TWM592123U - Intelligent system for inferring system or product quality abnormality - Google Patents

Abstract

An intelligent system for inferring system or product quality abnormality is provided. The system operates failure modes and effect analysis in a client system. A machine-learning algorithm is performed for collecting data and importing a database that records text reports, experts’ articles and various data leading the system abnormality. The data then undergoes a data-mining process for deleting the data not suitable for modeling and transforming unstructured data into structured data. A big-data analysis is performed on the data for learning correlations among the various data. An abnormality inference model is established. The model is used to infer abnormality of system. The abnormality inference model can be optimized through verification while receiving feedback data.

Description

Inference system or intelligent system with abnormal product quality

This work is about a technique to infer abnormal quality of a system or product, especially refers to a system that uses machine learning methods to infer an abnormal phenomenon of a system or product quality through intelligent methods such as data collection, text exploration and algorithms.

With the advancement of science and technology, the manufacturing process of an industrial product also becomes more complex and has more functions and a larger number of components as the product complexity becomes higher, making a manufacturing system more complicated. Therefore, when there is a problem with the product, find the system When it comes to sexual issues, it will be difficult to implement because of too much detail.

When the system is faced with failure, the conventional technology proposes the concept of Failure Modes and Effects Analysis (FMEA). FMEA is a method to gradually identify possible errors in the system and can be applied to the product manufacturing process or service process. , Used to check problems that may cause system failure. One of the conventional methods is to debug various links of the system in a tabular and documented manner. When a problem occurs, you can use the table lookup method to determine which link may be wrong.

However, the traditional FMEA still faces many pain points. When new errors or factors arise, such error elimination methods in the form of tables or documents need to be revised at any time, and rely on human interpretation and cognition of the text semantics. Everyone Interpretation of text semantics is different from cognition, and it is difficult to sum up a clear standard in evaluating numbers, so there is no standard and inefficiency; plus, as the complexity of the system increases, the complexity of multiple functions and a large number of components For products, after the system is decomposed, too many details are considered, and it is difficult to execute. If there are more complicated factors that cause the system to fail, multiple failure modes acting at the same time or affecting each other are difficult to analyze using tables or files.

Furthermore, when FMEA is applied in an enterprise, the enterprise may have problems such as inadequate analysis, incomplete measures, inaccurate risk assessment, and difficulty in teamwork during the implementation; the implementer may also implement it as a task and cannot produce actual benefits; the enterprise The ability level and knowledge habits of the teams participating in FMEA are different. As a result, the management of knowledge and risk cannot be unified with the fluctuation of personnel, ability and other factors, standardized data cannot be formed, and conventional FMEA will no longer meet the standard requirements.

According to the embodiments disclosed in the specification, an inference system implemented by a computer system or an intelligent system with abnormal product quality is proposed. One of its purposes is to propose an intelligent method using machine learning in view of the failure of traditional failure and impact judgment methods (such as FMEA). Methods for automatic induction, analysis, and elimination of failure effects.

The inference system or the intelligent system with abnormal product quality implements the above inference system or the intelligent method with abnormal product quality, which includes an algorithm module implemented by software or hardware, which includes a variety of machine learning algorithms, a machine The learning module analyzes the data provided by the client system with one of the machine learning algorithms, and learns and trains through a neural network, a model building module, and creates an abnormal inference model based on the training results of the machine learning module , And verify each abnormal inference model established by each machine learning algorithm, and then select a better machine learning algorithm, and perform evaluation and optimization of abnormal inference model, and a database module for searching data from the client system, Establish a knowledge base.

The methods implemented in the system mainly run on a client system that performs failure mode and impact analysis. In the system, a machine learning algorithm is run, including collecting data, and importing a knowledge base, and then collecting the collected data and knowledge The content of the library is prospected for text, and big data analysis can be performed on the data in the knowledge base to learn the relevance of the data and establish an abnormal inference model. After that, when the system receives new data, it inputs the abnormal inference model, performs abnormal inference, and outputs an inference result, and can verify and optimize the abnormal inference model after receiving the feedback message.

Among them, preferably, the knowledge base includes text reports (which may include quality inference reports, quality management analysis, problem solving and countermeasure documents, 8D documents (8D problem solving method (Eight Disciplines Problem Solving)), CAR documents (corrective action requirements (Corrective action request)), expert articles, existing FMEA documents, all kinds of information that affect the abnormal quality of the system and products, as well as system operation information and environmental information.

Further, the collected data further includes failure data obtained by inductive analysis of a failure mode effect and critical analysis method.

Further, the steps of text exploration of the obtained data or file description include the steps of screening and excluding data that is not favorable for the establishment of an abnormal inference model, and processing unstructured data into structured data, and may include The data is vocabulary unified to establish a vocabulary and select data as training samples to build abnormal inference models through training samples.

Preferably, when a plurality of machine learning algorithms are provided to select a machine learning algorithm to run the method, the method of selecting a machine learning algorithm includes: separately establishing individual abnormal inference models with multiple machine learning algorithms, and then The anomaly inference model is scored to select one of the machine learning algorithms.

Further, in order to verify the abnormal inference model based on the feedback information and the actual failure status, the parameters can be modified when needed, and a comparison table can be generated for speculating the system abnormality.

Preferably, the comparison table is used for comparison of abnormality solutions, in which various abnormality information is recorded, including severity, abnormality frequency, and ratio of abnormality detected.

In order to further understand the characteristics and technical content of this creation, please refer to the following detailed description and drawings of this creation. However, the drawings provided are for reference and explanation only, and are not intended to limit this creation.

The following are specific specific examples to illustrate the implementation of this creation, and those skilled in the art can understand the advantages and effects of this creation from the content disclosed in this specification. This creation can be implemented or applied through other different specific embodiments. The details in this specification can also be based on different views and applications, and various modifications and changes can be made without departing from the concept of this creation. In addition, the drawings in this creation are only a schematic illustration, not based on actual size, and are declared in advance. The following embodiments will further describe the relevant technical content of the creation, but the disclosed content is not intended to limit the protection scope of the creation.

It should be understood that although terms such as “first”, “second”, and “third” may be used herein to describe various elements or signals, these elements or signals should not be limited by these terms. These terms are mainly used to distinguish one component from another component, or one signal from another signal. In addition, the term "or" as used herein may include any combination of any one or more of the associated listed items, depending on the actual situation.

The manual proposes an inference system or an intelligent system of abnormal product quality, especially an intelligent system implemented by a computer system. This is a system that is suitable for the establishment of a knowledge base and needs to predict abnormal events. It is an intelligent method that uses machine learning to automatically Summarize, analyze, and eliminate the effects of failure. Such systems may generally have the ability to cope with general abnormal events, such as finding out solutions to abnormal events by looking up tables against special circumstances, or relying on experts in related fields to make abnormal judgments, but often still encounter problems because of The complexity of the system leads to the problem of difficult judgment, and the problems of inadequate analysis, incomplete measures, and inaccurate evaluation, plus the difficulty of handover because of the excessive reliance on experts. Therefore, the proposed inference system or intelligent system with abnormal product quality introduces machine learning technology of artificial intelligence (AI) into the knowledge base, and the ability to process large amounts of data through the computer system is constructed to pre-infer the abnormal quality of the system or product. Knowledge base and model, and at any time during the operation process, effectively optimize the knowledge base and modify the model based on the feedback information, to reduce errors caused by human factors, perform automatic and accurate inference of system or product quality abnormalities, and achieve the purpose of strengthening the prevention of system abnormalities in advance .

In order to import the client (such as various enterprises and factories) to intelligently conduct failure mode judgment, post-failure impact assessment, and search for the cause of failure, in order to establish a model of inference system or product quality abnormality, so that subsequent processing and build a knowledge base, The inference system or the method of abnormal product quality is introduced into an artificial intelligence engine, and a customized abnormal inference model is established by machine learning. Refer to FIG. 1 for a schematic diagram of an embodiment of an inference system or an intelligent system architecture of abnormal product quality.

The figure shows an inference system or an intelligent system 10 with abnormal product quality, which includes a plurality of artificial intelligence-related function modules implemented by software programs and hardware data processing capabilities, such as an algorithm module 101, in which a variety of Machine learning algorithms, combined with the machine learning technology implemented by the machine learning module 103, analyzes the data provided by the client system (customer one 111, customer two 112, and customer three 113) with one of the machine learning algorithms , And through neural network-like learning and training, the model building module 105 builds an abnormal inference model according to the training results of the machine learning module 103, and also verifies the abnormal inference models created by the machine learning algorithms. Afterwards, different algorithms can be used to verify the results of inference, select a better algorithm, and perform evaluation and optimization of abnormal inference models.

The inference system or the intelligent system 10 with abnormal product quality connects the knowledge base of each client through the network or a specific method, such as customer 111 (knowledge 115), customer 112 (knowledge 116), and customer three 113 (knowledge) Library 117), the inference system or the intelligent system 10 of abnormal product quality collects data from the client system through the database module 107, and establishes a system-side knowledge base, which can perform abnormal reasoning such as failure mode and impact analysis for different customers.

FIG. 2 shows a schematic diagram of an embodiment of an application inference system or an intelligent system with abnormal product quality. The inference system or the intelligent system with abnormal product quality is provided with a knowledge base 20. The items recorded therein include at least expert knowledge 201, historical record 202, abnormal factor 203 and comparison table 204, and the inference system or product quality executed in the system is abnormal In the intelligent method, the main steps include illustrated data collection 21, text exploration 22, index score 23, abnormal diagnosis 24 and optimization tracking 25.

Among them, the knowledge base 20 records information used for system abnormality judgment, such as expert knowledge 201, which refers to expert knowledge in the relevant field of the applied system, especially knowledge about how to eliminate system abnormalities. Sources include experts inside and outside the enterprise. Various unstructured text reports and articles imported into the system. The historical record 202 records records of past abnormal events of the system and records of how to resolve the abnormal events, and becomes the main data source for constructing abnormal inference models of inference systems or intelligent systems of abnormal product quality. The anomaly factor 203 records the relevant factors of the occurrence of a specific anomaly event and is used to control the anomaly solution. The knowledge base 20 derives a comparison table 204 of system anomalies and solutions, which records various anomaly information, including severity, frequency of anomalies, and the ratio of detected anomalies.

In the intelligent method of inference system or product quality abnormality realized by knowledge base 20, the goal of data collection 21 is to reduce the difficulty of data collection, but also to strengthen the ability to collect and clean up system data (such as process data), except for the source of knowledge In addition to the library 20, it also includes machine networking, MES, ERP, customer complaints, etc.

The step of text exploration 22 is to solve the problem of text cognition. The goal is to reduce the burden of staff's text work analysis and reduce inconsistencies caused by text cognition problems, that is, to be able to standardize some terms to facilitate operation. In one embodiment, a Support-Vector Machine (SVM) text analysis can be used to structure unstructured text into keywords to facilitate the establishment of an abnormal inference model.

The step of index score 23 is to solve the problem that the score difference is too large due to the subjective judgment of different personnel, and there is no objective score standard. According to an embodiment, a fuzzy decision system (Fuzzy Decision System), decision tree analysis, Bayesian can be used Sinet and a FTA-SVM (Fault Tree Analysis-Support Vector Machine) algorithm can successfully score SOD (severity-occurrence-detection). Among them, Bayesian network and FTA-SVM algorithm are used to build an intelligent failure diagnosis model, which is used to train a prediction model with high accuracy and strong self-adjusting ability.

In the step of anomaly diagnosis 24, to quickly analyze the causes of customer complaints online, and to propose steps to improve strategies to strengthen the ability to resolve failures.

The optimization tracking 25 is one of the technical purposes of the inference system or the intelligent system with abnormal product quality. It is used to optimize the knowledge base 20. The optimization goal can be to automatically optimize the traditional failure mode and impact analysis (FMEA) indicators and optimize the risk priority number (Risk Priority Number (RPN) or Action Priority (AP), which can dynamically analyze the action priority of various failure modes and continuously improve product quality.

It is worth mentioning that the above-mentioned 22 steps of text exploration are methods for processing text data, and the related process can refer to the text exploration process in the intelligent method applied to the inference system or product quality abnormality shown in FIG. 3.

At the beginning, in step S301, the failure data obtained by the analysis of a failure mode, effects and criticality analysis (FMECA) was collected. FMECA was added in addition to the failure mode and effects analysis (FMEA) For critical analysis, the probability of each failure mode corresponding to the consequences of different severity is listed to highlight the failure modes with higher probability and serious consequences, so that the remedial action of the failure mode can have the greatest effect.

Then, in step S303, the data collected that is not conducive to model establishment is filtered out and the data set for model training is removed, and in step S305, the unstructured data that is sorted is processed to be used as described Structured data for SVM text analysis. In step S307, the software program is used to select part of the pre-processed data as the training sample, and the remaining part is the test sample. Then, in step S309, the prediction model is established through the training sample, that is, the system or product quality is abnormal. Abnormal inference model in intelligence, and then make the data predict the output failure mode through this abnormal inference model.

Next, in step S311, the established abnormal inference model is verified. One of the methods is to verify the accuracy of the abnormal inference model by testing the data of the sample. If necessary, the best model can be adjusted by modifying the parameters. After that, relevant reports about system inferences are generated.

When running the intelligent method of the inference system or product quality anomaly, one of its characteristics is to verify various machine learning algorithms, where the operation of the algorithm has the same as shown in Figure 4 for the purpose of inferring the system or product quality anomaly. Main steps.

Choose one of the multiple algorithms provided by the system, such as step S401, execute the machine learning algorithm through the hardware processing circuit, and at the same time step S403, import each customer's knowledge base, through big data analysis, learn to obtain a tightly linked and repeated operation The correlation between the input (such as knowledge base, existing data, etc.) and the output data (such as inference results, customer feedback, user feedback, etc.), as in step S405, an abnormal inference model is established by learning the correlation between various data.

In step S407, the system receives the new data, and also inputs the abnormal inference model, as in step S409, performs abnormal inference, outputs the inference result (step S411), and receives the feedback message (step S413), then as in step S415, the system will The output results verify the anomaly inference model obtained by the machine learning algorithm selected this time. The verification method can already verify the anomaly inference model with the historical data of the system anomaly records, and can continue to optimize the anomaly inference model through correction. During the optimization process, various machine learning algorithms will learn about the anomalies mentioned in the feedback messages, and establish a correlation with the data, thereby highlighting the past known but minor factors, as well as the past unknown information related to system anomalies , To optimize the abnormal inference model.

For the machine learning calculation process, refer to FIG. 5 for a flowchart of an embodiment of an algorithm used in an inference system or an intelligent system with abnormal product quality.

At the beginning, as in step S501, the inference system or the intelligent system with abnormal product quality collects client data, including the client system's knowledge base, which can include text, audiovisual content, and records, including various text reports (including quality inference reports) , Quality management analysis, problem solving and countermeasure documents, 8D documents (8D problem solving method (Eight Disciplines Problem Solving)), CAR documents (corrective action request), expert articles, existing FMEA documents affecting the system Various information related to abnormal product quality, system operation information, environmental information, etc. Next, in step S503, to perform text exploration on the collected content, refer to the above embodiment. One of its main purposes is to perform text exploration and vocabulary unification on the received data, and establish a vocabulary database for each system.

Then, based on the results of the text exploration, multiple algorithms are executed, such as Algorithm One (Step S505), Algorithm Two (Step S506), and Algorithm Three (Step S507) in the illustrated process. For example, various machine learning methods (steps S508, S509, and S510) corresponding to various algorithms use neural-like networks to obtain the correlation between system abnormal data and various input data to establish an inference model of system abnormalities, that is, respectively The abnormal inference model one (step S511), the abnormal inference model two (step S512), and the abnormal inference model three (step S513) are established corresponding to each algorithm.

Next, in step S515, the system scores each abnormal inference model, and finally selects the algorithm (step S517). According to an embodiment, the scoring criteria are mainly for judging the machine learning algorithms suitable for a system based on the inference results generated by the abnormal inference model, such as classification metrics and regression metrics used by conventional technologies. .

