Multi-objective scheduling data processing method for distributed line cutting machine group

Abstract

This invention relates to the field of intelligent manufacturing and industrial data processing, specifically a multi-objective scheduling data processing method for distributed wire cutting machine clusters. The method includes: collecting task instruction data, equipment status time-series data, and actual execution time-series data containing energy consumption records; performing time alignment and data integration to generate a baseline scheduling dataset; constructing an ideal constraint model to generate an ideal scheduling time-series dataset containing task start time, end time, and resource consumption trajectory; correcting cycle time, waiting time, and resource usage parameters based on implicit loss factors and collaborative conflict factors, and simulating the scheduling state after disturbance; extracting the actual residual vector and the theoretical residual vector respectively; determining the cause of scheduling deviation based on the coupling similarity between the two, and outputting rescheduling, maintenance, hold, or verification instructions. This invention achieves the identification of real production bottlenecks and avoids excessive intervention in normal business fluctuations.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and industrial data processing, specifically a multi-objective scheduling data processing method for distributed wire cutting machine groups. Background Technology

[0002] In distributed wire cutting machine scenarios such as automotive wiring harness production, multiple wire cutting machines need to coordinate production scheduling around different wire type work orders, delivery date requirements and material resources. The scheduling results are directly related to equipment utilization, order on-time delivery rate and production process stability. Therefore, accurate data processing and deviation identification in the scheduling process are an important foundation for ensuring efficient operation of the workshop.

[0003] Existing production scheduling analysis methods have many problems. For example, they rely heavily on static rules, manual experience, or simple work order allocation results for judgment. They usually only focus on task planning and equipment availability, making it difficult to unify and integrate task instruction data, equipment status time series data, and actual execution time series data. Especially when hidden factors such as equipment cycle time decay, tool sub-health, material delivery delays, and competition for resources of the same specification persist, they cannot effectively characterize the sources of deviation between ideal production scheduling and actual execution. They also have difficulty distinguishing the differences between equipment wear and tear, coordination conflicts, and normal business disturbances, which can easily lead to a lack of targeted decisions for rescheduling, maintenance, or maintenance. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a multi-objective scheduling data processing method for distributed wire cutting machine clusters. Specifically, the technical solution of this invention includes:

[0005] Collect task instruction data, equipment status timing data, and actual execution timing data including energy consumption records, and perform time alignment and data integration to generate a baseline production scheduling dataset;

[0006] An ideal constraint model is constructed based on the benchmark production scheduling dataset to generate an ideal production scheduling time series dataset. The ideal production scheduling time series dataset includes task start time, task end time and resource consumption trajectory, as well as associated cycle time parameters, waiting parameters and resource usage parameters.

[0007] Based on preset implicit loss factors, collaborative conflict factors, and combinations thereof, the cycle time parameters, waiting parameters, and resource occupancy parameters in the ideal production scheduling time series dataset are corrected. Various perturbation-induced production scheduling states are generated through simulation. The implicit loss factor is used to characterize equipment cycle time decay and energy efficiency decline, while the collaborative conflict factor is used to characterize the waiting impact caused by material delivery delays and resource contention.

[0008] Based on the actual execution time series data and the ideal production scheduling time series dataset, the actual residual vector is extracted, and based on the production scheduling state after various disturbances and the ideal production scheduling time series dataset, various theoretical residual vectors are extracted. Among them, the actual residual vector and the various theoretical residual vectors include at least the time delay residual and the resource consumption residual.

[0009] Based on the coupling similarity between the actual residual vector and various theoretical residual vectors, the causes of production scheduling deviations are determined, and rescheduling instructions, maintenance instructions, hold instructions, or review instructions are output according to the determination results.

[0010] Preferably, before integrating the task instruction data, device status time series data, and actual execution time series data, the data is further preprocessed to obtain preprocessed data. The preprocessing includes missing value imputation, outlier detection, timestamp alignment, and data standardization.

