sequence maximum value, a sequence minimum value and a sequence peak Gu Chazhi, wherein a difference value smaller than 0.15 indicates smaller fluctuation. And (5) integrating statistics and time sequence characteristics, constructing seven-dimensional characteristic vectors, and providing high-quality input for subsequent classification. Dynamic characteristics of the stability index are comprehensively captured through multi-dimensional feature extraction.
S1082, classifying the seven-dimensional feature vectors by adopting a logistic regression algorithm, generating nitrogen charging completion probability and outputting a termination signal.
In the embodiment of the invention, a logistic regression algorithm is used for judging the nitrogen charging completion state. Seven-dimensional feature vectors are input, a model is trained by using 1000 groups of historical data, the optimization target is to minimize cross entropy loss, the learning rate is set to be 0.01, 1000 rounds of training are performed, and the classification accuracy reaches 95%. The classification result is mapped to a probability value by a sigmoid function, and the probability threshold value is set to be 0.9. When the probability values of five continuous sampling periods exceed the threshold value, the nitrogen charging process is stable, and a termination signal is generated. The termination signal triggers the device control command, which is executed by the proportional-integral controller, the proportional coefficient is 0.8, the integral time constant is 12.5 milliseconds, which is about one fourth of the sampling period, ensuring stable stop and avoiding air flow disturbance. The automatic termination is realized through probability classification and precise control, and the operation reliability is improved.
S1083, recording a process parameter data set, extracting a multidimensional feature vector, and evaluating production efficiency improvement by using a support vector machine.
In an embodiment of the invention, the data acquisition unit records nitrogen dosing process data at a sampling frequency of 200 hz, covering about 800 milliseconds, to generate a process parameter dataset. And extracting airflow speed uniformity index, vortex intensity stability index and streamline form continuity index from the data set to respectively reflect speed field consistency, vortex attenuation stability and streamline track smoothness, wherein values are respectively more than 0.9, more than 0.85 and more than 0.88, and constructing an eight-dimensional feature vector. The support vector machine is adopted for classification, a radial basis function is selected, the nuclear parameters are optimized through grid search, the search range is 0.01 to 10, the optimal parameters are 0.5, and the classification accuracy reaches 93%. The classification result shows that compared with the traditional method, the optimized nitrogen filling process has the advantages that the uniformity of the air flow speed is improved by 15%, the stability of the vortex strength is improved by 12%, the continuity of the streamline form is improved by 18%, and the comprehensive efficiency is improved by about 15%. The step quantifies the optimization effect through multidimensional feature analysis, and the practical value of the method is verified.
Compared with the traditional manual judgment, the method and the device have the advantages that the accuracy and the automation level of judging the nitrogen charging completion state are remarkably improved through sliding window analysis, logistic regression classification and support vector machine evaluation. The real-time data recording and the efficiency analysis provide data support for process optimization, reduce resource waste and improve the production efficiency and the stability of nitrogen charging processes in the industries of food, medicine and the like.
In the foregoing, only the specific embodiments of the present disclosure are described, and it will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the systems, modules and units described above may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein. It should be understood that the scope of the present disclosure is not limited thereto, and any equivalent modifications or substitutions can be easily made by those skilled in the art within the technical scope of the present disclosure, and these modifications or substitutions should be included in the scope of the present disclosure.

Claims (9)