After the selection of the algorithm is completed, a flowchart of one embodiment of the establishment of the abnormal inference model shown in FIG. 6 begins.

In step S601, the data is initially received, and the system-side knowledge base is established based on the client data or the existing knowledge base. In step S603, the historical data is sorted, including deleting data that is not conducive to model building, and unstructured data Process into structured data that can be easily analyzed. In step S605, you can start to learn the correlation in the data with the selected machine learning algorithm, analyze the data through the algorithm, learn the data features with a large amount of data and the algorithm, and establish an abnormal inference model. Among them, you can use a training Custom machine learning model (AutoML), which can automatically construct and deploy machine learning models on structured data for a given problem (marked data).

After that, in step S607, the generated abnormal inference model is verified, and it is evaluated whether the abnormal event of the system is accurately inferred. If the evaluation result is successful, that is, in step S609, the generated model is stored; otherwise, it is necessary to further adjust according to the verification result To calibrate the model parameters, such as step S611, when necessary, the steps of calibrating the model should be repeated.

After completing the establishment of the abnormal inference model, as shown in FIG. 7, data import (step S701) and clearing data (step S703) will begin, as described in the above embodiment, including deleting data that is not conducive to model establishment, and non-structural Process data into structured data that can be easily analyzed. Next is to perform text preprocessing (step S705) and data preprocessing (step S707), perform the adoption, classify the sample (step S709), input the abnormal inference model (step S711), and repeat steps S709 and S711 as needed, The abnormal inference model should be verified based on the feedback information and the actual failure status, and the parameters should be modified when necessary (step S713), before the comparison table finally used to speculate the system abnormality is generated (step S715).

According to the embodiment, the abnormal inference model established through the above steps can be used to predict the output failure mode. At the beginning of the operation of the client system, the client system should be understood, including the need to define a system failure item and confirm the failure factor of a system With related projects, it can analyze the potential failure mode based on each functional requirement in the system, and establish a database based on the causes and impact consequences of the failure mode. The next step is to collect historical experience and lessons, including test results, analysis data, and after-sales data for specific products to assist in the analysis of potential failure modes.

Regarding the evaluation of the inference model, this can be combined with the effect of failure to assess the severity and severity of the failure, as well as the frequency of occurrence of current preventive measures and the detection of current detection measures.

Finally, optimize the abnormal inference model, for example, take a series of improvement measures, re-evaluate the occurrence frequency and detection degree according to the implementation of the measures, thereby reducing the overall risk, and continuously optimize the loop until the RPN (AP) reaches an acceptable range.

In summary, according to the description, an embodiment of an inference system or an intelligent system with abnormal product quality is described. In view of the failure of the traditional failure and impact judgment methods, the intelligent method of machine learning is used to automatically summarize, analyze, and eliminate the effects of failure. The method, which combines the ability of text exploration and data exploration, and the knowledge established by the experience and creativity of people (such as experts), uses artificial intelligence machine learning technology to repeatedly verify and correct abnormal inference models to achieve the goal of preventing system abnormalities in advance.

The content disclosed above is only a preferred and feasible embodiment of the new model, and does not limit the scope of the patent application of the new model. Therefore, any equivalent technical changes made by using the description and drawings of the new model are included in the application of the new model. Within the scope of the patent.

10: Inference system or intelligent system with abnormal product quality 101: Algorithm module 103: Machine Learning Module 105: Model building module 107: Database module 111: Customer One 115: Knowledge Base 112: Customer 2 116: Knowledge Base 113: Customer Three 117: Knowledge Base 20: Knowledge base 201: Expert knowledge 202: History 203: Abnormal factor 204: comparison table 21: Data collection 22: Text exploration 23: Index score 24: Abnormal diagnosis 25: Optimized tracking Steps S301 to S311: a text exploration process applied to the inference system or product quality abnormality method Steps S401～S415: algorithm flow Steps S501-S517: flow for selecting algorithm Steps S601～S611: the process of establishing abnormal inference model Steps S701～S715: The process of running abnormal inference after importing data into artificial intelligence

