CN120971706A - Method and system for detecting oil state of gearbox of wind power generation equipment - Google Patents

Abstract

This invention discloses a method and system for detecting the state of gearbox oil in wind power generation equipment. The method acquires oil samples, uses an oil contamination analysis model to quantify solid, liquid, and gaseous contaminants, and simultaneously employs an oil physicochemical property analysis model to process parameters such as kinematic viscosity. The results from both methods are then fused using tensors, followed by feature decoupling and deep feature mining. Finally, a state assessment decision model determines the oil state. The system comprises six units, including sample acquisition and contamination analysis, to achieve these functions. This approach overcomes the shortcomings of traditional methods that rely on single-factor detection and lack correlation analysis. Instead of isolated detection of a single indicator, it comprehensively integrates contamination and physicochemical property information. Through tensor fusion, feature decoupling, and other techniques, it deeply analyzes the relationship between the two, achieving accurate assessment of the gearbox oil state and effectively avoiding misjudgments caused by incomplete detection or failure to consider interrelationships.

Description

Method and system for detecting oil state of gearbox of wind power generation equipment

Technical Field

The invention relates to the field of detection of gearboxes, in particular to a method and a system for detecting the oil state of a gearbox of wind power generation equipment.

Background

In the field of wind power generation, a gear box is used as a core transmission component, and the stable operation of the gear box directly relates to the efficiency and the reliability of the whole wind power generation system. The accurate detection of the oil state of the gear box is a key link for guaranteeing the normal operation of the gear box. With the development of wind power generation equipment to large-scale and intelligent directions, the traditional oil liquid state detection method is difficult to meet the actual demands.

The prior art has a plurality of defects in the aspect of oil liquid state detection. On the one hand, the detection means is single, and most of the detection means only detect one or more indexes of the oil, for example, only pay attention to the content of solid particle pollutants in the oil, or only analyze the kinematic viscosity of the oil, and cannot comprehensively reflect the overall state of the oil. The one-sided detection mode is easy to cause misjudgment on the oil state of the gear box, and the best maintenance and treatment time is missed. On the other hand, the existing detection method lacks in-depth analysis of the correlation between oil pollution and physicochemical properties. In fact, the physical and chemical properties of the pollutants in the oil can be influenced, and the influence of the pollutants can be aggravated by the change of the physical and chemical properties, so that the interaction of the pollutants and the physical and chemical properties can influence the running condition of the gearbox. However, the prior art fails to organically combine pollution degree analysis with physicochemical property analysis, so that the actual state of the gearbox oil is difficult to accurately evaluate, and powerful guarantee cannot be provided for efficient operation of wind power generation equipment.

Disclosure of Invention

In order to overcome the defects and shortcomings in the prior art, the invention provides a method and a system for detecting the oil state of a gearbox of wind power generation equipment.

The technical scheme adopted by the invention is that the method for detecting the oil state of the gearbox of the wind power generation equipment comprises the following steps:

Step S1, acquiring a gearbox oil sample, quantitatively analyzing solid particle pollutants, liquid pollutants and gas pollutants in the oil sample based on an oil pollution degree analysis model, and performing feature extraction and coding on particle size distribution and concentration information of different types of pollutants by constructing a multidimensional pollution factor matrix;

S2, utilizing an oil physicochemical property analysis model to establish physicochemical property parameter space for kinematic viscosity, acid value, moisture content and oxidation stability parameters of the gearbox oil, and converting the actually measured parameter value into a corresponding characteristic vector in a nonlinear mapping mode;

Step S3, tensor fusion is carried out on the multidimensional pollution factor matrix obtained in the step S1 and the physicochemical characteristic vector obtained in the step S2, so as to form a composite characteristic tensor containing pollution information and physicochemical characteristic information;

S4, separating the pollution related features and the physicochemical property related features in the composite feature tensor by utilizing a preset feature decoupling algorithm based on the composite feature tensor to obtain a pollution feature tensor and a physicochemical property feature tensor;

s5, carrying out deep feature mining on the pollution feature tensor and the physicochemical property feature tensor respectively, and extracting high-order pollution features and high-order physicochemical property features by constructing a multi-level feature extraction network;

And S6, inputting the extracted high-order pollution characteristics and high-order physicochemical property characteristics into a state evaluation decision model, comprehensively judging the oil state of the gear box through a preset decision rule, and outputting a detection result of the oil state of the gear box.

Further, in the step S1, a multidimensional pollution factor matrix constructed by the oil pollution degree analysis model is calculated by the following formula:

Wherein M _cF represents a multidimensional pollution factor matrix, n is the number of pollutant types, alpha _i is a weight coefficient of the i-th pollutant, the coefficient is determined according to an anti-pollution performance parameter of the gearbox oil, D _i is a particle size distribution vector of the i-th pollutant and comprises pollutant duty ratio information of different particle size intervals, C _i is a concentration scalar of the i-th pollutant, f _i is a characteristic coding function of the i-th pollutant, and a coding rule is adjusted according to a lubrication characteristic parameter of the gearbox oil.

Further, in the step S2, a physicochemical property parameter space conversion formula established by the oil physicochemical property analysis model is as follows:

V_PP＝β·ReLU(γ·[V,A,W,O]^T)

Wherein V _PP represents a physical and chemical property characteristic vector after conversion, beta is a scaling matrix, element values are set according to specification parameters of gear box oil, gamma is a weight matrix and is determined according to performance standard parameters of the gear box oil, V is a motion viscosity measured value of the gear box oil, A is an acid value measured value, W is a moisture content measured value, O is an oxidation stability index measured value, and ReLU is a linear rectification activation function for enhancing nonlinear expression capability of the characteristic vector.

Further, in the step S3, the tensor fusion process adopts a combination mode of tensor product and weighted summation, and the specific fusion formula is as follows:

Wherein T _CFP represents a composite characteristic tensor, m is the dimension of the physicochemical characteristic vector, omega _j is the weight coefficient of the j-th physicochemical characteristic vector in the fusion process, the coefficient is determined by the using working condition parameters of gearbox oil, and V _PPj is the j-th component of the physicochemical characteristic vector V _PP; representing tensor product operations.

Further, in the step S4, the feature decoupling algorithm decomposes the composite feature tensor based on the tensor decomposition principle by the following formula:

Wherein T _CF represents pollution characteristic tensor, T _PP represents physicochemical characteristic tensor, hadamard product operation of tensor, E is decomposition error tensor, the norm of the error tensor is minimized by iterative optimization algorithm, and parameter adjustment in the optimization process is carried out according to the stability parameters of the gearbox oil.

Further, in the step S5, the multi-level feature extraction network includes a plurality of feature extraction layers, and a feature extraction formula of each layer is:

F_l＝σ(θ_l·F_l-1+b_l)

Wherein F _l is the characteristic tensor extracted by the first layer, F _l-1 is the output characteristic tensor of the first-1 layer, theta _l is the weight tensor of the first layer, the parameter value is initialized according to the abrasion characteristic parameter of the gear box oil, b _l is the bias tensor of the first layer, sigma is an activation function, and a proper activation function type is selected according to the performance change trend of the gear box oil.