1. The quality evaluation method for the bottle nitrogen filling process by using artificial intelligence is characterized by comprising the following steps of:
Acquiring airflow speed distribution, vortex intensity and streamline form at the bottleneck of nitrogen charging equipment, and carrying out feature extraction and time sequence analysis to generate an airflow feature data set;
Extracting uniformity characteristics of airflow speed distribution, attenuation characteristics of vortex intensity and stability characteristics of streamline morphology in the airflow characteristic data set, and performing characteristic fusion processing to obtain multidimensional vector representation of bottleneck airflow characteristics;
Carrying out multidimensional vector representation on the airflow velocity distribution, the vortex intensity and the streamline morphology, constructing a pre-trained flow field dynamic analysis model, predicting the dynamic change rule of the airflow characteristics in a pressure balance state, and obtaining the stability index of the airflow characteristics of the bottle mouth through stability analysis;
if the stability of the airflow characteristics is lower than a preset threshold, predicting short-term change trends of airflow speed distribution, vortex strength and streamline form through a long-short-term memory network algorithm to obtain future evolution states of the airflow characteristics;
adjusting the nitrogen injection rate and the injection angle of the nitrogen charging equipment according to the future evolution state of the airflow characteristics to generate optimized nitrogen charging control parameters;
According to the optimized nitrogen filling control parameters, adjusting nitrogen filling parameters, updating the running state of nitrogen filling equipment in real time, and generating an updated airflow characteristic data set;
inputting the updated airflow characteristic data set into a flow field dynamic analysis model, recalculating characteristic vectors of airflow speed distribution, vortex strength and streamline form, and outputting an updated airflow characteristic stability index so as to determine the completion state of the nitrogen charging process;
if the air flow characteristic stability index reaches the threshold value, classifying and updating the index through a logistic regression algorithm, outputting the nitrogen charging completion probability, generating a termination signal according to the probability, controlling the nitrogen charging equipment to stop, recording an air flow characteristic data set, and determining the production efficiency improvement amplitude.
2. The method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein the steps of obtaining the airflow velocity distribution, the vortex intensity and the streamline form at the bottle mouth of the nitrogen filling device, extracting features and analyzing time series, and generating an airflow feature data set comprise the following steps:
acquiring air flow speed and air inlet pressure data at grid points divided by the diameter of the bottleneck by adopting an air sensor array, and carrying out noise reduction and standardization processing to obtain a first air flow data set;
Constructing an airflow velocity field distribution map according to the first airflow data set, scanning a bottleneck plane through a gas flow probe to obtain streamline track data, and then overlapping vortex degree contours to obtain a first vortex distribution map;
constructing a Bayesian probability network aiming at the first airflow data set and the gas concentration data acquired in the nitrogen filling process, and extracting a gas concentration distribution characteristic vector from an airflow velocity field distribution map;
And superposing a streamline track curve on the basis of the first vortex distribution diagram, calculating a gas diffusion coefficient according to the gas concentration distribution characteristic vector, and quantitatively describing the structure form of the flow field by adopting the diffusion coefficient to obtain a gas flow characteristic time sequence data set so as to form a gas flow characteristic data set.
3. The method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein the extracting the uniformity characteristic of the air flow velocity distribution, the attenuation characteristic of the vortex intensity and the stability characteristic of the streamline form in the air flow characteristic data set, performing characteristic fusion processing to obtain a multidimensional vector representation of the air flow characteristic of the bottle mouth, comprises:
cleaning the airflow characteristic data set by adopting a median filter, and segmenting the airflow characteristic data set by adopting a sliding window method to obtain a preprocessed airflow characteristic data set;
calculating airflow velocity field variance distribution of the bottleneck area according to the preprocessed airflow characteristic data set, performing gridding calculation on the variance distribution to obtain a first velocity field fluctuation value, and normalizing the fluctuation value through a maximum and minimum value to obtain a first airflow uniformity index;
