A machine learning-based process procedure quality fluctuation source identification method and device

Abstract

The application discloses a kind of process quality fluctuation source identification method and device based on machine learning, the method includes: obtaining the quality characteristic measurement data of process or part, establish abnormal mode matrix;Distinguish whether abnormal reason is accidental factor or system factor;If abnormal reason is system factor, then process quality fluctuation source identification is carried out, and fuzzy relation matrix is established;Possibility matrix of abnormal occurrence is constructed, each element in it is merged and processed, and abnormal reason diagnosis is carried out according to the size of credibility value;Potential quality risk of each item of "man-machine-material-method-ring" in production process is set process risk index, and the potential quality risk coefficient of process is calculated, to determine the total risk coefficient of process, to evaluate process risk;Read the process quality factor change information in process quality BOM, output display process quality fluctuation source identification result set by data integration mode.

Description

Technical Field

[0001] This invention relates to the field of quality diagnosis technology for aerospace product manufacturing processes, specifically to a method and apparatus for identifying sources of quality fluctuations in processes based on machine learning. Background Technology

[0002] The manufacturing of aerospace products is characterized by a wide variety of technologies, complex assembly structures, and discrete processes. During component production, if product quality abnormalities occur, only a small portion of the time is spent on quality monitoring and control, while 80% of the time is devoted to analyzing the source of the abnormalities, often consuming significant manpower and time to trace the root causes.

[0003] Traditional methods for analyzing quality fluctuations categorize the factors causing these fluctuations using the 5M1H framework, then employ cause-and-effect diagrams, failure counts, and other quality management tools for causal analysis, resulting in a set of factors contributing to the fluctuations. However, due to the large number of influencing factors and the complex hierarchical relationships between them, it is difficult to accurately identify key fluctuation sources. Furthermore, most current quality diagnostic studies are reactive, meaning that quality problems are only diagnosed after they have already occurred in the aerospace products. The aim is to eliminate these anomalies to ensure the next production process, which not only wastes significant costs but also impacts production efficiency.

[0004] In recent years, with the development of computer technology, expert systems, fuzzy neural networks and other technologies have been gradually introduced into process fault diagnosis. However, there are still problems such as knowledge acquisition "bottleneck", rule reasoning "combinatorial explosion", small scope of application and single diagnostic technology. At the same time, it is difficult to meet the real-time requirements of quality diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for identifying sources of process quality fluctuations based on machine learning. It combines neural networks, dynamic risk quantification assessment of processes, and process quality BOM technology to achieve online diagnosis and dynamic risk warning of sources of process quality fluctuations (man, machine, material, method, environment). To achieve the above objectives, the present invention employs the following technical solution: A machine learning-based method for identifying sources of quality fluctuations in a process includes: Acquire quality characteristic measurement data of processes or parts, identify control chart patterns of quality characteristics, and establish an abnormal pattern matrix if the pattern is identified as an abnormal control chart pattern. For abnormal control chart patterns other than the cyclical pattern, trend analysis is used to distinguish whether the cause of the abnormality is due to accidental factors or systemic factors. If the abnormal cause is a system factor, identify the sources of quality fluctuations in the process. Use the fuzzy statistics method or the expert experience given method to determine the membership degree between the abnormal pattern of the control chart and the abnormal cause. Finally, establish a fuzzy relation matrix. Based on the abnormal pattern matrix and the fuzzy relation matrix, construct a matrix of the probability of abnormality. Merge the elements in the probability matrix and diagnose the abnormal cause according to the size of the credibility value. Set process risk indicators for the potential quality risks of each item of "man, machine, material, method, and environment" in the production process and calculate the potential quality risk coefficient of the process, so as to determine the total process risk coefficient for evaluating the process risk. Read the process quality factor change information in the process quality BOM. The data acquisition condition is the latest change information before the production of the diagnosed process / part. Through data integration, output and display the result set of the identification of the sources of process quality fluctuations. The result set includes the diagnosis result and credibility value of the abnormal cause, the potential quality risk coefficient, the process risk level, the change of process quality factors, and preventive measures.

[0006] Furthermore, use the control chart pattern recognition method based on machine learning to recognize the control chart pattern of quality characteristics. The abnormal patterns of the control chart include upward trend pattern, downward trend pattern, upward step pattern, downward step pattern, and periodic pattern.

[0007] Furthermore, for the remaining abnormal patterns of the control chart except the periodic pattern, use the trend analysis method to distinguish whether the abnormal cause is a random factor or a system factor, including: First, obtain the quality characteristic observation curve in the control chart and calculate the curve slope: ; where, E(t n ) represents the quality characteristic measurement data caused by the abnormal cause at time t n ; k n represents the curve slope at time t n ; t m represents any moment when the process is in the stable state of the normal mode of the control chart, and E(t m ) is the quality characteristic measurement data at time t m ; When the process is in the stable state of the normal mode of the control chart, the maximum standard slope of the quality characteristic observation curve is k0. If the curve slope k n <k0, that is, the abnormal cause at time t n is a random factor, otherwise it is a system factor.

[0008] Furthermore, determine the membership degree between the abnormal pattern of the control chart and the abnormal cause, and finally establish a fuzzy relation matrix, including: Suppose we take a fixed element u0 on a universe of discourse U, where u0 represents the cause of the anomaly; consider a general set A of motion with variable boundaries in the universe of discourse U. * A * This represents the types of abnormal patterns in the control chart; after n trials, the calculation is as follows: ; Where η is the degree of membership of u0 to A; Establish a fuzzy relation matrix Q between the abnormal patterns in control charts and all abnormal causes under a certain abnormal cause category; the elements Q in the fuzzy relation matrix Q are... ij η represents the probability that the j-th abnormal cause will cause the i-th control chart abnormal pattern, that is, the degree of membership between the i-th control chart abnormal pattern and the j-th abnormal cause.

[0009] Furthermore, based on the anomaly pattern matrix and the fuzzy relation matrix, an anomaly occurrence probability matrix is ​​constructed. The elements in the probability matrix are merged, and anomaly cause diagnosis is performed based on the confidence value, including: Integrating the fuzzy relation matrix Q and the anomaly pattern matrix M yields the anomaly occurrence probability matrix T; the i-th element T in matrix T... i The value represents the confidence level of the cause of the anomaly. After calculating the probability matrix T of anomalies, each element T in the matrix is... i Perform a merge operation, dividing [0,1] into different intervals, and assigning a confidence value to each interval; then, perform a merge operation on each element T in the matrix. i The confidence level is represented by the corresponding confidence value based on the interval it falls into; Establish a correspondence table between confidence values ​​and probabilities, and include non-zero T values ​​in the merged matrix. i The value is compared with the corresponding relationship table; non-zero T i The order of values ​​in the probability matrix T corresponds to the sequence number of the cause of the anomaly, and the probability of the cause of the anomaly is determined from the correspondence table.

[0010] Furthermore, process risk indicators are set for the potential quality risks of each component of the "man, machine, material, method, and environment" in the production process, and the potential quality risk coefficient of the process is calculated to determine the total process risk coefficient, which is used to assess the process risk, including: By combining process risk indicators, the dynamic risk coefficients of each sub-item of "man, machine, material, method, and environment" are calculated and arranged in order of numerical value to obtain the potential quality risk coefficient υ. Based on the preset weight vectors of personnel, machines, materials, methods, and environment, the total risk coefficient σ of the process is calculated. Based on the magnitude of the total risk coefficient σ of the process, the process risk is divided into multiple levels, and different response plans are set according to different risk levels; Establish a mapping relationship between abnormal causes and process risk indicators, and conduct process abnormal cause analysis by combining process failure prediction causes and potential quality risk coefficients.

[0011] Furthermore, the process risk index refers to the risk index calculated for different influencing factors that may cause process risks under each of the sub-items of people, machines, materials, methods, and environment; the calculation is carried out by constructing a calculation model based on different variables in each process through data fitting.

[0012] A machine learning-based device for identifying sources of quality fluctuations in a production process includes: Anomaly pattern recognition unit is used to acquire quality characteristic measurement data of processes or parts, identify control chart patterns of quality characteristics, and establish anomaly pattern matrix if the pattern is identified as an anomaly pattern in the control chart. The anomaly cause identification unit is used to distinguish between random factors and systemic factors for anomaly modes in control charts other than the periodic mode, using trend analysis. The fuzzy relation matrix construction unit is used to identify the source of process quality fluctuation when the cause of the anomaly is a systemic factor. It uses fuzzy statistical methods or expert experience-based methods to determine the degree of membership between the anomaly pattern and the cause of the anomaly in the control chart, and finally establishes the fuzzy relation matrix. The anomaly cause diagnosis unit is used to construct an anomaly occurrence probability matrix based on the anomaly pattern matrix and the fuzzy relation matrix, merge the elements in the probability matrix, and diagnose the anomaly cause according to the confidence value. The risk assessment unit is used to set process risk indicators and calculate the potential quality risk coefficients of each sub-item of "man, machine, material, method, and environment" in the production process, thereby determining the total process risk coefficient for assessing process risk. The result output unit is used to read the process quality factor change information in the process quality BOM. The data acquisition condition is the most recent change information before the production of the diagnosed process / part. Through data integration, the output displays the process quality fluctuation source identification result set. The result set includes the abnormal cause diagnosis result and confidence value, potential quality risk coefficient, process risk level, process quality factor change, and preventive measures.

[0013] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, it implements the machine learning-based method for identifying sources of process quality fluctuations.

[0014] A computer-readable storage medium storing a computer program; when executed by a processor, the computer program implements the machine learning-based method for identifying sources of process quality fluctuations.

[0015] Compared with the prior art, the present invention has the following technical features: This invention uses machine learning to identify patterns in process quality characteristic control charts, enabling early warning of process anomalies. It employs trend analysis to analyze quality characteristic detection values, eliminating random factors and narrowing the scope of process quality diagnosis. By establishing a fuzzy relation matrix and performing numerical merging, diagnostic results are output based on the credibility of the cause. The invention calculates the dynamic quality risk coefficient of the process, combining process quality risk warnings with diagnostic results, and simultaneously outputting information on changes in process quality factors in real time. This improves the real-time diagnostic capability of process quality, providing support for further investigation of the root causes of process quality anomalies. It features high accuracy in identifying process quality fluctuation sources, fast diagnostic speed, and wide applicability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of abnormal control chart modes of the method of the present invention; wherein (a) is an upward trend mode, (b) is a downward trend mode, (c) is an upward step mode, (d) is a downward step mode, and (e) is a periodic mode. Figure 2 This is a trend analysis chart of the quality characteristic measurement values ​​in an embodiment of the present invention; Figure 3 This is a mapping diagram of the distribution of abnormal causes and process risk indicators in an embodiment of the present invention; Figure 4 This is a flowchart illustrating an embodiment of the present invention. Detailed Implementation

[0017] This invention provides a method for identifying sources of quality fluctuations in a process based on machine learning, comprising the following steps: Step 1: Obtain the quality characteristic measurement data of the process or part, identify the control chart pattern of the quality characteristics, and establish an abnormal pattern matrix if the control chart is identified as an abnormal pattern.

[0018] Step 101: For a certain quality characteristic measurement data, the control chart pattern of the quality characteristic is identified using a control chart pattern recognition method based on machine learning.

[0019] Step 102: After completing the control chart pattern recognition, if an abnormal control chart pattern appears, an abnormal pattern matrix M is established; the abnormal control chart patterns include five types: upward trend pattern, downward trend pattern, upward step pattern, downward step pattern, and periodic pattern. A schematic diagram of the abnormal control chart patterns in one embodiment is shown below. Figure 1 .

[0020] ; in This represents the probability of the i-th control chart's abnormal pattern occurring. .

[0021] ; For example, the thickness characteristic measurement data of a panel component is read, with 25 samples. The measurement dataset is test={18.3, 18.4, 18.2, 18.4, 18.3, 18.6, 18.3, 18.5, 18.6, 18.4, 18.5, 18.5, 18.6, 18.5, 18.5, 18.6, 18.5, 18.6, 18.5, 18.5, 18.6, 18.4, 18.5, 18.5, 18.6}. A control chart pattern recognition method based on neural networks is used for identification. The control chart pattern recognition result is "upward trend pattern," which is an abnormal control chart pattern. Therefore, an abnormal pattern matrix M={1, 0, 0, 0, 0} is established.

[0022] Step 2: For the abnormal control chart modes other than the periodic mode, use trend analysis to distinguish whether the cause of the abnormality is a random factor or a systemic factor. If it is determined to be a random factor, no fault diagnosis is required. If it is determined to be a systemic factor, the cause of the abnormality is further determined.

[0023] Step 201: For the upward trend pattern (I), downward trend pattern (II), upward step pattern (III), and downward step pattern (IV), use trend analysis to distinguish whether the abnormal cause is accidental or systematic. like Figure 2 As shown, first obtain the observed quality characteristic curves (curves that change over time) in the control chart, and then calculate the slope of the curves: ; Wherein, E(t) n ) represents t in the control chart n Quality characteristic measurement data caused by abnormal reasons at any time; k n Indicates t n The slope of the curve at time t m E(t) represents any moment when the process is in a stable state under the normal mode of the control chart. m ) for t m Mass characteristic measurement data at any given time.

[0024] Step 202: When the process is in a stable state under the normal control chart mode, the maximum standard slope of the quality characteristic observation curve is k0. If the curve slope k nIf k < k0, it is determined that the trend change of the quality characteristic measurement data is small and is only affected by inevitable accidental factors, that is, the abnormal cause at time t n is an accidental factor; if k n > k0, it is determined that the trend change of the quality characteristic measurement value is large, indicating that the personnel, equipment / tools, materials, processes, environment, etc. in the production process are in an abnormal state, and the process is affected by systematic factors, that is, the abnormal cause at time t n is a systematic factor, and further analysis and diagnosis of the abnormal cause should be carried out.

[0025] For example: Suppose the control chart pattern recognition results of samples 1 to 10 are in the normal control chart pattern, and the control chart recognition results of samples 4 to 16 are in the upward trend pattern in the abnormal control chart pattern; then the calculation result of the slope k of sample 16 is: ; Suppose the maximum standard slope of the quality characteristic observation curve is k0 = 0.025. Since k n > k0, the trend change of the quality characteristic measurement data at time t n is large, and the resulting abnormal cause is a systematic factor.

[0026] Step 3, if the abnormal cause is a systematic factor, identify the quality fluctuation sources in the process, and use the fuzzy statistics method or the expert experience given method to determine the membership degree (weight) of the abnormal control chart pattern and the abnormal cause, and finally establish a fuzzy relationship matrix.

[0027] Establish a fuzzy statistics calculation model; assume that a fixed element u0 is taken on a universe of discourse U, and u0 represents the abnormal cause; consider a moving boundary variable ordinary set A * (A * as the elastic domain of the fuzzy set A, which is expressed here as the type of abnormal control chart pattern), after n tests, calculate: ; where η is the membership degree (frequency) of u0 to A.

[0028] The membership degree of the abnormal control chart pattern and the abnormal cause can also be given by expert experience; finally, establish a fuzzy relationship matrix Q between the abnormal control chart pattern and all abnormal causes under a certain abnormal cause category.

[0029] ; In the formula: n represents the number of abnormal causes (i.e., the number of tests above), and Q ij represents the possibility that the jth abnormal cause causes the ith abnormal control chart pattern, that is, the membership degree η of the ith abnormal control chart pattern to the jth abnormal cause.

[0030] In one embodiment of the present invention, the fuzzy relation matrix Q constructed using fuzzy statistics is as follows: ; Step 4: Based on the anomaly pattern matrix M and the fuzzy relation matrix Q, construct the anomaly occurrence probability matrix T, merge the elements in the probability matrix T, and diagnose the cause of the anomaly based on the confidence value.

[0031] Step 401: Integrate the fuzzy relation matrix Q and the abnormal pattern matrix M to obtain the abnormality probability matrix T; common abnormal causes in the process of production of processes or parts are shown in Table 1.

[0032] ; In the formula, n represents the number of abnormal causes under a certain category, and the i-th element T in matrix T is... i The value represents the confidence level of the cause of the anomaly.

[0033] Table 1. Common causes of abnormalities in the production process of a step or part in one embodiment.

[0034] In one embodiment of the present invention, the constructed anomaly occurrence probability matrix T is as follows: .

[0035] Step 402: After calculating the probability matrix T of anomalies, process each element T in the matrix... i Perform a merge operation, dividing [0,1] into different intervals, and assigning a confidence value to each interval; then, perform a merge operation on each element T in the matrix. i The confidence level is represented by the corresponding confidence value based on the interval it falls into: ; Step 403: Establish a correspondence table between confidence values ​​and probabilities (see Table 2), and merge the non-zero T values ​​in the matrix. i The value is compared with the corresponding relationship table; non-zero T i The order of values ​​in the probability matrix T corresponds to the sequence number of the cause of the anomaly, and the probability of the cause of the anomaly is determined from the correspondence table.

[0036] Table 2. Correspondence between Abnormal Patterns and Potential Causes of Abnormalities

[0037] For example, in one embodiment of the present invention, taking "personnel" as the category of abnormal cause, after merging the probability matrix T of abnormal occurrence, T=[0, 0.6, ..., 0] is obtained; T iAfter comparing the value with Tables 1 and 2, since 0.6 is the second element in matrix T, corresponding to the second abnormal cause under the "Personnel" category in Table 1, it is concluded that the cause of this abnormal pattern "may" be the second abnormal cause, namely "operational error / mistake" in personnel factors.

[0038] Step 5: Set process risk indicators for the potential quality risks of each component of "man, machine, material, method, and environment" in the production process and calculate the potential quality risk coefficient of the process to determine the total process risk coefficient, which is used to assess the process risk.

[0039] Among them, the process risk index refers to the risk index calculated for different influencing factors that may cause process risks under each of the sub-items of people, machines, materials, methods, and environment. Specific calculation methods for these indicators can include expert assignment or calculation based on different variables (such as time, scores, etc.) in each process, constructed through data fitting (such as polynomial fitting) to create a calculation model. For example, in this embodiment of the invention, for the sub-item "people," the process risk index includes operator working status, operator's employment time, operator skill level, etc. Figure 3 As shown.

[0040] Step 501: Based on the process risk indicators, calculate the dynamic risk coefficient υ of each sub-item of "Man, Machine, Material, Method, Environment" for the process. i The values ​​are then arranged in order of magnitude to obtain the potential quality risk coefficient υ: ; In the formula, υ i Let be the dynamic risk coefficient of the i-th sub-item, i=1,2,…,5, which correspond to the dynamic risk coefficients of the sub-items of personnel, machinery, materials, methods, and environment, respectively. The risk coefficients are all between 0 and 1.

[0041] Step 502, calculate the total risk coefficient σ of the process. [0,1]: ; In the formula, the superscript T indicates transpose; O is the weight vector of the sub-items: person, machine, material, method, and environment.

[0042] Based on the total risk coefficient σ of the process, the process risk is divided into four levels: low risk (0 < σ < 0.25), moderate risk (0.25 ≤ σ < 0.5), relatively high risk (0.5 ≤ σ < 0.75), and high risk (0.75 ≤ σ ≤ 1). Different response plans are set according to different risk levels. For example, when σ ≥ 0.5, process adjustment measures are required, while when it is less than 0.5, no adjustment is required. However, moderate risk and relatively high risk should be continuously monitored.

[0043] In one embodiment of the present invention: Assuming the dynamic risk coefficients for the four sub-items "machine, material, method, and environment" are all 0, the dynamic risk coefficient υ1 for personnel is calculated based solely on the three process risk indicators of personnel factors. The three process risk indicators are the operator's working status (see formula a), the operator's working time (see formula b), and the operator's skill level (see formula c), resulting in υ1 = 0.55.

[0044] υ(P1)=2*10 -6 P1 3 +3*10 -4 P1 2 -0.0272P1+0.55670≤P1≤75(a) In the formula, P1 is the total working time (hours) within the statistical period, and υ(P1) is the dynamic risk coefficient of the product process for which the operator continues to work during that time. υ(P2) = -1*10 -5 P2 4 +8*10 -4 P2 3 -0.0175P2 2 +0.065P² + 0.84070 ≤ P² ≤ 36(b) In the formula, P2 represents the length of service (in years), and υ(P2) represents the dynamic risk coefficient of the product process for the operator during that length of service. υ(P3) = 1 / P3(c) In the formula, P3 is the operator's qualification exam score, and υ(P3) is the dynamic risk coefficient for the corresponding process processed by the operator who obtained the score.

[0045] Calculate the total risk coefficient σ of the process: ; Wherein, O is the weight value vector of each factor in the process (described as 18 factors in the embodiment), O1 is the weight value vector of the process risk index of the operator's working status, O2 is the weight value vector of the process risk index of the operator's working time, and O3 is the weight value vector of the process risk index of the operator's skill level.

[0046] Ultimately, since the total risk coefficient σ of the process is 0.55, the corresponding risk level of the process is "higher risk", so process adjustment measures need to be taken.

[0047] Step 503: Establish the mapping relationship between the causes of abnormalities and process risk indicators; see the following for an example of the mapping relationship. Figure 3 By combining the predicted causes of process failures with the potential quality risk coefficient, the causes of process abnormalities are analyzed.

[0048] Step 6: Based on the unique identifier of the process number / part number, read the process quality factor (man, machine, material, method, environment) change information in the process quality BOM. The data acquisition condition is the most recent change information before the production of the diagnosed process / part. Through data integration, output and display the process quality fluctuation source identification result set W. The result set includes the abnormal cause diagnosis results and confidence value, potential quality risk coefficient, process risk level, process quality factor changes, preventive measures, etc. The preventive measures are formulated according to the actual situation.

[0049] For example, after reading the process quality factor change information in the process quality BOM, it is found that the operator of a certain process has changed within the past 10 days, from Zhang San to Li Si. Since the quality of other processes has not changed, and operator Li Si has a short tenure of one year, the output process quality fluctuation source identification result W is: W = {Operational error / error R² (diagnosis result of abnormal cause), 0.6 (confidence value), 0.55 (potential quality risk coefficient), higher risk (process risk level), operator change for this process (process quality factor change), operator replacement recommended (preventive measures)} The method of this invention can significantly improve the efficiency of process quality anomaly diagnosis and traceability, reducing it from 2-3 person-days to 0.003 person-days, and improving traceability efficiency by two orders of magnitude. Practical analysis shows that the comprehensive diagnostic results have a high probability of covering the actual root cause, with an accuracy rate of over 90%. It can be widely applied to process quality diagnosis scenarios in the manufacturing industry.

[0050] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for identifying sources of quality fluctuations in a process based on machine learning, characterized in that, include: Acquire quality characteristic measurement data of processes or parts, identify control chart patterns of quality characteristics, and establish an abnormal pattern matrix if the pattern is identified as an abnormal control chart pattern. For abnormal control chart patterns other than the cyclical pattern, trend analysis is used to distinguish whether the cause of the abnormality is due to accidental factors or systemic factors. If the cause of the anomaly is a systemic factor, then the source of quality fluctuation in the process is identified. Fuzzy statistical methods or expert experience-based methods are used to determine the degree of membership between the anomaly pattern and the cause of the anomaly in the control chart, and finally a fuzzy relation matrix is ​​established. Based on the anomaly pattern matrix and the fuzzy relation matrix, an anomaly occurrence probability matrix is ​​constructed. The elements in the probability matrix are merged, and the anomaly cause is diagnosed according to the confidence value. For each component of the "man, machine, material, method, and environment" in the production process, process risk indicators are set and the potential quality risk coefficient of the process is calculated, thereby determining the total process risk coefficient, which is used to assess the process risk; Read the process quality factor change information from the process quality BOM. The data acquisition condition is the most recent change information before the production of the diagnosed process / part. Through data integration, output and display the process quality fluctuation source identification result set. The result set includes the abnormal cause diagnosis result and confidence value, potential quality risk coefficient, process risk level, process quality factor change, and preventive measures.

2. The method for identifying process quality fluctuation sources based on machine learning according to claim 1, characterized in that, A machine learning-based control chart pattern recognition method is used to identify control chart patterns of quality characteristics; the abnormal control chart patterns include upward trend patterns, downward trend patterns, upward step patterns, downward step patterns, and periodic patterns.

3. The method for identifying process quality fluctuation sources based on machine learning according to claim 1, characterized in that, For control chart anomalies other than cyclical patterns, trend analysis is used to distinguish whether the anomalies are caused by random or systemic factors, including: First, obtain the observed quality characteristic curves from the control charts and calculate the slope of the curves: ； Wherein, E(t) n ) represents t in the control chart n Quality characteristic measurement data caused by abnormal reasons at any time; k n Indicates t n The slope of the curve at time t m E(t) represents any moment when the process is in a stable state under the normal mode of the control chart. m ) for t m Mass characteristic measurement data at any given time; When the process is in a stable state of the normal mode of the control chart, the maximum standard slope of the quality characteristic observation curve is k0. If the curve slope k n < k0, that is, t n The cause of abnormality at the moment is random factor, otherwise it is systematic factor.

4. The method for identifying process quality fluctuation sources based on machine learning according to claim 1, characterized in that, Determine the degree of membership between the abnormal patterns and causes of the control charts, and finally establish a fuzzy relation matrix, including: Suppose we take a fixed element u0 on a universe of discourse U, where u0 represents the cause of the anomaly; consider a general set A of motion with variable boundaries in the universe of discourse U. * A * This represents the types of abnormal patterns in the control chart; after n trials, the calculation is as follows: ； Where η is the degree of membership of u0 to A; Establish a fuzzy relation matrix Q between the abnormal patterns in control charts and all abnormal causes under a certain abnormal cause category; the elements Q in the fuzzy relation matrix Q are... ij η represents the probability that the j-th abnormal cause will cause the i-th control chart abnormal pattern, that is, the degree of membership between the i-th control chart abnormal pattern and the j-th abnormal cause.