[0011] Preferably, the process of constructing an ideal constraint model based on the benchmark production scheduling dataset and generating an ideal production scheduling time series dataset includes: setting the equipment health status to correspond to the rated processing cycle time; setting the network transmission delay to a pre-calibrated benchmark transmission delay; setting the material switching time to a pre-calibrated standard switching time; and solving the integrated data based on mixed integer linear programming to generate the ideal production scheduling time series dataset.

[0012] Preferably, the process of generating the post-disturbance production scheduling state through simulation includes: correcting the cycle time parameters, waiting parameters, and resource occupancy parameters in the ideal production scheduling time series dataset according to preset implicit loss factors and collaborative conflict factors; determining the equipment health score based on tool wear data, fault interval duration, or cycle time fluctuation index, and mapping the equipment health score to a cycle time decay factor; mapping material delivery delay and resource contention status to a queue blocking factor; and performing forward extrapolation of the corrected production scheduling process based on discrete event simulation to generate the post-disturbance production scheduling state.

[0013] Preferably, the process of extracting the actual residual vector and the theoretical residual vector includes: extracting the actual residual vector based on the collected actual execution time series data and the ideal production scheduling time series dataset; and extracting the theoretical residual vector based on the perturbation-induced production scheduling state and the ideal production scheduling time series dataset; wherein, both the actual residual vector and the theoretical residual vector include time delay residual, resource consumption residual and energy consumption fluctuation residual.

[0014] Preferably, the coupling similarity between the actual residual vector and the theoretical residual vector is obtained in the following way: based on dynamic time warping, the time delay residual, resource consumption residual and energy consumption fluctuation residual are first normalized by using a pre-calibrated maximum tolerance threshold, and then the normalized data is time-series aligned; the aligned residual data is mapped to a topological structure representing the relationship between task nodes; the coupling similarity is calculated by weighting the time alignment distance and the topological structure matching degree.

[0015] Preferably, determining the causes of production scheduling deviations based on coupling similarity includes:

[0016] The decision boundary is extracted based on the distribution of historical abnormal production scheduling samples, and used as the first threshold and the second threshold, with the first threshold set to be greater than the second threshold.

[0017] When the coupling similarity is greater than or equal to the first threshold, the current deviation is determined to correspond to the latent loss factor and / or the cooperative conflict factor; if the coupling similarity between the actual residual vector and the theoretical residual vector corresponding to the latent loss factor among multiple theoretical residual vectors is the highest, the current deviation is determined to correspond to the latent loss factor, and a maintenance command is output.

[0018] If the coupling similarity between the actual residual vector and the theoretical residual vector corresponding to the cooperative conflict factor among multiple theoretical residual vectors is the highest, the current deviation is determined to correspond to the cooperative conflict factor, and a rescheduling instruction is output; when the current deviation is determined to correspond to both the implicit loss factor and the cooperative conflict factor, maintenance instructions and rescheduling instructions are output in parallel.

[0019] When the coupling similarity is less than or equal to the second threshold, the current deviation is determined to correspond to a business disturbance, and a hold instruction is output. The business disturbance includes new work orders, changes in work order priority, delivery date adjustments, or order cancellations.

[0020] When the coupling similarity is greater than the second threshold and less than the first threshold, a verification instruction is output, and differential extraction and coupling decision are re-executed based on the updated device state time series data.

[0021] Preferably, when the method is applied to a distributed wire cutting machine cluster scenario, the task instruction data includes work order data, wire type data, delivery weight data, wire harness length data, and quantity data; the equipment status timing data includes cutting cycle data, tool wear data, wire spool remaining data, and standby idle time data; the actual execution timing data includes actual start time, actual completion time, actual resource usage records, and actual energy consumption records.

[0022] Preferably, the collaborative conflict factors include material delivery delay factors and node resource preemption factors; the node resource preemption factor is used to characterize the competitive relationship between multiple execution nodes for materials of the same specification; the queue blocking factor is used to characterize the waiting time changes caused by the material delivery delay factor and the node resource preemption factor.