Extracting an eddy current intensity time sequence aiming at the preprocessed air current characteristic data set, calculating an eddy current intensity difference value at adjacent moments to obtain a first eddy current intensity change curve, and fitting the change curve by adopting an exponential decay function to obtain eddy current decay parameters;
and constructing a characteristic correlation matrix according to the first airflow uniformity index and the vortex attenuation parameter, calculating a pairwise relation number between characteristics to obtain a characteristic weight vector, and carrying out weighted summation on airflow characteristics by adopting the characteristic weight vector to obtain multidimensional vector representation of the airflow characteristics of the bottle mouth.
4. The method for evaluating the quality of the bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein the steps of performing multidimensional vector representation on the airflow velocity distribution, the vortex intensity and the streamline morphology, constructing a pre-trained dynamic flow field analysis model, predicting the dynamic change rule of the airflow characteristics in the pressure balance state, and obtaining the stability index of the bottle mouth airflow characteristics through stability analysis comprise the following steps:
Preprocessing air flow speed distribution data by adopting a normalization method, extracting air flow pulsation frequency spectrum characteristics by median filtering and discrete Fourier transformation, and superposing vortex intensity data according to the frequency spectrum characteristics to obtain a first multidimensional characteristic vector;
inputting a depth self-encoder aiming at the first multidimensional feature vector, and performing feature dimension reduction on the self-encoder by halving the number of intermediate layer nodes to obtain a second multidimensional feature vector;
constructing a residual neural network predictor according to the second multidimensional feature vector, extracting time sequence features by stacking residual blocks, forming a flow field dynamic analysis model, and obtaining an airflow dynamic prediction result;
and calculating a bottleneck pressure change sequence according to the air flow dynamic prediction result, calculating a maximum Lyapunov exponent according to the pressure change sequence, and dividing a stability interval by a characteristic sequence variance to obtain a stability index of bottleneck air flow characteristics.
5. The method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein if the stability of the air flow characteristics is lower than a preset threshold, predicting short-term variation trends of air flow speed distribution, vortex strength and streamline morphology by a long-term memory network algorithm to obtain future evolution states of the air flow characteristics, comprising:
if the airflow characteristic stability index is lower than a preset threshold, preprocessing airflow characteristic data by adopting median filtering, and extracting airflow speed distribution characteristics and vortex intensity time sequence characteristics through wavelet transformation to obtain a first characteristic vector;
generating a training data sequence by adopting a sliding window method according to the first feature vector, and training by using a long-short-term memory network predictor to obtain a first prediction result;
Extracting airflow speed distribution trend and vortex intensity change rule according to the first prediction result, and fusing by a weighted average method to obtain a second feature vector;
and constructing an airflow characteristic evolution sequence according to the second characteristic vector, processing the evolution sequence by adopting an exponential smoothing method, and acquiring a future evolution state of the airflow characteristic from the smoothed sequence.
6. The method for evaluating quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein the step of adjusting the nitrogen injection rate and the injection angle of the nitrogen filling device according to the future evolution state of the airflow characteristics to generate the optimized nitrogen filling control parameters comprises the following steps:
acquiring airflow characteristic evolution state data, carrying out normalization processing and median filtering on the airflow characteristic evolution state data, and obtaining a first evolution data set through a sliding window method;
Extracting an airflow speed distribution sequence according to the first evolution data set, carrying out Fourier transform operation on the airflow speed distribution sequence, and acquiring airflow fluctuation period data from a Fourier transform operation result;
carrying out reconstruction operation on the airflow fluctuation period data by adopting a phase space reconstruction method, and obtaining a first vortex evolution rule by calculating a radius of curvature of a reconstruction track;
And constructing an injection angle range and an injection rate range according to the first vortex evolution rule, and carrying out optimization operation on the injection angle range and the injection rate range by adopting a particle swarm algorithm to obtain optimized nitrogen filling control parameters.
