A defect part reprocessing decision method based on multi-objective balance

Abstract

This application provides a multi-objective balance-based decision-making method for reprocessing defective parts, including: acquiring real-time defect rate data and part distribution information from production, analyzing the trend of defect rate changes and fluctuations in part quantity, and setting a standard defect threshold based on industry standards and historical data; if the defect rate exceeds the standard defect threshold, identifying and fitting part performance parameters to obtain an average performance curve, calculating the slope change and abrupt change amplitude of the curve, and determining the time and scope of influence of the mutation point; acquiring the factory processing load, combining the priority allocation order with the target part grouping scheme and the factory processing load to obtain a reprocessing process configuration scheme, and identifying the balance point between resource consumption and reprocessing coverage through simulation; acquiring the real-time ratio of reprocessing coverage and resource consumption corresponding to the balance point, comparing it with the historical best ratio, and if it is worse than the historical best value, determining that there is a performance bottleneck risk and analyzing the causes of the risk.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a decision-making method for reprocessing defective parts based on multi-objective balancing. Background Technology

[0002] In the fields of manufacturing and industrial engineering, a multi-objective balance-based decision-making method for reprocessing defective parts is directly related to the optimization of production efficiency, resource utilization, and overall performance. As the global manufacturing industry increasingly pursues high quality and efficiency, how to achieve dynamic optimization of reprocessing when facing large volumes of defective parts has become one of the core issues for enhancing enterprise competitiveness. Research in this field not only affects production cost control but also has a profound impact on sustainable development and resource recycling. However, existing reprocessing decision-making methods often reveal limitations when dealing with parts with high defect rates. Traditional methods rely heavily on static role allocation and fixed resource scheduling, lacking the ability to dynamically respond to sudden performance changes and intensified resource competition. When defective parts flood in, these methods struggle to effectively adjust part grouping strategies and factory matching priorities, leading to decreased reprocessing coverage and increased resource waste. Especially in high defect rate scenarios, insufficient identification of abrupt changes in the average performance curve prevents optimization algorithms from adapting to changing load demands in a timely manner. The core challenges of this research lie in several key technical factors: first, how to accurately detect and adjust strategies based on the performance mutations caused by the increased proportion of high-defect-rate parts; second, how to dynamically balance part grouping and factory matching priorities under intensified resource competition; and third, the impact of these adjustments on the overall reprocessing coverage and resource consumption balance has not been fully quantified. These factors lead to performance bottlenecks and resource allocation imbalances in reprocessing when faced with a large number of defective parts, thus creating unique technical challenges. Therefore, designing a multi-objective balanced mutation detection mechanism that can dynamically adjust part grouping strategies and factory matching priorities when high-defect-rate parts enter in batches, and optimize the balance between reprocessing coverage and resource consumption, is a key problem that this research urgently needs to solve. Summary of the Invention

[0003] This invention provides a decision-making method for reprocessing defective parts based on multi-objective balance, mainly including:

[0004] Real-time defect rate data and part distribution information are obtained from production, defect rate trends and part quantity fluctuations are analyzed, and standard defect thresholds are set based on industry standards and historical data.

[0005] If the defect rate exceeds the standard defect threshold, the performance parameters of the part are identified and fitted to obtain the average performance curve. The slope change and abrupt change of the curve are calculated to determine the time and range of influence of the variation point.

[0006] Based on the defect rate and average performance curve, dynamic clustering of parts is performed to obtain the part group corresponding to the mutation point, and the target part group is determined. The parts include internal parts and cross-factory parts. Internal parts are allocated according to the processing capacity of each process and the real-time defect rate, and cross-factory parts are allocated according to geographical location, transportation cost, processing capacity and real-time defect rate.

[0007] Identify the resource allocation status and processing capacity of the factory after the target plan is executed, dynamically allocate parts between multiple processes within the factory based on real-time production load, obtain factory matching priority by combining historical production data and real-time production load, and allocate cross-factory parts based on factory matching priority.

[0008] The factory processing load is obtained, and the priority allocation order is combined with the target part grouping scheme and the factory processing load to obtain the reprocessing process configuration scheme. The balance point between resource consumption and reprocessing coverage is identified through simulation.

[0009] Obtain the real-time ratio of reprocessing coverage and resource consumption corresponding to the equilibrium point, compare it with the historical best ratio, and if it is worse than the historical best value, then it is determined that there is a performance bottleneck risk, and the cause of the risk is analyzed.

[0010] The part grouping strategy and factory matching priority are dynamically updated based on the risk causes to obtain the optimized reprocessing process configuration, obtain the optimized defect rate change and performance curve, and verify whether resource consumption is reduced and coverage is improved through iterative verification to determine the optimized stable state and form an executable improvement plan.

[0011] Furthermore, real-time defect rate data and part distribution information are obtained from production to analyze defect rate trends and part quantity fluctuations. Standard defect thresholds are set based on industry standards and historical data. This includes: obtaining real-time part location information and production quality record data from production execution; preprocessing part numbers, workstation numbers, production times, and inspection results to remove outliers and missing values; and obtaining a defect dataset by associating defect occurrence workstations and time points through real-time part traceability coding. For the preprocessed defect dataset, it is categorized and statistically analyzed according to workstation number, part type, and defect type. The real-time defect rate data for each workstation is calculated by dividing the number of defects by the total production volume within that time period. An automatic inspection device is used to randomly inspect products at each workstation on the production line at a preset frequency. The obtained inspection data is then compared with the real-time defect rate data, and the defect detection rate index is calculated using the Pearson correlation coefficient. Based on the defect detection rate index, the defect data is weighted, and exponential smoothing is used to perform time-series analysis on the weighted defect rate data to obtain a defect rate trend curve. Historical defect records are extracted from the product quality management database. Box plot analysis is used to identify outlier ranges in the defect rate. Defect rate warning thresholds are set based on industry standards to establish a defect rate monitoring baseline. Location distribution data for parts is collected at each workstation's inspection points. Density clustering algorithms are used to calculate the part distribution density. The correlation between workstation capacity and defect rate is analyzed using defect rate trend curves.

[0012] Furthermore, if the defect rate exceeds the standard defect threshold, the performance parameters of the part are identified and fitted to obtain an average performance curve. The slope change and abrupt change amplitude of the curve are calculated to determine the time and scope of influence of the variation point. This includes: acquiring performance parameter measurement data such as size, weight, surface finish, hardness, and temperature from the part inspection equipment; for workstations where the defect rate exceeds the preset threshold range, sensors are used to collect raw data of the part performance parameters in real time; the raw data is normalized to obtain a standardized performance parameter set; outlier removal is performed on the standardized performance parameter set; the mean and standard deviation of each parameter are calculated using the sliding window method; outliers are identified according to the three-standard-deviation principle; and after removing outlier data, a performance parameter time series is obtained. For the performance parameter time series, the least squares method is used to calculate the fitting curve equation; the parameter change rate at adjacent time points is calculated using the curve fitting equation to obtain the parameter slope series; a slope change threshold is set according to the historical performance parameter fluctuation range; the parameter slope series is traversed; when the slope difference between two adjacent points exceeds the set threshold, it is marked as a mutation point, and the corresponding time of the mutation point is recorded. Centered on the abrupt change point, the time series is extended forward and backward, and the deviation between the parameter values ​​and those at the abrupt change point is calculated. When the deviation falls back to a threshold range, that time point is recorded. Clustering is used to group the recorded time points, and the time interval between each group is calculated. The start and end times of the affected area are determined by the density distribution of these time intervals.

[0013] Furthermore, dynamic clustering of parts based on defect rate and average performance curves yields part groups corresponding to variation points, determining target part groups. Parts include internal parts and cross-factory parts. Internal parts are allocated based on processing capacity and real-time defect rate for each process, while cross-factory parts are allocated based on geographical location, transportation cost, processing capacity, and real-time defect rate. This includes: calculating the Euclidean distance between the slope of the part performance curve and the defect rate data using k-means clustering; generating a distance sequence by sorting distance values ​​from smallest to largest; and initially grouping parts using a minimum distance threshold to obtain a variation part identification table. For the part groups in the variation part identification table, the performance curve fluctuation amplitude is calculated using a curve fitting equation, and the part groups are further subdivided based on the real-time defect rate to obtain a target part group list. Precision inspection records, runtime records, and fault repair records of processing equipment are extracted from the production management database. Equipment capacity index is calculated using processing precision pass rate, equipment runtime, and fault interval time to obtain a process-level processing capacity evaluation table. For internally processed parts, equipment capacity index and real-time defect rate data are read, the remaining processing time for each process is calculated, and a process capacity allocation network is constructed using a maximum flow network algorithm to obtain an internal part allocation scheme. Inter-factory transportation distance data is obtained from a geographic information database. An inter-factory logistics cost matrix is ​​calculated based on the unit kilometer transportation cost. A cross-factory processing cost matrix is ​​generated using the factory capacity index and real-time defect rate. A multi-objective programming algorithm is used to allocate parts across factories. A set of constraint equations is constructed by combining the logistics cost matrix and the processing cost matrix. The optimal allocation scheme for cross-factory parts is obtained through genetic iterative computation.

