Fault prediction method and apparatus, model determination method and apparatus, and device and storage medium

Abstract

Provided in the present disclosure are a fault prediction method and apparatus, a model determination method and apparatus, and a device and a storage medium. A fault prediction model determination method comprises: on the basis of a partial least squares analysis method and a historical fault result, performing feature extraction on acquired historical feature data; and determining a relationship function between input feature extraction data and a fault prediction result on the basis of historical feature extraction data, so as to determine a fault prediction model on the basis of the relationship function. In the fault prediction model determination method, a mapping relationship between feature extraction data and a fault prediction result can be more deeply extracted by means of a partial least squares analysis method, and then a relationship function is determined on the basis of the extracted mapping relationship, and a fault prediction model with a higher fault prediction accuracy is determined on the basis of the relationship function.

Description

Fault prediction method, model determination method, device, equipment and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This disclosure claims priority to Chinese patent application number 2023115165699, filed with the Chinese Patent Office on November 15, 2023, entitled “Fault Prediction Method, Model Determination Method, Device, Equipment and Storage Medium,” the entire contents of which are incorporated by reference into this disclosure. Technical Field

[0003] The present disclosure relates to the field of fault prediction technology, and in particular to a fault prediction method, a model determination method, an apparatus, a device, and a storage medium. Background Art

[0004] In industrial production, the various components of automated test equipment such as integrated circuit automatic test equipment (ATE) inevitably experience many failures, such as electrical failures in motors, control failures, and mechanical failures. The occurrence of these failures may lead to interruptions in the test process and have an impact on the entire production test process.

[0005] However, in the related art, the method of predicting possible faults of ATE equipment based on a fault prediction model has a technical problem of poor fault prediction accuracy.

[0006] Summary of the Invention

[0007] The purpose of the embodiments of the present disclosure is to provide a fault prediction method, a model determination method, an apparatus, a device and a storage medium to solve the technical problem of poor fault prediction accuracy of fault prediction models in related technologies.

[0008] The present disclosure provides a method for determining a fault prediction model, the method comprising:

[0009] Acquire multiple sets of historical feature data of at least one tester during a communication process and historical fault results corresponding to the historical feature data; wherein each set of the historical feature data includes at least one of the tester's communication signal, tester temperature, tester pressure, communication board data exchange status, tester software running status, sorting progress of a sorter, and online line transmission rate;

[0010] Performing feature extraction on the historical feature data based on a partial least squares analysis method and the historical fault results to obtain historical feature extraction data;

[0011] Determining a relationship function between input feature extraction data and fault prediction results based on historical feature extraction data;

[0012] A fault prediction model is determined according to the relationship function.

[0013] In the above implementation process, the method for determining a fault prediction model extracts features from acquired historical feature data based on partial least squares analysis and historical fault results; determines a relationship function between the input feature extraction data and the fault prediction results based on the historical feature extraction data, and determines the fault prediction model based on this relationship function. This method for determining a fault prediction model uses partial least squares analysis to more deeply extract the mapping relationship between the feature extraction data and the fault prediction results, and then determines a relationship function based on the extracted mapping relationship. Based on this relationship function, a fault prediction model with high fault prediction accuracy is determined, thereby resolving the technical problem of poor fault prediction accuracy in fault prediction models in related technologies.

[0014] In addition, feature extraction is performed on historical feature data based on the partial least squares analysis method and historical fault results, and the "relationship function between input feature extraction data and fault prediction results" is determined based on the historical feature extraction data obtained by feature extraction. This can greatly reduce the amount of calculation in the relationship function determination process, thereby reducing the complexity of the fault prediction model training process.

[0015] Optionally, in an embodiment of the present disclosure, the input feature extraction data includes input time domain feature data and input frequency domain feature data; the feature extraction of the historical feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data includes: performing time domain feature extraction on the historical feature data to obtain historical time domain feature data; performing frequency domain feature extraction on the historical feature data to obtain historical frequency domain feature data; performing feature extraction on the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data.

[0016] In the above implementation process, through time domain feature extraction and frequency domain feature extraction, a more comprehensive analysis of the acquired historical feature data can be performed to obtain "feature parameters (historical time domain feature data and historical frequency domain feature data) with higher correlation with the fault category." Feature extraction of "historical time domain feature data and historical frequency domain feature data" based on partial least squares analysis and historical fault results can be performed to obtain historical feature extraction data with better correlation with the fault prediction results. Based on the historical feature extraction data, the "relationship function between the input feature extraction data and the fault prediction results" is determined, further improving the fault prediction accuracy of the fault prediction model.

[0017] Optionally, in an embodiment of the present disclosure, the feature extraction of the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data includes: determining an input variable matrix based on the historical time domain feature data and the historical frequency domain feature data; determining an output variable matrix based on the historical fault results; determining a feature vector set based on the input variable matrix, the output variable matrix, and a preset covariance relationship between the input variable matrix and the output variable matrix; determining the eigenvector in the eigenvector set that meets a preset eigenvalue condition as the historical feature extraction data; wherein the preset eigenvalue condition is related to the size of the eigenvalue corresponding to each eigenvector in the eigenvector set.

[0018] In the above implementation process, an input variable matrix carrying input variable information can be determined based on historical time-domain feature data and historical frequency-domain feature data; an output variable matrix carrying output variable information can be determined based on historical fault results; the correlation between the input variable matrix and the output variable matrix is ​​adjusted based on a preset covariance relationship to determine a set of eigenvectors corresponding to the output variable matrix; and, based on preset eigenvalue conditions, eigenvectors with strong explanatory power for the fault prediction results are selected from the set of eigenvectors as historical feature extraction data. Because historical feature extraction data has strong explanatory power for fault prediction results, the "relationship function between input feature extraction data and fault prediction results" determined based on historical feature extraction data can well reflect the mapping relationship between the input feature extraction data and the fault prediction results. Therefore, the fault prediction model obtained based on this fault prediction model determination method can reduce the computational complexity during model training while improving the model's fault prediction accuracy.