[0023] Preferably, after outputting the rescheduling instruction, maintenance instruction, hold instruction, or review instruction, the method further includes: writing the execution result back to the baseline production scheduling dataset; updating the implicit loss factor and collaborative conflict factor according to the preset recursive correction rule or sliding window statistical rule based on the written-back execution result; and regenerating the perturbation-induced production scheduling state based on the updated factors to form a closed-loop adaptive processing flow.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. This invention preprocesses, aligns, and integrates task instruction data, equipment status timing data, and actual execution timing data to form a unified benchmark production scheduling dataset. This reduces the interference of clock inconsistencies, missing fields, abnormal data collection, and differences in units on the production scheduling analysis results, and improves the reliability and comparability of the input data.

[0026] 2. This invention constructs an ideal constraint model based on the rated processing cycle time, the reference transmission delay, and the standard switching time, and generates an ideal production scheduling time series dataset. This avoids directly bringing historical execution deviations into the production scheduling benchmark, and enables subsequent deviation identification to be built on the rated performance benchmark of the system.

[0027] 3. This invention introduces implicit loss factors and collaborative conflict factors to correct cycle time parameters, waiting parameters and resource occupancy parameters, and combines discrete event simulation to generate production scheduling status after disturbance. It can transform implicit problems such as equipment cycle time decay, tool sub-health, material delivery delay and competition for resources of the same specification into calculable and deducible theoretical damage trajectories.

[0028] 4. This invention extracts the actual residual vector and the theoretical residual vector separately, and incorporates the time delay residual, resource consumption residual, and energy consumption fluctuation residual into a unified residual space. Then, it combines dynamic time warping and task-related topology matching to calculate coupling similarity, which effectively improves the accuracy and robustness of identifying the causes of production scheduling deviations. It can more accurately distinguish between equipment wear and tear, coordination conflicts and normal business disturbances such as new work orders, priority changes, delivery date adjustments, or order cancellations, and output rescheduling instructions, maintenance instructions, hold instructions, or review instructions in a targeted manner, thereby reducing the risk of misscheduling, mismaintenance, and missed judgments.

[0029] 5. This invention forms a closed-loop adaptive processing flow by writing back the execution results and updating relevant factors according to the recursive correction rules or sliding window statistical rules, so that the disturbance simulation and subsequent decisions can continuously closely approximate the real workshop state, thereby improving the production scheduling stability, on-time delivery guarantee capability and production process control level of the distributed wire cutting machine group during long-term operation. Attached Figure Description

[0030] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0031] Figure 1 This document provides a flowchart illustrating the multi-objective scheduling data processing method for distributed wire cutting machine clusters, as described in an embodiment of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0033] A multi-objective scheduling data processing method for distributed wire trimming machine clusters includes:

[0034] Collect task instruction data, equipment status timing data, and actual execution timing data including energy consumption records, and perform time alignment and data integration to generate a baseline production scheduling dataset;

[0035] An ideal constraint model is constructed based on the benchmark production scheduling dataset to generate an ideal production scheduling time series dataset. The ideal production scheduling time series dataset includes task start time, task end time and resource consumption trajectory, as well as associated cycle time parameters, waiting parameters and resource usage parameters.

[0036] Based on preset implicit loss factors, collaborative conflict factors, and combinations thereof, the cycle time parameters, waiting parameters, and resource occupancy parameters in the ideal production scheduling time series dataset are corrected. Various perturbation-induced production scheduling states are generated through simulation. The implicit loss factor is used to characterize equipment cycle time decay and energy efficiency decline, while the collaborative conflict factor is used to characterize the waiting impact caused by material delivery delays and resource contention.

[0037] Based on the actual execution time series data and the ideal production scheduling time series dataset, the actual residual vector is extracted, and based on the production scheduling state after various disturbances and the ideal production scheduling time series dataset, various theoretical residual vectors are extracted. Among them, the actual residual vector and the various theoretical residual vectors include at least the time delay residual and the resource consumption residual.