7. The method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein the adjusting the nitrogen injection parameter according to the optimized nitrogen filling control parameter updates the running state of the nitrogen filling device in real time, and generates an updated air flow characteristic data set, comprises:
Acquiring bottleneck airflow data, and performing Gaussian filtering processing on the airflow data to obtain a first airflow data set;
Calculating gridding variance distribution according to the first airflow data set, acquiring a first speed field uniformity index by adopting a regional block statistical method, and extracting adjacent sampling point difference values from the first speed field uniformity index to obtain a first attenuation rate curve;
Constructing a first response characteristic vector aiming at the first attenuation rate curve, and extracting features of the first response characteristic vector by adopting a support vector regressor to obtain a second response characteristic vector;
If the speed field uniformity index in the second response characteristic vector is larger than a preset threshold, judging that the speed distribution is uniform, and updating the speed distribution data in real time by adopting a sliding window method to obtain an airflow characteristic data set.
8. The method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein the steps of inputting the updated air flow characteristic data set into a dynamic flow field analysis model, recalculating characteristic vectors of air flow velocity distribution, vortex strength and streamline form, and outputting updated air flow characteristic stability indexes to determine the completion state of the nitrogen filling process comprise the following steps:
Carrying out abnormal value detection on the air flow characteristic data set by adopting a three-time standard deviation principle, and carrying out smoothing treatment on the air flow characteristic data set through a median filter to obtain a first treatment data set;
Performing wavelet transformation four-layer decomposition according to the first processing data set, and constructing a multidimensional speed distribution feature vector from wavelet transformation decomposition coefficients;
calculating the vortex intensity change rate aiming at the multidimensional speed distribution feature vector, and smoothing the vortex intensity change rate by an exponential weighting method to obtain a vortex feature vector;
and constructing a long-short time memory network according to the eddy characteristic vector to dynamically predict, analyzing a principal component of a prediction result to obtain an air flow characteristic stability index, and judging that the nitrogen charging process is completed if the air flow characteristic stability index is higher than a preset completion threshold value in ten continuous sampling periods.
9. The method for evaluating the quality of a bottle nitrogen filling process by using artificial intelligence according to claim 1, wherein if the air flow characteristic stability index reaches a threshold value, classifying and updating the index by a logistic regression algorithm, outputting a nitrogen filling completion probability, generating a termination signal according to the probability, controlling nitrogen filling equipment to stop, recording an air flow characteristic data set, and determining a production efficiency improvement amplitude, comprises:
Comparing the air flow characteristic stability index with a preset threshold value, adopting a sliding window to segment the stability index, and obtaining a seven-dimensional characteristic vector by calculating statistical characteristics and time sequence characteristics;
Classifying the seven-dimensional feature vectors by adopting a logistic regression algorithm, obtaining probability values through a sigmoid function, and obtaining a termination signal according to the probability values continuously exceeding a preset probability threshold;
constructing a device control instruction aiming at the termination signal, executing the control instruction by adopting a proportional-integral controller, and recording an airflow characteristic data set;
and extracting an airflow speed uniformity index, a vortex intensity stability index and a streamline form continuity index from the airflow characteristic data set, constructing an eight-dimensional characteristic vector, classifying the eight-dimensional characteristic vector by adopting a support vector machine, and extracting an efficiency improvement index from a classification result.
CN202510628010.8A 2025-05-15 2025-05-15 Quality assessment method for bottle nitrogen filling process by using artificial intelligence Active CN120541419B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202510628010.8A CN120541419B (en) 2025-05-15 2025-05-15 Quality assessment method for bottle nitrogen filling process by using artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202510628010.8A CN120541419B (en) 2025-05-15 2025-05-15 Quality assessment method for bottle nitrogen filling process by using artificial intelligence