[0019] Optionally, in an embodiment of the present disclosure, determining a set of eigenvectors according to the input variable matrix, the output variable matrix, and a preset covariance relationship between the input variable matrix and the output variable matrix includes:

[0020] Performing centralization processing on the input variable matrix and the output variable matrix, and obtaining a regression coefficient matrix by calculating the covariance matrix of the input variable matrix and the output variable matrix;

[0021] Based on the input variable matrix and the regression coefficient matrix, a projection matrix of the input variable matrix under the regression coefficient matrix is ​​constructed; based on the output variable matrix and the regression coefficient matrix, a projection matrix of the output variable matrix under the regression coefficient matrix is ​​constructed; and the eigenvector set is determined based on the constraint conditions and the preset covariance relationship.

[0022] Optionally, in an embodiment of the present disclosure, the preset eigenvalue condition includes selecting the eigenvector with the largest eigenvalue as historical feature extraction data; the number of the eigenvectors is determined by arranging each eigenvector in the eigenvector set according to the size of the eigenvalue and using the inflection point method.

[0023] Optionally, in an embodiment of the present disclosure, the relationship function between input feature extraction data and fault prediction results is determined based on historical feature extraction data, including: grouping the historical feature extraction data to obtain a training extraction data group and a test extraction data group; wherein, the training extraction data group includes multiple training feature data, and the test extraction data group includes multiple test feature data; calculating the distance value between the training feature data and each of the test feature data; determining multiple neighboring feature data in the test feature data based on the distance value; wherein, the number of the neighboring feature data is determined based on the number of feature vectors in the feature vector set that meet the preset feature value condition; determining the fault classification result of the training feature data based on the historical fault results corresponding to the neighboring feature data; wherein, the fault classification result includes whether there is a fault and the type of fault; determining the relationship function between the input feature extraction data and the fault prediction result based on the training feature data and the fault classification result corresponding to each of the training feature data.

[0024] In the above implementation process, by dividing the historical feature extraction data into a training extraction data group and a test extraction data group; determining the fault classification result of each training feature data based on the test extraction data in the test extraction data group and its corresponding historical fault results (that is, determining the fault classification result of each training feature data through K-nearest neighbor classification); the fault classification result corresponding to the training feature data can be determined more accurately, thereby improving the accuracy of the relationship function determined based on the "training feature data and the fault classification result corresponding to each training feature data", and improving the fault prediction accuracy of the determined fault prediction model.

[0025] Optionally, in an embodiment of the present disclosure, determining a relationship function between input feature extraction data and the fault prediction result based on the training feature data and the fault classification result corresponding to each training feature data includes:

[0026] The fault classification result corresponding to each of the training feature data is used as the fault identification of the training feature data, and the training feature data is used as a variable. The regression coefficient of each variable is calculated using the least squares method or the gradient descent method to obtain the relationship function between the input feature extraction data and the fault prediction result.

[0027] Optionally, in an embodiment of the present disclosure, the historical fault results include: no fault or at least one of upper computer communication abnormality, resource board abnormality, test machine communication abnormality and information interaction error; the acquisition of multiple sets of historical feature data of at least one test machine during the communication process and the historical fault results corresponding to the historical feature data include: acquiring multiple sets of historical feature data of at least one test machine during the communication process; determining the first data exchange volume between the upper computer and the test machine, the second data exchange volume of the resource board, the third data exchange volume between the test machine and the sorting machine, and the classification matching degree between the test machine and the sorting machine based on the historical feature data; determining whether the upper computer communication abnormality occurs based on the first data exchange volume and the first preset exchange volume threshold; determining whether the resource board abnormality occurs based on the second data exchange volume and the second preset exchange volume threshold; determining whether the test machine communication abnormality occurs based on the third data exchange volume and the third preset exchange volume threshold; determining whether an information interaction error occurs based on the changing state of the classification matching degree.

[0028] In the above implementation process, by determining the abnormality judgment data corresponding to each set of historical feature data (the first data exchange volume between the host computer and the test machine, the second data exchange volume of the resource board, the third data exchange volume between the test machine and the sorting machine, and the classification matching degree between the test machine and the sorting machine), and determining whether a host computer communication abnormality, a resource board abnormality, a test machine communication abnormality, or an information interaction error occurs based on the abnormality judgment data and the preset abnormality judgment conditions, the historical fault results corresponding to the historical feature data can be obtained. In addition, the existing fault prediction technology can only locate whether a fault has occurred, but cannot accurately locate what kind of fault has occurred. The disclosed solution can not only determine whether a fault has occurred in the target test machine during the communication process, but also predict the type of fault that occurred in the target test machine during the communication process, thereby providing a reference basis for subsequent fault handling and improving the efficiency of fault handling.

[0029] Optionally, in the embodiment of the present disclosure, determining whether the host computer communication abnormality occurs according to the first data exchange volume and a first preset exchange volume threshold includes:

[0030] When the first data exchange volume is less than the first preset exchange volume threshold, determining that the host computer communication abnormality occurs;

[0031] The determining whether the resource board abnormality occurs according to the second data exchange volume and a second preset exchange volume threshold includes:

[0032] When the second data exchange volume is less than a second preset exchange volume threshold, determining that the resource board is abnormal;

[0033] The determining whether a tester communication abnormality occurs according to the third data exchange volume and a third preset exchange volume threshold includes:

[0034] When the third data exchange amount is less than the third preset exchange amount threshold, determining that a test machine communication abnormality occurs;

[0035] The determining whether an information interaction error occurs according to the change state of the classification matching degree includes:

[0036] When the classification matching degree decreases or is lower than a matching degree threshold, it is determined that an information interaction error occurs.

[0037] The present disclosure also provides a fault prediction method, which includes:

[0038] Acquire communication characteristic data of the target tester during the communication process; wherein the communication characteristic data includes at least one of the communication signal of the target tester, the temperature of the target tester, the pressure of the target tester, the data exchange status of the communication board, the running status of the tester software, the classification progress of the sorting machine, and the online line transmission rate;

[0039] performing feature extraction on the communication feature data to obtain feature extraction data;

[0040] The feature extraction data is input into a fault prediction model, and a fault prediction result corresponding to the communication feature data is determined according to the output of the fault prediction model; wherein the fault prediction model is determined according to any of the above-mentioned methods for determining a fault prediction model.

[0041] The present disclosure also provides a fault prediction device, comprising:

[0042] a communication characteristic data acquisition module, configured to acquire communication characteristic data of the target tester during the communication process; wherein the communication characteristic data includes at least one of the communication signal of the target tester, the target tester temperature, the target tester pressure, the communication board data exchange status, the tester software running status, the sorting progress of the sorter, and the online line transmission rate;

[0043] A feature extraction module, configured to extract features from the communication feature data to obtain feature extraction data;

[0044] A fault prediction module is used to input the feature extraction data into a fault prediction model and determine a fault prediction result corresponding to the communication feature data based on the output of the fault prediction model; wherein the fault prediction model is determined according to any of the above-mentioned methods for determining a fault prediction model.

[0045] The present disclosure also provides a device for determining a fault prediction model, the device comprising:

[0046] A historical data acquisition module is configured to acquire historical characteristic data of at least one tester during a communication process and historical fault results corresponding to the historical characteristic data; wherein the historical characteristic data includes at least one of the tester's communication signal, tester temperature, tester pressure, communication board data exchange status, tester software running status, sorting progress of a sorter, and online line transmission rate;

[0047] A historical feature extraction module, configured to extract features from the historical feature data based on a partial least squares analysis method to obtain historical feature extraction data;

[0048] a relationship function determination module, configured to determine a relationship function between input feature extraction data and the fault result based on the historical feature extraction data and the historical fault result;

[0049] The model determination module is used to determine the fault prediction model according to the relationship function.

[0050] An embodiment of the present disclosure further provides an electronic device, comprising: a memory and a processor, wherein the memory stores a computer program executable by the processor, and when the computer program is executed by the processor, the determination method of the fault prediction device or the fault prediction method is executed.

[0051] An embodiment of the present disclosure further provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the determination method of the fault prediction device or the fault prediction method is executed.

[0052] The beneficial effects of the present disclosure are: using partial least squares analysis, a more in-depth mapping relationship between feature extraction data and fault prediction results can be extracted, and a relationship function can be determined based on the extracted mapping relationship. Based on this relationship function, a fault prediction model with high fault prediction accuracy is determined, thus resolving the technical problem of "existing fault prediction models having poor fault prediction accuracy." By extracting features from historical feature data based on partial least squares analysis and historical fault results, and determining the "relationship function between input feature extraction data and fault prediction results" based on the historical feature extraction data obtained through feature extraction, the amount of computation required to determine the relationship function can be significantly reduced, thereby reducing the complexity of the fault prediction model training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0054] FIG1 is a flow chart of a method for determining a fault prediction model provided by an embodiment of the present disclosure;

[0055] FIG2 is a characteristic value deviation diagram provided by an embodiment of the present disclosure;

[0056] FIG3 is a flow chart of a fault prediction method provided by an embodiment of the present disclosure;

[0057] FIG4 is a schematic structural diagram of a device for determining a fault prediction model provided by an embodiment of the present disclosure;

[0058] FIG5 is a schematic structural diagram of a fault prediction device provided by an embodiment of the present disclosure;

[0059] FIG6 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0060] The following embodiments of the technical solution of the present disclosure are described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present disclosure and are therefore only examples and are not intended to limit the scope of protection of the present disclosure.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this disclosure.

[0062] In the description of the embodiments of the present disclosure, technical terms such as "first" and "second" are used solely to distinguish between different objects and should not be understood to indicate or imply relative importance or to implicitly specify the quantity, specific order, or primary and secondary relationship of the technical features indicated. In the description of the embodiments of the present disclosure, "plurality" means two or more, unless otherwise specifically defined.

[0063] Please refer to Figure 1, which shows a flow chart of a method for determining a fault prediction model provided by an embodiment of the present disclosure. The method for determining a fault prediction model may include the following steps:

[0064] Step 101: Acquire multiple sets of historical characteristic data of at least one tester during a communication process and historical fault results corresponding to the historical characteristic data; wherein each set of the historical characteristic data includes at least one of the tester's communication signal, tester temperature, tester pressure, communication board data exchange status, tester software running status, sorting progress of a sorter, and online line transmission rate;

[0065] Step 102: extracting features from the historical feature data based on a partial least squares analysis method and the historical fault results to obtain historical feature extraction data;

[0066] Step 103: Determine a relationship function between the input feature extraction data and the fault prediction result based on the historical feature extraction data;

[0067] Step 104: Determine a fault prediction model according to the relationship function.

[0068] In step 101 provided in the embodiment of the present disclosure, the tester may be an automated testing device such as an ATE device or an online tester.

[0069] Optionally, multiple sets of historical feature data of a test machine at different time points during operation and after repair can be randomly selected, or multiple sets of historical feature data of multiple test machines of the same model or the same performance series at different time points during operation and after repair can be selected.

[0070] In an example, 1,000 groups of historical feature data at different time points may be extracted, or 1,500 groups of historical feature data at different time points may be extracted. The number of groups of historical feature data extracted may be adjusted according to actual model accuracy requirements.

[0071] In step 102 provided in the embodiment of the present disclosure, the partial least squares analysis method can project the independent variable and the dependent variable into a new space to maximize the correlation between the two. The partial least squares analysis method can more deeply extract the mapping relationship between the feature extraction data and the fault prediction results, and then determine a relationship function based on the extracted mapping relationship. Based on this relationship function, a fault prediction model with high fault prediction accuracy is determined.

[0072] In step 103 provided in the embodiment of the present disclosure, the fault prediction result may include whether a fault occurs, or may include whether a fault occurs and the type of fault that occurs.

[0073] In one possible implementation, a "relationship function between input feature extraction data and fault prediction results" can be determined based on historical feature extraction data and historical fault results. In another possible implementation, the historical feature data can be divided into two groups (a training group and a test group). Based on the historical feature data and corresponding historical fault results of the test group, the fault classification results of the historical feature data of the training group are determined. Based on the historical feature data of the training group and the determined fault classification results, the "relationship function between input feature extraction data and fault prediction results" is determined.

[0074] In step 104 provided in the embodiment of the present disclosure, the fault prediction model may determine a fault prediction result corresponding to the input feature extraction data according to the relationship function.

[0075] It can be seen that the method for determining the fault prediction model provided by the embodiment of the present disclosure can more deeply extract the mapping relationship between feature extraction data and fault prediction results through the partial least squares analysis method, and then determine the relationship function according to the extracted mapping relationship, and determine the fault prediction model with higher fault prediction accuracy based on the relationship function, thereby solving the technical problem of poor fault prediction accuracy of the fault prediction model in the related technology.

[0076] In some optional embodiments, the input feature extraction data includes input time domain feature data and input frequency domain feature data; step 102, based on the partial least squares analysis method and the historical fault results, feature extraction is performed on the historical feature data to obtain historical feature extraction data, including: performing time domain feature extraction on the historical feature data to obtain historical time domain feature data; performing frequency domain feature extraction on the historical feature data to obtain historical frequency domain feature data; performing feature extraction on the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data.

[0077] Optionally, feature extraction may be performed on the historical feature data based on the time domain and the frequency domain respectively to obtain feature parameters (historical time domain feature data and historical frequency domain feature data) that may be directly related to the fault category.

[0078] In one example, using the communication signal of a tester as an example, in the time domain, signal characteristics such as the amplitude mean, variance, peak value, and peak-to-peak value of the communication signal can be measured. In the frequency domain, communication power in different frequency ranges can be calculated, for example, communication power in the low-frequency range (which can be 10-20 Hz), communication power in the medium-frequency range (which can be 40-60 Hz), communication power in the high-frequency range (greater than 100 Hz), or peak frequency. The number of groups of extracted historical time-domain feature data and historical frequency-domain feature data can be 48, 36, or other values. The specific number of groups can be determined based on the actual model accuracy requirements.

[0079] In some optional embodiments, the above-mentioned feature extraction of the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data includes: determining the input variable matrix based on the historical time domain feature data and the historical frequency domain feature data; determining the output variable matrix based on the historical fault results; determining the feature vector set based on the input variable matrix, the output variable matrix and the preset covariance relationship between the input variable matrix and the output variable matrix; determining the eigenvector in the eigenvector set that meets the preset eigenvalue condition as the historical feature extraction data; wherein the preset eigenvalue condition is related to the size of the eigenvalue corresponding to each eigenvector in the eigenvector set.

[0080] In this embodiment, the “preset covariance relationship between the input variable matrix and the output variable matrix” may be that the covariance between the input variable matrix and the output variable matrix is ​​as large as possible.

[0081] In one possible implementation, the input variable matrix X can be determined based on the historical time domain feature data and the historical frequency domain feature data, and the output variable matrix Y can be determined based on the historical fault results. The matrix X and the matrix Y are centered, and the regression coefficient matrix W is obtained by calculating the covariance matrix of the matrix X and the matrix Y; wherein, ‖W‖=1; by T=W T The projection matrix T of the matrix X under the regression coefficient matrix W is calculated, and the projection matrix U of the matrix Y under the regression coefficient matrix W is calculated by U=WY; the covariance matrix between the obtained projection matrices T and U is recalculated, and the eigenvector set is determined according to the “eigenvectors included in the projection matrix T when the rank of the covariance matrix is ​​maximized”.

[0082] Specifically, the projection matrix T=W can be constructed based on the input variable matrix T X, construct the projection matrix U=C based on the output variable matrix T Y; Based on the constraints (‖W‖=1, ‖C‖=1, W T=C) and a preset covariance relationship (cov(T, U)→MAX) to determine the feature vector set.

[0083] Optionally, the preset eigenvalue condition may be to select k eigenvectors with the largest eigenvalues ​​as historical feature extraction data. Each eigenvector in the eigenvector set may be arranged according to the size of the eigenvalue, and the specific value of k may be determined according to the inflection point method.

[0084] Please refer to Figure 2, which is an eigenvalue deviation diagram provided by an embodiment of the present disclosure. Among them, the horizontal axis of Figure 2 is the sequence number of the eigenvector after sorting (arranged from large to small according to the size of the eigenvalue), and the vertical axis is the eigenvalue corresponding to the eigenvector with different sequence numbers. Taking the eigenvalue deviation diagram shown in Figure 2 as an example, it can be seen that after arranging according to the size of the eigenvalue, the first three eigenvalues ​​are large but decay very quickly, while the eigenvalues ​​after the third eigenvalue are basically very small and decay very slowly. According to the inflection point method, k=3 is determined.

[0085] In some optional embodiments, step 103, determining the relationship function between input feature extraction data and fault prediction results based on historical feature extraction data, includes: grouping the historical feature extraction data to obtain a training extraction data group and a test extraction data group; wherein, the training extraction data group includes multiple training feature data, and the test extraction data group includes multiple test feature data; calculating the distance value between the training feature data and each of the test feature data; determining multiple neighboring feature data in the test feature data based on the distance value; wherein, the number of the neighboring feature data is determined according to the number of feature vectors in the feature vector set that meet the preset feature value condition; determining the fault classification result of the training feature data based on the historical fault results corresponding to the neighboring feature data; wherein, the fault classification result includes whether there is a fault and the type of fault; determining the relationship function between the input feature extraction data and the fault prediction result based on the training feature data and the fault classification result corresponding to each of the training feature data.

[0086] Optionally, the test extraction data group can account for 15% or 20% of the number of groups of historical feature extraction data obtained, and the data ratio relationship between the test extraction data group and the training extraction data group can be adjusted according to the actual model accuracy requirements. Taking the number of groups of historical feature extraction data as 1500 and the test extraction data group accounting for 20% of the number of groups of historical feature extraction data obtained as an example, the test extraction data group is 300 groups and the training extraction data group is 1200 groups. By calculating the Euclidean distance between each training feature data in the training extraction data group and all the test feature data in the test extraction data group, the several test feature data closest to the training feature data in the test extraction data group can be found and used as the neighboring feature data. Among them, the number of neighboring feature data can be equal to "the number of feature vectors that meet the preset feature value conditions in the feature vector set".

[0087] Optionally, the historical fault results may include no fault or at least one of a host computer communication anomaly, a resource board anomaly, a test machine communication anomaly, and an information exchange error. Taking the number of neighbor feature data as an example, if the historical fault results of two of the neighbor feature data are host computer communication anomalies, the fault classification result of the corresponding training feature data is host computer communication anomalies; if the historical fault results of two of the neighbor feature data are no fault, the fault classification result of the corresponding training feature data is no fault; if the historical fault results of two of the neighbor feature data are both host computer communication anomalies and resource board anomalies, the fault classification result of the corresponding training feature data is both host computer communication anomalies and resource board anomalies.

[0088] In an example, taking the historical fault results including: no fault or at least one of the host computer communication abnormality, resource board abnormality, test machine communication abnormality and information interaction error as an example, the historical fault results specifically include no fault (1 possible situation), only one fault occurred (4 possible situations such as host computer communication abnormality, resource board abnormality, test machine communication abnormality or information interaction error), two faults occurred at the same time (6 possible situations, not listed in detail here), three faults occurred at the same time (4 possible situations, not listed in detail here) and four faults occurred at the same time (host computer communication abnormality, resource board abnormality, test machine communication abnormality and information interaction error fault at the same time, corresponding to 1 possible situation), etc., a total of 16 possible fault classification results; the above 16 fault classification results can be labeled (for example, labeled with 0, 1, 2...15 respectively).

[0089] In this embodiment, the "training feature data and the fault classification results corresponding to each of the training feature data (which can be the fault classification results represented by the above-mentioned fault identifiers)" can be input into the regression algorithm for regression relationship fitting to obtain a fault prediction model including a relationship function.

[0090] Specifically, the training feature data can be determined as variables, and the regression coefficient of each variable can be calculated using the least squares method or the gradient descent method to obtain the "relationship function between the input feature extraction data and the fault prediction result."

[0091] In some optional embodiments, the historical fault results include: no fault or at least one of upper computer communication abnormality, resource board abnormality, test machine communication abnormality and information interaction error; step 101, obtain multiple sets of historical feature data of at least one test machine during the communication process and historical fault results corresponding to the historical feature data, including: obtaining multiple sets of historical feature data of at least one test machine during the communication process; determining the first data exchange volume between the upper computer and the test machine, the second data exchange volume of the resource board, the third data exchange volume between the test machine and the sorting machine, and the classification matching degree between the test machine and the sorting machine based on the historical feature data; determining whether the upper computer communication abnormality occurs based on the first data exchange volume and the first preset exchange volume threshold; determining whether the resource board abnormality occurs based on the second data exchange volume and the second preset exchange volume threshold; determining whether the test machine communication abnormality occurs based on the third data exchange volume and the third preset exchange volume threshold; determining whether an information interaction error occurs based on the changing state of the classification matching degree.

[0092] Optionally, the first preset exchange rate threshold can be 2 bit / s. If the first data exchange rate is less than 2 bit / s, a host computer communication abnormality fault is determined to have occurred. The first preset exchange rate threshold can also be 3 bit / s or 5 bit / s. The second preset exchange rate threshold can be 4 bit / s. If the second data exchange rate is less than 4 bit / s, a resource board abnormality fault is determined to have occurred. The second preset exchange rate threshold can also be 2 bit / s or 5 bit / s. The third preset exchange rate threshold can be 5 bit / s. If the third data exchange rate is less than 5 bit / s, a tester communication abnormality fault is determined to have occurred. The third preset exchange rate threshold can also be 3 bit / s or 4 bit / s. If the classification matching degree decreases or falls below the matching degree threshold, an information exchange error fault can be determined to have occurred. The specific values ​​of the first preset exchange rate threshold, the second preset exchange rate threshold, the third preset exchange rate threshold, and the matching degree threshold can be adjusted according to the actual application scenario (for example, the model of the ATE equipment or the actual operating environment of the ATE equipment).

[0093] The embodiment of the present disclosure uses the partial least squares analysis method to extract the mapping relationship between the feature extraction data and the fault prediction result, and then determines the relationship function based on the extracted mapping relationship, and determines a fault prediction model with a higher fault prediction accuracy based on the relationship function. Compared with the method in the related art of using the principal component analysis method to extract the mapping relationship between the feature extraction data and the fault prediction result, and then determining the relationship function based on the extracted mapping relationship, and determining the principal component fault prediction model based on the relationship function, the fault prediction model of the present disclosure has a higher accuracy. Moreover, in the specific fault classification process, the K nearest neighbor classifier is used to determine the fault classification result for each training feature data. Compared with the "vector machine fault classification result for each training feature data determined by using the support vector machine", the fault classification result obtained based on the implementation method provided by the present disclosure has a higher accuracy.

[0094] Specifically, classification models were established for comparison based on two feature extraction methods (principal component analysis and partial least squares analysis), as well as two classification methods (support vector machine and k-nearest neighbor). A binary classification model was developed for the binary fault prediction problem. Initially, the support vector machine and k-nearest neighbor classifiers were constructed using the original dataset (which could be 48 features). All 48 features were then filtered using principal component analysis and partial least squares analysis, respectively. The classifiers were then tested using the optimal filtered feature set (the feature set can be tested using the same platform and standard). The results are shown in Tables 1 and 2 below. It can be seen that the fault prediction model constructed using partial least squares analysis significantly improves in accuracy compared to the principal component fault prediction model. The algorithm accuracy of the fault prediction model constructed using partial least squares analysis increased by 3-4%, and its running speed increased by over 50%. Furthermore, the results in Table 1 show that the fault prediction model based on the k-nearest neighbor classifier can achieve an optimal correct classification rate of up to 98.5% for the binary classification problem.

[0095] Table 1

[0096] Table 2

[0097] For the sixteen-class classification problem, the correct classification rates of the models determined using the aforementioned modeling method are shown in Table 3 below. The results demonstrate that the fault prediction model trained using partial least squares analysis and the K-nearest neighbor classifier is more effective at classifying single or multiple faults, achieving an average accuracy of approximately 95%. However, the results indicate that the support vector machine is not suitable for multi-classification problems, achieving an accuracy of only approximately 63% for the sixteen-class classification problem. This may be due to noise or overfitting caused by the pursuit of a perfect fit (i.e., a perfect fit of the training data does not result in good predictive performance).

[0098] Table 3

[0099] Please refer to Figure 3, which is a flow chart of a fault prediction method provided by an embodiment of the present disclosure. The fault prediction method may include the following steps:

[0100] Step 201: Acquire communication characteristic data of the target tester during the communication process; wherein the communication characteristic data includes at least one of the communication signal of the target tester, the target tester temperature, the target tester pressure, the communication board data exchange status, the tester software running status, the sorting progress of the sorter, and the online line transmission rate;

[0101] Step 202: extracting features from the communication feature data to obtain feature extraction data;

[0102] Step 203: Input the feature extraction data into a fault prediction model, and determine a fault prediction result corresponding to the communication feature data based on an output of the fault prediction model; wherein the fault prediction model is determined based on any of the above-described methods for determining a fault prediction model.

[0103] Optionally, the feature extraction data may include time domain feature data and frequency domain feature data; step 202, performing feature extraction on the communication feature data to obtain feature extraction data, may include: performing time domain feature extraction on the communication feature data to obtain the time domain feature data; performing frequency domain feature extraction on the communication feature data to obtain the frequency domain feature data.

[0104] In some optional embodiments, the fault prediction result includes: no fault occurs or at least one of the following: abnormal communication with the host computer, abnormal resource board, abnormal communication with the test machine, and information interaction error; in step 203, the feature extraction data is input into the fault prediction model, and the fault prediction result corresponding to the communication feature data is determined according to the output of the fault prediction model. The method also includes: performing maintenance inspection on the target test machine and related equipment according to the fault prediction result.

[0105] Optionally, the aforementioned fault prediction method can accurately diagnose and predict communication faults in the target tester, effectively improving the safety and reliability of the target tester and reducing the risk of catastrophic accidents. This fault prediction method, through continuous online status monitoring and data analysis of the target tester, can also diagnose and predict the target tester's fault development trends, enabling the development of predictive maintenance plans and the implementation of inspection and repair actions in advance. This method can also effectively reduce equipment downtime for maintenance, identifying potential faults early and preventing them from escalating.

[0106] Please refer to FIG4 , which is a schematic diagram of the structure of a fault prediction device provided by an embodiment of the present disclosure. The fault prediction device includes:

[0107] The communication characteristic data acquisition module 301 is used to acquire the communication characteristic data of the target tester during the communication process; wherein the communication characteristic data includes at least one of the communication signal of the target tester, the target tester temperature, the target tester pressure, the communication board data exchange status, the tester software running status, the sorting progress of the sorter, and the online line transmission rate;

[0108] A feature extraction module 302 is used to extract features from the communication feature data to obtain feature extraction data;

[0109] The fault prediction module 303 is used to input the feature extraction data into a fault prediction model and determine the fault prediction result corresponding to the communication feature data based on the output of the fault prediction model; wherein the fault prediction model is determined according to any of the above-mentioned fault prediction model determination methods.

[0110] In some optional embodiments, the fault prediction result includes: no fault or at least one of the following: abnormal communication with the host computer, abnormal resource board, abnormal communication with the test machine, and information interaction error; the fault prediction device also includes: a maintenance inspection module, which is used to perform maintenance and inspection on the target test machine and related equipment according to the fault prediction result.

[0111] Please refer to FIG5 , which is a schematic diagram of a structure of a device for determining a fault prediction model provided by an embodiment of the present disclosure. The device for determining a fault prediction model includes:

[0112] The historical data acquisition module 401 is configured to acquire historical characteristic data of at least one tester during the communication process and historical fault results corresponding to the historical characteristic data; wherein the historical characteristic data includes at least one of the tester's communication signal, tester temperature, tester pressure, communication board data exchange status, tester software running status, sorting progress of the sorter, and online line transmission rate;

[0113] A historical feature extraction module 402 is configured to extract features from the historical feature data based on a partial least squares analysis method to obtain historical feature extraction data;

[0114] A relationship function determination module 403 is configured to determine a relationship function between input feature extraction data and fault results based on the historical feature extraction data and the historical fault results;

[0115] The model determination module 404 is configured to determine a fault prediction model according to the relationship function.

[0116] In some optional embodiments, the input feature extraction data includes input time domain feature data and input frequency domain feature data; the historical feature extraction module 402 is specifically used to: perform time domain feature extraction on the historical feature data to obtain historical time domain feature data; perform frequency domain feature extraction on the historical feature data to obtain historical frequency domain feature data; perform feature extraction on the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data.

[0117] In some optional embodiments, the input feature extraction data includes input time domain feature data and input frequency domain feature data; the historical feature extraction module 402 is specifically used to: perform time domain feature extraction on the historical feature data to obtain historical time domain feature data; perform frequency domain feature extraction on the historical feature data to obtain historical frequency domain feature data; perform feature extraction on the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data.

[0118] In some optional embodiments, the historical feature extraction module 402 is further specifically used to: determine the input variable matrix based on the historical time domain feature data and the historical frequency domain feature data; determine the output variable matrix based on the historical fault results; determine the feature vector set based on the input variable matrix, the output variable matrix and the preset covariance relationship between the input variable matrix and the output variable matrix; determine the feature vector in the feature vector set that meets the preset eigenvalue condition as the historical feature extraction data; wherein the preset eigenvalue condition is related to the size of the eigenvalue corresponding to each eigenvector in the feature vector set.

[0119] In some optional embodiments, the historical feature extraction module 402 is further specifically used to: perform centralization processing on the input variable matrix and the output variable matrix, and obtain the regression coefficient matrix by calculating the covariance matrix of the input variable matrix and the output variable matrix; construct a projection matrix of the input variable matrix under the regression coefficient matrix based on the input variable matrix and the regression coefficient matrix, and construct a projection matrix of the output variable matrix under the regression coefficient matrix based on the output variable matrix and the regression coefficient matrix, and determine the feature vector set based on the constraint conditions and the preset covariance relationship.

[0120] In some optional embodiments, the preset eigenvalue condition includes selecting the eigenvector with the largest eigenvalue as historical feature extraction data; the number of the eigenvectors is determined by arranging each eigenvector in the eigenvector set according to the size of the eigenvalue and using the inflection point method.

[0121] In some optional embodiments, the relationship function determination module 403 is specifically used to: group the historical feature extraction data to obtain a training extraction data group and a test extraction data group; wherein the training extraction data group includes multiple training feature data, and the test extraction data group includes multiple test feature data; calculate the distance value between the training feature data and each of the test feature data; determine multiple neighbor feature data in the test feature data based on the distance value; wherein the number of the neighbor feature data is determined according to the number of feature vectors in the feature vector set that meet the preset feature value conditions; determine the fault classification result of the training feature data based on the historical fault results corresponding to the neighbor feature data; wherein the fault classification result includes whether there is a fault and the type of fault; determine the relationship function between the input feature extraction data and the fault prediction result based on the training feature data and the fault classification result corresponding to each of the training feature data.

[0122] In some optional embodiments, the relationship function determination module 403 is specifically used to: use the fault classification result corresponding to each of the training feature data as the fault identification of the training feature data, and use the training feature data as a variable, use the least squares method or the gradient descent method to calculate the regression coefficient of each of the variables, and obtain the relationship function between the input feature extraction data and the fault prediction result.

[0123] In some optional embodiments, the historical fault results include: no fault or at least one of the following: abnormal communication with the host computer, abnormal resource board, abnormal communication with the test machine, and an information interaction error; the historical data acquisition module 401 is specifically used to: obtain multiple sets of historical feature data of at least one test machine during the communication process; determine the first data exchange volume between the host computer and the test machine, the second data exchange volume of the resource board, the third data exchange volume between the test machine and the sorting machine, and the classification matching degree between the test machine and the sorting machine based on the historical feature data; determine whether the abnormal communication with the host computer occurs based on the first data exchange volume and the first preset exchange volume threshold; determine whether the abnormal resource board occurs based on the second data exchange volume and the second preset exchange volume threshold; determine whether the abnormal communication with the test machine occurs based on the third data exchange volume and the third preset exchange volume threshold; determine whether an information interaction error occurs based on the changing state of the classification matching degree.

[0124] In some optional embodiments, the historical data acquisition module 401 is specifically used to: determine that the host computer communication abnormality occurs when the first data exchange volume is less than the first preset exchange volume threshold; determine that the resource board abnormality occurs when the second data exchange volume is less than the second preset exchange volume threshold; determine that the test machine communication abnormality occurs when the third data exchange volume is less than the third preset exchange volume threshold; and determine that an information interaction error occurs when the classification matching degree decreases or is lower than the matching degree threshold.

[0125] It should be understood that the apparatus for determining a fault prediction model (fault prediction apparatus) corresponds to the aforementioned method for determining a fault prediction model (fault prediction method) embodiment and is capable of executing each of the steps involved in the aforementioned method embodiment. The specific functions of the apparatus for determining a fault prediction model (fault prediction apparatus) can be found in the description above. To avoid repetition, a detailed description is omitted herein. The apparatus for determining a fault prediction model (fault prediction apparatus) includes at least one software functional module that can be stored in a memory in the form of software or firmware or embedded in the operating system (OS) of the device.

[0126] Please refer to Figure 6, which is a schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. An electronic device 500 provided in an embodiment of the present disclosure includes: a processor 501 and a memory 502. These components are interconnected and communicate with each other via a communication bus 503 and / or other forms of connection mechanisms (not shown). Memory 502 stores a computer program executable by processor 501. When executed by processor 501, the computer program performs the above-described method for determining a fault prediction model or fault prediction method.

[0127] The embodiment of the present disclosure further provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by the processor 501 , the method for determining a fault prediction model or the fault prediction method as described above is executed.

[0128] The storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage device, flash memory, magnetic disk or optical disk.

[0129] In the several embodiments provided in the embodiments of the present disclosure, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to the multiple embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in a different order than the order marked in the drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0130] In addition, the functional modules in each embodiment of the present disclosure may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0131] The above description is only an optional implementation of the embodiment of the present disclosure, but the protection scope of the embodiment of the present disclosure is not limited to this. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in the embodiment of the present disclosure, and they should all be covered by the protection scope of the embodiment of the present disclosure. Industrial Applicability

[0132] In summary, the present disclosure provides a fault prediction method, model determination method, device, equipment and storage medium, which can determine a fault prediction model with high fault prediction accuracy, thereby solving the technical problem of poor fault prediction accuracy of the fault prediction model in the related art.

Claims

1. A method for determining a fault prediction model, characterized in that: The method comprises: Acquire multiple groups of historical feature data of at least one test machine during the communication process and historical fault results corresponding to the historical feature data; wherein each group of the historical feature data includes at least one of the communication signal of the test machine, the temperature of the test machine, the pressure of the test machine, the data exchange status of the communication board, the running status of the test machine software, the classification progress of the sorting machine, and the online line transmission rate; Performing feature extraction on the historical feature data based on a partial least squares analysis method and the historical fault results to obtain historical feature extraction data; Determine a relationship function between input feature extraction data and fault prediction results based on historical feature extraction data; A fault prediction model is determined according to the relationship function.

2. The method according to claim 1, characterized in that in, The input feature extraction data includes input time domain feature data and input frequency domain feature data; The extracting features from the historical feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data includes: Performing time domain feature extraction on the historical feature data to obtain historical time domain feature data; Performing frequency domain feature extraction on the historical feature data to obtain historical frequency domain feature data; Based on the partial least squares analysis method and the historical fault results, feature extraction is performed on the historical time domain feature data and the historical frequency domain feature data to obtain historical feature extraction data.

3. The method according to claim 2, characterized in that The extracting the historical time domain feature data and the historical frequency domain feature data based on the partial least squares analysis method and the historical fault results to obtain historical feature extraction data includes: Determine an input variable matrix according to the historical time domain feature data and the historical frequency domain feature data; Determine an output variable matrix according to the historical fault results; Determining a set of eigenvectors according to the input variable matrix, the output variable matrix, and a preset covariance relationship between the input variable matrix and the output variable matrix; The feature vectors in the feature vector set that meet the preset feature value condition are determined as the historical feature extraction data; wherein the preset feature value condition is related to the size of the feature value corresponding to each feature vector in the feature vector set.

4. The method according to claim 3, characterized in that The determining of the feature vector set according to the input variable matrix, the output variable matrix, and a preset covariance relationship between the input variable matrix and the output variable matrix includes: The input variable matrix and the output variable matrix are centralized, and the input variable matrix is ​​calculated. matrix and the covariance matrix of the output variable matrix to obtain the regression coefficient matrix; Based on the input variable matrix and the regression coefficient matrix, a projection matrix of the input variable matrix under the regression coefficient matrix is ​​constructed; based on the output variable matrix and the regression coefficient matrix, a projection matrix of the output variable matrix under the regression coefficient matrix is ​​constructed; and based on the constraints and the preset covariance relationship, the eigenvector set is determined.

5. The method according to claim 3, characterized in that: The preset eigenvalue condition includes selecting the eigenvector with the largest eigenvalue as historical feature extraction data; the number of the eigenvectors is determined by arranging each eigenvector in the eigenvector set according to the eigenvalue size and using the inflection point method.

6. The method according to claim 3, characterized in that The determining of the relationship function between the input feature extraction data and the fault prediction result based on the historical feature extraction data includes: The historical feature extraction data is grouped to obtain a training extraction data group and a test extraction data group; wherein the training extraction data group includes a plurality of training feature data, and the test extraction data group includes a plurality of test feature data; Calculating the distance value between the training feature data and each of the test feature data; Determine a plurality of neighbor feature data in the test feature data according to the distance value; wherein the number of the neighbor feature data is determined according to the number of feature vectors in the feature vector set that meet the preset feature value condition; Determine the fault classification result of the training feature data according to the historical fault results corresponding to the neighbor feature data; wherein the fault classification result includes whether there is a fault and the type of fault; According to the training feature data and the fault classification result corresponding to each of the training feature data, a relationship function between the input feature extraction data and the fault prediction result is determined.

7. The method according to claim 6, characterized in that The determining, based on the training feature data and the fault classification result corresponding to each of the training feature data, a relationship function between the input feature extraction data and the fault prediction result comprises: The fault classification result corresponding to each of the training feature data is used as the fault identification of the training feature data, and the training feature data is used as a variable. The regression coefficient of each variable is calculated using the least squares method or the gradient descent method to obtain the relationship function between the input feature extraction data and the fault prediction result.

8. The method according to any one of claims 1 to 7, characterized in that: The historical fault results include: no fault or at least one of abnormal communication with the host computer, abnormal resource board, abnormal communication with the test machine, and information interaction error; the acquisition of multiple groups of historical feature data of at least one test machine during the communication process and the historical fault results corresponding to the historical feature data include: Acquire multiple groups of historical characteristic data of at least one test machine during the communication process; Determine, based on the historical feature data, a first data exchange amount between the host computer and the tester, a second data exchange amount of the resource board, a third data exchange amount between the tester and the sorter, and a classification matching degree between the tester and the sorter; Determining whether the host computer communication abnormality occurs according to the first data exchange volume and a first preset exchange volume threshold; Determining whether the resource board is abnormal according to the second data exchange volume and a second preset exchange volume threshold; Determining whether a test machine communication abnormality occurs according to the third data exchange volume and a third preset exchange volume threshold; Whether an information interaction error occurs is determined according to a change state of the classification matching degree.

9. The method according to claim 8, characterized in that The determining whether the host computer communication abnormality occurs according to the first data exchange volume and the first preset exchange volume threshold comprises: determining that the host computer communication abnormality occurs when the first data exchange volume is less than the first preset exchange volume threshold; The determining whether the resource board is abnormal according to the second data exchange volume and a second preset exchange volume threshold comprises: When the second data exchange amount is less than a second preset exchange amount threshold, determining that the resource board is abnormal; The determining whether a test machine communication abnormality occurs according to the third data exchange volume and the third preset exchange volume threshold comprises: When the third data exchange amount is less than the third preset exchange amount threshold, determining that a test machine communication abnormality occurs; The determining whether an information interaction error occurs according to the change state of the classification matching degree includes: When the classification matching degree decreases or is lower than a matching degree threshold, it is determined that an information interaction error occurs.

10. A fault prediction method, characterized in that: The method comprises: Acquire the communication characteristic data of the target tester during the communication process; wherein the communication characteristic data includes at least one of the communication signal of the target tester, the temperature of the target tester, the pressure of the target tester, the data exchange status of the communication board, the running status of the tester software, the classification progress of the sorting machine, and the online line transmission rate; Performing feature extraction on the communication feature data to obtain feature extraction data; The feature extraction data is input into a fault prediction model, and a fault prediction result corresponding to the communication feature data is determined according to an output of the fault prediction model; wherein the fault prediction model is determined according to the method according to any one of claims 1 to 9.

11. A fault prediction device, characterized in that: The device comprises: A communication characteristic data acquisition module, used to acquire the communication characteristic data of the target test machine during the communication process; wherein the communication characteristic data includes at least one of the communication signal of the target test machine, the temperature of the target test machine, the pressure of the target test machine, the data exchange status of the communication board, the running status of the test machine software, the classification progress of the sorting machine, and the online line transmission rate; A feature extraction module, used to extract features from the communication feature data to obtain feature extraction data; A fault prediction module is used to input the feature extraction data into a fault prediction model, and determine the fault prediction result corresponding to the communication feature data according to the output of the fault prediction model; wherein the fault prediction model is determined according to the method described in any one of claims 1-9.

12. A device for determining a fault prediction model, characterized in that: The device comprises: A historical data acquisition module, used to acquire historical characteristic data of at least one tester during the communication process and historical fault results corresponding to the historical characteristic data; wherein the historical characteristic data includes at least one of the communication signal of the tester, the temperature of the tester, the pressure of the tester, the data exchange status of the communication board, the running status of the tester software, the classification progress of the sorting machine, and the online line transmission rate; A historical feature extraction module, used to extract features from the historical feature data based on a partial least squares analysis method to obtain historical feature extraction data; A relationship function determination module, used to determine the relationship function between the input feature extraction data and the fault result based on the historical feature extraction data and the historical fault result; A model determination module is used to determine a fault prediction model according to the relationship function.

13. An electronic device, characterized in that: The electronic device comprises: Memory; processor; The memory stores a computer program executable by the processor, and when the computer program is executed by the processor, the method according to any one of claims 1 to 9 or the method according to claim 10 is performed.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1 to 9 or the method according to claim 10 is executed.

Images