5. The method for identifying process quality fluctuation sources based on machine learning according to claim 1, characterized in that, Based on the anomaly pattern matrix and fuzzy relation matrix, an anomaly occurrence probability matrix is ​​constructed. The elements in the probability matrix are merged, and the anomaly causes are diagnosed based on the confidence values, including: Integrating the fuzzy relation matrix Q and the anomaly pattern matrix M yields the anomaly occurrence probability matrix T; the i-th element T in matrix T... i The value represents the confidence level of the cause of the anomaly. After calculating the probability matrix T of anomalies, each element T in the matrix is... i Perform a merge operation, dividing [0,1] into different intervals, and assigning a confidence value to each interval; then, perform a merge operation on each element T in the matrix. i The confidence level is represented by the corresponding confidence value based on the interval it falls into; Establish a correspondence table between confidence values ​​and probabilities, and include non-zero T values ​​in the merged matrix. i The value is compared with the corresponding relationship table; non-zero T i The order of values ​​in the probability matrix T corresponds to the sequence number of the cause of the anomaly, and the probability of the cause of the anomaly is determined from the correspondence table.

6. The method for identifying process quality fluctuation sources based on machine learning according to claim 1, characterized in that, For each component of the "man, machine, material, method, and environment" in the production process, process risk indicators are set and process risk coefficients are calculated to determine the total process risk coefficient, which is used to assess process risk, including: By combining process risk indicators, the dynamic risk coefficients of each sub-item of "man, machine, material, method, and environment" are calculated and arranged in order of numerical value to obtain the potential quality risk coefficient υ. Based on the preset weight vectors of personnel, machines, materials, methods, and environment, the total risk coefficient σ of the process is calculated. Based on the magnitude of the total risk coefficient σ of the process, the process risk is divided into multiple levels, and different response plans are set according to different risk levels; Establish a mapping relationship between abnormal causes and process risk indicators, and conduct process abnormal cause analysis by combining process failure prediction causes and potential quality risk coefficients.

7. The method for identifying process quality fluctuation sources based on machine learning according to claim 6, characterized in that, The process risk index refers to the risk index calculated for different factors that may cause process risks under each of the sub-items of people, machines, materials, methods, and environment; the calculation is carried out by constructing a calculation model based on different variables in each process through data fitting.

8. A machine learning-based device for identifying sources of quality fluctuations in a production process, characterized in that, include: Anomaly pattern recognition unit is used to acquire quality characteristic measurement data of processes or parts, identify control chart patterns of quality characteristics, and establish anomaly pattern matrix if the pattern is identified as an anomaly pattern in the control chart. The anomaly cause identification unit is used to distinguish between random factors and systemic factors for anomaly modes in control charts other than the periodic mode, using trend analysis. The fuzzy relation matrix construction unit is used to identify the source of process quality fluctuation when the cause of the anomaly is a systemic factor. It uses fuzzy statistical methods or expert experience-based methods to determine the degree of membership between the anomaly pattern and the cause of the anomaly in the control chart, and finally establishes the fuzzy relation matrix. The anomaly cause diagnosis unit is used to construct an anomaly occurrence probability matrix based on the anomaly pattern matrix and the fuzzy relation matrix, merge the elements in the probability matrix, and diagnose the anomaly cause according to the confidence value. The risk assessment unit is used to set process risk indicators and calculate the potential quality risk coefficients of each sub-item of "man, machine, material, method, and environment" in the production process, thereby determining the total process risk coefficient for assessing process risk. The result output unit is used to read the process quality factor change information in the process quality BOM. The data acquisition condition is the most recent change information before the production of the diagnosed process / part. Through data integration, the output displays the process quality fluctuation source identification result set. The result set includes the abnormal cause diagnosis result and confidence value, potential quality risk coefficient, process risk level, process quality factor change, and preventive measures.

9. A terminal device, comprising a processor, a memory, and a computer program stored in the memory; characterized in that, When the processor executes the computer program, it implements the machine learning-based method for identifying sources of process quality fluctuations as described in any one of claims 1-8.

10. A computer-readable storage medium storing a computer program; characterized in that, When the computer program is executed by a processor, it implements the machine learning-based method for identifying sources of process quality fluctuations as described in any one of claims 1-8.

Images