[0038] Based on the coupling similarity between the actual residual vector and various theoretical residual vectors, the causes of production scheduling deviations are determined, and rescheduling instructions, maintenance instructions, hold instructions, or review instructions are output according to the determination results.

[0039] This embodiment provides a multi-objective scheduling data processing mechanism for distributed thread cutting machine clusters, such as... Figure 1 As shown; specifically, the scenario is set as follows: within the same production shift in an automotive wiring harness factory, three distributed wire cutting machines, M1, M2, and M3, each handle different wire type work orders. After the manufacturing execution system issues the work orders, the system continuously receives three types of data streams: one is task instruction data, mainly including work order number, target wire type, delivery weight, quantity, and priority; another is equipment status time-series data, mainly including cycle time of each machine, tool status, wire reel remaining, and idle time; the third is actual execution time-series data, mainly including actual start-up, actual completion, actual wire reel occupied, actual energy consumption, and waiting time records; the system aligns the three types of data according to a unified time base and forms a baseline production scheduling dataset;

[0040] In the specific implementation process, each task can be abstracted as a task node T. For ease of explanation, it is assumed that there are currently only three work orders: T1 for 100 pieces of type A wire harness, T2 for 80 pieces of type B wire harness, and T3 for 60 pieces of type A wire harness. After integration, the following baseline data is obtained: M1 can currently process type A wire harnesses, M2 can process both type A and type B wire harnesses, and M3 can process type B wire harnesses; the real-time cycle time of M1 is 5 seconds per piece, M2 is 4 seconds per piece, and M3 is 6 seconds per piece; the system records that the current wire spool inventory can only support M1 and M2 to process one of the type A wire harnesses at the same time.

[0041] Based on this benchmark data, the system constructs an ideal constraint model, which initially excludes implicit equipment losses and coordination conflicts, retaining only process constraints, equipment capacity constraints, and delivery target constraints to generate an ideal production scheduling time series dataset. For example, under ideal conditions, T1 is assigned to M2, starting at 08:00 and ending at 08:06:40; T2 is assigned to M3, starting at 08:00 and ending at 08:08:00; and T3 is assigned to M1, starting at 08:00 and ending at 08:05:00. Simultaneously, the ideal resource consumption trajectory for each type is recorded; for example, the ideal consumption for type A wire is 50 meters and 30 meters respectively, and the ideal consumption for type B wire is 48 meters.

[0042] Based on this, the system further introduces two types of disturbance parameters: one is the implicit loss factor, which is used to characterize the decrease in cycle time and the deterioration of energy efficiency caused by the degradation of equipment performance; the other is the cooperative conflict factor, which is used to characterize the waiting caused by material delivery delays and competition for resources of the same specification.

[0043] For ease of understanding, we can assume that when the wear of the M2 tool reaches a first preset threshold, the corresponding cycle time decay coefficient is 1.10, meaning that the original 4 seconds per piece is corrected to 4.4 seconds per piece; we can also assume that the delivery delay of the A-type wire spool is 40 seconds, which corresponds to an increase of 40 seconds in the queue waiting time; the system corrects the cycle time parameters, waiting parameters, and resource occupancy parameters in the ideal timing accordingly, and extrapolates forward through discrete event simulation to obtain the production scheduling state after the disturbance; for example, in the simulation, the end time of T1 changes from 08:06:40 to 08:07:20, and T3 is delayed to start at 08:00:40 and end at 08:05:40 due to contention for A-type wire;

[0044] Enter the double difference extraction process; the actual residual vector is obtained by subtracting the ideal production scheduling time series data from the actual execution time series data; the theoretical residual vector is obtained by subtracting the ideal production scheduling time series data from the perturbation-induced production scheduling state; taking the above three work orders as examples, if the actual completion time of T1 is 08:07:18, then the actual time delay residual of T1 is 38 seconds; if the actual consumption of type A wire is 2 meters more than the ideal value, then the resource consumption residual is 2 meters;

[0045] Correspondingly, theoretical simulation yields a theoretical delay of 40 seconds for T1 and an additional theoretical resource consumption of 1.8 meters; thus, the actual residual vectors can be generated. =[38 seconds, 2 meters, ...] and theoretical residual vector =[40 seconds, 1.8 meters, ...]; The system then identifies the cause based on the coupling similarity of these two sets of residuals; If the two are very close, it means that the current production scheduling deviation is not purely from temporary business changes, but is consistent with the preset hidden losses or coordination conflicts; If the difference is large, it means that it is more likely to be a business disturbance such as order insertion or delivery date adjustment.

[0046] As an exception handling mechanism, if a work order lacks a completion time on the actual execution side, the system can first mark the work order as an unclosed stage and only use the collected start time, resource usage, and energy consumption segments to participate in the stage residual calculation; if the ideal production scheduling model cannot directly obtain a feasible solution due to constraint conflicts, low-priority work orders can be released from high to low according to the delivery option weight to generate a temporary feasible benchmark, and then continue the subsequent comparison; if the data of a certain device is interrupted for a long time, the system can replace the device status time series corresponding to the device with the most recent window statistical value and add a review mark to the final output to avoid directly generating maintenance conclusions;

[0047] During the morning shift (08:00-10:00) of the automotive wiring harness factory, the system detected that the A-type wiring harness work orders received by M2 were consistently delayed by 30 to 45 seconds beyond the ideal plan. At the same time, the deviation of the A-type wire consumption from the ideal trajectory was within the preset tolerance range. Moreover, these deviations were highly similar to the theoretical residuals generated by slight tool wear and material delivery delays on both the time and resource axes. Therefore, the system determined that the current deviations were mainly due to sub-health of the equipment and collaborative waiting, rather than the single variable of order insertion disturbance, and output corresponding rescheduling, maintenance, hold, or review instructions accordingly.

[0048] The purpose of this step is to structurally decompose the deep evolution mechanism of production scheduling deviation from the result display level to the cause location level by using ideal benchmarks, disturbance simulation and dual residual coupling, so that the system can both identify real production bottlenecks and avoid excessive intervention in normal production disturbances.

[0049] Furthermore, before integrating the task instruction data, device status timing data, and actual execution timing data, the process also includes: preprocessing the task instruction data, device status timing data, and actual execution timing data to obtain preprocessed data. The preprocessing includes missing value imputation, outlier detection, timestamp alignment, and data standardization.

[0050] This embodiment provides a preprocessing step for data purification before production scheduling. Specifically, in the workshop scenario of the previous embodiment, although the three types of data all come from the same production process, they come from different systems, often resulting in problems such as inconsistent clocks, missing fields, and inconsistent units. If the ideal modeling and residual calculation are directly entered, the collected noise is easily misjudged as production scheduling deviation.

[0051] In the specific implementation process, preprocessing can be performed continuously in four steps; the first step is to fill in missing values; for example, if the delivery weight of T2 in the task instruction flow is missing, but the default delivery weight of the same batch of work orders from the same customer is 0.8, then the system can fill in 0.8; another example is that M1 did not upload the tool wear value between 08:12:00 and 08:12:10, but the two records before and after were 0.20 and 0.24 respectively, then linear interpolation can be used to fill in 0.22; the second step is to detect outliers; for example, if M3 uploads a cycle time of 0.3 seconds per piece at a certain moment, which is significantly lower than the equipment process limit of 3 seconds, then this value is marked as outlier and replaced with the neighborhood median of 4.8 seconds;

[0052] The third step is to align the timestamps. Assuming the manufacturing execution system clock reference is 08:00:00, the programmable logic controller (PLC) is 5 seconds fast and the energy consumption acquisition gateway is 3 seconds slow, the system will uniformly roll back the PLC data by 5 seconds and move the gateway data forward by 3 seconds, so that the same event can correspond on the same timeline. The fourth step is to standardize the data. Since the cycle time is measured in seconds, the coil balance is measured in meters, and the energy consumption is measured in kilowatt-hours, the system can map each feature to a unified numerical range, such as 0 to 1, so that the weight of each dimension can be controlled in the subsequent similarity calculation.

[0053] To illustrate, a specific application scenario is provided: Work order T1 is recorded as starting at 08:05:00 in the Manufacturing Execution System (MES), with a corresponding start signal time of 08:05:04 in the Programmable Logic Controller (PLC), and a power ramp-up time of 08:04:58 in the Energy Gateway. After timestamp alignment, all three timestamps are uniformly adjusted to approximately 08:05:00, allowing for a device start-up error window of no more than 2 seconds. If the error exceeds this window, the system checks whether it indicates an acquisition anomaly or device preheating behavior. This processing accurately binds the task node of T1 to the device status segment.

[0054] As an anomaly handling mechanism, if the duration of continuous missing data exceeds a preset threshold, such as if a device fails to upload cycle data for 10 minutes, interpolation will no longer be used. Instead, the segment will be marked as a low-confidence segment, and its weight will be reduced in subsequent models. If multiple conflict sources occur simultaneously for the same field, such as the actual completion time differing by more than 5 minutes between the manufacturing execution system and the device, the shutdown completion event on the device side will take precedence, and the manufacturing execution system record will be retained as an audit copy. If the maximum and minimum values ​​of a certain dimension are the same within the current window during the standardization process, resulting in a denominator of zero, the dimension will be directly mapped to a constant of 0.5 or kept at its original value with a static label to avoid calculation anomalies.

[0055] During the production window from 08:00 to 08:30, tool wear data for M2 was lost twice, and M3 experienced a sudden jump in cycle time. There was a 4-second clock deviation between the manufacturing execution system and the programmable logic controller. After preprocessing, the three types of data were organized into a stable data table under a unified timeline and then used for ideal production scheduling modeling and residual extraction. The deviation judgment obtained in this way is no longer affected by the high-frequency abnormal pulses collected.

[0056] The purpose of this step is to provide reliable input for subsequent ideal benchmark construction, perturbation simulation and coupling decision, and to reduce misjudgments caused by insufficient quality of the original data.

[0057] Furthermore, an ideal constraint model is constructed based on the benchmark production scheduling dataset to generate an ideal production scheduling time series dataset, including: setting the equipment health status to correspond to the rated processing cycle time; setting the network transmission delay to a pre-calibrated benchmark transmission delay; setting the material switching time to a pre-calibrated standard switching time; and solving the integrated data based on mixed integer linear programming to generate the ideal production scheduling time series dataset.

[0058] This embodiment provides a construction step for an ideal constraint model. Specifically, in the aforementioned scenario, if the historical average cycle time is directly used as the benchmark, it will introduce the cumulative deviation in the historical operation process, causing the production scheduling benchmark to deviate from the rated performance of the equipment. Therefore, this embodiment uses the rated processing cycle time, the benchmark network transmission delay, and the standard switching time as a unified reference to construct an ideal constraint model that is close to zero interference.

[0059] In the specific implementation process, rated capacity parameters can be set for each piece of equipment first; for example, the rated cycle time of M1 is 4.5 seconds per piece, M2 is 4 seconds per piece, and M3 is 5.5 seconds per piece; the baseline transmission delay from the manufacturing execution system to the equipment control layer is uniformly set to 1 second; the standard line changeover time between different line types is set to 20 seconds; at this time, for each task node ,in A positive integer index representing the task sequence number, and for each device ,in Define a decision variable for whether to assign a device serial number to a positive integer index.

[0060] For ease of understanding, a simplified verification model can be used: Of the three work orders T1, T2, and T3, T1 and T3 use type A wiring harnesses, and T2 uses a type B wiring harness; M1 only supports type A, M3 only supports type B, and M2 supports both. Therefore, the model must simultaneously satisfy the following relationships: each work order can only be executed by one compatible device; two work orders on the same device cannot overlap in time; if adjacent work orders have different wiring harnesses, a 20-second switching time must be added; the overall objective can comprehensively consider the shortest completion time, minimum delay, and lowest resource consumption.

[0061] For example, the ideal timing sequence might be obtained after solving: T3 starts at 08:00:01 and ends at 08:04:31 on M1; T2 starts at 08:00:01 and ends at 08:07:21 on M3; T1 starts at 08:00:01 and ends at 08:06:41 on M2; here, 01 seconds is the earliest start point after the reference transmission delay is converted; if T1 is executed by M1 instead, the overall delivery risk may increase because the cycle time of M1 is lower than the preset standard cycle time. Therefore, the solver will automatically select the allocation that makes the objective function more optimal.

[0062] Since this embodiment uses mixed integer linear programming, discrete variables are used to characterize whether a task is assigned to a certain device, and continuous variables are used to characterize the start time, end time, and cumulative consumption. This not only keeps the constraints clear, but also facilitates the subsequent generation of a structured ideal production scheduling time series dataset, in which each record corresponds to the task number, device number, start time, end time, and resource consumption trajectory.

[0063] As an exception handling mechanism, if the total number of work orders within a certain production window is too large, causing the time required for a single solution to be too long, the work orders can be first layered according to delivery date, forming the main problem for work orders with near-delivery dates and the secondary problems for work orders with far-delivery dates. The ideal scheduling result for the main problem is obtained first, and then the secondary problems are inserted in a rolling manner. If a work order has no equipment compatibility, it is immediately marked as an unschedulable node and upstream process verification is triggered, rather than forcibly putting it into the model. If the target values ​​of multiple feasible solutions are close, the group with fewer line changes is selected first to improve the stability of the ideal baseline.

[0064] In a batch of automotive door wiring harness orders, both Type A wiring harness work orders and Type B wiring harness work orders entered the 08:00 window simultaneously. The system solved the problem based on the rated cycle time, a 1-second issuance delay, and a 20-second standard line changeover time, obtaining an ideal production schedule that is unaffected by equipment wear and material coordination. All subsequent actual and theoretical deviations are based on this schedule as a unified reference.

[0065] The purpose of this step is to establish a theoretically optimal benchmark that is not affected by historical inefficiencies, so that subsequent residuals truly reflect the degree of deviation from the ideal, rather than deviating from a historical average that has introduced systematic bias.

[0066] Furthermore, the production scheduling state after disturbance is generated through simulation, including: correcting the cycle time parameters, waiting parameters, and resource occupancy parameters in the ideal production scheduling time series dataset according to preset implicit loss factors and collaborative conflict factors; determining the equipment health score based on tool wear data, fault interval duration, or cycle time fluctuation index, and mapping the equipment health score to the cycle time decay factor; mapping material delivery delay and resource contention status to queue blocking factors; and performing forward extrapolation of the corrected production scheduling process based on discrete event simulation to generate the production scheduling state after disturbance.

[0067] This embodiment provides a disturbance injection and simulation generation step; specifically, relying solely on ideal production scheduling results is still insufficient to explain the hidden losses that slowly accumulate in the workshop but have not yet formed downtime-level faults; therefore, based on ideal production scheduling, the system introduces known hidden losses and cooperative conflict parameters into the model to generate a theoretical production scheduling evolution state that characterizes the influence of the above-mentioned disturbance factors.

[0068] In the specific implementation process, an equipment health score is constructed. Considering the differences in the physical attributes of each parameter, the system performs normalization processing on the tool wear, fault interval duration, and cycle time fluctuation index based on the reference boundary and maps them to between 0 and 1. The higher the score, the closer the equipment operating status is to the rated standard. The specific normalization processing rule is as follows: for negative characteristic parameters such as tool wear and cycle time fluctuation, the system calculates the difference between the preset maximum tolerable physical extreme value and the actual collected value, and divides it by the absolute span value of the allowable change set by the system for the monitoring item to obtain the calibration margin.