Further, in the step S6, the state evaluation decision model adopts a mode of combining a decision tree with a rule base, and the decision process is performed by the following rules:

If the oil liquid state of the gear box is H _CF>τ_CF and H _PP>τ_PP, judging that the oil liquid state of the gear box is seriously abnormal;

If the oil liquid state of the gear box is H _CF>τ_CF and H _PP≤τ_PP, judging that the oil liquid state of the gear box is pollution abnormality;

If the oil liquid state of the gear box is H _CF≤τ_CF and H _PP>τ_PP, judging that the oil liquid state of the gear box is abnormal in physicochemical property;

If the oil level is H _CF≤τ_CF and H _PP≤τ_PP, judging that the oil level of the gear box is normal;

The method comprises the steps of extracting high-order pollution characteristics from gearbox oil, wherein H _CF is a comprehensive measurement value of the extracted high-order pollution characteristics and is calculated according to a cleanliness standard parameter of the gearbox oil, H _PP is a comprehensive measurement value of high-order physicochemical characteristics and is determined according to a quality standard parameter of the gearbox oil, tau _CF and tau _PP are respectively judging thresholds of the pollution characteristics and the physicochemical characteristics, and the thresholds are set by carrying out statistical analysis on historical detection data of a large number of gearbox oil samples and combining with performance limit parameters of the gearbox oil.

Further, before step S1, the method further includes a step of determining a collection position of the oil sample of the gearbox, where the collection position is determined by the following method:

According to the distribution characteristics of the internal flow field of the gearbox, combining the liquidity parameters of the gearbox oil, establishing an oil sample acquisition position optimization model, and determining an optimal acquisition position by calculating representative indexes R _I of oil samples at different positions, wherein the calculation formula is as follows:

The method comprises the steps of setting p as the number of candidate collecting positions, setting delta _k as the weight coefficient of the kth candidate collecting position, determining the coefficient by structural parameters of a gear box, setting Corr (S _k,S_ref) as the correlation measurement value of an oil sample of the kth candidate collecting position and a reference sample, constructing the reference sample according to the standard performance parameters of gear box oil, and selecting the position with the largest representative index as the oil sample collecting position by comparing the representative index R _I of each candidate collecting position.

Further, after step S6, a confidence evaluation step of the detection result is further included, the confidence evaluation being calculated by the following formula:

The method comprises the steps of detecting a gear box oil, wherein C _E represents the confidence coefficient of the detection result, R is the number of reference indexes used for evaluating the confidence coefficient, lambda _q is the weight coefficient of the q-th reference index, the coefficient is set according to the detection precision requirement parameter of the gear box oil, conf (R _q) is the confidence coefficient value corresponding to the q-th reference index, the reference index comprises the repeatability of detection data and the consistency with historical detection data, and the confidence coefficient value is calculated according to the stability parameter of the gear box oil and the performance parameter of detection equipment.

Wind power generation equipment gear box fluid state detecting system includes:

The oil liquid sample acquisition and pollution degree quantitative analysis unit is used for acquiring an oil liquid sample of the gearbox, and carrying out quantitative analysis on solid particle pollutants, liquid pollutants and gas pollutants in the oil liquid sample based on an oil liquid pollution degree analysis model to construct a multidimensional pollution factor matrix;

The physicochemical property parameter analysis and characteristic vector conversion unit is used for establishing a physicochemical property parameter space based on an oil physicochemical property analysis model for the kinematic viscosity, acid value, moisture content and oxidation stability parameters of the gearbox oil, and converting the actually measured parameter value into a corresponding characteristic vector;

the pollution-physical and chemical characteristic tensor fusion unit is used for tensor fusion of the multidimensional pollution factor matrix and the physical and chemical characteristic tensor to form a composite characteristic tensor containing pollution information and physical and chemical characteristic information;

The characteristic decoupling and sub-tensor separating unit is used for separating the pollution related characteristic and the physicochemical property related characteristic in the composite characteristic tensor based on a preset characteristic decoupling algorithm to obtain a pollution characteristic sub-tensor and a physicochemical property characteristic sub-tensor;

the multi-level feature extraction and high-order feature mining unit is used for respectively carrying out deep feature mining on the pollution feature tensor and the physicochemical property feature tensor and extracting high-order pollution features and high-order physicochemical property features;

And the oil liquid state comprehensive judgment and result output unit is used for comprehensively judging the oil liquid state of the gear box through a preset decision rule based on the extracted high-order pollution characteristic and the high-order physicochemical property characteristic and outputting the detection result of the oil liquid state of the gear box.

The method and the system for detecting the oil state of the gearbox of the wind power generation equipment have the beneficial effects that in the detection means, the method is not limited to single index analysis, solid particles, liquid and gas pollutants in an oil sample are quantitatively analyzed at the same time, a multidimensional pollution factor matrix and physicochemical property feature vectors are constructed by combining multiple physicochemical property parameters such as kinematic viscosity, acid value and the like, and the comprehensive detection of the oil state is realized by fusing tensors to form a composite feature tensor. In the aspect of analysis logic, the system does not stand alone for pollution and physicochemical properties, but separates pollution and physicochemical property characteristics in a composite characteristic tensor by utilizing a characteristic decoupling algorithm, extracts high-order characteristics through a multi-level characteristic extraction network, comprehensively judges based on a decision tree and a rule base, fully considers interaction relation of the two, and accurately evaluates the oil state. In addition, the system also comprises links such as sample acquisition position optimization, detection result confidence evaluation and the like, so that the representativeness and reliability of the detection result are ensured. The method and the system comprehensively improve the accuracy and the effectiveness of the detection of the oil state of the gearbox, provide powerful guarantee for the stable operation of the wind power generation equipment, discover potential problems of oil in time and avoid equipment faults caused by misjudgment or incomplete detection.

Drawings

FIG. 1 is a flow chart of method steps of the present invention;

Fig. 2 is a diagram showing the system unit constitution of the present invention.

Detailed Description

It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other, and the present application will be further described in detail with reference to the drawings and the specific embodiments.

As shown in fig. 1, the method for detecting the oil state of the gearbox of the wind power generation equipment comprises the following steps:

Step S1, acquiring a gearbox oil sample, quantitatively analyzing solid particle pollutants, liquid pollutants and gas pollutants in the oil sample based on an oil pollution degree analysis model, and performing feature extraction and coding on particle size distribution and concentration information of different types of pollutants by constructing a multidimensional pollution factor matrix;

specifically, the main task of step S1 is to obtain a gearbox oil sample, and quantitatively analyze pollutants in the sample based on an oil pollution degree analysis model. In the embodiment, when an oil sample is obtained, a special sampler is generally adopted, and under the condition that a gear box of the wind power generation equipment is in a stable running state, the oil sample is extracted from a specific sampling port, and the position of the sampling port is optimally designed, so that the collected sample can be ensured to be representative. The method comprises the steps of obtaining a sample, detecting solid particle pollutants by using a particle counter, accurately measuring the particle number in different particle size intervals by using a laser scattering principle, further determining the particle size distribution and concentration of the solid particle pollutants, identifying the types of the liquid pollutants according to the separation characteristics of different substances in a chromatographic column by using a chromatographic analyzer, determining the mixing proportion of the liquid pollutants, and analyzing the gas components and the content dissolved in oil liquid by using a gas sensor by using a gas analyzer. And finally, constructing a multidimensional pollution factor matrix according to the detection results and completing feature extraction and coding of pollutant information.