[0014] Furthermore, real-time defect rate data for internal parts and transportation costs for parts across factories are acquired. Initial part groups are generated based on the real-time defect rate data. Processing capacity vectors for each process group are identified based on the equipment status of each factory. A target internal part allocation scheme is generated based on these vectors. The target internal part allocation scheme is then fused with the cross-factory part allocation scheme to obtain a global part scheduling instruction set. This includes: acquiring part processing records and quality inspection data from the production execution database; calculating the real-time defect rate for internal parts; extracting cross-factory part transportation distances and freight rates from the logistics management database; and using a three-tiered grouping rule to classify defect rates less than 1% as low-risk, 1% to 3% as medium-risk, and greater than 3% as high-risk, thus generating an initial part grouping table. Operating status data of processing equipment in each factory is collected from the equipment management database. The percentage of continuous operating time is weighted by 0.3, the equipment accuracy inspection pass rate by 0.4, and the equipment maintenance interval by 0.3. A weighted summation is then used to calculate a comprehensive equipment status score. A process processing capability vector is constructed based on the comprehensive equipment status score. The vector dimensions include processing accuracy, processing cycle time, and equipment stability indicators. The least squares method is used to calculate the process equipment status feature values. For each part identifier in the internal parts grouping table, the matching degree between the part processing requirements and the process processing capability vector is calculated using Euclidean distance. The process equipment with the highest matching degree is selected as the target process. A priority ranking method is used to allocate high-risk group parts to process equipment with the best processing capability, medium-risk group parts to process equipment with medium processing capability, and low-risk group parts to the remaining process equipment, resulting in an internal parts allocation scheme. The cross-factory parts allocation scheme is read from the factory collaboration platform. A correspondence table between internal and cross-factory parts is established through part coding, and parts are sorted according to their risk level in the risk level correspondence table. A processing priority sequence is generated based on the part risk level ranking results, and a process scheduling timetable is generated corresponding to the priority sequence, forming a global parts scheduling instruction dataset.

[0015] Furthermore, the resource allocation status and processing capacity of the factory after implementing the target plan are identified. Parts are dynamically allocated between multiple processes based on real-time production load. Factory matching priorities are obtained by combining historical production data and real-time production load. Cross-factory parts are then allocated based on these priorities, including: obtaining process scheduling data and equipment operation records from the production management platform; calculating equipment processing load using a 4-hour fixed time window; obtaining the equipment load rate by dividing the number of online processed parts by the equipment's standard capacity; and generating a process equipment load data table. The remaining available working hours for each piece of equipment in the load data table are calculated. Based on the process sequence specified in the part processing procedure, the upper and lower limits of equipment capacity are calculated using linear programming, resulting in a process-level remaining capacity data table. The actual cycle time records for each process over the past 30 days are extracted from the historical production database. The standard processing cycle time for each process is calculated using a weighted average method. A process processing capacity scoring table is generated based on the deviation between the standard and actual cycle times. The data in the process processing capacity scoring table is normalized. Combined with the real-time load rate calculation results, a process capacity prediction curve is constructed using a multilayer perceptron, resulting in a process-level capacity prediction data table. Based on process capacity forecast data, a minimum cost flow algorithm is used to allocate internal parts across multiple processes. The priority allocation order is determined by calculating the ratio of unit processing time per part to remaining capacity in the process. A production equipment configuration list is retrieved from the factory resource database, and the equipment capacity and actual load data for each factory are summarized. The analytic hierarchy process (AHP) is used to calculate the overall factory capability score, generating a factory matching priority sequence. A maximum flow algorithm is then used to allocate parts across factories. Based on the factory priority sequence, high-priority factory groups and low-priority factory groups are divided, and cross-factory part allocation schemes are generated according to the proximity principle.

[0016] Furthermore, the factory processing load is obtained, and the priority allocation order is combined with the target part grouping scheme and the factory processing load to obtain a reprocessing process configuration scheme. The balance point between resource consumption and reprocessing coverage is identified through simulation, including: obtaining real-time factory processing load and equipment operation data from the manufacturing execution database; using the moving average method to calculate the process equipment uptime and capacity utilization rate; dividing the process load into high-load, medium-load, and low-load intervals based on capacity utilization rate; and generating a factory load distribution data table. Based on the priority sequence in the part grouping scheme, process equipment uptime, capacity utilization rate, and maintenance interval duration are extracted as feature parameters, and the correlation score between the priority sequence and load features is calculated using the random forest algorithm. The correlation scores are sorted, and matching processes are filtered by setting a correlation threshold. A reprocessing process path table is generated based on the processing sequence constraints between processes. A discrete event simulator is used to construct the reprocessing process scenario, inputting process equipment parameters, processing cycle time, and changeover time into the simulation environment to generate a simulation running dataset. Equipment running time and maintenance interval duration are extracted from the simulation running dataset, and the resource consumption rate curve is calculated by dividing the cumulative running time by the standard running time. The process records of part handling during the reprocessing workflow are statistically analyzed, and the coverage curve is calculated by dividing the number of completed parts by the planned quantity. A piecewise regression method is used to fit the resource consumption rate curve and the coverage curve, and the process parameter combinations corresponding to the intersection points are calculated. Based on the process parameter settings at the intersection points, a reprocessing balance configuration scheme, including equipment configuration, processing cycle time, and changeover time, is generated.

[0017] Furthermore, the reprocessing process configuration schemes are categorized to obtain a set of configuration schemes. Resource consumption data and reprocessing coverage rates for different configuration schemes are obtained from this set. If resource consumption data exceeds a preset consumption threshold, the configuration schemes are adjusted to obtain an optimized set of configuration schemes. Based on the optimized set, a linear regression algorithm is used to predict the relationship between resource consumption and reprocessing coverage rate, obtaining equilibrium point data. This includes: extracting features from process parameters, equipment configurations, and processing cycle times in the reprocessing process configuration schemes; calculating the Euclidean distance between schemes using k-means clustering; and dividing the schemes into high-similarity, medium-similarity, and low-similarity groups based on the distance values, generating a configuration scheme grouping table. Equipment operation records for each configuration scheme are read from the production execution database. Equipment utilization is calculated based on the standard working hours per shift. Resource consumption indicators are calculated using the ratio of cumulative equipment operating time to maintenance intervals. The reprocessing coverage rate indicator is calculated by dividing the number of online completed parts by the planned quantity. A resource consumption baseline value is obtained from the historical production database, and a resource consumption threshold is set at 1.2 times the baseline value. When the resource consumption indicator exceeds the threshold, processing cycle time parameters are adjusted first. If the cycle time still exceeds the threshold after adjustment, the number of devices is increased, and the capacity redundancy is calculated by multiplying the number of devices by the cycle time. A genetic algorithm is used to optimize the adjustment scheme, setting the capacity utilization rate as the fitness function, and iterative calculation is performed with the resource consumption index being less than the threshold as a constraint. Resource consumption and coverage indices are extracted from the optimized configuration scheme, and a linear regression algorithm is used to fit the two sets of index data to generate consumption rate and coverage curves. The coordinates of the intersection of the two curves are calculated using the least squares method to obtain the resource consumption and coverage values ​​corresponding to the equilibrium point. Cross-validation is used to verify the equilibrium point data, and the stability of the equilibrium point parameter combination is verified using historical data, generating an equilibrium parameter verification report.

[0018] Furthermore, the real-time ratio of reprocessing coverage and resource consumption corresponding to the equilibrium point is obtained and compared with the historical best ratio. If it is worse than the historical best value, a performance bottleneck risk is identified, and the causes of the risk are analyzed, including: reading the reprocessing data records corresponding to the equilibrium point from the production execution database; calculating the coverage value by dividing the number of completed parts by the planned quantity; calculating the resource consumption value by dividing the equipment runtime by the standard man-hours; and obtaining the real-time performance ratio by dividing the coverage rate by the resource consumption. Production records for the most recent 30 days are extracted from the historical database, and the daily performance ratios are sorted. The top 10% of records are selected as the optimal ratio sample, and a performance prediction curve is established using support vector regression. An early warning threshold is set based on the performance prediction curve. When the real-time performance ratio is lower than the early warning threshold, it is marked as a risk warning state, and classified into three levels: mild, moderate, and severe warning, according to the degree of deviation from the threshold. For the warning data, operating parameters, including processing accuracy, runtime, and maintenance interval, are extracted from the equipment management database. The parameters are normalized to generate equipment status feature vectors. Hierarchical clustering is used to group equipment state feature vectors, and abnormal parameter combinations are identified by calculating inter-class distances, generating a parameter anomaly feature table. The confidence level between parameter anomalies and performance degradation is calculated based on an association rule algorithm, and high-risk parameter combinations are screened using confidence thresholds. High-risk parameter combinations are categorized into equipment accuracy, equipment lifespan, and equipment maintenance categories, generating a risk cause classification table. A decision tree algorithm is used to construct a risk warning model, and the risk causes are validated using historical fault data, outputting a risk warning report.

[0019] Furthermore, based on the risk causes, the part grouping strategy and factory matching priority are dynamically updated to obtain the optimized reprocessing process configuration. The optimized defect rate change and performance curve are obtained. Through iterative verification, it is determined whether resource consumption is reduced and coverage is improved, and the optimized stable state is judged, forming an executable improvement plan. This includes: extracting equipment status characteristics and performance degradation records from the risk cause database; setting three risk levels (high risk, medium risk, and low risk) based on historical performance data; using deep learning algorithms to predict and classify parts for risk, generating an updated part grouping table. Equipment operating status is scored, with weights of 0.4 for equipment status, 0.4 for processing accuracy, and 0.2 for maintenance cycle. A weighted summation is used to calculate the factory's overall score. Based on the factory's overall score, the equipment is ranked, with the top 30% of equipment selected as the highest priority group, the middle 40% as the medium priority group, and the remaining equipment as the lowest priority group. Based on the priority grouping results, a genetic algorithm is used to optimize process parameters, using the product of equipment load rate and capacity utilization rate as the optimization objective function. The optimized process parameters are verified using the capacity calculation module, which calculates the capacity matching degree between each process and generates a new table of processing cycle time and changeover cycle parameters. Optimized production data is monitored, using a fixed 4-hour time window to statistically analyze defect rate changes and performance curve fluctuations. The percentage reduction in resource consumption and the percentage increase in coverage are calculated; a stable state is determined when the index fluctuation is less than 5% over three consecutive monitoring periods. Cross-validation is used to verify the stable state data; the stability of the optimization scheme is verified by calculating the index fluctuation range of the validation sample, generating an improvement scheme verification report.

[0020] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0021] This invention discloses a multi-objective balance-based decision-making method for reprocessing defective parts. By acquiring real-time production defect rate data and part distribution information, it analyzes defect rate trends and sets standard thresholds. When the defect rate exceeds the threshold, it identifies and fits part performance parameters to determine the points of variation and their impact range. Parts are dynamically clustered and grouped based on defect rate and performance curves, and internal and cross-factory parts are allocated based on factors such as process capability, geographical location, and transportation costs. This invention also identifies the balance point between resource consumption and reprocessing coverage through simulation, dynamically updating the part grouping strategy and factory matching priority to form an executable improvement plan. This method can effectively reduce resource consumption, improve reprocessing coverage, and achieve intelligent configuration of the part reprocessing process. Attached Figure Description

[0022] Figure 1 This is a flowchart of a decision-making method for reprocessing defective parts based on multi-objective balance according to the present invention.

[0023] Figure 2 This is a schematic diagram of a decision-making method for reprocessing defective parts based on multi-objective balance according to the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In this embodiment of the invention, for ease of understanding, the relevant terms are briefly explained: Defective parts refer to products that do not meet standards due to dimensional deviations, surface defects, or other quality problems during the production process, and require reprocessing to restore their function. Defective parts can be divided into internal parts and cross-factory parts, corresponding to parts that are repaired within the same factory or need to be transported to other factories for reprocessing, respectively. A point of variation refers to the time point or state in which performance changes significantly due to a sudden increase in the defect rate or abnormal part distribution, usually identified by changes in the slope or magnitude of abrupt changes in the performance curve. Reprocessing coverage refers to the proportion of defective parts that have been successfully reprocessed to the total number of defective parts, and is a key indicator for measuring reprocessing efficiency.

[0026] In this embodiment of the invention, by collecting production data in real time, analyzing defect rate trends, and dynamically optimizing scheduling strategies, the performance bottleneck problem of traditional methods in high defect rate scenarios is solved. The following, in conjunction with... Figure 1 The technical solutions of the embodiments of the present invention will be described in detail below.

[0027] S101. Obtain real-time defect rate data and part distribution information from production, analyze the trend of defect rate changes and the fluctuation of part quantity, and set standard defect thresholds based on industry standards and historical data.

[0028] In this embodiment of the invention, when the reprocessing process for defective parts is initiated, real-time data is first acquired from the manufacturing execution and analyzed to determine the defect rate trend and part distribution characteristics. The specific operation includes the following steps:

[0029] like Figure 2 S1011. Obtain real-time part location information and production quality record data from the production and manufacturing execution, and generate a defect dataset by associating the defect occurrence station and time point with the part traceability code.

[0030] In this embodiment of the invention, the manufacturing process records the processing status and quality information of parts at each workstation. Using a unique traceability code for each part, the workstation it passed through on the production line, the processing time, and the inspection results are linked to generate a defect dataset containing the workstation and time point where the defect occurred. For example, in an automotive engine assembly line, parts at the cylinder block processing station are identified by barcodes, and the processing parameters and quality inspection results for each part are recorded. If a workstation detects dimensional deviations, quality inspectors enter the non-conformance information through a data acquisition terminal, generating a corresponding defect dataset based on the traceability code. This step does not impose excessive restrictions on the specific data acquisition method, which can be determined by technical personnel according to the actual scenario.

[0031] S1012. Perform data preprocessing on the defect dataset, remove outliers and missing values, classify and statistically analyze the defect data according to workstation number and defect type, and calculate the real-time defect rate data for each workstation.

[0032] In this embodiment of the invention, the collected defect dataset is cleaned to remove records with missing part numbers, incorrect workstation numbers, or incomplete inspection results. Subsequently, it is grouped and statistically analyzed according to workstation number and defect type. The real-time defect rate for each workstation is calculated by dividing the number of defects by the total production volume within that time period. For example, if a workstation produces 500 products in a day, and 10 of them are judged to be defective due to dimensional deviations, then the real-time defect rate for that workstation is 2%. This step provides a reliable data foundation for subsequent trend analysis through categorized statistical analysis.

[0033] S1013. Automatic detection devices are used to acquire station detection data, and the defect detection rate index is calculated using the Pearson correlation coefficient. The defect rate data is then weighted.

[0034] In this embodiment of the invention, an automatic inspection device is used to randomly inspect products at each workstation of the production line at a preset frequency. The acquired inspection data is then compared with real-time defect rate data. A defect detection rate index is generated by calculating the Pearson correlation coefficient to evaluate the reliability of the inspection results. For example, a coordinate measuring machine is used for dimensional inspection at the cylinder block machining station, with one piece randomly inspected out of every 50 pieces. The inspection data is uploaded to the quality management system in real time. If the detection rate is below 90%, the inspection parameters will be adjusted. When weighting the defect rate data, different weights are assigned according to the severity of the defect type; for example, a weight of 1.5 is given for critical dimension deviations, and a weight of 1.0 is given for surface defects.

[0035] S1014. Use the exponential smoothing method to perform time series analysis on the weighted defect rate data and generate a defect rate change trend curve.

[0036] In this embodiment of the invention, the weighted defect rate data is subjected to exponential smoothing for time-series analysis to generate a trend curve reflecting changes in the defect rate. Recent data is assigned a higher weight; for example, the weight coefficient for data from the most recent week is 0.3, while the weight of historical data decays over time. The trend curve can predict the defect rate change range for the next 24 hours, providing a basis for subsequent anomaly detection. This step does not impose excessive limitations on the specific parameters of the smoothing algorithm and can be adjusted according to actual needs.

[0037] S1015. Identify defect rate outlier ranges using box plot analysis and set standard defect thresholds based on industry standards and historical data.

[0038] In this embodiment of the invention, defect records from the past three months are extracted from the product quality management database, and box plot analysis is used to identify outlier ranges in the defect rate. For example, the normal defect rate fluctuation for a certain workstation is between 0.5% and 1.5%, with the upper limit being the quartile plus 1.5 times the interquartile range. If the real-time defect rate exceeds 2%, an alert is triggered. A standard defect threshold is set based on industry standards and historical data, serving as a baseline for subsequent anomaly detection. This step ensures the reasonableness of the threshold through outlier analysis.

[0039] S1016. Analyze the location distribution data of parts at each workstation, calculate the distribution density of parts using density clustering algorithm, and analyze the correlation between workstation capacity and defect rate by combining the defect rate change trend curve.

[0040] In this embodiment of the invention, real-time position information of parts on the production line is collected by workstation sensors. The spacing between adjacent parts and the production cycle time are calculated, and a density clustering algorithm is used to generate the part distribution density. For example, if the inspection cycle time of a certain assembly station is 90 seconds, and the upstream workstation's delivery speed is too fast, resulting in a part spacing of less than 85 seconds, the production cycle time will be adjusted to ensure inspection quality. The correlation between workstation capacity and defect rate is analyzed by combining the defect rate trend curve, providing data support for dynamically optimizing part grouping.

[0041] In this embodiment of the invention, the above steps enable the efficient acquisition of real-time defect rate data and part distribution information from production processes, and the setting of standard defect thresholds through trend analysis and outlier detection. These operations ensure the accuracy of subsequent anomaly detection and part grouping strategies, significantly improving reprocessing coverage, optimizing resource utilization efficiency, and meeting the urgent needs of the manufacturing industry for high-quality production.

[0042] S102. If the defect rate exceeds the standard defect threshold, the performance parameters of the parts are collected from production, an average performance curve is generated by fitting, the slope change and abrupt change of the curve are analyzed, and the time point of the variation and its range of influence are determined.

[0043] In this embodiment of the invention, when the defect rate at a certain workstation exceeds a preset standard defect threshold, a performance parameter acquisition and analysis process is triggered. By fitting the part's performance curve and detecting its abrupt change characteristics, the variation points and their impact range are accurately located, providing a basis for part grouping and reprocessing optimization, ensuring rapid response to abnormal fluctuations in high defect rate scenarios. This embodiment of the invention does not limit the specific type or acquisition method of the performance parameters, which can be determined by technical personnel based on the actual production scenario.

[0044] S1021. Obtain multi-dimensional performance parameter data from the part inspection equipment, normalize the data, and generate a standardized performance parameter set.

[0045] In this embodiment of the invention, a distributed sensor network is used to collect multi-dimensional performance parameters in real time from the part inspection equipment, such as key indicators like dimensions, weight, surface finish, hardness, and temperature. Taking the machining of automotive engine pistons as an example, the collected parameters include piston outer diameter, pin hole coaxiality, ring groove runout, and surface roughness. When the defect rate at a certain workstation exceeds 2.5%, a full inspection procedure is initiated, and the sensors collect raw data at a high frequency. Since the parameters have different dimensions (e.g., dimensions are in millimeters, hardness in HRC), a maximum-minimum value normalization method is used to map all parameter values ​​to a unified range of 0 to 1, generating a standardized performance parameter set. This step eliminates dimensional differences through normalization.

[0046] S1022. For the standardized performance parameter set, the local mean and standard deviation are calculated using the sliding window method. Outliers are removed based on the principle of three times the standard deviation, and a time series of performance parameters is generated.

[0047] In this embodiment of the invention, outlier processing is performed on the standardized performance parameter set to ensure data reliability. A sliding window method is used, with 20 data points as a window, to calculate the local mean and standard deviation of each parameter. If a parameter value deviates from the mean by more than three times the standard deviation, it is identified as an outlier and removed. For example, jump values ​​in the piston outer diameter parameter series caused by equipment vibration will be removed. After removing outliers, a continuous performance parameter time series is generated, providing high-quality data for subsequent curve fitting. This step does not strictly limit the window size or outlier judgment criteria and can be adjusted according to the production environment.

[0048] S1023. Based on the time series of performance parameters, the least squares method is used to fit the average performance curve, calculate the rate of change of parameters at adjacent time points, and generate a parameter slope sequence.

[0049] In this embodiment of the invention, the least squares method is used to perform curve fitting on the time series of performance parameters, generating an average performance curve reflecting the trend of parameter changes over time. Taking the piston outer diameter as an example, the fitted curve describes the fluctuation pattern of the dimensional parameter during the production process. The rate of change of the parameter at adjacent time points on the curve is calculated at 10-second intervals, forming a parameter slope sequence. Under normal production conditions, the rate of change of the piston outer diameter typically fluctuates within ±0.002 mm / s. This step, through fitting and slope calculation, quantifies the dynamic characteristics of parameter changes, laying the foundation for abrupt change detection.

[0050] S1024. Analyze the slope sequence of parameters, determine whether the slope difference between adjacent time points exceeds the preset threshold, mark the mutation point and record its time point.

[0051] In this embodiment of the invention, based on historical performance parameter fluctuation statistics, a slope change threshold is set to 5 times the normal fluctuation range; for example, the piston outer diameter slope change threshold is 0.01 mm / s. The parameter slope sequence is traversed, and the slope difference between adjacent time points is calculated. If the difference exceeds the threshold, that time point is marked as a mutation point and its timing is recorded. For example, when the piston outer diameter slope abruptly changes to 0.012 mm / s at a certain moment, that point is marked as a mutation point.

[0052] S1025. Extend from the mutation point to the time series before and after, calculate the parameter deviation, determine the start and end time points of the affected range, and confirm the time interval through cluster analysis.

[0053] In this embodiment of the invention, the deviation between the parameter value at each time point and the parameter value at the mutation point is calculated, centered on the marked mutation point and extending forward and backward along the time series. When the deviation value falls back to the normal fluctuation range, the corresponding time point is recorded. For example, when the deviation of the piston outer diameter mutation point falls back to ±0.002 mm / s, this moment is recorded as the boundary of the influence range. Multiple mutation points and boundary time points are grouped using a density clustering algorithm, and the density distribution of the time intervals of each group is calculated. For example, if five dense mutation points are detected between 9:30 and 9:45, density distribution analysis determines that the influence range is from 9:25 to 9:50, indicating that there is a significant anomaly in the production process during this time period.

[0054] In this embodiment of the invention, the above steps enable efficient identification of performance parameter mutation points in high-defect-rate scenarios, and determination of their timing and impact range. This step fully utilizes the dynamic characteristics of multidimensional performance parameters, employing normalization, outlier removal, curve fitting, and cluster analysis to ensure the accuracy and robustness of mutation point detection. The results provide reliable data support for subsequent part grouping and factory matching optimization, significantly improving the adaptability and efficiency of reprocessing, ultimately achieving a higher-quality production process.

[0055] S103. Based on defect rate data and average performance curves, dynamically cluster parts to generate part groups corresponding to mutation points, determine target part groups, including internal parts and cross-factory parts, optimize part allocation scheme by combining process processing capacity, real-time defect rate, geographical location and transportation cost, and generate a global part scheduling instruction set through equipment status assessment and matching degree calculation.

[0056] In this embodiment of the invention, when the production system detects that the defect rate at a certain workstation exceeds the standard defect threshold, the parts are grouped using a dynamic clustering method based on the defect rate data and performance curves. This distinguishes between internal parts and parts from different factories, and the allocation scheme is optimized based on factors such as processing capacity and transportation costs. Simultaneously, a global scheduling instruction set is generated through equipment status assessment and processing requirement matching to ensure the efficiency and stability of the reprocessing process.

[0057] S1031. Based on the slope of the part performance curve and the defect rate data, the Euclidean distance is calculated using the k-means clustering method to generate a variable part identification table, and the target part grouping list is determined through curve fluctuation analysis.

[0058] In this embodiment of the invention, the slope of the performance curve of a part and the real-time defect rate are used as feature parameters. K-means clustering is used to calculate the Euclidean distance between parts, generating a distance sequence. Initial grouping is then performed based on a preset minimum distance threshold to obtain a list of mutated parts. Taking the machining of automotive engine cylinder blocks as an example, if 15 out of 100 cylinder blocks in a batch exhibit dimensional deviations, with a performance curve slope of 0.015 mm / s, far exceeding the normal batch's 0.002 mm / s, this batch is marked as a mutated part group using a distance threshold of 0.01. Subsequently, the performance curve of the mutated part group is fitted using the least squares method to calculate the fluctuation amplitude. If the fluctuation amplitude reaches 0.08 mm, significantly higher than the normal batch's 0.02 mm, and the real-time defect rate is 5%, secondary clustering is performed based on the fluctuation amplitude and defect rate, dividing the parts into a high-risk group and a medium-risk group. The high-risk group has a fluctuation amplitude exceeding 0.05 mm and a defect rate greater than 3%. This step, through clustering and fluctuation analysis, ensures the accuracy and relevance of part grouping.

[0059] S1032. Extract the operation records of processing equipment from the production management database, calculate the equipment capacity index, construct the process capacity allocation network, and generate an internal parts allocation scheme.

[0060] In this embodiment of the invention, the accuracy inspection records, runtime records, and fault repair records of the processing equipment are obtained from the production management database to calculate the equipment capacity index. For example, a processing center has a monthly accuracy pass rate of 98.5%, a cumulative operating time of 480 hours, and an average fault-free operating time of 96 hours. Based on the accuracy pass rate, runtime, and fault interval, the calculated capacity index is 0.92, which is higher than the benchmark value of 0.85. For internal parts, real-time defect rate data and the remaining processing time of each process are read. Taking the roughing process as an example, the remaining capacities of the three processing centers are 120 hours, 80 hours, and 60 hours, respectively, with defect rates of 1.2%, 1.5%, and 1.8%, respectively. A maximum flow network algorithm is used to construct a process capacity allocation network, allocating 80% of high-risk parts to the processing center with the lowest defect rate to ensure processing quality. This step maximizes internal processing efficiency by quantifying equipment capacity and optimizing allocation.

[0061] S1033. Optimize cross-factory parts allocation schemes by combining geographical information and cost data.

[0062] In this embodiment of the invention, the inter-factory transportation distance is extracted from a geographic information database, and a logistics cost matrix is ​​calculated by combining it with the unit transportation cost. For example, the distances between the main factory and two collaborating factories are 120 km and 180 km, respectively, and the unit transportation cost is 5 yuan / km, generating the corresponding logistics cost matrix. Simultaneously, a processing cost matrix is ​​generated based on the collaborating factories' capacity indices of 0.88 and 0.85, and real-time defect rates of 2.1% and 2.3%. A multi-objective programming algorithm is used to construct a set of constraint equations including logistics costs and processing costs. Through iterative optimization using a genetic algorithm, under the conditions that transportation costs do not exceed 1 million yuan and capacity utilization is not less than 85%, an allocation scheme is determined: the closer collaborating factories undertake 60% of the cross-factory parts, the more distant factories undertake 25%, and the main factory retains 15% of the critical parts processing.

[0063] S1034. Based on part processing records and equipment status data, a global part scheduling instruction set is generated through risk level grouping and processing capacity matching.

[0064] In this embodiment of the invention, part processing records, including quality inspection data and defect rate data, are extracted from the production execution database. Parts are then divided into three groups using a three-tier grouping rule: a low-risk group (defect rate less than 1%), a medium-risk group (defect rate 1% to 3%), and a high-risk group (defect rate greater than 3%), generating an initial part grouping table. Taking cylinder block processing as an example, in a batch of 100 cylinder blocks with a defect rate of 4.2%, 25 parts are placed in the high-risk group, 50 in the medium-risk group, and 25 in the low-risk group. Processing equipment operating status data is collected from the equipment management database. A comprehensive equipment status score is calculated using a weighting of 0.3 for continuous operating time percentage, 0.4 for accuracy inspection pass rate, and 0.3 for maintenance interval duration. For example, a processing center with 85% continuous operation, 98% accuracy pass rate, and 96 hours of trouble-free operation receives a score of 0.92. Based on the score, a process processing capability vector is constructed, including three dimensions: processing accuracy, processing cycle time, and equipment stability. The process equipment status feature value is calculated using the least squares method. For each group of parts, the Euclidean distance between the part's processing requirements and the process's processing capacity vector is calculated, and the process equipment with the highest matching degree is selected. High-risk group parts are preferentially assigned to the equipment with the highest status score, medium-risk group parts are assigned to medium-risk equipment, and low-risk group parts are assigned to the remaining equipment. Internal and cross-plant part allocation schemes are integrated, and a processing priority sequence is generated by associating parts with part codes. A global scheduling instruction set containing processing equipment, time period, and process sequence is generated.

[0065] In this embodiment of the invention, the above steps enable dynamic clustering of parts based on defect rates and performance curves, optimizing internal and cross-factory part allocation, and generating a global scheduling instruction set. This method, through multi-dimensional data analysis and algorithm optimization, significantly improves the adaptability of reprocessing and resource utilization efficiency, providing stable support for production processes in high-defect-rate scenarios.

[0066] S104. Based on the factory resource status and processing capacity after the target allocation scheme is implemented, combined with real-time production load and historical data, dynamically optimize the allocation of parts between multiple processes, and determine the matching priority through comprehensive factory capability assessment to allocate cross-factory parts to improve reprocessing efficiency.

[0067] In this embodiment of the invention, after executing the target allocation scheme, the part allocation strategy among multiple internal processes is dynamically adjusted by real-time monitoring of factory resource status and processing capacity, combined with production load data and historical production records. Furthermore, a matching priority sequence is generated based on a comprehensive factory capability assessment to optimize cross-factory part allocation. This embodiment of the invention does not strictly limit the data acquisition frequency or algorithm parameters, which can be adjusted by technical personnel according to the actual production scenario.

[0068] S1041. Obtain process scheduling data and equipment operation records from the production management platform, calculate equipment load rate, generate process-level load data table, and determine the upper and lower limits of equipment capacity through linear programming method.

[0069] In this embodiment of the invention, process scheduling data and equipment operation records are extracted from the production management platform. Using a fixed time window of 4 hours, the ratio of the number of parts processed online to the standard capacity of the equipment is calculated to generate equipment load rate data. For example, on an automotive engine block production line, the standard capacity of a roughing machine is 8 parts per hour, and currently 6 parts are being processed, resulting in a load rate of 75%. The load rate data is summarized into a process-level load data table. Based on the process sequence specified in the process specification, such as the cylinder block processing requiring roughing, finishing, and honing in sequence, a linear programming method is used, combined with the equipment operating status and load rate, to calculate the upper and lower limits of the capacity of each process's equipment. For example, the roughing machine has 120 minutes of remaining time, the finishing machine 180 minutes, and the honing machine 150 minutes, generating a process-level remaining capacity data table. This step, through the quantification of load rate and capacity range, provides a precise basis for subsequent allocation.

[0070] S1042. Extract historical production data, calculate the standard processing cycle time of each process, generate a processing capacity score based on the cycle time deviation, and predict the process capacity trend through a multilayer sensor.

[0071] In this embodiment of the invention, actual cycle time records for the past 30 days of processes are extracted from a historical production database, and a weighted average method is used to calculate the standard processing cycle time for each process. For example, the standard cycle time for a certain finishing machine is 15 minutes / piece, and the actual average cycle time is 16.5 minutes / piece, with a deviation rate of 10%. Based on this, the processing capacity score is calculated to be 0.90. Similarly, the scores for roughing and honing processes are 0.95 and 0.92, respectively. The score data is normalized and combined with the real-time load rate, then input into a multilayer perceptron model to predict the process capacity trend for the next 4 hours. The model takes the load rate, processing capacity score, and historical capacity data as input, and the output shows that the capacity utilization rate for roughing is expected to reach 85%, finishing 78%, and honing 70%. This step enhances the foresight of allocation decisions through cycle time analysis and prediction.

[0072] S1043. The minimum cost flow algorithm is used to optimize the allocation of internal multi-process parts, and the priority allocation order is determined based on the ratio of unit processing time to remaining capacity.

[0073] In this embodiment of the invention, for the parts to be processed, such as 50 cylinder blocks, a minimum cost flow algorithm is used for multi-process allocation. The ratio of the unit processing time of each part to the remaining capacity of the process is calculated: 0.15 for roughing, 0.12 for finishing, and 0.10 for honing. Based on these ratios, 20 cylinder blocks are preferentially allocated to the honing process, 18 to the finishing process, and 12 to the roughing process to minimize processing costs and balance process load. This step, through the minimum cost flow algorithm, ensures efficient and balanced part allocation and optimizes resource utilization between internal processes.

[0074] S1044. Through factory equipment configuration and comprehensive capability assessment, generate a factory matching priority sequence and use the maximum flow algorithm to optimize cross-factory part allocation.

[0075] In this embodiment of the invention, a production equipment configuration list is obtained from a factory resource database, and the equipment capacity and real-time load data of each factory are summarized. The comprehensive capability score of each factory is calculated using the analytic hierarchy process (AHP). For example, the main factory is equipped with 4 precision machining centers and 6 honing machines, scoring 0.95; cooperating factory A is equipped with 3 precision machining centers and 4 honing machines, scoring 0.88; and cooperating factory B is equipped with 2 precision machining centers and 3 honing machines, scoring 0.82. A factory matching priority sequence is generated based on the scores, and a maximum flow algorithm is used to allocate parts across factories. The main factory prioritizes processing high-precision cylinder blocks, cooperating factory A handles medium-precision parts, and cooperating factory B processes ordinary-precision parts. Combining the principle of proximity allocation, transportation routes are optimized to ensure cost-effectiveness of cross-factory allocation. This step achieves efficient collaboration of cross-factory resources through comprehensive capability assessment and the maximum flow algorithm.

[0076] S105. Combine the factory's real-time processing load, target part grouping scheme, and priority sequence to generate a reprocessing process configuration scheme and identify the balance point between resource consumption and reprocessing coverage.

[0077] In this embodiment of the invention, a reprocessing process configuration scheme is generated by integrating real-time factory processing load data, target part grouping schemes, and priority sequences. Simulation technology and optimization algorithms are then used to analyze the balance between resource consumption rate and reprocessing coverage. This embodiment of the invention does not strictly limit the data acquisition frequency or algorithm parameters, allowing for flexible adjustments based on actual production needs.

[0078] S1051. Obtain process equipment operation data from the manufacturing execution database, divide the load range, use the random forest algorithm to extract the correlation between load features and priority sequence, and generate a reprocessing process path table.

[0079] In this embodiment of the invention, the uptime and capacity utilization data of the process equipment are extracted from the manufacturing execution database. The real-time load status is calculated using a moving average method, and the process load is divided into a high load range (above 80%), a medium load range (60% to 80%), and a low load range (below 60%) based on the capacity utilization rate. Taking the machining of automotive engine cylinder blocks as an example, the uptime of the roughing process is 85%, and the capacity utilization rate is 75%, falling within the medium load range. The uptime, capacity utilization rate, and maintenance interval are extracted as feature parameters and input into a random forest algorithm to calculate the correlation score between the priority sequence and the load features. For example, the correlation score for roughing is 0.82, for finishing is 0.88, and for honing is 0.75. A correlation threshold of 0.80 is set, and roughing and finishing are selected as key processes. Based on the process sequence constraints, a reprocessing flow path table is generated to ensure that roughing is executed before finishing. This step provides accurate input for simulation through feature extraction and path planning.

[0080] S1052. Construct a reprocessing scenario using a discrete event simulator, calculate resource consumption rate and coverage curves, and determine the equilibrium point parameters through piecewise regression analysis.

[0081] In this embodiment of the invention, a discrete event simulator is used. Based on the reprocessing process path table, input process parameters such as processing cycle time and changeover time are used to construct a virtual production scenario. Taking cylinder block processing as an example, the roughing cycle time is set to 15 minutes and the changeover time to 30 minutes; the finishing cycle time is 20 minutes and the changeover time to 45 minutes. After 8 hours of simulation, the roughing center has accumulated 420 minutes of operation, with a standard operating time of 480 minutes and a resource consumption rate of 87.5%; the finishing center has accumulated 400 minutes of operation, with a resource consumption rate of 83.3%. During the same period, 28 roughing operations were completed, out of a planned 30, resulting in a coverage rate of 93.3%; 24 finishing operations were completed, resulting in a coverage rate of 80%. By fitting the resource consumption rate curve and the coverage rate curve using a piecewise regression method, it was found that the values ​​of both curves at the intersection point of the 6th hour are close to 85%. The intersection point parameters are extracted: the roughing cycle time is adjusted to 18 minutes, the finishing cycle time to 22 minutes, and the changeover time is shortened to 40 minutes. This step, through simulation and regression analysis, accurately locates the balance point between resource consumption and coverage rate.

[0082] S1053. The k-means clustering algorithm is used to classify the configuration schemes, and the resource consumption and coverage indicators are calculated based on the equipment operation records. The configuration schemes are then optimized using a genetic algorithm.

[0083] In this embodiment of the invention, process parameters, equipment configuration, and processing cycle time are extracted from the reprocessing flow configuration scheme. The K-means clustering algorithm is used to calculate the Euclidean distance between schemes, classifying them into high-similarity, medium-similarity, and low-similarity groups. For example, schemes with a distance less than 0.2 are classified into the high-similarity group. Equipment operation records are obtained from the production execution database, and equipment utilization and resource consumption indicators are calculated. Using an 8-hour shift as a standard, a certain piece of equipment runs for 6.5 hours, achieving a utilization rate of 81.25%; with a cumulative operation of 240 hours and a maintenance interval of 48 hours, the resource consumption indicator is 5. The plan is to reprocess 100 cylinder blocks, with 85 completed, achieving a coverage rate of 85%. If the resource consumption indicator exceeds 1.2 times the historical average of 4.8 (i.e., 5.76), the cycle time is adjusted first, for example, increasing it from 15 minutes to 18 minutes. If it still exceeds the threshold, one more piece of equipment is added, and the capacity redundancy is calculated to be 6%. A genetic algorithm was used for optimization, with a fitness function set at 85% capacity utilization. After 200 generations, the optimal solution was obtained: 2 machines, a cycle time of 16 minutes, and resource consumption index reduced to 5.5. This step ensured the efficiency of the configuration solution through clustering and optimization.

[0084] S1054. By fitting the relationship between resource consumption and coverage through linear regression, determine the equilibrium point, verify the stability of parameters, and generate an optimized configuration scheme.

[0085] In this embodiment of the invention, resource consumption and coverage indices are extracted from the optimized configuration scheme. A linear regression algorithm is used for fitting, generating a resource consumption curve y = 0.8x + 1.2 and a coverage curve y = -0.6x + 4.5. The intersection coordinates, calculated using the least squares method, are 2.3 and 3.0, indicating that when the resource consumption index is 2.3, the coverage reaches 90%. Cross-validation is used to randomly select 50 historical configuration schemes to verify the stability of the equilibrium point. The results show that 90% of the samples have resource consumption fluctuations within ±0.2 and coverage fluctuations within ±5%.

[0086] Internally, based on the equilibrium point parameters, an optimized configuration scheme is generated: 2 machines, processing cycle time of 16 minutes, and changeover time of 30 minutes. This step, through regression analysis and verification, ensures the stability and repeatability of the configuration scheme, providing reliable support for efficient reprocessing.

[0087] S106. Calculate the real-time performance ratio of reprocessing coverage to resource consumption based on equilibrium point data, compare it with the historical best performance ratio, identify performance bottleneck risks and analyze their causes, and optimize the reprocessing process.

[0088] In this embodiment of the invention, the real-time performance ratio of coverage to resource consumption is calculated using the reprocessing data corresponding to the equilibrium point, and compared with the historical best value. If it is lower than the warning threshold, potential performance bottleneck risks are identified. Through equipment status analysis and risk classification, the causes of anomalies are traced, providing a basis for optimizing the reprocessing process. This step, through a data-driven risk warning mechanism combined with multi-dimensional feature analysis and historical data verification, significantly improves production stability in high-defect-rate scenarios. This embodiment of the invention does not strictly limit the data collection method or threshold setting, and can be adjusted according to the actual production environment.

[0089] S1061. Extract balance point data from the production execution database, calculate coverage and resource consumption rate, generate real-time performance ratio, and compare it with the historical best value.

[0090] In this embodiment of the invention, reprocessing data corresponding to the equilibrium point is obtained from the production execution database. Coverage is calculated by dividing the number of completed parts by the planned quantity, and resource consumption rate is calculated by dividing equipment runtime by standard working hours. The real-time performance ratio is then obtained by dividing the coverage rate by the resource consumption rate. Taking automotive engine block processing as an example, a batch planned to reprocess 100 cylinder blocks, actually completed 85, resulting in a coverage rate of 85%; the standard working hours were 8 hours, but actual operation was 6.8 hours, resulting in a resource consumption rate of 85% and a performance ratio of 1.0. Production records from the past 30 days are extracted from the historical database, and the top 10% of samples by performance ratio are selected, with an average optimal value of 1.25. If the real-time performance ratio is lower than the historical average of 1.05, further analysis is triggered. This step quantifies reprocessing efficiency through performance ratio, providing a data foundation for risk identification.

[0091] S1062. A performance prediction model is established using the support vector regression method, an early warning threshold is set, the risk level is marked, and the equipment status feature vector is extracted.

[0092] In this embodiment of the invention, a performance prediction curve is constructed using support vector regression based on historical production records. Input parameters include daily performance ratio, equipment runtime, and maintenance interval. The prediction curve reflects the normal fluctuation range of the performance ratio, and warning thresholds are set accordingly: below 0.9 indicates a mild warning, below 0.85 indicates a moderate warning, and below 0.8 indicates a severe warning. For example, if the performance ratio of a machining center drops to 0.82, a moderate warning is triggered. Operating parameters are extracted from the equipment management database, including machining accuracy (0.92, below the normal value of 0.95), continuous runtime (180 hours, exceeding the maintenance interval of 160 hours), and tool life (10 hours remaining, approaching the replacement cycle). These parameters are normalized to generate an equipment status feature vector. This step accurately identifies risk states through the prediction model and threshold grading.

[0093] S1063. Use hierarchical clustering to analyze the device state feature vector, identify abnormal parameter combinations, and mine the causes of risks through association rules.

[0094] In this embodiment of the invention, a hierarchical clustering method is used to group the equipment state feature vectors, calculate the Euclidean distance between classes, and identify abnormal parameter combinations. If the feature vector of a certain machining center is 0.35 away from the normal operating condition, exceeding the threshold of 0.3, it is marked as abnormal. Further, an association rule algorithm is used to calculate the confidence level between abnormal parameters and performance degradation. For example, the confidence level between exceeding the continuous running time limit and performance degradation is 85%, and the confidence level between insufficient tool life and decreased machining accuracy is 90%. Abnormal parameter combinations are classified into equipment accuracy categories such as dimensional deviations, equipment life categories such as spindle wear, and equipment maintenance categories such as abnormal lubricating oil pressure, generating a parameter abnormality feature table. This step, through clustering and association analysis, clarifies the source of risk causes.

[0095] S1064. Construct a risk early warning model based on the decision tree algorithm, verify the risk classification and generate a risk cause report to guide equipment maintenance decisions.

[0096] In this embodiment of the invention, a risk warning model is constructed using a decision tree algorithm. An abnormal parameter feature table and historical fault data are input, and risk types are classified and verified. For example, a machining center simultaneously faces equipment lifespan risks such as tool wear and maintenance risks such as insufficient coolant concentration. Historical data shows that similar parameter combinations have previously caused batch-related quality problems. A risk cause report is generated, recommending immediate equipment maintenance, tool replacement, and coolant parameter adjustment to restore machining accuracy to above 0.95. This step, through decision tree verification and report generation, provides precise guidance for maintenance decisions, preventing the escalation of quality problems.

[0097] S107. Based on the risk cause analysis results, dynamically adjust the part grouping strategy and factory matching priority, optimize the reprocessing process configuration, and verify the effects of reduced resource consumption and improved coverage by monitoring the defect rate and performance curve changes in real time. Confirm the stable state of the optimization scheme and generate an executable improvement scheme.

[0098] In this embodiment of the invention, based on the analysis results of the risk cause database, the part grouping strategy and factory matching priority are dynamically updated to generate an optimized reprocessing process configuration scheme. Through real-time data monitoring and cyclical verification, it is ensured that resource consumption is reduced while reprocessing coverage is improved, ultimately forming a stable improvement scheme. This achieves intelligent optimization of the production process in high defect rate scenarios, significantly improving production efficiency and resource utilization. This embodiment of the invention does not strictly limit the monitoring cycle or threshold standards, and can be flexibly adjusted according to actual production needs.

[0099] S1071. Extract equipment status and performance degradation records from the risk cause database, set risk thresholds, use deep learning algorithms to predict part risk classification, and update the part grouping strategy.

[0100] In this embodiment of the invention, equipment status characteristics and performance degradation records are obtained from a risk cause database. Risk thresholds are set based on historical data: performance degradation exceeding 10% is considered high risk, 5% to 10% is medium risk, and less than 5% is low risk. Taking automotive engine cylinder block machining as an example, a machining center ran continuously for 180 hours, exceeding its maintenance cycle, and its machining accuracy dropped to 0.92, below the standard value of 0.95, triggering a high-risk warning. Using a deep learning algorithm, with equipment runtime, machining accuracy, and maintenance interval as inputs, the risk distribution of a batch of 100 cylinder blocks is predicted. The results show 35 high-risk, 45 medium-risk, and 20 low-risk parts, generating an updated parts grouping table.

[0101] S1072. Calculate the overall factory score based on equipment operating status, processing accuracy, and maintenance cycle, prioritize equipment, and optimize process parameters to maximize the product of load rate and capacity utilization.

[0102] In this embodiment of the invention, the operating status of the equipment is scored, with operating status weighted at 0.4, machining accuracy weighted at 0.4, and maintenance cycle weighted at 0.2. The overall factory score is calculated by weighted summation. The operating status score is based on continuous operating time, failure interval, and maintenance plan execution rate; the machining accuracy score includes dimensional pass rate, surface roughness, and geometric accuracy; the maintenance cycle score covers equipment lifespan, lubrication status, and cleanliness. For example, the overall scores of five machining centers in a workshop are 0.92, 0.88, 0.85, 0.82, and 0.78, respectively. The top 30% of the equipment (0.92 and 0.88) are classified as high-priority, the middle 40% (0.85 and 0.82) as medium-priority, and the remainder as low-priority. For high-priority equipment, a genetic algorithm is used to optimize process parameters, with the product of equipment load rate and capacity utilization rate as the objective function. For example, a piece of equipment originally had a cycle time of 15 minutes, a load rate of 85%, and a utilization rate of 80%. After optimization, the cycle time was adjusted to 18 minutes, the load rate dropped to 75%, and the utilization rate increased to 85%, with the product increasing from 0.68 to 0.74.

[0103] S1073. Verify the optimized process parameters, generate new processing cycles and changeover cycles, and evaluate the defect rate and performance curve changes through real-time monitoring.

[0104] In this embodiment of the invention, the optimized process parameters are verified through a capacity calculation module to evaluate the capacity matching degree between each process. For example, the optimized daily capacities of the five machines are 26, 24, 22, 20, and 18 units, respectively, with the capacity difference between processes controlled within 10%. The changeover cycle is extended from 4 hours to 6 hours to reduce changeover frequency and improve production continuity. A 4-hour time window is used to monitor production data; the defect rate in the first cycle decreases from 4.2% to 3.5%, a performance improvement of 15%; the defect rate in the second cycle is 3.3%, a performance improvement of 18%; and the defect rate in the third cycle is 3.4%, a performance improvement of 16%. This step ensures the effectiveness of the optimization scheme through real-time monitoring and parameter verification.

[0105] S1074. The proportion of reduced computing resource consumption and the proportion of increased coverage are confirmed to be in a stable state, and an improvement scheme verification report is generated through cross-validation.

[0106] In this embodiment of the invention, the optimized resource consumption is reduced by 12% and the coverage is increased by 16%, and the fluctuation of the indicators is verified to be less than 5% over three consecutive monitoring periods. The results show that the fluctuation in each period did not exceed the threshold, confirming that production has entered a stable state. Using cross-validation, 10 sets of historical optimization schemes were randomly selected for comparison, and the fluctuation range of the current scheme is better than the historical average. The optimized parameter combination, including a cycle time of 18 minutes and a changeover cycle of 6 hours, is recorded in the improvement scheme library as a reference for subsequent production. This step, through stability and verification analysis, ensures the reliability and repeatability of the optimization scheme.

[0107] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A method for multi-objective balancing based reprocessing decision of defective parts, characterized in that, The method comprises: acquiring real-time defect rate and part distribution information, setting a standard defect threshold, identifying and fitting part performance parameters if the defect rate exceeds the standard defect threshold, obtaining an average performance curve, calculating the slope change and mutation amplitude of the curve, determining the time and influence range of the variation point; based on the defect rate and performance curve, the parts are dynamically clustered to obtain part grouping, the target part grouping is determined, the parts include internal parts and cross-factory parts, the internal parts are distributed according to the processing capacity and defect rate, and the cross-factory parts are distributed according to the geographical location, cost, processing capacity and real-time defect rate; the resource consumption of the factory after executing the target part grouping is identified, the internal parts are distributed according to the load, the factory matching priority is obtained by combining the historical data, and the cross-factory parts are distributed accordingly; the processing load is acquired, the priority distribution sequence is combined with the part grouping scheme and the load to obtain a reprocessing process configuration scheme, and the reprocessing process configuration scheme is simulated to identify the balance point of resource consumption and coverage rate; the coverage rate and resource consumption ratio corresponding to the balance point are compared with the historical optimal ratio, if the balance point is worse than the historical optimal ratio, it is determined that there is a risk of system performance bottleneck, and the risk causes are analyzed; the part grouping strategy and the factory matching priority are dynamically updated according to the risk causes to obtain an optimized reprocessing process configuration, the optimized defect rate change and performance curve are obtained, it is verified whether the consumption is reduced and the coverage rate is improved, the stable state is judged, and an executable improvement scheme is formed.

2. The method of claim 1, wherein, The acquisition of real-time defect rate and part distribution information, and the setting of a standard defect threshold, comprise: acquiring part position information and quality record table data, obtaining a defect station data set through part traceability coding association; the defect station data set is classified and counted according to the station number, and the real-time defect rate data of the station is obtained by dividing the number of defects by the total production quantity; the station detection data is acquired by using an automatic detection device, and the defect detection rate index is calculated by using the Pearson correlation coefficient; the defect rate data is weighted processed according to the defect detection rate index, the defect rate trend curve is obtained by using the exponential smoothing method, and the defect rate abnormal value interval is identified by using the box plot analysis method.

3. The method of claim 1, wherein, If the defect rate exceeds the standard defect threshold, the part performance parameters are identified and fitted to obtain an average performance curve, the slope change and mutation amplitude of the curve are calculated, and the time and influence range of the variation point are determined, which comprises: acquiring measurement data of part detection equipment and performing normalization processing to obtain a standardized performance parameter set; for the standardized performance parameter set, the sliding window method is used to calculate the mean and standard deviation, and the performance parameter time series is obtained by removing abnormal data points according to the three-sigma rule based on the mean and standard deviation; according to the performance parameter time series, the least square method is used to calculate the fitting curve equation, and the parameter slope sequence is obtained by calculating the parameter change rate at adjacent time points according to the fitting curve equation; for the parameter slope sequence, it is judged whether the slope difference value of adjacent two points exceeds the preset threshold, if it exceeds the preset threshold, it is marked as a variation point, and the variation points are grouped to obtain the start and end time points of the influence range by using the clustering method.

4. The method of claim 1, wherein, The dynamic clustering of the parts based on the defect rate and performance curve obtains a part grouping, determines a target part grouping, the parts include internal parts and cross-factory parts, the internal parts are distributed according to the processing capacity and the defect rate, and the cross-factory parts are distributed according to the geographical position, the cost, the processing capacity and the real-time defect rate, and the method comprises the following steps: calculating the Euclidean distance according to the performance curve slope value and the defect rate data, and obtaining a variation part identification table through comparison of the Euclidean distance with a preset minimum distance threshold; for the part group in the variation part identification table, calculating the performance curve fluctuation amplitude by using a curve fitting equation, and obtaining a target part grouping list through correlation analysis of the fluctuation amplitude and the real-time defect rate; obtaining the precision detection record and the running time record and the fault maintenance record of the processing equipment according to the target part grouping list, and calculating the equipment productivity index through the recorded data; for the equipment productivity index and the real-time defect rate data, constructing a process capacity distribution network to obtain an internal part distribution scheme.

5. The method of claim 4, wherein, Also includes: Obtaining the real-time defect rate data of the internal parts and the transportation cost of the cross-factory parts, generating an initial part grouping according to the real-time defect rate data, identifying the process processing capacity vector in the initial part grouping according to the equipment state of each factory, generating a target internal part distribution scheme according to the process processing capacity vector, and obtaining a global part scheduling instruction set by data fusion of the target internal part distribution scheme and the cross-factory part distribution scheme, specifically comprising: obtaining part processing records from a production execution database record, the part processing records containing quality inspection data and defect rate data, and obtaining a part risk level grouping table through three-layer grouping rules; obtaining the running state data of the processing equipment by using the equipment management database, and obtaining the equipment state comprehensive score through the weighted calculation of the three indexes of the continuous running time length proportion of the equipment, the precision inspection qualified rate and the maintenance interval length; constructing a process processing capacity vector according to the equipment state comprehensive score, the process processing capacity vector containing the processing precision index, the processing beat index and the equipment stability index, and obtaining the process equipment state characteristic value through the least square method; calculating the matching degree between the part processing requirements and the process processing capacity vector, and distributing the process equipment according to the part risk level grouping table to obtain a part processing distribution scheme; establishing a corresponding relationship table of internal parts and cross-factory parts through part coding correlation, sorting the parts in the risk level corresponding relationship table, generating a processing priority sequence according to the part risk level sorting result, generating a process scheduling time table corresponding to the priority sequence, and forming a global part scheduling instruction data set.

6. The method of claim 1, wherein, The recognition execution target part grouping factory resource consumption, according to the load distribution internal parts, combined with historical data to get factory matching priority, according to which the allocation of cross factory parts, including: from the production management platform to obtain process scheduling data and equipment operation record, through the calculation of on-line processing parts quantity divided by the standard capacity of equipment to get equipment load rate data;According to the device load rate data and process sequence specified in the process specification, the upper limit value and the lower limit value of the device capacity are calculated by using linear programming method;The actual tact time record is processed by weighted average method to obtain the standard processing tact of process, and the process processing capacity score is generated according to the deviation value of the standard processing tact and the actual tact time record;The internal parts are allocated to multiple processes, and the priority allocation order is determined according to the ratio of part unit processing time to residual capacity of process.

7. The method of claim 1, wherein, The acquisition processing load, the priority allocation order and the part grouping scheme and the load are combined to obtain the reprocessing flow configuration scheme, and the reprocessing flow configuration scheme is simulated, so as to identify the balance point of resource consumption and coverage rate, including: obtaining the process equipment start-up rate data and capacity utilization rate data in the manufacturing execution database, dividing the high load interval data, the medium load interval data and the low load interval data according to the capacity utilization rate data;The random forest algorithm is used to extract the features of the process equipment start-up rate data, the capacity utilization rate data and the maintenance interval time data, and the correlation degree score of the priority sequence and the load characteristics is obtained;According to the correlation degree score, the correlation threshold is set, the matching process data is filtered through the correlation threshold, and the reprocessing flow path table is generated according to the processing sequence constraint between processes;The discrete event simulator is used to build the reprocessing flow scene, the reprocessing flow path table is input into the simulation environment, the resource consumption rate curve is obtained by dividing the cumulative running time by the standard running time, the coverage rate curve is obtained by dividing the finished part quantity by the planned quantity, and the intersection parameters of the resource consumption rate curve and the coverage rate curve are calculated by using the piecewise regression method.

8. The method of claim 7, wherein, Also includes: The reprocessing process configuration scheme is classified to obtain a configuration scheme set, resource consumption data and reprocessing coverage of different configuration schemes are obtained from the configuration scheme set, if the resource consumption data exceeds a preset consumption threshold, the configuration scheme is adjusted to obtain an optimized configuration scheme set, and a linear regression algorithm is used to predict the relationship between resource consumption and reprocessing coverage according to the optimized configuration scheme set to obtain balance point data, which specifically includes: obtaining process parameters, equipment configuration and processing cycle in the reprocessing process configuration scheme, and using a k-means clustering algorithm to obtain a configuration scheme grouping table according to the process parameters; obtaining equipment operation records from a production execution database according to the configuration scheme grouping table, and calculating equipment utilization and resource consumption indicators through the equipment operation records and single-shift standard working hours; optimizing the equipment utilization, wherein the resource consumption indicator is less than a preset threshold as a constraint condition to obtain an optimized configuration scheme; extracting resource consumption indicators and coverage indicators for the optimized configuration scheme, fitting the resource consumption indicators and coverage indicators, and obtaining resource consumption values and coverage values corresponding to the balance point.

9. The method of claim 1, wherein, The balance point corresponding coverage and resource consumption ratio value is compared with the historical optimal value, if it is worse than the historical optimal value, it is determined that there is a risk of system performance bottleneck, and the risk causes are analyzed, including: obtaining the ratio of finished part quantity to planned quantity to obtain coverage value, obtaining the ratio of equipment running time to standard working hours to obtain resource consumption value, and obtaining real-time performance ratio according to the coverage value divided by the resource consumption value; extracting production records from a historical database, and establishing a performance prediction curve using a support vector regression method according to the production records; if the real-time performance ratio is lower than a warning threshold, it is marked as a risk warning state, and device state feature vectors are obtained from a device management database according to the warning state; the device state feature vectors are grouped by using a hierarchical clustering method, and parameter abnormality feature tables are obtained by calculating the inter-class distance to identify abnormal parameter combinations.

10. The method of claim 1, wherein, The part grouping strategy and factory matching priority are dynamically updated according to the risk causes to obtain an optimized reprocessing process configuration, the optimized defect rate change and performance curve are obtained, it is verified whether the consumption is reduced and the coverage is improved, the stable state is judged, and an executable improvement scheme is formed, including: obtaining device state feature records and performance reduction records from a risk cause database, setting high-risk threshold, medium-risk threshold and low-risk threshold according to historical performance data; obtaining part risk prediction classification results by using a deep learning algorithm according to the device state feature records and the performance reduction records; scoring the device running state, and calculating a factory comprehensive score by using state weight value, precision weight value and maintenance cycle weight value; dividing the devices into three priority levels according to the factory comprehensive score, and calculating process parameter optimization values that maximize the product of equipment load rate and capacity utilization rate.

Images