Multi-agent driven multi-algorithm fusion production parameter quality analysis and prediction method

Abstract

This application provides a multi-agent-driven, multi-algorithm fusion method for production parameter quality analysis and prediction. The method includes: acquiring the raw production data of the object; aligning sensor detection data, total defective products, and total finished products according to delay parameters to obtain multiple sets of initial sample production data; performing zeroing detection and incremental calculation on the total defective products and total finished products to obtain quality parameters; adding quality labels to each set of initial sample production data to obtain multiple sets of target sample production data; and obtaining a quality report of the object during the production process based on multiple pre-trained agents and multiple preset data processing algorithms. This application enables data-driven decision-making and intelligent closed-loop control in the production process, significantly reducing defect rates and human error, improving the production efficiency of production equipment, and enhancing the reliability, robustness, and economy of the production process.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing and quality control technology, and more specifically, to a multi-agent-driven, multi-algorithm fusion method for production parameter quality analysis and prediction. Background Technology

[0002] Molded bottles, as important packaging containers widely used in the pharmaceutical, food, and other industries, have their production quality directly affecting the safety and sealing of their contents. Therefore, the requirements for controlling defect rates are extremely stringent. Accurately identifying key influencing factors, predicting potential quality risks, and providing interpretable optimization suggestions from massive amounts of heterogeneous production data has become a core challenge for improving production line yield and operational efficiency.

[0003] Currently, quality analysis for multi-parameter production processes mainly relies on two technical approaches: one is the traditional method based on statistical process control (SPC), such as using X-bar control charts, CPK analysis, or Pearson correlation coefficients to monitor parameter stability and correlation; the other is to introduce machine learning models for predictive analysis, such as using regression models, support vector machines, or deep neural networks to establish the mapping relationship between parameters and defect rates.

[0004] However, existing technologies still have problems such as difficulty in adapting to the differences in cycle time and dynamic transition processes of different product models, large deviations in data alignment, and neglect of the inherent differences in quality distribution under different mold, material, and process conditions. Summary of the Invention

[0005] The purpose of this application is to provide a multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method to address the shortcomings of the prior art. This method solves the problems of existing technologies, such as difficulty in adapting to the differences in cycle time and dynamic transition processes of different product models, large deviations in data alignment, and neglect of the inherent differences in quality distribution under different mold, material, and process conditions.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, one embodiment of this application provides a multi-agent-driven, multi-algorithm fusion method for analyzing and predicting the quality of production parameters, the method comprising: Obtain the original production data of the current model object during the production process. The production process includes multiple production stages. The original production data includes: sensor detection data of each sensor at each time window during the production process, the total number of defective products passing through each production stage, and the total number of products passing through each production stage. Determine the delay parameter of the object, and based on the delay parameter, align the sensor detection data, the total number of defective products, and the total number of finished products to obtain multiple sets of initial sample production data sorted by time; The total number of defective products and the total number of finished products are reset and incrementally calculated to obtain the quality parameters of the object at each time point in the production process. Based on the quality parameters of the object at each time point in the production process, quality labels are added to each group of initial sample production data to obtain multiple groups of target sample production data. Based on multiple pre-trained agents and multiple preset data processing algorithms, the production data of the target sample is analyzed for quality and production is predicted to obtain a quality report of the object in the production process. The quality report includes: quality evaluation information and quality improvement recommendation information.

[0007] Secondly, another embodiment of this application provides a multi-agent driven, multi-algorithm fusion production parameter quality analysis and prediction device, the device comprising: The acquisition module is used to acquire the original production data of the current model object during the production process. The production process includes multiple production stages. The original production data includes: sensor detection data of each sensor in each time window during the production process, the total number of defective products passing through each production stage, and the total number of products passing through each production stage. An alignment module is used to determine the delay parameters of the object, and according to the delay parameters, to align the sensor detection data, the total number of defective products and the total number of products to obtain multiple sets of initial sample production data sorted by time. The detection module is used to perform zeroing detection and incremental calculation on the total number of defective products and the total number of finished products to obtain the quality parameters of the object at each time point in the production process. An addition module is used to add quality labels to each group of initial sample production data based on the quality parameters of the object at each time point in the production process, so as to obtain multiple groups of target sample production data. The analysis module is used to perform quality analysis and production prediction on the target sample production data based on multiple pre-trained agents and multiple preset data processing algorithms, and to obtain a quality report of the object in the production process. The quality report includes: quality evaluation information and quality improvement recommendation information.

[0008] Thirdly, another embodiment of this application provides an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of any of the methods described in the first aspect above.

[0009] Fourthly, another embodiment of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of any of the methods described in the first aspect above.

[0010] The beneficial effects of this application are as follows: By acquiring the original production data of the current model object during the production process, the delay parameters of the object are determined. Based on the delay parameters, the sensor detection data, the total number of defective products, and the total number of finished products are aligned to obtain multiple sets of initial sample production data sorted by time. This achieves preprocessing of the original data collected by each sensor during the production process, fundamentally eliminating the temporal semantic misalignment between heterogeneous sensor data and discrete quality events, and providing a causal, reliable, physically traceable, and model-adaptive data foundation for high-precision quality modeling. Simultaneously, by performing zeroing detection and incremental calculation on the total number of defective products and the total number of finished products, the quality parameters of the object at each time point in the production process are obtained, and based on the... By adding quality labels to the initial sample production data at various time points in the production process, multiple sets of target sample production data are obtained, thus achieving data labeling. This solves the technical bottleneck of physical distortion of quality labels in precision manufacturing. Based on multiple pre-trained agents and preset data processing algorithms, the system performs quality analysis and production prediction on the target sample production data, generating a quality report of the object in the production process. It can perceive, understand, reason, and make suggestions on various production parameters in the production process, thereby realizing data-driven decision-making and intelligent closed-loop control in the production process. This significantly reduces the defect rate and human judgment error, improves the production efficiency of production equipment, and enhances the reliability, robustness, and economy of the production process. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1A flowchart illustrating a multi-agent-driven, multi-algorithm fusion method for production parameter quality analysis and prediction provided in an embodiment of this application; Figure 2 This is a partial schematic diagram of sensor detection data from each sensor at each time window during the production process, provided in the original production data for embodiments of this application. Figure 3 A partial schematic diagram of the total number of defective products passing through each production stage and the total number of products passing through each production stage in the original production data provided for the embodiments of this application. Figure 4 This is a flowchart illustrating the process of obtaining multiple sets of initial sample production data sorted by time in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in the embodiments of this application. Figure 5 This is a flowchart illustrating the process of determining the sensor feature vector of a sensor in each time window in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. Figure 6 This is a flowchart illustrating the process of determining the delay parameters of an object in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. Figure 7 This is a flowchart illustrating the process of obtaining the delay parameters of an object in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment; Figure 8 This is a flowchart illustrating the process of obtaining the quality parameters of an object at various time points during the production process in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. Figure 9 This is a flowchart illustrating the process of obtaining multiple sets of target sample production data in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment; Figure 10 This is a flowchart illustrating the process of determining quality thresholds corresponding to multiple types of quality labels in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment; Figure 11 This is a flowchart illustrating the process of obtaining a quality report of an object during the production process in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment; Figure 12 This is a schematic diagram of the electronic device structure provided in an embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0014] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0015] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0016] Currently, quality analysis for multi-parameter production processes mainly relies on two technical approaches: one is the traditional method based on statistical process control (SPC), such as using X-bar control charts, CPK analysis, or Pearson correlation coefficients to monitor parameter stability and correlation; the other is to introduce machine learning models for predictive analysis, such as using regression models, support vector machines, or deep neural networks to establish the mapping relationship between parameters and defect rates.

[0017] However, existing technologies still have problems such as difficulty in adapting to the differences in cycle time and dynamic transition processes of different product models, large deviations in data alignment, and neglect of the inherent differences in quality distribution under different mold, material, and process conditions.

[0018] Based on the aforementioned problems, this application proposes a multi-agent-driven, multi-algorithm fusion method for production parameter quality analysis and prediction. By acquiring the raw production data of the current model of the object during the production process, the delay parameters of the object are determined. Based on these delay parameters, sensor detection data, the total number of defective products, and the total number of finished products are aligned to obtain multiple sets of initial sample production data sorted by time. This preprocesses the raw data collected by various sensors during the production process. Simultaneously, by zeroing out the total number of defective products and the total number of finished products and performing incremental calculations, the quality parameters of the object at each time point in the production process are obtained. Based on these quality parameters, quality labels are added to each set of initial sample production data to obtain multiple sets of target sample production data, achieving data labeling. Then, based on multiple pre-trained agents and preset data processing algorithms, quality analysis and production prediction are performed on the target sample production data, resulting in a quality report of the object during the production process. This method can perceive, understand, reason, and make suggestions on various production parameters during the production process, thereby achieving data-driven decision-making and intelligent closed-loop control in the production process. This significantly reduces the defect rate and human judgment errors, and improves the production efficiency of production equipment.

[0019] First, the application scenarios of the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in the embodiments of this application will be described.

[0020] It is understood that the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment can be applied to industrial production processes, such as the production process of precision injection molded products like molded bottles. By executing the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment, a production parameter quality intelligent analysis system that can autonomously understand the production scenario, flexibly call upon analysis tools, and comprehensively output decision suggestions can be constructed, realizing a leap from "post-event attribution" to "pre-event prediction and in-event optimization".

[0021] Specifically, the production process of precision injection molded products such as molded bottles usually involves multiple stages (such as raw material preparation, molding, cooling, and packaging), and each stage includes a large number of sensor parameters, such as temperature, pressure, current, and voltage.

[0022] The following describes in detail the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application, using multiple embodiments.

[0023] Figure 1 This is a flowchart illustrating a multi-agent-driven, multi-algorithm fusion method for production parameter quality analysis and prediction provided in an embodiment of this application. (Refer to...) Figure 1As shown, the subject executing this method can be any electronic device with processing capabilities, and the method includes: S101. Obtain the original production data of the current model object during the production process.

[0024] Optionally, the original production data of the current model object during the production process can be obtained.

[0025] The object refers to the product produced during the production process. The production process includes multiple production stages, and the original production data includes: sensor detection data of each sensor at each time window during the production process, the total number of defective products passing through each production stage, and the total number of products passing through each production stage.

[0026] Specifically, each sensor can be installed in different stages of the production line. Each sensor may include temperature sensors, pressure sensors, power sensors, current sensors, voltage sensors, flow sensors, displacement sensors, etc.

[0027] Specifically, each time window refers to multiple discrete time periods in the production process, and the length of each time window can be determined based on the time length of the corresponding production stage. For example, a time window can be 1 second.

[0028] For example, Figure 2 This is a partial schematic diagram of sensor detection data from each sensor at different time windows during the production process, provided in the original production data for embodiments of this application. (Refer to...) Figure 2 As shown, taking a molded bottle as an example, the production stages can include: the production stage corresponding to the main material channel integrated power supply cabinet, the production stage corresponding to the material channel electrode power supply cabinet, the production stage corresponding to the furnace pool, and the production stage corresponding to the liquid flow channel riser power supply cabinet. The sensor detection data of each sensor in each production stage and each time window can include: temperature sensor detection data, power sensor detection data, etc.

[0029] For example, the total number of defective products passing through each production stage refers to the cumulative total number of products judged as defective in each production stage, while the total number of products passing through each production stage refers to the cumulative total number of products judged as defective in each production stage. Figure 3 A partial schematic diagram of the total number of defective products passing through each production stage and the total number of finished products passing through each production stage in the original production data provided in this application embodiment, with reference to... Figure 3 As shown, taking molded bottles as an example, the original production data can include the total number of defective products and the total number of finished products passing through multiple production stages. Specifically, this includes the total number of products passing inspection from the three inspection lines, the total number of defective products, and the accumulation of air bubbles and stones on the bottle body, bottle neck, and bottle bottom. For example, both the total number of defective products and the total number of finished products can be represented as a count sequence that changes over time.

[0030] S102. Determine the delay parameters of the object, and based on the delay parameters, align the sensor detection data, the total number of defective products and the total number of products to obtain multiple sets of initial sample production data sorted by time.

[0031] It is understandable that during the production process, there is a time delay in the recording of parameters for the same product at different production stages, and this delay is related to the product model, which is manifested as different delay constants corresponding to different models.

[0032] Optionally, after obtaining the raw production data, the delay parameters of the object can be determined, and based on the delay parameters, the sensor detection data, the total number of defective products passing through each production stage, and the total number of products passing through each production stage can be aligned to obtain multiple sets of initial sample production data sorted by time.

[0033] For example, for the current model of object, mutual information matching can be performed on sensor detection data, the total number of defective products passing through each production stage, and the total number of products passing through each production stage to obtain mutual information matching results. From the mutual information matching results, the delay that maximizes the cross-stage feature correlation can be determined as the delay parameter for the current model of object.

[0034] Specifically, the object's delay parameter characterizes the systematic time delay between the application of process parameters and the response of quality results under a specific production line configuration, and is used to achieve time alignment of asynchronous data across stages.

[0035] For example, after obtaining the delay parameters of the object, the sensor detection data, the total number of defective products passing through each production stage, and the total number of products passing through each production stage can be matched according to the delay parameters of the object to obtain multiple sets of sample data. The multiple sets of sample data can then be sorted by time to obtain multiple sets of initial sample production data sorted by time.

[0036] Optionally, after obtaining multiple sets of initial sample production data, the data legality of the multiple sets of initial sample production data can be checked and cleaned, including at least: excluding data segments with incomplete records, abnormal range of change, long-term constant, and data segments that do not meet basic consistency constraints after alignment, and retaining traceable marks on the cleaning results for subsequent quality investigation.

[0037] Optionally, the data can be standardized based on the distribution of the production parameters themselves. In this application, standardized parameters can be pre-calculated for different models, and the production data of multiple initial samples can be normalized using these parameters, so that the differences in statistical scales between different models do not mask the actual process variations. Specifically, normalization can include normalizing the mean to 0 and the variance to 1.

[0038] S103. Perform zeroing detection and incremental calculation on the total number of defective products and the total number of finished products to obtain the quality parameters of the object at each time point in the production process.

[0039] It is understandable that since the total number of defective products and the total number of finished products are reset to zero at fixed time points, and since the reset time is often not aligned with the hour and may have slight drift in different shifts, the total number of defective products and the total number of finished products can be reset and incrementally calculated to obtain the quality parameters of the object at each time point in the production process.

[0040] Optionally, the total number of defective products and the total number of products can be reset to zero, and multiple zeroing points can be obtained from the total number of defective products and the total number of products. Incremental calculations can be performed based on each zeroing point, and the quality parameters of the object at each time point in the production process can be obtained based on the incremental calculation results.

[0041] Among them, the quality parameters of the object at various time points in the production process are used to characterize the quality status of the production process. Specifically, the quality parameters may include the defect rate, the pass rate, the defect rate per unit time, and the defect rate per unit output.

[0042] It is understood that the execution order of S102 and S103 is not unique and can be adjusted according to the actual situation.

[0043] S104. Based on the quality parameters of the object at each time point in the production process, add quality labels to the initial sample production data of each group to obtain multiple groups of target sample production data.

[0044] Optionally, after obtaining the quality parameters of the object at each time point in the production process, quality labels at each time point can be generated based on the quality parameters of the object at each time point in the production process, and corresponding quality labels can be added to each group of initial sample production data to obtain multiple groups of target sample production data.

[0045] For example, the quality parameters of the object at each time point in the production process can be judged according to a preset quality threshold to obtain the quality label at each time point, thereby obtaining multiple sets of target sample production data.

[0046] Specifically, the quality threshold can include quality thresholds corresponding to multiple types of quality labels, such as: good product quality threshold range, defective product quality threshold range, and neutral product quality threshold range.

[0047] S105. Based on multiple pre-trained agents and multiple preset data processing algorithms, perform quality analysis and production prediction on the target sample production data to obtain a quality report of the object in the production process.

[0048] Optionally, after obtaining the target sample production data, the target sample production data can be analyzed for quality and production prediction based on multiple pre-trained agents and multiple preset data processing algorithms to obtain quality evaluation information and quality improvement recommendation information, and generate a quality report.

[0049] The data processing algorithms include: Pearson correlation coefficient analysis algorithm, deep learning model prediction algorithm, database retrieval-based data comparison analysis algorithm, and correlation coefficient analysis algorithm based on linear discriminant analysis dimensionality reduction.

[0050] The quality report includes quality evaluation information and quality improvement recommendations. Specifically, the quality evaluation information includes a quantitative assessment and attribution analysis of the product quality status during the current or historical production process, while the quality improvement recommendations are used to guide users or control systems in adjusting parameters.

[0051] In one example, a first intelligent agent can analyze the target sample production data to determine the target data processing algorithm among multiple data processing algorithms. The target data processing algorithm is then used to perform quality analysis on the target sample production data to obtain quality evaluation information. This quality evaluation information is then input into a second intelligent agent, which performs production prediction to obtain quality improvement recommendation information.

[0052] In another example, the first agent can be fine-tuned in advance based on multiple data processing algorithms. The fine-tuned first agent can then perform quality analysis on the target sample production data to obtain quality evaluation information. This quality evaluation information is then input into the second agent, which performs production prediction to obtain quality improvement recommendation information.

[0053] In this embodiment, by acquiring the original production data of the current model object during the production process, the object's delay parameters are determined. Based on the delay parameters, the sensor detection data, the total number of defective products, and the total number of finished products are aligned to obtain multiple sets of initial sample production data sorted by time. This achieves preprocessing of the original data collected by each sensor during the production process, fundamentally eliminating the temporal semantic misalignment between heterogeneous sensor data and discrete quality events, and providing a causal, physically traceable, and model-adaptive data foundation for high-precision quality modeling. Simultaneously, by performing zeroing detection and incremental calculations on the total number of defective products and the total number of finished products, the quality parameters of the object at each time point in the production process are obtained, and based on the object... By adding quality labels to the initial sample production data at various time points in the production process, multiple sets of target sample production data are obtained, thus achieving data labeling. This solves the technical bottleneck of physical distortion of quality labels in precision manufacturing. Based on multiple pre-trained agents and preset data processing algorithms, the target sample production data is used for quality analysis and production prediction, resulting in a quality report of the object in the production process. It can perceive, understand, reason, and make suggestions on various production parameters in the production process, thereby realizing data-driven decision-making and intelligent closed-loop control in the production process. This significantly reduces the defect rate and human judgment error, improves the production efficiency of production equipment, and enhances the reliability, robustness, and economy of the production process.

[0054] In one possible implementation, Figure 4 This is a flowchart illustrating the process of obtaining multiple sets of initial sample production data sorted by time in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 4 As shown, in step S102 above, the sensor detection data, the total number of defective products, and the total number of finished products are aligned according to the delay parameter to obtain multiple sets of initial sample production data sorted by time, including: S401. Based on the sensor detection data, determine the sensor feature vector for each time window.

[0055] Optionally, after obtaining the sensor detection data, the sensor detection data can be calculated to obtain the sensor feature vector of the sensor in each time window.

[0056] For example, taking the sensor detection data of a sensor in each time window during the production process as an example, the sensor detection data corresponding to the sensor can be adaptively averaged according to each time window, and the sensor's mean, variance, quantile, slope, steady-state ratio and other features in each time window can be calculated as the sensor feature vector corresponding to the sensor.

[0057] S402. Based on the delay parameter, the sensor feature vector, sensor detection data, total number of defective products and total number of finished products are aligned to obtain multiple sets of initial sample production data sorted by time.

[0058] Optionally, after obtaining the delay parameter, the sensor feature vector can be combined with the sensor detection data to obtain sensor combination data. The total quantity of defective products can be combined with the total quantity of products to obtain total quantity data. According to the delay parameter, the corresponding data can be extracted from the sensor combination data and the total quantity data as an initial sample production data. The initial sample production data can be sorted by time to obtain multiple sets of initial sample production data sorted by time.

[0059] In one possible implementation, Figure 5 This is a flowchart illustrating the process of determining the sensor feature vectors of sensors in each time window in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 5 As shown, in step S301 above, the sensor feature vector for each time window is determined based on the sensor detection data, including: S501. Perform steady-state detection and segmentation on the sensor detection data to obtain multiple transition segments and one steady-state segment corresponding to the sensor.

[0060] Optionally, steady-state detection and segmentation can be performed on the sensor detection data, dividing the sensor detection data into steady-state segments and transition segments.

[0061] For example, by using the Cumulative Sum Algorithm (CUSUM) to detect changes and determine steady-state conditions in the sensor data, multiple transition segments and one steady-state segment corresponding to the sensor are obtained. Specifically, there can be two transition segments and one steady-state segment.

[0062] S502. Perform first dynamic window generation processing on the sensor detection data of the sensor in each transition section to obtain multiple first time windows corresponding to the sensor in each transition section, and determine the sensor feature vector of the sensor in each first time window.

[0063] Optionally, the sensor detection data in each transition segment is processed by a first dynamic window generation process to obtain multiple first time windows corresponding to each transition segment. The first dynamic window generation process is a short-time dynamic window generation process, thereby preserving the process changes of the sensor within each transition segment. For example, the short-time dynamic window can be a few seconds or tens of seconds.

[0064] Optionally, after obtaining multiple first time windows, the sensor's mean, variance, quantile, slope, steady-state ratio, and other characteristics in each first time window are calculated and used as the sensor feature vector in each first time window.

[0065] S503. Perform second dynamic window generation processing on the sensor detection data in the steady-state section to obtain multiple second time windows corresponding to the sensor in the steady-state section, and determine the sensor feature vector of the sensor in each second time window.

[0066] Optionally, a second dynamic window generation process is applied to the sensor detection data in each steady-state segment to obtain multiple second time windows corresponding to each steady-state segment. This second dynamic window generation process is a long-term dynamic window generation process, thereby reducing noise from the sensor within the steady-state segment. For example, the long-term dynamic window can be several minutes or even longer.

[0067] Optionally, after obtaining multiple second time windows, the sensor's mean, variance, quantile, slope, steady-state ratio, and other characteristics in each second time window are used as the sensor feature vector in each second time window.

[0068] By identifying steady-state and transition segments, and generating short-term dynamic windows for transition segments and long-term dynamic windows for steady-state segments, we can maintain the data compression effect while avoiding the homogenization of data in key transition segments with long time windows. This also ensures that different types of objects can obtain observable feature expressions under different time cycles, guaranteeing the accuracy and authenticity of the obtained sensor feature vectors.

[0069] In one possible implementation, Figure 6 This is a flowchart illustrating the process of determining the delay parameters of an object in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 6 As shown, the delay parameters for determining the object in S102 above include: S601. Based on the object, determine whether the object's delay parameters exist from the preset delay repository.

[0070] Optionally, you can search the preset delay repository to see if the delay parameters for the current model exist.

[0071] The delay repository can store objects of various models and their corresponding delay parameters.

[0072] S602. If so, read the object's delay parameters from the delay repository.

[0073] Optionally, if the delay parameters for an object exist in the delay repository, the delay parameters for the current model of the object are read from the delay repository.

[0074] S603. If not, obtain the preset delay range of the object, and within the preset delay range of the object, perform mutual information matching based on sensor detection data, total number of defective products and total number of products to obtain the delay parameters of the object.

[0075] Optionally, if the delay parameters of an object are not present in the delay repository, the preset delay range of the object can be obtained, and within the preset delay range of the object, mutual information matching is performed based on sensor detection data, the total number of defective products and the total number of products to obtain the delay parameters of the object.

[0076] The preset delay range of an object refers to the maximum acceptable delay range, which can be pre-configured by the user.

[0077] For example, within a preset delay range of the object, the sensor detection data, the total number of defective products and the total number of products can be traversed and mutual information matching can be performed to obtain the mutual information matching result. Based on the mutual information matching result, the object's delay parameter can be obtained.

[0078] Optionally, the delay parameters of the current model of the object can also be stored in the delay repository.

[0079] In one possible implementation, Figure 7 This is a flowchart illustrating the process of obtaining the delay parameters of an object in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 7 As shown, in S603 above, within the preset delay range of the object, mutual information matching is performed based on sensor detection data, the total number of defective products passed, and the total number of finished products passed to obtain the object's delay parameters, including: S701. Determine the current processing data from the sensor detection data, the total number of defective products, and the total number of finished products.

[0080] Optionally, the current processing data can be determined from sensor detection data, total defective products, and total products in a preset order.

[0081] The current data being processed includes: current upstream data and the current downstream data corresponding to the current upstream data.

[0082] For example, taking a production process that includes two production stages as an example, when the current upstream data is the sensor detection data of the first production stage, the current downstream data corresponding to the current upstream data is the sensor detection data of the second production stage; when the current upstream data is the sensor detection data of the second production stage, the current downstream data corresponding to the current upstream data is the total number of defective products passing through each production stage and the total number of products passing through each production stage.

[0083] Specifically, when the current upstream data is sensor detection data from the first production stage, the current upstream data can be detection data from at least one sensor among the sensors in the first production stage. The corresponding current downstream data is sensor detection data from the second production stage, and the current downstream data can also be detection data from at least one sensor among the sensors in the second production stage. For example, if the current upstream data is pressure sensor detection data from the first production stage, the corresponding current downstream data could be temperature sensor detection data, pressure sensor detection data, or speed sensor detection data, etc.

[0084] For example, the current model can be... Current upstream data Compared with current downstream data Window aggregation yields a window sequence. .in, This refers to the current moment.

[0085] S702. Based on the current upstream data, the current downstream data, and the preset delay range, calculate the set of candidate data pairs.

[0086] Optionally, a set of candidate data pairs can be calculated based on the current upstream data, the current downstream data, and a preset delay range.

[0087] For example, multiple candidate data pairs are obtained by combining the current upstream data and the current downstream data according to a preset delay range, thereby obtaining a candidate data pair set.

[0088] For example, a set of candidate data pairs .in, A candidate data pair includes an upstream data point in the current upstream data and a downstream data point in the current downstream data, and the upstream data and the downstream data satisfy a preset delay range.

[0089] S703. Perform mutual information matching on the candidate data set to obtain the mutual information parameters of each candidate data pair in the candidate data pair set, and determine the candidate delay parameters corresponding to the current processing data based on the mutual information parameters of each candidate data pair.

[0090] Optionally, mutual information matching can be performed on each candidate data pair in the candidate data pair set to calculate the mutual information parameter of each candidate data pair.

[0091] For example, joint distribution can be used. With marginal distribution Calculate each candidate data pair mutual information parameters :

[0092] in, This is the current model. For joint distribution, It is distributed on the periphery. For the current upstream data in the candidate data pair, This refers to the current downstream data in the candidate data pair.

[0093] Optionally, after obtaining each candidate data pair mutual information parameters Then, the candidate data can be analyzed. mutual information parameters Determine the optimal delay This serves as the candidate delay parameter corresponding to the current data being processed.

[0094] Specifically, optimal delay .

[0095] Optionally, after obtaining the candidate delay parameters corresponding to the current processing data, the workstation sequence and output consistency can be used for constraint verification, that is, the upstream data and downstream data are consistent within a reasonable range within the same time window. If the verification fails, the delay parameters will be re-determined.

[0096] For example, after obtaining the candidate delay parameters corresponding to the current processing data, the yield difference error after alignment is calculated.

[0097] Specifically, set up windows Incremental upstream data is Downstream data increment is .

[0098] Get alignment parameters .

[0099] And summed up as consistency error .

[0100] For example, a reasonable range constraint could be: given an allowable deviation threshold. ,like If the delay parameter is inconsistent, it needs to be re-determined.

[0101] Specifically, the process of re-determining the delay parameters includes: performing a local re-evaluation near the optimal delay. For details, please refer to: .

[0102] Based on this, perform a consistency check again. If it still fails, output a "low confidence alignment" flag for subsequent steps to reduce the weight of the batch of windows or remove them.

[0103] S704. Determine the delay parameters of the object based on the candidate delay parameters corresponding to all the processed data.

[0104] Optionally, after obtaining all the candidate delay parameters corresponding to the processed data, the candidate delay parameters corresponding to all the processed data can be used as the delay parameters of the object.

[0105] By analyzing sensor detection data, total defective products, and total products, the current processing data is determined, along with the corresponding candidate delay parameters. Based on these candidate delay parameters, the delay parameters of the object are determined, enabling the identification of delay parameters for each object model. This allows for accurate capture of cross-stage dynamic response delays and highly sensitive detection of nonlinear correlations.

[0106] In one possible implementation, the quality parameters include: defect rate; Figure 8 This is a flowchart illustrating the process of obtaining the quality parameters of an object at various time points during the production process in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 8 As shown, in S103 above, the total number of defective products and the total number of finished products are reset and incrementally calculated to obtain the quality parameters of the object at each time point in the production process, including: S801. According to the preset time interval, the total number of defective products is cleared and detected, multiple first clearing points are identified in the total number of defective products, and the total number of defective products is divided into multiple first sub-count data according to each first clearing point.

[0107] Optionally, according to a preset time interval, the total number of defective products passing through is scanned within the preset time interval, and the moment when the total number of defective products passing through suddenly jumps from an increasing mode to a low value (or close to 0 value) is identified to obtain the first zeroing point.

[0108] For example, the preset time interval can be an hourly interval. The total amount of defective products passing through can be scanned within the hourly interval to identify the moment when the total amount of defective products suddenly jumps from an increasing mode to a low value (or close to 0 value), thus obtaining multiple first zeroing points.

[0109] For example, within an hourly interval, based on a first zeroing point, the defective products within the hourly interval are divided into multiple first sub-count data by total quantity.

[0110] S802. According to the preset time interval, the total number of products passing through is reset to zero, multiple second reset points are identified in the total number of products passing through, and the total number of products passing through is divided into multiple second sub-count data according to each second reset point.

[0111] Optionally, according to a preset time interval, the total amount of products passing through is scanned within the preset time interval, and the moment when the total amount of products passing through suddenly jumps from an increasing mode to a low value (or close to 0 value) is identified to obtain the second zeroing point.

[0112] For example, the preset time interval can be an hourly interval. The total amount of products passing through can be scanned within the hourly interval to identify the moment when the total amount of products passing through suddenly jumps from an increasing mode to a low value (or close to 0 value), thus obtaining multiple second zeroing points.

[0113] For example, within an hourly interval, according to a second zeroing point, the products within the hourly interval are divided into multiple second sub-count data based on the total amount.

[0114] S803. Based on each first sub-count data, determine the defective product throughput increment for each first sub-count data, and based on the defective product throughput increment for each first sub-count data, determine the total defective product throughput increment within a preset time interval.

[0115] Optionally, after obtaining the first sub-count data within the hour interval, the defective product throughput increment of each first sub-count data is calculated, and the defective product throughput increments of each first sub-count data are summed to obtain the total defective product throughput increment within the preset time interval.

[0116] For example, for each first sub-count data, the difference between the final value and the initial value of the first sub-count data is calculated, and the difference is used as the increment for the defect of the first sub-count data.

[0117] For example, after obtaining the defective product throughput increment of all first sub-count data, the defective product throughput increment of each first sub-count data can be summed to obtain the total defective product throughput increment within a preset time interval.

[0118] S804. Based on each second sub-count data, determine the product throughput increment for each second sub-count data, and based on the product throughput increment for each second sub-count data, determine the total product throughput increment within the preset time interval.

[0119] Optionally, after obtaining the second sub-count data within the hour interval, the product throughput increment of each second sub-count data is calculated, and the product throughput increments of each second sub-count data are summed to obtain the total product throughput increment within the preset time interval.

[0120] For example, for each second sub-count data, the difference between the final value and the initial value of the second sub-count data is calculated, and the product of the second sub-count data is incremented.

[0121] For example, after obtaining the product pass increments of all second sub-count data, the product pass increments of each second sub-count data can be summed to obtain the total product pass increment within a preset time interval.

[0122] S805. Based on the total incremental amount of defective products within a preset time interval and the total incremental amount of products within a preset time interval, determine the defect rate of the object at each time point within the preset time interval.

[0123] Optionally, the ratio of the total increase in defective products within a preset time interval to the total increase in product output within a preset time interval can be calculated, and the obtained ratio can be used as the defect rate of the object at each time point within the preset time interval.

[0124] By resetting the total number of defective products and the total number of products passing through within a preset time interval, and calculating the total increment of defective products and the total increment of products passing through, the defect rate of the object at each time point within the preset time interval can be determined. This can accurately locate all resetting points, completely eliminate false jumps and noise caused by counter reset, and ensure the physical authenticity of quality parameter calculation. At the same time, it can also realize quality measurement under dynamic time windows, adapt to cycle fluctuations and non-steady-state processes, and improve the reliability of quality parameter calculation.

[0125] In one possible implementation, Figure 9 This is a flowchart illustrating the process of obtaining multiple sets of target sample production data in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 9 As shown, in step S104 above, quality labels are added to each group of initial sample production data based on the quality parameters of the object at each time point in the production process, resulting in multiple groups of target sample production data, including: S901. Based on the quality parameters of the object at various time points during the production process, determine the quality thresholds corresponding to multiple types of quality labels.

[0126] Optionally, data processing can be performed based on the quality parameters of the object at various points in time during the production process to dynamically generate quality thresholds corresponding to multiple types of quality labels.

[0127] In one example, the quality parameters of an object at various points in time during the production process can be input into a pre-trained data analysis model to generate quality thresholds corresponding to multiple types of quality labels.

[0128] S902. Based on the defect rate of the object at each time point in the production process and the quality threshold corresponding to each quality label, add quality labels to multiple sets of initial sample production data to obtain target sample production data.

[0129] Optionally, after obtaining the quality threshold corresponding to each quality label, the defect rate of the object at each time point in the production process can be judged based on the quality threshold corresponding to each quality label, the quality label of the object at each time point in the production process can be obtained, and the quality label can be added to the production data of each group of initial sample to obtain the target sample production data.

[0130] Optionally, the defect rate at each time point can be added to the production data of each target sample group.

[0131] By analyzing the quality parameters of an object at various time points during the production process, the quality thresholds corresponding to multiple types of quality labels are determined. Based on the defect rate of the object at various time points during the production process and the quality thresholds corresponding to each quality label, quality labels are added to multiple sets of initial sample production data to obtain target sample production data. This method can achieve automated setting of quality thresholds while taking into account training usability and separability.

[0132] In one possible implementation, Figure 10 This is a flowchart illustrating the process of determining quality thresholds corresponding to multiple types of quality labels in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 10 As shown, in S902 above, based on the quality parameters of the object at various time points during the production process, the quality thresholds corresponding to multiple types of quality labels are determined, including: S1001. Sort the objects according to their quality parameters at each time point in the production process to obtain a sorted sequence of objects, and generate multiple candidate threshold pairs based on the sorted sequence of objects.

[0133] Optionally, the objects are sorted according to their quality parameters at different points in time during the production process to obtain a sorted sequence of the defect rate of the objects.

[0134] Optionally, after obtaining the sorted sequence of the defect rate of the objects, candidate threshold pairs can be generated from the sorted sequence of the objects according to the preset quality threshold range of each type of quality label.

[0135] The candidate threshold pairs include a good product threshold and a defective product threshold. That is, when the defect rate is greater than 0 and less than the defective product threshold, the quality label is defective; when the defect rate is greater than or equal to the defective product threshold and less than or equal to the good product threshold, the quality label is neutral; when the defect rate is greater than the good product threshold, the quality label is good.

[0136] S1002. Based on the preset sample size constraints, each candidate threshold pair is screened to obtain multiple selectable threshold pairs.

[0137] Optionally, after obtaining each candidate threshold pair, a good product sample set, a defective product sample set, and a neutral product sample set corresponding to each candidate threshold pair can be generated based on each candidate threshold pair. According to the preset sample size constraint, the number of good product sample sets, defective product sample sets, and neutral product sample sets corresponding to each candidate threshold pair can be filtered to obtain multiple optional threshold pairs.

[0138] For example, if the number of good samples in the good sample set corresponding to a certain candidate threshold does not meet the preset sample size constraint, then the candidate threshold will be eliminated.

[0139] S1003. Calculate the inter-class divergence and intra-class divergence corresponding to each optional threshold pair.

[0140] Optionally, after obtaining each optional threshold pair, the inter-class divergence and intra-class divergence corresponding to each optional threshold pair can be calculated.

[0141] For example, taking an optional threshold pair as an example, the optional threshold pair includes a good product threshold. and defect threshold The inter-class divergence of the optional threshold pair can be the inter-class mean of the optional threshold pair, including: the first inter-class mean. and the mean between the second category Specifically, it can be:

[0142] in, Let G be the production parameter combination corresponding to the optional threshold pair, and B be the number of good products corresponding to the optional threshold pair.

[0143] For example, the intra-class divergence of the optional threshold pair for:

[0144] S1004. Based on the inter-class divergence and intra-class divergence of each optional threshold pair, determine the quality threshold corresponding to multiple types of quality labels.

[0145] Optionally, after obtaining the inter-class divergence and intra-class divergence corresponding to each optional threshold pair, the quality thresholds corresponding to multiple types of quality labels can be calculated.

[0146] For example, the intra-class scatter can be assessed first. Regularization is performed to obtain the regularized intra-class divergence. This avoids collinearity or pathological conditions, specifically:

[0147] in, The preset regularization threshold is used to control the degree of regularization. Regularization terms are used to prevent ill-conditioned combinations of production parameters within a class. Optional regularization terms include L1 regularization and L2 regularization.

[0148] For example, based on the inter-class mean Inter-class mean and the intra-class divergence after regularization The feasibility score was calculated. The details are as follows:

[0149] For example, the optional threshold pairs are combined with the corresponding production parameters. Scaling to , in the above formula Change to .

[0150] For example, after obtaining the feasibility scores of each optional threshold pair, the quality thresholds corresponding to multiple types of quality labels are determined from each optional threshold pair according to the preset threshold pair selection rules. The preset threshold for the selection rule can be referenced in the following formula:

[0151] in, It is a set consisting of each pair of optional thresholds.

[0152] Optionally, to avoid selecting extremes... We can also introduce a target function with a penalty, as shown below:

[0153] in, To achieve the desired neutral product ratio, The preset threshold for penalty items is used to control the degree of effect of the penalty items. G represents the number of neutral products corresponding to the optional threshold pair, G represents the number of good products corresponding to the optional threshold pair, and B represents the number of defective products corresponding to the optional threshold pair.

[0154] By determining each optional threshold pair and its corresponding inter-class and intra-class divergence, and using the inter-class and intra-class divergence of each optional threshold pair, quality thresholds for multiple types of quality labels are determined. This allows the system to automatically learn quality thresholds from current production data, enabling them to adaptively adjust to product models and real-time operating conditions. This improves classification accuracy, eliminates reliance on human experience, and enhances the system's generalization ability and robustness. Furthermore, optimization of inter-class and intra-class divergence achieves optimal separability.

[0155] In one possible implementation, Figure 11 This is a flowchart illustrating the process of obtaining a quality report of an object during the production process in the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method provided in this application embodiment. (Refer to...) Figure 11 As shown, in S105 above, based on multiple pre-trained agents and multiple preset data processing algorithms, quality analysis and production prediction are performed on the target sample production data to obtain a quality report of the object during the production process, including: S1101. Based on the target sample production data and the first intelligent agent obtained through pre-training, screen multiple data processing algorithms to determine the target data processing algorithm.

[0156] Optionally, the target sample production data can be input into a pre-trained first agent, which then makes decisions based on the target sample production data, filters multiple data processing algorithms, and determines the target data processing algorithm.

[0157] For example, the target sample production data and the pre-constructed first prompt word can be input into the first intelligent agent, which can then filter multiple data processing algorithms to obtain the target data processing algorithm.

[0158] Specifically, the first prompt could include: "You are a senior molding bottle production line analysis expert who can make decisions on algorithm calls based on the characteristics of production data and user needs, so that the corresponding algorithm outputs relevant results that adapt to the current data format and meet the user's exact requirements. The following is the relevant information."

[0159] Production data information: Overall production data, such as data volume, ratio of good to defective products. Representative examples of production data, such as the meaning of each parameter and the range of each parameter.

[0160] User input: ... Algorithm library: Algorithm 1, Algorithm name, Algorithm function, Applicable scenarios; Algorithm 2, Algorithm name, Algorithm function, Applicable scenarios; ... The result format requires only the algorithm number to be output. If multiple algorithms need to be called, separate the different algorithm numbers with commas. Please note that the analysis process and algorithm name do not need to be output. S1102. Based on the target data processing algorithm, perform quality analysis on the target sample production data to obtain the quality evaluation information corresponding to the target sample production data.

[0161] Optionally, after obtaining the target data processing algorithm, the target sample production data can be analyzed using the target data processing algorithm to obtain the quality evaluation information corresponding to the target sample production data.

[0162] In one example, using the Pearson correlation coefficient analysis algorithm as the target data processing algorithm, the Pearson correlation coefficient analysis algorithm can calculate the Pearson product-moment correlation coefficient of each production parameter with respect to changes in the inspection machine parameter. The Pearson correlation coefficient between two variables is defined as the quotient of the covariance and standard deviation between the two variables, as shown below:

[0163] in, , and These are samples The standard score, sample mean, and sample standard deviation.

[0164] Pearson correlation coefficient analysis algorithm is not only used to set a certain production parameter as a variable The defect rate is set as a variable. The Pearson correlation coefficient analysis is then performed. Building upon this, the "correlation analysis" is transformed from static correlation to correlation with alignment. Specifically, the time delay corresponding to the maximum absolute value of the Pearson correlation coefficient can be calculated for each candidate data pair given in S703. The corresponding improved formula is as follows: .

[0165] In another example, taking the target data processing algorithm as a deep learning model prediction algorithm, the deep learning model prediction algorithm may include: constructing a deep neural network to predict the defect rate under different parameter combinations. This neural network is a three-layer multilayer perceptron. The activation functions of the first two layers are linear rectified functions, and the activation function of the last layer is an sigmoid growth curve function. The loss function is the mean squared error loss function. The dimensions of the two intermediate hidden layers are set to 128 and 64, respectively, the number of training epochs is 500, and the batch size is 32. After training, by inputting the new production parameter combination after normalization into the deep model, the corresponding defect rate prediction can be output.

[0166] Building upon this, a conditional multilayer perceptron based on product type can be added to the neural network. This network can be trained using a shared backbone (neural network modules except the last layer) and multiple model prediction heads (the last neural network module). This allows for the sharing of common features among data while also distinguishing production differences between different models. Furthermore, for input combinations that do not meet production constraints, feasible region projection is performed before prediction to avoid meaningless predictions for combinations of unproducible parameters.

[0167] In another example, taking a database-retrieval-based data comparison and analysis algorithm as the target data processing algorithm, the idea behind this algorithm is to analyze how to adjust the production parameter set of a defective product to the production parameter set of a good product at the lowest cost, facilitating subsequent targeted parameter adjustments. This is done within a good product feature space database. In this study, a high-dimensional index KD-tree is used to retrieve data related to the product under constraints. Find the set of parameters of the nearest good product cluster. Then calculate the vector difference from A to the center point of the good product cluster. Define the cost function. ,in To adjust the difficulty level. Finally, for The components in the calculation are sorted in descending order, and the parameters whose adjustment range crosses the threshold or contribute the most to the cost function are identified as key sensitive parameters that affect quality.

[0168] In another example, taking the target data processing algorithm as a correlation coefficient analysis algorithm based on linear discriminant analysis (LDA) for dimensionality reduction, this algorithm extends Pearson correlation coefficient analysis from a single production parameter to a weighted average of multiple production parameters. To extend to multiple production parameters, the weighting of these parameters must first be determined. This weighting is determined using the LDA method. Since LDA is a supervised projection method, each piece of production data must be classified.

[0169] Building upon this, to address the common problems of "collinearity" and "singularity of the covariance matrix due to small samples" in industrial data, a method for handling the within-class scatter matrix can be introduced. The regularization term to be modified, i.e. .in, The regularization coefficient is . It is the identity matrix. This correction ensures the projection direction. The solution demonstrates stability in high-dimensional, small-sample scenarios, avoiding weight distortion caused by individual noise parameters. Projection vector The optimal solution construction algorithm determines the optimal weighted vector for multiple production parameters by solving the problem of maximizing the generalized Rayleigh quotient. The problem is transformed into an eigenvalue decomposition problem using the Lagrange multiplier method. The resulting eigenvectors... This refers to the contribution weight of each production parameter in the classification. This set of weights not only reflects the fluctuation of a single parameter, but also the degree of contribution of the coupling of multiple parameters to the "good / defective" classification.

[0170] For example, the quality evaluation information corresponding to the target sample production data can also be different when the target data processing algorithm is different. For the Pearson correlation coefficient analysis algorithm, it can provide the linear correlation coefficient value of each production parameter with respect to the defect rate. The value ranges from [-1, 1], and the corresponding key parameters can be given according to the high linear relationship. For the deep learning algorithm, it can predict the defect rate of the new combination of production parameters. The value ranges from [0, 1]. For the data comparison analysis algorithm based on database retrieval, it can provide the adjustment range of each parameter when the new production parameters are adjusted to the good product production parameters, and the corresponding key parameters can be given according to the adjustment probability. For the correlation coefficient analysis algorithm based on linear discriminant analysis dimensionality reduction, it can provide the linear correlation after the combination of each production parameter. The value ranges from [-1, 1], and the corresponding key parameters can be given according to the parameter weights of the combination.

[0171] S1103. Based on the quality evaluation information, the target data processing algorithm, the target sample production data, and the pre-trained second agent, determine the quality improvement recommendation information.

[0172] Optionally, after obtaining the quality evaluation information, the quality evaluation information, the target data processing algorithm, and the target sample production data can be input into a pre-trained second agent, which can then infer quality improvement recommendation information.

[0173] For example, quality evaluation information, target data processing algorithm, target sample production data, and second prompt words can be input into a pre-trained second agent, which can then infer quality improvement recommendation information.

[0174] Specifically, the second cue word may include: "You are a senior algorithm analysis expert for molded bottle production lines, capable of comprehensively analyzing production data based on the results of various analysis algorithms to provide users with diverse and accurate auxiliary analysis results. The following is the relevant information."

[0175] Production data information: Overall production data, such as data volume, ratio of good to defective products. Representative examples of production data, such as the meaning of each parameter and the range of each parameter.

[0176] User input: ... Algorithm Results: Algorithm ID used, and corresponding results; ... Please thoroughly understand and analyze the results of each algorithm, and, in conjunction with current production data, formulate possible improvement strategies and comprehensive solutions. By using multiple pre-trained agents and pre-set data processing algorithms, the system performs quality analysis and production prediction on the target sample production data, and obtains a quality report of the object in the production process. It can limit the large model to a structured extractor of intent and constraints, and realize reproducible and auditable automated algorithm invocation decision-making for subsequent automatic execution by the system.

[0177] Based on the same inventive concept, this application also provides a multi-agent-driven multi-algorithm fusion production parameter quality analysis and prediction device corresponding to the multi-agent-driven multi-algorithm fusion production parameter quality analysis and prediction method. Since the principle of the device in this application is similar to the multi-agent-driven multi-algorithm fusion production parameter quality analysis and prediction method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0178] The device includes: an acquisition module, an alignment module, a detection module, an addition module, and an analysis module; The acquisition module is used to acquire the original production data of the current model object during the production process. The production process includes multiple production stages. The original production data includes: sensor detection data of each sensor in each time window during the production process, the total number of defective products passing through each production stage, and the total number of products passing through each production stage. The alignment module is used to determine the delay parameters of the object, and based on the delay parameters, to align the sensor detection data, the total number of defective products and the total number of finished products to obtain multiple sets of initial sample production data sorted by time. The detection module is used to perform zeroing detection and incremental calculation on the total number of defective products and the total number of finished products to obtain the quality parameters of the object at various time points in the production process. The module adds quality labels to each set of initial sample production data based on the quality parameters of the object at each time point in the production process, thereby obtaining multiple sets of target sample production data. The analysis module is used to perform quality analysis and production prediction on the target sample production data based on multiple pre-trained agents and multiple preset data processing algorithms, and to obtain a quality report of the object in the production process. The quality report includes: quality evaluation information and quality improvement recommendation information.

[0179] In one possible implementation, the alignment module is specifically used for: Based on the sensor detection data, determine the sensor feature vector for each time window; Based on the delay parameter, the sensor feature vector, sensor detection data, total number of defective products, and total number of finished products are aligned to obtain multiple sets of initial sample production data sorted by time.

[0180] In one possible implementation, the alignment module is specifically used for: Steady-state detection and segmentation are performed on the sensor detection data to obtain multiple transition segments and one steady-state segment corresponding to the sensor. The sensor detection data of the sensor in each transition section is processed by the first dynamic window generation process to obtain multiple first time windows corresponding to the sensor in each transition section, and the sensor feature vector of the sensor in each first time window is determined. The sensor detection data in the steady-state region is processed by a second dynamic window generation process to obtain multiple second time windows corresponding to the steady-state region, and the sensor feature vector of the sensor in each second time window is determined.

[0181] In one possible implementation, the alignment module is specifically used for: Based on the object, determine whether the object's delay parameters exist from the preset delay repository; If so, read the object's delay parameters from the delay repository; If not, obtain the object's preset delay range, and within the object's preset delay range, perform mutual information matching based on sensor detection data, the total number of defective products and the total number of products to obtain the object's delay parameters.

[0182] In one possible implementation, the alignment module is specifically used for: The current processing data is determined from the sensor detection data, the total number of defective products and the total number of finished products. The current processing data includes the current upstream data and the current downstream data corresponding to the current upstream data. Based on the current upstream data, the current downstream data, and the preset delay range, a set of candidate data pairs is calculated; The mutual information of the candidate data set is matched to obtain the mutual information parameters of each candidate data pair in the candidate data set. Based on the mutual information parameters of each candidate data pair, the candidate delay parameters corresponding to the current processing data are determined. Determine the delay parameters of the object based on the candidate delay parameters corresponding to all the processed data.

[0183] In one possible implementation, the quality parameters include: defect rate; and a detection module, specifically used for: According to the preset time interval, the total number of defective products is cleared and detected, multiple first clearing points are identified in the total number of defective products, and the total number of defective products is divided into multiple first sub-count data according to each first clearing point. According to the preset time interval, the total number of products passing through is reset to zero, multiple second reset points are identified in the total number of products passing through, and the total number of products passing through is divided into multiple second sub-count data according to each second reset point; Based on each first sub-count data, determine the defective product throughput increment for each first sub-count data, and based on the defective product throughput increment for each first sub-count data, determine the total defective product throughput increment within the preset time interval; Based on each second sub-count data, determine the product throughput increment for each second sub-count data, and based on the product throughput increment for each second sub-count data, determine the total product throughput increment within the preset time interval. Based on the total increase in defective products and the total increase in product output within a preset time interval, the defect rate of the object at each time point within the preset time interval is determined.

[0184] In one possible implementation, a module is added, specifically for: Based on the quality parameters of the object at various time points during the production process, determine the quality thresholds corresponding to multiple types of quality labels; Based on the defect rate of the object at each time point in the production process and the quality threshold corresponding to each quality label, quality labels are added to multiple sets of initial sample production data to obtain target sample production data.

[0185] In one possible implementation, a module is added, specifically for: The objects are sorted according to their quality parameters at various points in time during the production process to obtain a sorted sequence. Based on the sorted sequence, multiple candidate threshold pairs are generated. Based on the preset sample size constraints, each candidate threshold pair is screened to obtain multiple selectable threshold pairs; Calculate the inter-class divergence and intra-class divergence for each of the optional threshold pairs; Based on the inter-class divergence and intra-class divergence of each optional threshold pair, determine the quality thresholds corresponding to multiple types of quality labels.

[0186] In one possible implementation, the analysis module is specifically used for: Based on the target sample production data and the pre-trained first intelligent agent, multiple data processing algorithms are screened to determine the target data processing algorithm; Based on the target data processing algorithm, the quality analysis of the target sample production data is performed to obtain the quality evaluation information corresponding to the target sample production data. Based on quality evaluation information, target data processing algorithms, target sample production data, and a pre-trained second agent, quality improvement recommendation information is determined.

[0187] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0188] This application also provides an electronic device, such as... Figure 12 As shown, Figure 12 The schematic diagram of the electronic device structure provided in this application embodiment includes: a processor 1201 and a memory 1202, and optionally, a bus 1203. The memory 1202 stores machine-readable instructions executable by the processor 1201. When the electronic device is running, the processor 1201 and the memory 1202 communicate via the bus 1203. The processor 1201 executes the machine-readable instructions to perform the steps of the above-described multi-agent driven multi-algorithm fusion production parameter quality analysis and prediction method.

[0189] This application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described multi-agent-driven multi-algorithm fusion production parameter quality analysis and prediction method.

[0190] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0191] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0192] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-agent-driven, multi-algorithm fusion method for production parameter quality analysis and prediction, characterized in that, include: Obtain the original production data of the current model object during the production process. The production process includes multiple production stages. The original production data includes: sensor detection data of each sensor at each time window during the production process, the total number of defective products passing through each production stage, and the total number of products passing through each production stage. Determine the delay parameter of the object, and based on the delay parameter, align the sensor detection data, the total number of defective products, and the total number of finished products to obtain multiple sets of initial sample production data sorted by time; The total number of defective products and the total number of finished products are reset and incrementally calculated to obtain the quality parameters of the object at each time point in the production process. Based on the quality parameters of the object at each time point in the production process, quality labels are added to each group of initial sample production data to obtain multiple groups of target sample production data. Based on multiple pre-trained agents and multiple preset data processing algorithms, the production data of the target sample is analyzed for quality and production is predicted to obtain a quality report of the object in the production process. The quality report includes: quality evaluation information and quality improvement recommendation information.

2. The method according to claim 1, characterized in that, The process involves aligning the sensor detection data, the total number of defective products, and the total number of finished products according to the delay parameter to obtain multiple sets of initial sample production data sorted by time, including: Based on the sensor detection data, determine the sensor feature vector for each time window; Based on the delay parameter, the sensor feature vector, the sensor detection data, the total number of defective products, and the total number of finished products are aligned to obtain the multiple sets of initial sample production data sorted by time.

3. The method according to claim 2, characterized in that, The step of determining the sensor feature vector for each time window based on the sensor detection data includes: The sensor detection data is subjected to steady-state detection and segmentation to obtain multiple transition segments and one steady-state segment corresponding to the sensor; The sensor detection data of the sensor in each of the transition sections is processed by a first dynamic window generation process to obtain multiple first time windows corresponding to the sensor in each of the transition sections, and the sensor feature vector of the sensor in each of the first time windows is determined. The sensor detection data of the sensor in the steady-state section is processed by a second dynamic window generation process to obtain multiple second time windows corresponding to the sensor in the steady-state section, and the sensor feature vector of the sensor in each second time window is determined.

4. The method according to claim 1, characterized in that, Determining the delay parameters of the object includes: Based on the object, determine whether the object's delay parameters exist from a preset delay repository; If so, then read the delay parameters of the object from the delay repository; If not, then obtain the preset delay range of the object, and within the preset delay range of the object, perform mutual information matching based on the sensor detection data, the total number of defective products and the total number of products to obtain the delay parameters of the object.

5. The method according to claim 4, characterized in that, Within the preset delay range of the object, the delay parameters of the object are obtained by mutual information matching based on the sensor detection data, the total number of defective products, and the total number of finished products, including: The current processing data is determined from the sensor detection data, the total number of defective products, and the total number of finished products. The current processing data includes: the current upstream data and the current downstream data corresponding to the current upstream data. Based on the current upstream data, the current downstream data, and the preset delay range, a set of candidate data pairs is calculated; Mutual information matching is performed on the candidate data set to obtain the mutual information parameters of each candidate data pair in the candidate data pair set, and the candidate delay parameters corresponding to the current processing data are determined based on the mutual information parameters of each candidate data pair. The delay parameters of the object are determined based on the candidate delay parameters corresponding to all the processed data.

6. The method according to claim 1, characterized in that, The quality parameters include: defect rate; The process of resetting and incrementally calculating the total number of defective products and the total number of finished products to obtain the quality parameters of the object at various time points in the production process includes: According to a preset time interval, the total number of defective products passing through is reset to zero, multiple first reset points are identified in the total number of defective products passing through, and the total number of defective products passing through is divided into multiple first sub-count data according to each first reset point. According to a preset time interval, the total number of products passing through is reset to zero, multiple second reset points are identified in the total number of products passing through, and the total number of products passing through is divided into multiple second sub-count data according to each second reset point; Based on each of the first sub-count data, determine the defective product passing increment of each of the first sub-count data, and based on the defective product passing increment of each of the first sub-count data, determine the total defective product passing increment within the preset time interval; Based on each of the second sub-count data, determine the product passing increment for each of the second sub-count data, and based on the product passing increment for each of the second sub-count data, determine the total product passing increment within the preset time interval; Based on the total increase in defective products within the preset time interval and the total increase in product products within the preset time interval, the defect rate of the object at each time point within the preset time interval is determined.

7. The method according to claim 1, characterized in that, The process involves adding quality labels to each set of initial sample production data based on the quality parameters of the object at various time points during the production process, resulting in multiple sets of target sample production data, including: Based on the quality parameters of the object at various time points in the production process, determine the quality thresholds corresponding to multiple types of quality labels; Based on the defect rate of the object at each time point in the production process and the quality threshold corresponding to each quality label, quality labels are added to the multiple sets of initial sample production data to obtain target sample production data.

8. The method according to claim 7, characterized in that, The step of determining quality thresholds corresponding to multiple types of quality labels based on the quality parameters of the object at various time points during the production process includes: The objects are sorted according to their quality parameters at various time points in the production process to obtain a sorted sequence of the objects, and multiple candidate threshold pairs are generated based on the sorted sequence of the objects. Based on the preset sample size constraints, each candidate threshold pair is screened to obtain multiple selectable threshold pairs; Calculate the inter-class divergence and intra-class divergence for each of the optional threshold pairs; Based on the inter-class divergence and intra-class divergence corresponding to each optional threshold pair, the quality thresholds corresponding to the multiple types of quality labels are determined.

9. The method according to claim 1, characterized in that, The method, based on multiple pre-trained agents and multiple preset data processing algorithms, performs quality analysis and production prediction on the target sample production data to obtain a quality report of the object during the production process, including: Based on the target sample production data and the pre-trained first intelligent agent, the multiple data processing algorithms are screened to determine the target data processing algorithm; Based on the target data processing algorithm, the target sample production data is subjected to quality analysis to obtain the quality evaluation information corresponding to the target sample production data. The quality improvement recommendation information is determined based on the quality evaluation information, the target data processing algorithm, the target sample production data, and the pre-trained second agent.

10. An electronic device, characterized in that, include: The device includes a processor and a memory, the memory storing machine-readable instructions executable by the processor. When the electronic device is running, the processor executes the machine-readable instructions to perform the steps of the multi-agent-driven, multi-algorithm fusion production parameter quality analysis and prediction method as described in any one of claims 1 to 9.

Images