Accurate quantitative pollutant information is a core basis for judging the pollution degree of oil in a gear box. In the long-term operation process of the wind power generation equipment, foreign matters such as external dust, gravel and the like possibly invade oil, and the friction and abrasion of parts in the gear box also can generate pollutants such as metal particles. The different types of pollutants have different damage modes and degrees on abrasion, corrosion and the like of gear box components, and the specific condition of oil pollution can be mastered through comprehensive and careful analysis and coding, so that potential threats to normal operation of the gear box are evaluated, and important data support is provided for subsequent detection and maintenance decisions. If neglecting this step, the pollutant information cannot be accurately obtained, the problem of oil pollution cannot be found in time, so that the components of the gearbox are worn out too early, the performance is reduced, even the equipment is broken down, and the stable operation of the wind power generation equipment is affected.

S2, utilizing an oil physicochemical property analysis model to establish physicochemical property parameter space for kinematic viscosity, acid value, moisture content and oxidation stability parameters of the gearbox oil, and converting the actually measured parameter value into a corresponding characteristic vector in a nonlinear mapping mode;

Specifically, in the step S2, an oil physicochemical property analysis model is used for developing and analyzing a plurality of important parameters such as kinematic viscosity, acid value, moisture content, oxidation stability and the like of gearbox oil. The method comprises the steps of measuring the kinematic viscosity by a capillary viscometer, measuring the flowing time of oil through the capillary at a specified temperature according to the poiseuille law, further calculating the kinematic viscosity value, measuring the acid value by a potentiometric titration method, reacting an appropriate titrant with acidic substances in the oil by a measuring electrode potential change to determine a titration end point, obtaining the acid value, detecting the moisture content by a Karl Fischer titration method, chemically reacting a Karl Fischer reagent with water, determining the moisture content by measuring the consumption of the reagent, measuring the oxidation stability by a rotary oxygen bomb method, contacting the oil sample with oxygen at a specified temperature and pressure, and recording the time required for reaching a specified pressure drop value, thereby evaluating the oxidation stability of the oil. After obtaining actual measurement values of various parameters, a physicochemical property parameter space is established through a specially designed software algorithm, and the parameters are converted into corresponding feature vectors by using a nonlinear mapping mode, so that the digitalized expression of the physicochemical property of the oil is realized.

The physicochemical properties of the oil directly affect the lubrication, corrosion resistance and other performances of the oil, so as to determine the operation reliability of the gear box. The proper kinematic viscosity is directly related to whether each part of the gearbox can be well lubricated, the friction resistance among the parts is increased due to the fact that the viscosity is too high, energy consumption is increased, an effective lubricating oil film cannot be formed due to the fact that the viscosity is too low, abrasion of the parts is increased, the fact that the acid value is increased means that oil is possibly subjected to oxidative deterioration or pollution to corrode the gearbox parts, the oil is emulsified due to the fact that the water content is too high, the lubricating performance of the oil is damaged, and the fact that the oxidation stability is poor accelerates the degradation of the oil performance. Through converting the physical and chemical property parameters which are related and are mutually influenced into feature vectors, the comprehensive performance state of the oil can be systematically analyzed, the variation trend of each parameter is comprehensively captured, and a key basis is provided for accurately judging whether the oil is suitable for continuous use. If the step is skipped, the intrinsic performance change of the oil cannot be deeply known, the normal operation of the gear box is difficult to ensure, equipment faults can be caused due to poor oil performance, and maintenance cost and downtime are increased.

Step S3, tensor fusion is carried out on the multidimensional pollution factor matrix obtained in the step S1 and the physicochemical characteristic vector obtained in the step S2, so as to form a composite characteristic tensor containing pollution information and physicochemical characteristic information;

Specifically, in step S3, tensor fusion is performed on the multidimensional pollution factor matrix obtained in step S1 and the physicochemical characteristic feature vector obtained in step S2, so as to form a composite characteristic tensor. In the specific implementation process, fusion operation is realized by means of a data processing platform of a high-performance computer and by adopting a specially written tensor operation program. When tensor product operation is carried out, operation is carried out on each dimension element of the multidimensional pollution factor matrix and the physicochemical property feature vector according to tensor operation rules strictly, and a primary fusion result is obtained. And then, according to the using working condition parameters of the gearbox oil, such as the running temperature, the load size, the rotating speed and the like, corresponding weight coefficients are given to the physicochemical property feature vectors with different dimensions through a pre-established weight distribution model. The weight distribution model is constructed based on a large amount of experimental data and actual operation experience, and can accurately reflect the influence degree of each physicochemical property parameter on the oil liquid state under different working conditions. And finally, carrying out weighted summation on the primary fusion result to obtain a composite characteristic tensor containing pollution information and physicochemical property information, and realizing the high integration of oil liquid state data.

This step has a key role in accurately assessing the oil condition. The pollution condition and physicochemical properties of the oil are closely related and mutually influenced, the existence of pollutants can accelerate the oxidation process of the oil, so that the change of physicochemical properties such as the increase of acid value, the change of kinematic viscosity and the like is caused, and the change of physicochemical properties such as the reduction of viscosity can influence the suspension and dispersion state of the pollutants in the oil, so that the abrasion degree of the pollutants on gear box components is influenced. Through tensor fusion, the complex correlations can be presented in a data form, so that subsequent analysis can be performed from an overall point of view, and the phenomenon that the pollution or the physicochemical property of oil is treated in isolation is avoided. The comprehensive data integration lays a foundation for grasping the actual state of oil more accurately, provides a more reliable and comprehensive basis for the maintenance and management of the gear box, is beneficial to operation and maintenance personnel to formulate a more scientific and reasonable maintenance strategy, and ensures the stable operation of the wind power generation equipment.

S4, separating the pollution related features and the physicochemical property related features in the composite feature tensor by utilizing a preset feature decoupling algorithm based on the composite feature tensor to obtain a pollution feature tensor and a physicochemical property feature tensor;

specifically, step S4 separates the pollution-related features and the physicochemical-property-related features thereof by using a preset feature decoupling algorithm based on the composite feature tensor obtained in step S3, and obtains a pollution feature tensor and a physicochemical-property feature tensor. In the implementation mode, a characteristic decoupling algorithm is realized by utilizing computer programming, and the algorithm is based on a tensor decomposition principle and adopts an iterative optimization calculation mode. Firstly, according to stability parameters of gearbox oil, such as oxidation resistance stability, emulsification resistance stability and the like of oil, initial parameters in an algorithm are reasonably set. Then, the relevant parameters in the decomposition process are continuously adjusted through multiple iterative calculations, and in each iteration, the norm of the decomposition error tensor is calculated, and parameter optimization is performed with the aim of minimizing the norm. In the specific calculation process, mathematical methods such as matrix operation, vector operation and the like are utilized to gradually decompose the composite characteristic tensor into two relatively independent sub tensors. In the iteration process, the accuracy and reliability of the decomposition result are monitored in real time, and when the norm of the decomposition error tensor meets the preset precision requirement, iteration is stopped to obtain the final pollution characteristic tensor and physicochemical property characteristic tensor.

Although the composite characteristic tensor integrates rich oil liquid state information, in the actual analysis and evaluation process, the pollution characteristic and the physicochemical property characteristic are separated, so that the respective change condition and interaction mechanism of the pollution characteristic and the physicochemical property characteristic can be clearly known. For example, when the pollution characteristic tensor is analyzed, the change trend of different types of pollutants and the influence degree of the pollution on the abrasion of gear box parts can be studied, so that whether treatment measures such as oil filtration or replacement are needed or not can be judged, and when the physicochemical characteristic tensor is analyzed, whether the lubricating performance, the corrosion resistance and the like of the oil meet the running requirements of the gear box or not can be determined, and whether additives or other treatment modes are needed to be added to improve the physicochemical properties of the oil can be facilitated. The separation operation improves the pertinence and the accuracy of analysis, avoids the analysis difficulty caused by pollution and physical and chemical property information mixing, and provides clear and definite data support for the follow-up more accurate evaluation of the oil state and the establishment of maintenance decisions.

S5, carrying out deep feature mining on the pollution feature tensor and the physicochemical property feature tensor respectively, and extracting high-order pollution features and high-order physicochemical property features by constructing a multi-level feature extraction network;

Specifically, in step S5, depth feature mining is performed on the pollution feature tensor and the physicochemical property feature tensor respectively, and the implementation is achieved by constructing a multi-level feature extraction network. When the multi-level feature extraction network is actually built, the weight tensor of each layer in the network is initialized and set according to the abrasion characteristic parameters of gear box oil, such as gear material characteristics, abrasion rate in the running process and the like, based on a deep learning framework, such as TensorFlow or PyTorch. Meanwhile, according to the performance change trend of the oil under different working conditions, a proper activation function, such as a ReLU function, a Sigmoid function and the like, is selected. In the network operation process, each layer receives the characteristic tensor output by the previous layer as input, multiplies the input characteristic tensor with the weight tensor of the layer through matrix multiplication operation, adds the offset tensor to obtain a preliminary calculation result, and then inputs the preliminary calculation result into an activation function to perform nonlinear transformation to obtain the output characteristic tensor of the layer. By such calculation and activation operations of a plurality of layers, the high-order pollution characteristics and the high-order physicochemical characteristics are gradually extracted from the pollution characteristic tensor and the physicochemical characteristic tensor. In the network training process, a large amount of historical oil liquid detection data is used as a training sample, and weight parameters of all layers in the network are continuously adjusted through a back propagation algorithm, so that errors between a predicted result and an actual result are minimized, and the accuracy and the effectiveness of network extraction characteristics are improved.

Only basic pollution and physicochemical property information of oil are acquired, and the actual state of the oil under complex working conditions is difficult to comprehensively and accurately evaluate. The network is extracted through the multi-level features to mine the high-order features, so that the subtle trend and potential law of the oil liquid state change can be captured. For example, in the early wear stage of a gear box, the high-order pollution characteristic can find abnormal changes of the quantity and the particle size distribution of some tiny particle pollutants, the changes are not obvious in basic analysis, but are indicative of potential wear problems, and for the oxidation process of oil, the high-order physicochemical property characteristic can reflect the degree and the speed of performance degradation of the oil more accurately, so that the basis is provided for taking measures in advance to prevent the further degradation of the oil performance. The high-order features provide more valuable information for subsequent state evaluation, so that the detection result is more prospective and reliable, potential problems of the oil liquid state can be found in time by operation and maintenance personnel, a maintenance plan is formulated in advance, the probability of equipment failure is reduced, and stable and efficient operation of the wind power generation equipment is ensured.

And S6, inputting the extracted high-order pollution characteristics and high-order physicochemical property characteristics into a state evaluation decision model, comprehensively judging the oil state of the gear box through a preset decision rule, and outputting a detection result of the oil state of the gear box.

Specifically, step S6 inputs the high-order pollution feature and the high-order physicochemical property feature extracted in step S5 into a state evaluation decision model, comprehensively determines the oil state of the gearbox according to a preset decision rule, and outputs a detection result of the oil state of the gearbox. In implementation, the state evaluation decision model is constructed based on a computer software system, and a framework combining decision trees and rule bases is adopted. The decision rule is set based on statistical analysis of historical detection data of a large number of gearbox oil samples, and is combined with a cleanliness standard parameter, a quality standard parameter and a performance limit parameter of the gearbox oil. When the decision tree is specifically constructed, key parameters of high-order pollution characteristics and high-order physicochemical property characteristics are used as node division basis, and the optimal node division mode is determined by continuously learning the relation between the characteristics in the historical data and the oil liquid state, so that a decision tree structure with distinct layers is formed. The rule base stores specific rules for judging the oil state of the gearbox under various conditions, for example, when the comprehensive measurement value of the high-order pollution characteristic is higher than the pollution characteristic judgment threshold value and the comprehensive measurement value of the high-order physicochemical characteristic is also higher than the physicochemical characteristic judgment threshold value, the rules for judging the oil state of the gearbox as serious abnormality are corresponding to the rules in the rule base, otherwise, if the two are lower than the threshold value, the rules are judged as normal according to the corresponding rules. After the high-order pollution characteristics and the high-order physicochemical property characteristics are input, the decision model firstly carries out preliminary classification through a decision tree, then carries out accurate judgment by combining rules in a rule base, and finally outputs a detection result of the oil state of the gear box.

The detection result output by the step is directly related to the maintenance decision and operation safety of the gear box. By combining the high-order characteristics with a scientifically set decision rule, the oil liquid state can be objectively and accurately estimated. Based on the detection result, operation and maintenance personnel can take corresponding measures in time, for example, when the detection result shows that the oil state is seriously abnormal, the oil is immediately replaced, the fault of a gear box caused by the oil problem is avoided, and if the oil state is slightly abnormal, the filtering treatment can be carried out or the monitoring frequency can be enhanced. The method enables the whole detection flow to form a complete closed loop, and obtains the final state evaluation and result output from sample acquisition, pollutant analysis, physicochemical property analysis, feature fusion and decoupling and deep feature mining, thereby providing comprehensive and reliable technical support for the management of the oil state of the gearbox, being beneficial to improving the operation efficiency of wind power generation equipment, reducing the maintenance cost and ensuring the stable operation of a wind power generation system.