Publications (2)

Publication Number Publication Date
CN120541419A CN120541419A (en) 2025-08-26
CN120541419B true CN120541419B (en) 2025-11-04

Family

ID=96786873

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202510628010.8A Active CN120541419B (en) 2025-05-15 2025-05-15 Quality assessment method for bottle nitrogen filling process by using artificial intelligence

Country Status (1)

Country Link
CN (1) CN120541419B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109268130A (en) * 2017-07-18 2019-01-25 福特全球技术公司 Method and system for active casing processing control
CN115879396A (en) * 2023-03-02 2023-03-31 中国航发四川燃气涡轮研究院 Flow one-dimensional pneumatic design method for air inlet front chamber of high-altitude simulation test bed

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007047367A2 (en) * 2005-10-14 2007-04-26 Flowsafe Inc. Converting existing prior art fume hoods into high performance low airflow stable vortex fume hoods
CN103043311B (en) * 2013-01-11 2014-08-20 申正军 Nitrogen charging sealing method of drugs
WO2021260082A1 (en) * 2020-06-25 2021-12-30 Redwave Medical GmbH Method for operating a blood pressure measuring apparatus, and apparatus for carrying out the method
US20250131353A1 (en) * 2023-09-27 2025-04-24 Hydrology Bureau of Yangtze River Water Resources Commission Scheduling method and system for operation of reservoirs to recharge freshwater for repelling saltwater intrusion under changing conditions
CN118536785B (en) * 2024-07-26 2024-10-25 南通智恒塑胶制品有限公司 PVC ball automatic inflation regulation and control system
CN118602276B (en) * 2024-08-02 2024-10-29 陕西黑石绿能能源科技有限公司 Hydrogen filling control method of hydrogen filling machine, electronic equipment and storage medium
CN119178100A (en) * 2024-10-11 2024-12-24 云南电网有限责任公司楚雄供电局 Air inflation control method and device for aerial working clothes, equipment and storage medium
CN119550591A (en) * 2024-10-29 2025-03-04 深圳市同元科技有限公司 Intelligent processing method and system for plastic mold process parameters
CN119271976B (en) * 2024-12-11 2025-02-25 南京比逊弥特智能科技有限公司 Stacked bottle filling control method and system integrating image and photoelectric sensors

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109268130A (en) * 2017-07-18 2019-01-25 福特全球技术公司 Method and system for active casing processing control
CN115879396A (en) * 2023-03-02 2023-03-31 中国航发四川燃气涡轮研究院 Flow one-dimensional pneumatic design method for air inlet front chamber of high-altitude simulation test bed

Also Published As

Publication number Publication date
CN120541419A (en) 2025-08-26

Similar Documents

Publication Publication Date Title
CN119598410B (en) Data processing method, device and equipment based on artificial intelligence technology and data model
CN118226313B (en) Storage battery remaining life assessment method and system based on data analysis
CN119247151B (en) Golf cart battery capacity prediction method, device, equipment and storage medium
CN119884694B (en) Vortex flowmeter monitoring method, device, medium and product based on deep learning
CN118557162B (en) Sphygmomanometer calibration method and system
CN118693887B (en) An optimization method for controlling wind power prediction error during black start of overcapacity energy storage
CN112288137A (en) LSTM short-term load prediction method and device considering electricity price and Attention mechanism
CN119024215B (en) Method for predicting residual capacity of lead-acid storage battery of direct-current system
CN119179404A (en) Touch screen pressure sensing method, device, equipment and storage medium
CN111260082B (en) Spatial object motion trail prediction model construction method based on neural network
CN118800428B (en) Intelligent medical diagnostic bed control system and method based on Internet of Things
CN120180276A (en) Transformer fault prediction method based on time-space-spectrum multi-source information fusion
CN119720818A (en) Barrel ablation wear prediction method and device based on degradation path approximation
CN119868711A (en) Automatic control method and system for field intelligent injection pump based on physiological parameter monitoring
CN120541419B (en) Quality assessment method for bottle nitrogen filling process by using artificial intelligence
CN120089833A (en) Performance optimization test method and system for new energy batteries
CN118964964B (en) Fuzzy semantic coding method, device, equipment and medium based on electroencephalogram signals
CN118521837B (en) Rapid iteration method of intelligent detection model for defects of contact net
CN119361085A (en) An emotion recognition method and system based on artificial intelligence
CN117436476A (en) Method for predicting battery RUL based on EIS enhanced data
CN117052970A (en) Intelligent control system and method for pneumatic ball valve assembly
CN120372455B (en) Urban pipe network operation state identification method
CN114742320B (en) Geographic element prediction model construction method, prediction method and related device
CN120577710B (en) Energy storage system state evaluation method based on neural network
CN115688050B (en) A method for detecting time series change points of multi-sensor information based on fusion confidence strategy

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant