Quality risk prediction method and device for equipment production and manufacturing, terminal equipment and storage medium
By acquiring parameters from the equipment production process, calculating statistical features and performing feature fusion, and using a random forest model for quality risk prediction, the problem of inaccurate detection caused by single-parameter judgment is solved, and more accurate quality risk judgment is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ELECTRIC POWER RES INST OF GUANGDONG POWER GRID CO LTD
- Filing Date
- 2026-04-17
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies use a fixed statistical threshold for a single parameter as the judgment standard, which leads to inaccurate quality risk detection results during the equipment manufacturing process.
By acquiring production process parameters and environmental parameters during equipment production, calculating statistical features, performing feature fusion, using a random forest model for quality risk prediction, and determining quality risk through time series smoothing and continuous over-limit judgment.
It effectively reduces false alarms and false negatives in quality risk detection, improves the accuracy of detection results, and can accurately identify equipment quality risks under noise and operating condition fluctuations.
Smart Images
Figure CN122047658A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of quality inspection technology, and in particular to a method, apparatus, terminal equipment, and storage medium for predicting quality risks in equipment manufacturing. Background Technology
[0002] With the advancement of digital transformation and standardization of quality control in the manufacturing of main power grid equipment, enterprises are gradually forming a multi-source heterogeneous data system with MES, PLM, and ERP as the backbone for equipment such as gas-insulated switches throughout the entire process of "design-raw materials-processing-assembly-debugging-type testing-shipment". They are also carrying out practices such as parameter specification verification, data standardization and integration, and status assessment based on test data on enterprise-level platforms to achieve quality risk detection of equipment.
[0003] Existing solutions typically use a fixed statistical threshold for a certain parameter (such as the partial discharge of equipment) as the criterion for quality risk detection. However, in reality, the measurement of these parameters is subject to unavoidable measurement noise. At the same time, fluctuations in operating conditions throughout the production process also affect the measurement results of these parameters. Therefore, fixed thresholds cannot adapt to the influence of variables at the mechanistic level, resulting in inaccurate quality risk detection results. Summary of the Invention
[0004] This invention provides a method, apparatus, terminal equipment, and storage medium for predicting quality risks in equipment manufacturing, which can solve the problem of inaccurate quality risk detection results caused by using a fixed statistical threshold of a single parameter as the judgment standard in the prior art.
[0005] An embodiment of the present invention provides a method for predicting quality risks in equipment manufacturing, comprising: Acquire the values of production process parameters and environmental parameters of the equipment under test throughout the entire production process; Based on the values of the above production process parameters and environmental parameters, the statistical characteristic quantities of the above-mentioned equipment under test in different time periods are calculated. For each time period, feature fusion is performed based on all statistical features of the current time period to obtain the fused feature vector for each time period. Each fused feature vector is input into a preset quality detection model to obtain the predicted quality risk value for each time period. The predicted quality risk values for all time periods are smoothed to obtain smoothed predicted quality risk values. Based on the smoothed predicted quality risk values for all time periods, risk prediction trajectories are generated in chronological order. If, in the above risk prediction trajectory, there are more than k consecutive time periods for smoothed predicted quality risk values that are all greater than the corresponding preset risk threshold, then the above-mentioned device under test is determined to have quality risk; otherwise, the above-mentioned device under test is determined not to have quality risk; where k is a positive integer.
[0006] Furthermore, based on the values of the aforementioned production process parameters and environmental parameters, the statistical characteristic quantities of the equipment under test are calculated for different time periods, including: After aligning the values of the above production process parameters and environmental parameters by time, the aligned values of the production process parameters and environmental parameters are segmented according to a preset sliding window size to obtain the values of the production process parameters and environmental parameters in different time periods; wherein, one time period corresponds to a set of production process parameter values and a set of environmental parameter values; For each time period, based on the values of the production process parameters and environmental parameters within that time period, the corresponding statistical characteristic quantities for the production process parameters and environmental parameters are calculated.
[0007] Furthermore, after obtaining the values of production process parameters and environmental parameters at different time periods, the process also includes: For each time period, determine the missing values of production process parameters and environmental parameters within that time period; Perform consistent imputation on all missing values.
[0008] Furthermore, for each time period, based on the values of the production process parameters and environmental parameters within that time period, the statistical characteristic quantities corresponding to the production process parameters and environmental parameters for that time period are calculated, including: For each time period, based on the values of the production process parameters within the current time period, the first mean, first extreme value, first slope, and first median of the production process parameters are calculated. Based on the values of environmental parameters within the current time period, the second mean, second standard deviation, second extreme value, second slope, and second median of the environmental parameters are calculated. The first mean, first extreme value, first slope, or first median mentioned above are used as statistical characteristic quantities of the production process parameters in the corresponding time period. The second mean, second extreme value, second slope, or second median are used as statistical characteristic quantities of environmental parameters for the corresponding time period.
[0009] Furthermore, the training of the aforementioned pre-defined quality detection model includes: Acquire the first historical values of production process parameters, the second historical values of environmental parameters, the true quality category labels, and the total number of all samples for several sample devices throughout the entire historical production process; wherein, the aforementioned true quality category labels are used to indicate whether the corresponding sample device has quality risks; Based on the first and second historical values mentioned above, the sample fusion feature vectors of each sample device are constructed. Based on the number of samples corresponding to the above-mentioned true quality category labels and the above-mentioned total number, the label weight factor corresponding to the true quality category label indicating the existence of quality risk and the label weight factor corresponding to the true quality category label indicating the absence of quality risk are calculated. Using the sample fusion feature vector of the aforementioned sample devices as input, the aforementioned real quality category labels as supervision quantities, and the aforementioned label weight factors as training weights for the corresponding sample devices, the quality detection model to be trained based on the random forest model is trained to obtain the aforementioned preset quality detection model.
[0010] Furthermore, before smoothing the predicted quality risk values for all time periods to obtain the smoothed predicted quality risk values, the following steps are also included: Obtain the preset calibration parameters; The predicted quality risk values for each time period are monotonically calibrated according to preset calibration parameters.
[0011] Furthermore, the determination of the aforementioned preset risk threshold includes: After dividing the monotonic calibrated predicted quality risk values into the above time periods, the average value and standard deviation of the monotonic calibrated predicted quality risk values in each time period are calculated. The preset risk threshold for the corresponding time period is calculated based on the above average value and standard deviation.
[0012] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments; This invention provides a quality risk prediction device for equipment manufacturing, comprising: The module includes a data acquisition module, a statistical feature calculation module, a feature fusion module, a risk prediction module, a risk trajectory smoothing module, and a quality prediction module. The aforementioned data acquisition module is used to acquire the values of production process parameters and environmental parameters of the equipment under test throughout the entire production process. The aforementioned statistical feature calculation module is used to calculate the statistical feature quantities of the equipment to be tested in different time periods based on the values of the aforementioned production process parameters and environmental parameters. The aforementioned feature fusion module is used to perform feature fusion operations on all statistical features of the current time period for each time period, so as to obtain the fused feature vector for each time period. The aforementioned risk prediction module is used to input each fused feature vector into a preset quality detection model to obtain the predicted quality risk value for each time period. The aforementioned risk trajectory smoothing module is used to smooth the predicted quality risk values for all time periods to obtain smoothed predicted quality risk values, and to generate risk prediction trajectories in chronological order based on the smoothed predicted quality risk values for all time periods. The aforementioned quality prediction module is used to determine that the device under test has a quality risk if, in the aforementioned risk prediction trajectory, there are more than k consecutive time periods of smoothed predicted quality risk values that are all greater than the corresponding preset risk threshold; otherwise, it is determined that the device under test does not have a quality risk; where k is a positive integer.
[0013] Based on the above method embodiments, the present invention provides a corresponding terminal device embodiment; The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the quality risk prediction method for device manufacturing described in any embodiment of the present invention.
[0014] Based on the above method embodiments, the present invention provides a corresponding storage medium embodiment; The present invention provides a storage medium including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the quality risk prediction method for equipment manufacturing described in any embodiment of the present invention.
[0015] The embodiments of the present invention have the following beneficial effects: This invention provides a method, apparatus, terminal device, and storage medium for predicting quality risks in equipment manufacturing. The method includes: acquiring the values of production process parameters and environmental parameters of the equipment under test throughout the entire production process; calculating statistical feature quantities of the equipment under test in different time periods based on the values of the production process parameters and environmental parameters; performing feature fusion operation on all statistical feature quantities of the current time period for each time period to obtain a fused feature vector for each time period; inputting each fused feature vector into a preset quality detection model to obtain a predicted quality risk value for each time period; smoothing the predicted quality risk values for all time periods to obtain smoothed predicted quality risk values, and generating a risk prediction trajectory in chronological order based on the smoothed predicted quality risk values for all time periods; if, in the risk prediction trajectory, there are more than k consecutive time periods where the smoothed predicted quality risk values are all greater than the corresponding preset risk threshold, then the equipment under test is determined to have quality risks; otherwise, the equipment under test is determined not to have quality risks; where k is a positive integer. Therefore, in this invention, a fusion feature vector is first constructed based on the values of the equipment's production process parameters and environmental parameters. Then, a preset quality detection model is used to initially detect the predicted quality risk value. Finally, by combining time series smoothing and continuous over-limit judgment conditions, the quality risk is determined. This can eliminate misjudgments caused by noise and operating condition fluctuations, effectively reduce false alarms and false negatives, and improve the accuracy of quality risk detection results. Attached Figure Description
[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0017] Figure 1 This is a flowchart illustrating a method for predicting quality risks in equipment manufacturing, provided by an embodiment of the present invention.
[0018] Figure 2 This is a schematic diagram of the structure of a quality risk prediction device for equipment manufacturing provided in an embodiment of the present invention. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] 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 application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.
[0021] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.
[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0023] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.
[0024] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).
[0025] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.
[0026] See Figure 1To address the problem of inaccurate quality risk detection results caused by using a fixed statistical threshold for a single parameter as the judgment standard in existing technologies, an embodiment of the present invention provides a method for predicting quality risks in equipment manufacturing, comprising: Step S101: Obtain the values of the production process parameters and environmental parameters of the equipment under test throughout the entire production process; Specifically, the aforementioned production process parameters are derived from the process quantities and bench test quantities during the manufacturing of the main power grid equipment. Process quantities include, for example, torque values during assembly, pressure changes during sealing and pressure holding, and cleanliness monitoring counts. Bench test quantities include, for example, leakage rate during pressure resistance testing and discharge magnitude during partial discharge testing. The aforementioned environmental parameters include external conditions such as temperature, humidity, and vibration intensity at the workstation during production. All collected data is linked to a timeline to ensure traceability of the equipment at each stage of production.
[0027] Step S102: Based on the values of the above-mentioned production process parameters and environmental parameters, calculate the statistical characteristic quantities of the above-mentioned equipment under test in different time periods; Specifically, the aforementioned statistical characteristics refer to the statistical quantities of production process parameters and environmental parameters within the corresponding time period. These statistical quantities can be the mean, extreme values, median, slope, etc.
[0028] In a preferred embodiment, the calculation of statistical characteristic quantities of the device under test in different time periods based on the values of the production process parameters and environmental parameters includes: After aligning the values of the above production process parameters and environmental parameters by time, the aligned values of the production process parameters and environmental parameters are segmented according to a preset sliding window size to obtain the values of the production process parameters and environmental parameters in different time periods; wherein, one time period corresponds to a set of production process parameter values and a set of environmental parameter values; Specifically, the aforementioned preset sliding window is determined based on the production line cycle time during the production process. Since the production line cycle time may differ when different parameters are acquired, a unified time reference is required.
[0029] For each time period, based on the values of the production process parameters and environmental parameters within that time period, the corresponding statistical characteristic quantities for the production process parameters and environmental parameters are calculated.
[0030] Specifically, the calculated statistical characteristic is used as the equivalent window representative value for that time period.
[0031] Specifically, the process of determining the value represented by the equivalent window is expressed by the following formula: In the formula, This represents the equivalent window value of parameter i within time period j. Represents a timestamp. Indicates the preset sliding window length. This represents the original acquired value of parameter i at time τ (the observation value before interpolation). This indicates that the mean, extreme values, slope, or robust median can be taken, and the parameters are fixed in the feature dictionary to ensure reproducibility across stages.
[0032] In this preferred embodiment, statistical characteristic quantities of the device under test in different time periods are calculated based on the values of production process parameters and environmental parameters within the divided time periods.
[0033] In another preferred embodiment, after obtaining the values of the production process parameters and environmental parameters at different time periods, the method further includes: For each time period, determine the missing values of production process parameters and environmental parameters within that time period; Perform consistent imputation on all missing values.
[0034] Specifically, to improve the completeness of the final statistical features, it is necessary to impute missing points within the time period. The imputation process for missing values is achieved using the following formula: In the formula, Indicates that parameter i is in The value after time interpolation This indicates a verified interpolation operator, such as piecewise linear interpolation or forward hold operator, where the piecewise strategy is preferentially enabled at process changeover boundaries to prevent the introduction of spurious trends.
[0035] In this preferred embodiment, before calculating the statistical features, the missing values of each parameter within the time period are imputed to obtain complete data.
[0036] In another preferred embodiment, for each time period, based on the values of the production process parameters and environmental parameters within the current time period, the statistical characteristic quantities corresponding to the production process parameters and environmental parameters for the current time period are calculated, including: For each time period, based on the values of the production process parameters within the current time period, the first mean, first extreme value, first slope, and first median of the production process parameters are calculated. Based on the values of environmental parameters within the current time period, the second mean, second standard deviation, second extreme value, second slope, and second median of the environmental parameters are calculated. The first mean, first extreme value, first slope, or first median mentioned above are used as statistical characteristic quantities of the production process parameters in the corresponding time period. The second mean, second extreme value, second slope, or second median are used as statistical characteristic quantities of environmental parameters for the corresponding time period.
[0037] Specifically, within each time period, each parameter will have several values. By calculating the average, extreme values, slope, and median of these values, the corresponding statistical characteristics of the parameter can be obtained. It should be noted that the specific statistical characteristic to be chosen from the average, extreme values, slope, and median can be determined based on the actual situation.
[0038] In this preferred embodiment, statistical characteristic quantities are obtained by calculating the statistical characteristics corresponding to the production process parameters and environmental parameters within a time period.
[0039] Step S103: For each time period, perform feature fusion operation based on all statistical features of the current time period to obtain the fused feature vector for each time period; Specifically, based on the feature dictionary, all statistical feature quantities are mapped to process quantities (i.e., process quantities in the production process parameters), test quantities (i.e., bench test quantities in the production process parameters), and environmental quantities (i.e., values of environmental parameters) according to production process parameters and environmental parameters, and then fused in a unified space: In the formula, This represents the fused feature vector corresponding to time period j. This represents the process of concatenating and robustly transforming the various sub-vectors (i.e., process variables, experimental variables, and environmental variables). This represents the process quantity corresponding to time period j. This represents the number of tests corresponding to time period j. This represents the environmental quantity corresponding to time period j.
[0040] Specifically, the above-mentioned word vector concatenation process includes quantile clipping of the corresponding dimensionless deviations and Huber transformation of the extrema. This concatenation process can improve resistance to abnormal interference. The aforementioned dimensionless deviations can be calculated using the following formula: In the formula, This represents the standardized value of parameter i over time period j. This represents the value of parameter i within time period j. This represents the mean of parameter i. This represents the standard deviation of parameter i. This represents the dimensionless deviation of parameter i over time period j.
[0041] Preferably, the above standardization process can eliminate dimensional differences and measure relative offset.
[0042] Step S104: Input each fused feature vector into the preset quality detection model to obtain the predicted quality risk value for each time period; Specifically, the aforementioned preset quality detection model is a random forest model, which includes... Tree-based learners (i.e., T trees).
[0043] Specifically, the aforementioned preset quality inspection model obtains the predicted quality risk value using the following formula: In the formula, This represents the predicted quality risk value, where T represents the total number of base learners. Indicates the first A tree.
[0044] In a preferred embodiment, the training of the aforementioned preset quality detection model includes: Acquire the first historical values of production process parameters, the second historical values of environmental parameters, the true quality category labels, and the total number of all samples for several sample devices throughout the entire historical production process; wherein, the aforementioned true quality category labels are used to indicate whether the corresponding sample device has quality risks; Specifically, after uniformly collecting the first and second historical values of sample devices throughout the entire historical production process, a sample data set is generated within a sliding time window: In the formula, Represents a set of sample data. Indicates time window The above sample data obtained by statistical calculation of m parameters eigenvectors, Indicates the length of the time window. Representing sample data timestamp, Representing sample data The corresponding actual quality category label.
[0045] It should be noted that each sample data corresponds to a structured record of the sample device or its key components within a time window.
[0046] Based on the first and second historical values mentioned above, the sample fusion feature vectors of each sample device are constructed. Specifically, by performing missing value repair, window representative value calculation, and feature fusion on the first and second historical values, the above-mentioned sample fusion feature vector is obtained, and each sample data corresponds to a sample fusion feature vector.
[0047] Based on the number of samples corresponding to the above-mentioned true quality category labels and the above-mentioned total number, the label weight factor corresponding to the true quality category label indicating the existence of quality risk and the label weight factor corresponding to the true quality category label indicating the absence of quality risk are calculated. Specifically, since sample data with quality risks is relatively scarce, it can easily lead to data imbalance in the entire sample dataset. Therefore, it is necessary to construct training weights for sample data according to different label categories in order to enhance the influence of a few risky samples in the weighted split and weighted loss calculation of random forest.
[0048] Specifically, the label weight factor is calculated using the following formula: In the formula, Representing sample data The corresponding training weights, Representing sample data The corresponding label weight factor for the actual quality category label. The weight factor representing the category c of the true quality category label (e.g., c=0 indicates no quality risk, c=1 indicates quality risk). This represents the total number mentioned above. This represents the number of samples with the true quality category label set to category c.
[0049] Specifically, for each sample data j', based on its true quality category label... ,from The corresponding label weight factor is selected as the training weight. For example, if the true quality category label corresponding to a certain sample data is "quality risk exists", then its training weight is the label weight factor corresponding to the true quality category label "quality risk exists".
[0050] Using the sample fusion feature vector of the aforementioned sample devices as input, the aforementioned real quality category labels as supervision quantities, and the aforementioned label weight factors as training weights for the corresponding sample devices, the quality detection model to be trained based on the random forest model is trained to obtain the aforementioned preset quality detection model.
[0051] Specifically, the model is trained using sample fusion feature vectors as input and true quality category labels as supervision. During training, the entire sample dataset is first divided into training and validation sets. In the training set, the total number of samples with true quality category labels for each category is counted, label weight factors are calculated, and sample data is weighted according to these factors. Within each node to be split, the weighted frequency corresponding to the true quality category label for each category is calculated based on the training weights. This yields the Gini impurity, and the split that maximizes the reduction in impurity is selected. This process is then repeated to train all trees and ensembled to obtain the final model. Therefore, the difference between this invention and existing technologies lies in embedding the category weight mechanism into the split statistics and training process to adapt to sample data with scarce risk samples and class imbalance.
[0052] Specifically, the weighted frequency and Gini impurity are calculated using the following formula: In the formula, G represents the weighted frequency of category c, and G represents the Gini impurity. Representing sample data The corresponding category, e, represents the subset of samples within the decision tree to be split in the quality inspection model.
[0053] Preferably, by employing sample reweighting during the model training phase, generalization performance is guaranteed even with a small number of risky samples.
[0054] In this preferred embodiment, a preset quality detection model based on random forest is trained by using the first historical values of production process parameters, the second historical values of environmental parameters, the true quality category labels, and the total number of all samples from several sample devices throughout the entire historical production process.
[0055] In another preferred embodiment, before smoothing the predicted quality risk values for all time periods to obtain the smoothed predicted quality risk values, the method further includes: Obtain the preset calibration parameters; The predicted quality risk values for each time period are monotonically calibrated according to preset calibration parameters.
[0056] Specifically, to improve probabilistic interpretability and cross-batch stability, the predicted quality risk values obtained from the preset quality inspection model are monotonically calibrated: In the formula, This represents the calibrated predicted quality risk value corresponding to time period j. Represents a logical function. and This represents the preset calibration parameters obtained by fitting the validation set. Represents the fused feature vector The corresponding predicted quality risk value.
[0057] In this preferred embodiment, the calibrated prediction quality risk value is obtained by calibrating the prediction results output by the model.
[0058] Step S105: Smooth the predicted quality risk values for all time periods to obtain smoothed predicted quality risk values, and generate risk prediction trajectories in chronological order based on the smoothed predicted quality risk values for all time periods. Specifically, in order to suppress occasional noise, the exponential smoothing method is used in this invention to achieve smoothing: In the formula, Indicates time Below, the smoothed predicted quality risk value, This represents the smoothing coefficient, with a value of [0,1].
[0059] Preferably, the aforementioned risk prediction trajectory can be used for early deviation identification and subsequent continuous risk assessment.
[0060] Preferably, traditional methods mostly provide results at the inspection stage, making it difficult to respond promptly to process deviations. The risk prediction trajectory generated by this invention can capture rising inflection points and slow drifts, and then trigger review points and handling suggestions according to rules, so that discovery and intervention are moved to the process and bench testing stages, shortening the closed-loop handling cycle and improving the certainty of work-in-process release.
[0061] Step S106: If, in the above risk prediction trajectory, there are more than k consecutive time periods for smoothed predicted quality risk values that are all greater than the corresponding preset risk threshold, then the above-mentioned device under test is determined to have quality risk; otherwise, the above-mentioned device under test is determined not to have quality risk; where k is a positive integer.
[0062] Specifically, k is used to characterize the minimum number of consecutive time periods required to determine the existence of quality risk. Preferably, k can be 3 in this invention. The reason is that quality risk determination should not be made directly based on the result of exceeding the threshold for a single time period or a very short duration, because the predicted quality risk value of a single or two adjacent time periods may still be affected by instantaneous noise, short-term operating condition fluctuations, and sampling disturbances. This setting can better distinguish between occasional anomalies and persistent anomalies, reducing the false alarm rate while ensuring the timeliness of risk warning. If the value of k is too small, it is easy to misjudge instantaneous fluctuations as risks; if the value of k is too large, it will lead to a lag in risk identification. Therefore, in this embodiment, k is preferably 3.
[0063] Specifically, the system uses smoothed risk and standard boundaries to jointly determine the quality risk of the equipment under test. If the smoothed predicted quality risk value for more than k consecutive time periods is greater than the corresponding preset risk threshold, an alarm will be triggered. In the formula, Indicates an alarm event. Indicates an indicator function, This indicates a preset risk threshold.
[0064] Specifically, the above alarm determination formula means: if in the case of... Within the last k consecutive time intervals ( to All of them have ,but =1, therefore =1; otherwise =0.
[0065] Preferably, once an alarm is triggered, the corresponding dimensionless deviation of the device, the influencing factors and their time location will be simultaneously displayed, along with boundary items such as nominal values, tolerance bands, and capability indices, forming an auditable basis for handling. This chain of evidence and location of influencing factors allows the final judgment to be reviewed and audited, facilitating on-site execution and liability determination.
[0066] Preferably, this invention uses the posterior risk probability (i.e., the predicted quality risk value) output by random forest as the core, and combines time series smoothing and continuous over-limit criteria to form a stable risk trajectory and significance determination, effectively reducing false alarms and false negatives, and maintaining consistency in determination under batch differences and environmental disturbances.
[0067] In a preferred embodiment, determining the aforementioned preset risk threshold includes: After dividing the monotonic calibrated predicted quality risk values into the above time periods, the average value and standard deviation of the monotonic calibrated predicted quality risk values in each time period are calculated. The preset risk threshold for the corresponding time period is calculated based on the above average value and standard deviation.
[0068] Specifically, based on a scrolling window of length L (i.e., the aforementioned time period). The predicted quality risk value after monotonic calibration is used to calculate the mean and standard deviation. A preset risk threshold is then calculated based on the mean and standard deviation. In the formula, This represents the average value. Indicates standard deviation, This represents the risk threshold coefficient, which is set by the target underreporting rate.
[0069] In this preferred embodiment, a preset risk threshold for the corresponding time period is calculated based on the predicted quality risk value after monotonic calibration within the rolling window.
[0070] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.
[0071] like Figure 2 As shown, an embodiment of the present invention provides a quality risk prediction device for equipment manufacturing, comprising: The module includes a data acquisition module, a statistical feature calculation module, a feature fusion module, a risk prediction module, a risk trajectory smoothing module, and a quality prediction module. The aforementioned data acquisition module is used to acquire the values of production process parameters and environmental parameters of the equipment under test throughout the entire production process. The aforementioned statistical feature calculation module is used to calculate the statistical feature quantities of the equipment to be tested in different time periods based on the values of the aforementioned production process parameters and environmental parameters. The aforementioned feature fusion module is used to perform feature fusion operations on all statistical features of the current time period for each time period, so as to obtain the fused feature vector for each time period. The aforementioned risk prediction module is used to input each fused feature vector into a preset quality detection model to obtain the predicted quality risk value for each time period. The aforementioned risk trajectory smoothing module is used to smooth the predicted quality risk values for all time periods to obtain smoothed predicted quality risk values, and to generate risk prediction trajectories in chronological order based on the smoothed predicted quality risk values for all time periods. The aforementioned quality prediction module is used to determine that the device under test has a quality risk if, in the aforementioned risk prediction trajectory, there are more than k consecutive time periods of smoothed predicted quality risk values that are all greater than the corresponding preset risk threshold; otherwise, it is determined that the device under test does not have a quality risk; where k is a positive integer.
[0072] It should be noted that the device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort. The above schematic diagrams are merely examples of a quality risk prediction device for equipment manufacturing and do not constitute a limitation on a quality risk prediction device for equipment manufacturing. It may include more or fewer components than illustrated, or combine certain components, or use different components.
[0073] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.
[0074] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the quality risk prediction method for device manufacturing described in any embodiment of the present invention.
[0075] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the device. The aforementioned terminal devices may be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These devices may include, but are not limited to, processors and memory. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the device, connecting various parts of the device via various interfaces and lines. The aforementioned memory can be used to store the aforementioned computer programs and / or modules. The aforementioned processor implements various functions of the aforementioned device by running or executing the computer programs and / or modules stored in the aforementioned memory, and by calling data stored in the memory. The aforementioned memory may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0076] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.
[0077] Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute the quality risk prediction method for equipment manufacturing described in any embodiment of the present invention.
[0078] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.
[0079] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for predicting quality risks in equipment manufacturing, characterized in that, include: Acquire the values of production process parameters and environmental parameters of the equipment under test throughout the entire production process; Based on the values of the production process parameters and environmental parameters, the statistical characteristic quantities of the equipment under test in different time periods are calculated. For each time period, feature fusion is performed based on all statistical features of the current time period to obtain the fused feature vector for each time period. Each fused feature vector is input into a preset quality detection model to obtain the predicted quality risk value for each time period. The predicted quality risk values for all time periods are smoothed to obtain smoothed predicted quality risk values. Based on the smoothed predicted quality risk values for all time periods, risk prediction trajectories are generated in chronological order. If, in the risk prediction trajectory, there are more than k consecutive time periods for smoothed predicted quality risk values that are all greater than the corresponding preset risk threshold, then the device under test is determined to have quality risk; otherwise, the device under test is determined not to have quality risk; where k is a positive integer.
2. The method for predicting quality risks in equipment manufacturing according to claim 1, characterized in that, The step of calculating the statistical characteristic quantities of the equipment under test in different time periods based on the values of the production process parameters and environmental parameters includes: After aligning the values of the production process parameters and environmental parameters by time, the aligned values of the production process parameters and environmental parameters are segmented according to a preset sliding window size to obtain the values of the production process parameters and environmental parameters in different time periods; wherein, one time period corresponds to a set of production process parameter values and a set of environmental parameter values; For each time period, based on the values of the production process parameters and environmental parameters within that time period, the corresponding statistical characteristic quantities for the production process parameters and environmental parameters are calculated.