Preferably, in the step S1, the multidimensional pollution factor matrix constructed by the oil pollution degree analysis model is calculated by the following formula:

Wherein M _CF represents a multidimensional pollution factor matrix, n is the number of pollutant types, alpha _i is a weight coefficient of the i-th pollutant, the coefficient is determined according to an anti-pollution performance parameter of the gearbox oil, D _i is a particle size distribution vector of the i-th pollutant and comprises pollutant duty ratio information of different particle size intervals, C _i is a concentration scalar of the i-th pollutant, f _i is a characteristic coding function of the i-th pollutant, and a coding rule is adjusted according to a lubrication characteristic parameter of the gearbox oil.

Specifically, the technology of constructing the multidimensional pollution factor matrix in the step S1 is refined. In practice, various types of contaminants in gearbox oil, such as solid particles, liquid impurities, and gas components, need to be considered. For each type of pollutant, key information is acquired, for example, the proportion distribution of solid particles with different sizes is determined, and the content of liquid and gas pollutants is defined. Meanwhile, different weights are given to various pollutants according to the anti-pollution performance of the gearbox oil. The oil with excellent anti-pollution performance is provided, and certain pollutants have relatively small influence on the oil, so that the weight is set lower. In practice, various contaminant data are accurately determined by means of specialized detection equipment, such as particle counters, chromatographic analyzers and gas analyzers. And then integrating the data according to preset rules and procedures to construct a multidimensional pollution factor matrix. The matrix can comprehensively and systematically present the oil pollution condition, and provides a key data basis for the subsequent deep analysis of the oil condition and judgment of whether the oil condition affects the normal operation of the gearbox.

Preferably, in the step S2, a physicochemical property parameter space conversion formula established by the oil physicochemical property analysis model is as follows:

V_PP＝β·ReLU(γ·[V,A,W,O]^T)

Wherein V _PP represents a physical and chemical property characteristic vector after conversion, beta is a scaling matrix, element values are set according to specification parameters of gear box oil, gamma is a weight matrix and is determined according to performance standard parameters of the gear box oil, V is a motion viscosity measured value of the gear box oil, A is an acid value measured value, W is a moisture content measured value, O is an oxidation stability index measured value, and ReLU is a linear rectification activation function for enhancing nonlinear expression capability of the characteristic vector.

Specifically, the physical and chemical properties of the gear box oil include a plurality of important indexes such as kinematic viscosity, acid value, moisture content and oxidation stability. Different physicochemical property indexes reflect the performance characteristics of oil in different aspects and are related to each other. In the implementation process, a special detection instrument such as a capillary viscometer is used for measuring the kinematic viscosity, a potentiometric titrator is used for measuring the acid value, and the like, so that accurate parameter values are obtained. Then, the measured parameters are processed according to the specification standard and the performance requirement of the oil. These parameters are converted into feature vectors by specific algorithms and procedures. In the conversion process, the numerical range of the characteristic vector is adjusted according to the oil specification, and the importance degree of each parameter is determined according to the performance standard, so that the key physicochemical property index is highlighted. The finally obtained feature vector presents the originally dispersed physicochemical property parameters in a digital form which is more convenient to analyze and process, and is convenient for subsequent systematic research on the physicochemical property state of the oil and the change trend thereof.

Preferably, in the step S3, the tensor fusion process combines tensor product and weighted summation, and the specific fusion formula is as follows:

Wherein T _CFP represents a composite characteristic tensor, m is the dimension of the physicochemical characteristic vector, omega _j is the weight coefficient of the j-th physicochemical characteristic vector in the fusion process, the coefficient is determined by the using working condition parameters of gearbox oil, and V _PPj is the j-th component of the physicochemical characteristic vector V _PP; representing tensor product operations.

Specifically, the tensor fusion in the step S3 needs to deeply fuse the oil pollution information and the physicochemical property information after the independent analysis of the two parts of information is completed. When the method is used, the influence degree of various physicochemical property parameters on the oil state is different under different use conditions of the gearbox oil, for example, under the working conditions of high temperature and high load, the importance of certain physicochemical property indexes can be obviously improved. Therefore, according to the actual working conditions, corresponding weights are given to physicochemical property characteristics of different dimensions. When the method is implemented, the multidimensional pollution factor matrix and the physicochemical characteristic vector obtained in the previous step are firstly obtained, and tensor product and weighted summation operation is carried out on the pollution information matrix and the physicochemical characteristic vector according to a specific fusion rule by utilizing a high-performance computer and professional data processing software. By such operation, a complex characteristic tensor is formed which contains the pollution information and the physicochemical property information and the correlations thereof. The composite characteristic tensor integrates comprehensive information of the oil state, and provides comprehensive data support for more accurately evaluating the oil state and formulating a reasonable maintenance strategy.

Preferably, in the step S4, the feature decoupling algorithm decomposes the composite feature tensor based on the tensor decomposition principle by the following formula:

Wherein T _CF represents pollution characteristic tensor, T _PP represents physicochemical characteristic tensor, hadamard product operation of tensor, E is decomposition error tensor, the norm of the error tensor is minimized by iterative optimization algorithm, and parameter adjustment in the optimization process is carried out according to the stability parameters of the gearbox oil.

Specifically, because the fused composite characteristic tensor contains mixed information of oil pollution and physicochemical properties, in order to more clearly and accurately analyze the influence of the mixed information and the mixed information on the oil state, pollution-related characteristics and physicochemical property-related characteristics in the composite characteristic tensor are required to be separated. In carrying out this step, initial parameters of the feature decoupling algorithm are set according to stability-related characteristics of the gearbox oil, such as antioxidant stability, demulsification stability, etc. Then, by utilizing computer programming and through repeated iterative computation, various parameters in the decomposition process are continuously adjusted. In each iteration, the accuracy of the decomposition result is monitored to ensure that the decomposition error is as small as possible. And stopping iteration after the preset precision requirement is met, and successfully obtaining the pollution characteristic tensor and the physicochemical property characteristic tensor. The separated sub tensors can be used for carrying out in-depth analysis on pollution characteristics and physicochemical property characteristics respectively, so that the oil liquid state can be accurately judged, and a more targeted basis is provided for the maintenance of the gear box.

Preferably, in step S5, the multi-level feature extraction network includes a plurality of feature extraction layers, and a feature extraction formula of each layer is:

F_l＝σ(θ_l·F_l-1+b_l)

Wherein F _l is the characteristic tensor extracted by the first layer, F _l-1 is the output characteristic tensor of the first-1 layer, theta _l is the weight tensor of the first layer, the parameter value is initialized according to the abrasion characteristic parameter of the gear box oil, b _l is the bias tensor of the first layer, sigma is an activation function, and a proper activation function type is selected according to the performance change trend of the gear box oil.

Specifically, in order to extract more valuable information which can reflect the oil liquid state essence from the pollution characteristic tensor and the physicochemical property characteristic tensor, a multi-level characteristic extraction network is constructed. When the network is constructed, the weight tensor of each layer of the network is initialized according to the gear material characteristics applied by the gear box oil, the wear characteristic parameters such as the wear rate in the running process and the like. Meanwhile, by combining the trend of performance change of the oil under different working conditions, a proper activation function is selected. In the implementation process, the pollution and physicochemical property characteristic tensor obtained by decoupling is input into a network, each layer of the network can process the input characteristic tensor, and the high-order pollution characteristic and the high-order physicochemical property characteristic are gradually extracted through operations such as matrix multiplication, bias addition, function operation activation and the like. With the increase of network level, the extracted features are more abstract and can reflect the deep change rule of oil state. The high-order features provide a data support with depth and value for accurately evaluating the oil state and predicting the oil performance change trend.

Preferably, in step S6, the state evaluation decision model adopts a mode of combining a decision tree with a rule base, and the decision process is performed by the following rules:

If the oil liquid state of the gear box is H _CF>τ_CF and H _PP>τ_PP, judging that the oil liquid state of the gear box is seriously abnormal;

If the oil liquid state of the gear box is H _CF>τ_CF and H _PP≤τ_PP, judging that the oil liquid state of the gear box is pollution abnormality;

If the oil liquid state of the gear box is H _CF≤τ_CF and H _PP>τ_PP, judging that the oil liquid state of the gear box is abnormal in physicochemical property;

If the oil level is H _CF≤τ_CF and H _PP≤τ_PP, judging that the oil level of the gear box is normal;

The method comprises the steps of extracting high-order pollution characteristics from gearbox oil, wherein H _CF is a comprehensive measurement value of the extracted high-order pollution characteristics and is calculated according to a cleanliness standard parameter of the gearbox oil, H _PP is a comprehensive measurement value of high-order physicochemical characteristics and is determined according to a quality standard parameter of the gearbox oil, tau _CF and tau _PP are respectively judging thresholds of the pollution characteristics and the physicochemical characteristics, and the thresholds are set by carrying out statistical analysis on historical detection data of a large number of gearbox oil samples and combining with performance limit parameters of the gearbox oil.

Specifically, after the high-order pollution characteristic and the high-order physicochemical property characteristic are obtained, the state of the gearbox oil liquid needs to be judged according to certain standards and rules. In actual operation, according to relevant parameters such as cleanliness standard, quality standard and performance limit of the gearbox oil, comprehensive measurement values of high-order pollution characteristics and high-order physicochemical property characteristics are calculated, so that the degree of pollution and physicochemical property characteristics is quantitatively reflected. Meanwhile, by carrying out systematic statistical analysis on a large number of gear box oil sample historical detection data and combining oil performance limit parameters, a judgment threshold value of pollution and physicochemical property characteristics is set. And after the calculated comprehensive measurement value is compared with the judgment threshold value, judging the oil liquid state according to a preset mode of combining the decision tree with the rule base. If the specific condition is met, judging the oil liquid state as serious abnormality, pollution abnormality, physical and chemical property abnormality or normal. The clear judgment rules provide clear basis for operation and maintenance personnel to accurately judge the oil state and make reasonable maintenance decisions in time.

Preferably, before step S1, the method further includes a step of determining a collection position of the oil sample of the gearbox, where the collection position is determined by the following method:

According to the distribution characteristics of the internal flow field of the gearbox, combining the liquidity parameters of the gearbox oil, establishing an oil sample acquisition position optimization model, and determining an optimal acquisition position by calculating representative indexes R _I of oil samples at different positions, wherein the calculation formula is as follows:

The method comprises the steps of setting p as the number of candidate collecting positions, setting delta _k as the weight coefficient of the kth candidate collecting position, determining the coefficient by structural parameters of a gear box, setting Corr (S _k,S_ref) as the correlation measurement value of an oil sample of the kth candidate collecting position and a reference sample, constructing the reference sample according to the standard performance parameters of gear box oil, and selecting the position with the largest representative index as the oil sample collecting position by comparing the representative index R _I of each candidate collecting position.

Specifically, in order to ensure that the collected oil sample can truly and comprehensively reflect the overall state of the oil in the gearbox, the sample collecting position needs to be scientifically and reasonably determined. In actual operation, firstly, the flow field distribution condition in the gear box is studied, the flow rule of oil in the gear box is known, and meanwhile, the fluidity characteristics of the gear box oil are combined. Then, a plurality of candidate acquisition positions are determined in the gearbox, and for each candidate position, relevant parameters of the oil sample are measured. A reference sample is constructed based on standard performance parameters of the gearbox oil. And calculating the correlation between each candidate position sample and the reference sample, and determining the weight coefficient of each candidate position by combining the structural parameters of the gearbox, so as to calculate the representative index of each position. Finally, comparing the representative index sizes of the candidate positions, and selecting the position with the maximum index as the final sample acquisition position. The collected sample can represent the oil liquid state in the gear box to the greatest extent, and a reliable data basis is provided for subsequent accurate detection and analysis.

Preferably, after step S6, a confidence evaluation step of the detection result is further included, the confidence evaluation being calculated by the following formula:

The method comprises the steps of detecting a gear box oil, wherein C _E represents the confidence coefficient of the detection result, R is the number of reference indexes used for evaluating the confidence coefficient, lambda _q is the weight coefficient of the q-th reference index, the coefficient is set according to the detection precision requirement parameter of the gear box oil, conf (R _q) is the confidence coefficient value corresponding to the q-th reference index, the reference index comprises the repeatability of detection data and the consistency with historical detection data, and the confidence coefficient value is calculated according to the stability parameter of the gear box oil and the performance parameter of detection equipment.

Specifically, after the detection and determination of the oil state of the gearbox are completed, in order to enable operation and maintenance personnel to better understand the reliability of the detection result, the confidence evaluation needs to be performed on the detection result. In actual operation, reference indexes for evaluation, such as repeatability of detection data, consistency with historical detection data, and the like, are determined first. Then, according to the specific requirements of the oil detection precision of the gearbox, corresponding weight coefficients are set for each reference index, and the higher the detection precision requirements are, the larger the weights of certain key indexes are. And then, calculating the confidence value of each reference index according to the stability of the gearbox oil and the performance parameters of the detection equipment. And finally, substituting the data into a preset evaluation formula, and calculating the confidence coefficient of the detection result. Through the confidence value, operation and maintenance personnel can intuitively judge the credibility of the detection result, so that a reasonable maintenance decision is made more scientifically according to the detection result, and the accuracy and the effectiveness of maintenance work are improved.

As shown in fig. 2, the oil state detection system of the gearbox of the wind power generation equipment comprises:

The oil liquid sample acquisition and pollution degree quantitative analysis unit is used for acquiring an oil liquid sample of the gearbox, and carrying out quantitative analysis on solid particle pollutants, liquid pollutants and gas pollutants in the oil liquid sample based on an oil liquid pollution degree analysis model to construct a multidimensional pollution factor matrix;

The physicochemical property parameter analysis and characteristic vector conversion unit is used for establishing a physicochemical property parameter space based on an oil physicochemical property analysis model for the kinematic viscosity, acid value, moisture content and oxidation stability parameters of the gearbox oil, and converting the actually measured parameter value into a corresponding characteristic vector;

the pollution-physical and chemical characteristic tensor fusion unit is used for tensor fusion of the multidimensional pollution factor matrix and the physical and chemical characteristic tensor to form a composite characteristic tensor containing pollution information and physical and chemical characteristic information;

The characteristic decoupling and sub-tensor separating unit is used for separating the pollution related characteristic and the physicochemical property related characteristic in the composite characteristic tensor based on a preset characteristic decoupling algorithm to obtain a pollution characteristic sub-tensor and a physicochemical property characteristic sub-tensor;

the multi-level feature extraction and high-order feature mining unit is used for respectively carrying out deep feature mining on the pollution feature tensor and the physicochemical property feature tensor and extracting high-order pollution features and high-order physicochemical property features;

The oil liquid state comprehensive judgment and result output unit is used for comprehensively judging the oil liquid state of the gear box through a preset decision rule based on the extracted high-order pollution characteristic and the high-order physicochemical property characteristic and outputting a detection result of the oil liquid state of the gear box;

The oil liquid sample acquisition and pollution degree quantitative analysis unit is connected with the first input end of the pollution-physical and chemical characteristic tensor fusion unit, the physical and chemical property parameter analysis and characteristic vector conversion unit is connected with the second input end of the pollution-physical and chemical characteristic tensor fusion unit, the output end of the pollution-physical and chemical characteristic tensor fusion unit is connected with the input end of the characteristic decoupling and sub-tensor separation unit, the two output ends of the characteristic decoupling and sub-tensor separation unit are respectively connected with the two input ends of the multi-level characteristic extraction and high-order characteristic excavation unit, and the output end of the multi-level characteristic extraction and high-order characteristic excavation unit is connected with the input end of the oil liquid state comprehensive judgment and result output unit.

According to the invention, the oil liquid state is accurately grasped through multidimensional depth analysis and innovative model construction. The advantages are described in terms of detection comprehensiveness, relationship analysis, and detection optimization.

Firstly, aiming at the problem of single traditional detection means, the detection method and the detection system realize multi-dimensional comprehensive detection. The method is not limited to detecting one or more indexes of the oil liquid, but quantitatively analyzes solid particle pollutants, liquid pollutants and gas pollutants in an oil liquid sample, and comprehensively considers a plurality of physicochemical property parameters such as kinematic viscosity, acid value, moisture content, oxidation stability and the like. By constructing a multidimensional pollution factor matrix and physicochemical characteristic vectors and fusing tensors, a composite characteristic tensor containing rich information is formed, the oil pollution and physicochemical characteristic states are covered comprehensively, erroneous judgment caused by detection of one side is avoided, and a more comprehensive and accurate data basis is provided for oil state evaluation of a gear box.

Secondly, on the basis of solving the problem that the prior art lacks in-depth analysis of the correlation between oil pollution and physicochemical properties, the method utilizes a characteristic decoupling algorithm to separate pollution-related characteristics and physicochemical property-related characteristics in the composite characteristic tensor, and then extracts high-order characteristics through a multi-level characteristic extraction network. Based on a state evaluation decision model combining a decision tree and a rule base, fully considering the interaction relationship of the decision tree and the rule base, and comprehensively judging the oil state of the gear box. The method breaks the limitation of traditional isolation analysis, can evaluate the actual state of oil liquid more accurately, discover potential problems in time, and effectively avoid the situation that maintenance treatment time is missed due to failure to consider interaction.

In addition, the detection method and the detection system are optimized in the links of sample acquisition and result evaluation. By carrying out confidence evaluation on the detection result, comprehensively considering reference indexes such as repeatability of detection data, consistency with historical data and the like, and by combining the stability of gearbox oil and the performance parameters of detection equipment, the reliability of the detection result is evaluated, the accuracy and the effectiveness of detection are further improved, and solid guarantee is provided for the stable operation of the wind power generation equipment.

In the description of the present invention, unless explicitly stated or limited otherwise, the terms "disposed," "mounted," "connected," and "secured" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected, mechanically connected, electrically connected, directly connected, indirectly connected via an intervening medium, or in communication between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art in a specific case.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (10)

1. The method for detecting the oil state of the gearbox of the wind power generation equipment is characterized by comprising the following steps of:

Step S1, acquiring a gearbox oil sample, quantitatively analyzing solid particle pollutants, liquid pollutants and gas pollutants in the oil sample based on an oil pollution degree analysis model, and performing feature extraction and coding on particle size distribution and concentration information of different types of pollutants by constructing a multidimensional pollution factor matrix;

S2, utilizing an oil physicochemical property analysis model to establish physicochemical property parameter space for kinematic viscosity, acid value, moisture content and oxidation stability parameters of the gearbox oil, and converting the actually measured parameter value into a corresponding characteristic vector in a nonlinear mapping mode;

Step S3, tensor fusion is carried out on the multidimensional pollution factor matrix obtained in the step S1 and the physicochemical characteristic vector obtained in the step S2, so as to form a composite characteristic tensor containing pollution information and physicochemical characteristic information;

S4, separating the pollution related features and the physicochemical property related features in the composite feature tensor by utilizing a preset feature decoupling algorithm based on the composite feature tensor to obtain a pollution feature tensor and a physicochemical property feature tensor;

s5, carrying out deep feature mining on the pollution feature tensor and the physicochemical property feature tensor respectively, and extracting high-order pollution features and high-order physicochemical property features by constructing a multi-level feature extraction network;

And S6, inputting the extracted high-order pollution characteristics and high-order physicochemical property characteristics into a state evaluation decision model, comprehensively judging the oil state of the gear box through a preset decision rule, and outputting a detection result of the oil state of the gear box.

2. The method for detecting the oil state of the gearbox of the wind power generation equipment according to claim 1, wherein in the step S1, a multidimensional pollution factor matrix constructed by an oil pollution degree analysis model is calculated by the following formula:

Wherein M _CF represents a multidimensional pollution factor matrix, n is the number of pollutant types, alpha _i is a weight coefficient of the i-th pollutant, the coefficient is determined according to an anti-pollution performance parameter of the gearbox oil, D _i is a particle size distribution vector of the i-th pollutant and comprises pollutant duty ratio information of different particle size intervals, C _i is a concentration scalar of the i-th pollutant, f _i is a characteristic coding function of the i-th pollutant, and a coding rule is adjusted according to a lubrication characteristic parameter of the gearbox oil.