FIG. 1 shows a schematic diagram of an embodiment of an intelligent system architecture that infers an abnormal quality of a system or product;

2 shows a schematic diagram of an embodiment of an inference system or an intelligent system with abnormal product quality;

Figure 3 shows the flow chart of the text exploration method applied to the method running in the inference system or the intelligent system with abnormal product quality;

FIG. 4 is a diagram showing an embodiment of an algorithm process running on an inference system or an abnormal purpose of product quality;

FIG. 5 shows a flowchart of an embodiment of an algorithm selected for use in an inference system or an intelligent system with abnormal product quality;

6 shows a flowchart of an embodiment of establishing an abnormal inference model in an inference system or an intelligent system with abnormal product quality; and

7 shows a flowchart of an embodiment of abnormal inference after data is imported into artificial intelligence.

10: Inference system or intelligent system with abnormal product quality

101: Algorithm module

103: Machine Learning Module

105: Model building module

107: Database module

111: Customer One

115: Knowledge Base

112: Customer 2

116: Knowledge Base

113: Customer Three

117: Knowledge Base

Claims (10)

An inference system or intelligent system with abnormal product quality, including: An algorithm module, which has a variety of machine learning algorithms; A machine learning module, using one of the machine learning algorithms to analyze the data provided by a client system, and learn and train through a neural network-like; A model building module, creating an abnormal inference model based on the training results of the machine learning module; and A database module, the inference system or the intelligent system with abnormal product quality searches the data from the client system through the database module to establish a knowledge base; The intelligent system in which the inference system or product quality is abnormal performs an intelligent method in which the inference system or product quality is abnormal, including: Collect the data with the selected machine learning algorithm and import it into the knowledge base; Perform text exploration on the collected data and the knowledge base; Perform big data analysis on the data after text exploration, learn the correlation of the data, and establish the abnormal inference model; When the system receives new data, input the abnormal inference model, perform abnormal inference, and output an inference result; and After receiving the feedback message, the abnormal inference model obtained by the machine learning algorithm is verified to optimize the abnormal inference model. The inference system or the intelligent system of abnormal product quality as described in claim 1, wherein the knowledge base includes text reports, expert articles, FMEA documents, various information that affects the abnormality of the system, and information on the operation of the system and environmental information. The inference system or the intelligent system with abnormal product quality as described in claim 2, wherein the items recorded in the knowledge base further include a comparison table. The inference system or the intelligent system with abnormal product quality as described in claim 3, wherein the abnormal inference model is verified based on the feedback information and the actual failure status, and the parameters are modified as necessary to generate the Chart. The inference system or the intelligent system of abnormal product quality as described in claim 4, wherein the comparison table is used to control the abnormal solution, which records various abnormal information, including the severity, the abnormal frequency and the proportion of detected abnormalities . The inference system or the intelligent system with abnormal product quality as described in claim 3, wherein the collected data further includes failure data obtained by induction analysis by a failure mode effect and criticality analysis method. The inference system or the intelligent system with abnormal product quality as described in claim 1, wherein the model building module further verifies the abnormal inference models created by each machine learning algorithm, and then selects a better machine learning algorithm from it, and executes Scoring and optimizing the abnormal inference model. The inference system or the intelligent system with abnormal product quality as described in claim 7, wherein the scoring uses a classification index and a regression index to judge the inference result generated by each abnormal inference model to be suitable for the machine of the client system applied Learning algorithms. The inference system or the intelligent system with abnormal product quality as described in claim 7, wherein the purpose of optimizing the abnormal inference model is to optimize FMEA indicators, including optimizing risk priority or action priority to an acceptable range. The inference system or the intelligent system with abnormal product quality as described in any one of claims 1 to 9, wherein in the intelligent system with the inferential system or product quality abnormality, the step of performing text exploration on the acquired data includes: Screen and remove data that is not favorable for the establishment of the abnormal inference model; Process unstructured data into structured data; Unify the vocabulary of the received data and establish a vocabulary; and Select the data as training samples to build the abnormal inference model through the training samples.

Images