3. The method for predicting quality risks in equipment manufacturing according to claim 2, characterized in that, After obtaining the values of production process parameters and environmental parameters at different time periods, the process further includes: For each time period, determine the missing values of production process parameters and environmental parameters within that time period; Perform consistent imputation on all missing values.
4. The method for predicting quality risks in equipment manufacturing according to claim 3, characterized in that, For each time period, based on the values of the production process parameters and environmental parameters within that time period, the statistical characteristic quantities corresponding to the production process parameters and environmental parameters for that time period are calculated, including: For each time period, based on the values of the production process parameters within the current time period, the first mean, first extreme value, first slope, and first median of the production process parameters are calculated. Based on the values of environmental parameters within the current time period, the second mean, second standard deviation, second extreme value, second slope, and second median of the environmental parameters are calculated. The first mean, first extreme value, first slope, or first median are used as statistical characteristic quantities of the production process parameters in the corresponding time period. The second mean, second extreme value, second slope, or second median are used as statistical characteristic quantities of environmental parameters for the corresponding time period.
5. The method for predicting quality risks in equipment manufacturing according to claim 4, characterized in that, The training of the preset quality detection model includes: The process involves obtaining the first historical values of production process parameters, the second historical values of environmental parameters, the true quality category labels, and the total number of all samples for a number of sample devices throughout the entire historical production process; wherein, the true quality category labels are used to indicate whether the corresponding sample device has a quality risk. Based on the first historical value and the second historical value, the sample fusion feature vector of each sample device is constructed; Based on the number of samples corresponding to the true quality category labels and the total number, the label weight factor corresponding to the true quality category label indicating the existence of quality risk and the label weight factor corresponding to the true quality category label indicating the absence of quality risk are calculated. Using the sample fusion feature vector of the sample device as input, the real quality category label as supervision quantity, and the label weight factor as the training weight of the corresponding sample device, the quality detection model to be trained based on the random forest model is trained to obtain the preset quality detection model.
6. The method for predicting quality risks in equipment manufacturing according to claim 5, characterized in that, Before smoothing the predicted quality risk values for all time periods to obtain the smoothed predicted quality risk values, the process also includes: Obtain the preset calibration parameters; The predicted quality risk values for each time period are monotonically calibrated according to preset calibration parameters.
7. The method for predicting quality risks in equipment manufacturing according to claim 6, characterized in that, The determination of the preset risk threshold includes: After dividing the monotonic calibrated predicted quality risk values according to the time period, the average value and standard deviation of the monotonic calibrated predicted quality risk values in each time period are calculated. The preset risk threshold for the corresponding time period is calculated based on the average value and standard deviation.
8. A quality risk prediction device for equipment manufacturing, characterized in that, include: The module includes a data acquisition module, a statistical feature calculation module, a feature fusion module, a risk prediction module, a risk trajectory smoothing module, and a quality prediction module. The data acquisition module is used to acquire the values of production process parameters and environmental parameters of the equipment under test throughout the entire production process; The statistical feature calculation module is used to calculate the statistical feature quantities of the equipment to be tested in different time periods based on the values of the production process parameters and the environmental parameters. The feature fusion module is used to perform feature fusion operation on all statistical features of the current time period for each time period to obtain the fused feature vector for each time period. The risk prediction module is used to input each fused feature vector into a preset quality detection model to obtain the predicted quality risk value for each time period. The risk trajectory smoothing module is used to smooth the predicted quality risk values for all time periods to obtain smoothed predicted quality risk values, and generate risk prediction trajectories in chronological order based on the smoothed predicted quality risk values for all time periods. The quality prediction module is used to determine that the device under test has a quality risk if, in the risk prediction trajectory, there are more than k consecutive time periods of smoothed predicted quality risk values that are all greater than the corresponding preset risk threshold; otherwise, it determines that the device under test does not have a quality risk; where k is a positive integer.
9. A terminal device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a method for predicting quality risks in the production and manufacturing of equipment as described in any one of claims 1 to 7.
10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform a quality risk prediction method for equipment manufacturing as described in any one of claims 1 to 7.