3. The method for detecting the oil state of the gearbox of the wind power generation equipment according to claim 1, wherein in the step S2, a physicochemical property parameter space conversion formula established by an oil physicochemical property analysis model is as follows:

V_PP＝β·ReLU(γ·[V,A,W,O]^T)

Wherein V _PP represents a physical and chemical property characteristic vector after conversion, beta is a scaling matrix, element values are set according to specification parameters of gear box oil, gamma is a weight matrix and is determined according to performance standard parameters of the gear box oil, V is a motion viscosity measured value of the gear box oil, A is an acid value measured value, W is a moisture content measured value, O is an oxidation stability index measured value, and ReLU is a linear rectification activation function for enhancing nonlinear expression capability of the characteristic vector.

4. The method for detecting the oil state of the gearbox of the wind power generation equipment according to claim 1, wherein in the step S3, a tensor fusion process adopts a mode of combining tensor product and weighted summation, and a specific fusion formula is as follows:

Wherein T _CFP represents a composite characteristic tensor, m is the dimension of the physicochemical characteristic vector, omega _j is the weight coefficient of the j-th physicochemical characteristic vector in the fusion process, the coefficient is determined by the using working condition parameters of gearbox oil, and V _PPj is the j-th component of the physicochemical characteristic vector V _PP; representing tensor product operations.

5. The method for detecting the oil state of the gearbox of the wind power generation equipment according to claim 1, wherein in the step S4, the characteristic decoupling algorithm decomposes the composite characteristic tensor based on the tensor decomposition principle by the following formula:

wherein T _CF represents a pollution characteristic tensor, T _PP represents a physicochemical characteristic tensor; E is decomposing an error tensor, minimizing the norm of the error tensor through an iterative optimization algorithm, and adjusting parameters in the optimization process according to the stability parameters of the gearbox oil.

6. The method for detecting the oil state of the gearbox of the wind power generation equipment according to claim 1, wherein in the step S5, the multi-level feature extraction network comprises a plurality of feature extraction layers, and a feature extraction formula of each layer is as follows:

F_l＝σ(θ_l·F_l-1+b_l)

Wherein F _l is the characteristic tensor extracted by the first layer, F _l-1 is the output characteristic tensor of the first-1 layer, theta _l is the weight tensor of the first layer, the parameter value is initialized according to the abrasion characteristic parameter of the gear box oil, b _l is the bias tensor of the first layer, sigma is an activation function, and a proper activation function type is selected according to the performance change trend of the gear box oil.

7. The method for detecting the oil state of the gearbox of the wind power generation equipment according to claim 1, wherein in the step S6, the state evaluation decision model adopts a mode of combining a decision tree with a rule base, and the decision process is performed by the following rules:

If the oil liquid state of the gear box is H _CF>τ_Cf and H _PP>τ_PP, judging that the oil liquid state of the gear box is seriously abnormal;

If the oil liquid state of the gear box is H _CF>τ_CF and H _PP≤τ_PP, judging that the oil liquid state of the gear box is pollution abnormality;

If the oil liquid state of the gear box is H _CF≤τ_CF and H _PP>τ_PP, judging that the oil liquid state of the gear box is abnormal in physicochemical property;

If the oil level is H _CF≤τ_CF and H _PP≤τ_PP, judging that the oil level of the gear box is normal;

The method comprises the steps of extracting high-order pollution characteristics from gearbox oil, wherein H _CF is a comprehensive measurement value of the extracted high-order pollution characteristics and is calculated according to a cleanliness standard parameter of the gearbox oil, H _PP is a comprehensive measurement value of high-order physicochemical characteristics and is determined according to a quality standard parameter of the gearbox oil, tau _CF and tau _PP are respectively judging thresholds of the pollution characteristics and the physicochemical characteristics, and the thresholds are set by carrying out statistical analysis on historical detection data of a large number of gearbox oil samples and combining with performance limit parameters of the gearbox oil.

8. The method for detecting the oil state of a gearbox of a wind power generation device according to claim 1, further comprising the step of determining the collection position of the oil sample of the gearbox before the step S1, wherein the collection position is determined by the following method:

According to the distribution characteristics of the internal flow field of the gearbox, combining the liquidity parameters of the gearbox oil, establishing an oil sample acquisition position optimization model, and determining an optimal acquisition position by calculating representative indexes R _I of oil samples at different positions, wherein the calculation formula is as follows:

The method comprises the steps of obtaining a number of candidate collecting positions, wherein p is the number of the candidate collecting positions, delta _k is a weight coefficient of the j-th candidate collecting position, the weight coefficient is determined by structural parameters of a gear box, corr (S _k,S_ref _is a correlation measurement value of an oil sample of the k-th candidate collecting position and a reference sample, the reference sample is constructed according to standard performance parameters of gear box oil, and a position with the largest representative index is selected as the oil sample collecting position by comparing representative indexes R _I of the candidate collecting positions.

9. The method for detecting the oil state of a gearbox of a wind power generation device according to claim 1, further comprising a confidence evaluation step of the detection result after step S6, wherein the confidence evaluation is calculated by the following formula:

The method comprises the steps of detecting a gear box oil, wherein C _E represents the confidence coefficient of the detection result, R is the number of reference indexes used for evaluating the confidence coefficient, lambda _q is the weight coefficient of the q-th reference index, the coefficient is set according to the detection precision requirement parameter of the gear box oil, conf (R _q) is the confidence coefficient value corresponding to the q-th reference index, the reference index comprises the repeatability of detection data and the consistency with historical detection data, and the confidence coefficient value is calculated according to the stability parameter of the gear box oil and the performance parameter of detection equipment.

10. Wind power generation equipment gear box fluid state detecting system, its characterized in that includes:

The oil liquid sample acquisition and pollution degree quantitative analysis unit is used for acquiring an oil liquid sample of the gearbox, and carrying out quantitative analysis on solid particle pollutants, liquid pollutants and gas pollutants in the oil liquid sample based on an oil liquid pollution degree analysis model to construct a multidimensional pollution factor matrix;

The physicochemical property parameter analysis and characteristic vector conversion unit is used for establishing a physicochemical property parameter space based on an oil physicochemical property analysis model for the kinematic viscosity, acid value, moisture content and oxidation stability parameters of the gearbox oil, and converting the actually measured parameter value into a corresponding characteristic vector;

the pollution-physical and chemical characteristic tensor fusion unit is used for tensor fusion of the multidimensional pollution factor matrix and the physical and chemical characteristic tensor to form a composite characteristic tensor containing pollution information and physical and chemical characteristic information;

The characteristic decoupling and sub-tensor separating unit is used for separating the pollution related characteristic and the physicochemical property related characteristic in the composite characteristic tensor based on a preset characteristic decoupling algorithm to obtain a pollution characteristic sub-tensor and a physicochemical property characteristic sub-tensor;

the multi-level feature extraction and high-order feature mining unit is used for respectively carrying out deep feature mining on the pollution feature tensor and the physicochemical property feature tensor and extracting high-order pollution features and high-order physicochemical property features;

And the oil liquid state comprehensive judgment and result output unit is used for comprehensively judging the oil liquid state of the gear box through a preset decision rule based on the extracted high-order pollution characteristic and the high-order physicochemical property characteristic and outputting the detection result of the oil liquid state of the gear box.