[0069] For positive characteristic parameters such as fault interval duration, the fault-free time that has been achieved at each workstation is directly obtained and divided by the manufacturer's recommended reliable operating time ratio. If the calculated ratio is greater than 1, it is forcibly truncated and assigned the maximum limit value of 1, thereby ensuring the transparency of the calculation model structure and avoiding the calculation process from being invisible due to the encapsulation of formulas.

[0070] For ease of explanation, let's assume the normalized value of tool wear for M2, calculated using the above methods, is 0.6; the most recent fault interval meets the preset excellent standard score of 0.8; and the cycle time fluctuation score is 0.7. Then, a weighted average method can be used, assuming weights of 0.4, 0.3, and 0.3 respectively, to obtain a comprehensive health score of 0.69, approximated as 0.7. This equipment health score is then mapped to a cycle time decay factor; specifically, the following mapping rule can be used:

[0071] ;

[0072] in, The beat decay factor, Rate the health of the equipment. This is the sensitivity coefficient to equipment performance degradation; the sensitivity coefficient to equipment performance degradation These are empirical constants pre-determined through linear regression analysis of historical maintenance records and operating cycle data of the same model of equipment; assuming Set to 0.33, when the health score is 1, the calculated attenuation factor is 1.00; when the health score is 0.7, the calculated attenuation factor is approximately 1.10, indicating a 10% increase in processing time; thus, the M2 principle's cycle time of 4 seconds / piece will be corrected to 4.4 seconds / piece.

[0073] The system maps collaborative waiting; assuming an average delay of 35 seconds in the delivery of type A cable reels, and M1 and M2 competing for the same type A cable with a resource contention intensity at the preset second level; the system can convert these two into a queue blocking factor, for example, 0.25, meaning an additional 25% waiting amplification before the relevant task starts; if the ideal waiting time for T3 is 0 seconds, it can be converted into a queue waiting of about 40 seconds under blocked conditions; if a task already has a 20-second switching wait, it can be corrected to 25 seconds after being blocked;

[0074] After the correction is completed, the system uses discrete event simulation to extrapolate forward. The events in the simulation can include work order issuance, equipment occupation, start of production line arrival, equipment release, task completion, etc. Taking three work orders as an example, under the ideal state, T1, T2, and T3 all start around 08:00. After the disturbance is injected, the event flow becomes: 08:00:01 T1 requests M2; 08:00:01 M2 enters processing, but the total processing time increases due to cycle time decay; 08:00:01 T3 requests M1, while waiting for the A-type production line to arrive; 08:00:41 Production line arrives, T3 starts; 08:07:20 T1 ends. The resulting production scheduling state after the disturbance not only has new start and end times, but also the corrected resource occupation trajectory and waiting chain.

[0075] As an anomaly handling mechanism, if the health score of a device is abnormally low, for example below 0.2, and is close to the point of shutdown, the system can directly mark the device as a high-risk node and add a failure shutdown event to the simulation instead of using mild decay simulation. If material delivery delays cannot be directly obtained from the existing system, they can be approximated by the sliding window average of recent delivery records. If multiple coordination conflicts occur simultaneously and cause circular waiting, the simulation engine should set a maximum number of simulation steps and deadlock release rules, such as prioritizing the release of the work order with the highest delivery weight, to avoid infinite simulation blocking.

[0076] During the morning shift, although M2 did not report a fault, the tool wear gradually increased, resulting in a cycle time 10% slower than the rated value. At the same time, the delivery delay of the A-type wire reel from the warehouse to the machine increased significantly during this period. After the system injected these two implicit factors into the ideal timing sequence, it derived a set of theoretical damage trajectories through forward extrapolation. It was found that the trajectories were characterized by a combination of M2 continuously being delayed below the first preset time threshold and M1 waiting to be stretched at the front end.

[0077] The purpose of this step is to transform implicit losses and node cooperative conflicts from abstract experience into perturbation states that can be calculated and reproduced experimentally, so as to provide a theoretical reference state for subsequent cause matching.

[0078] Furthermore, the actual residual vector and the theoretical residual vector are extracted, including: extracting the actual residual vector based on the collected actual execution time series data and the ideal production scheduling time series dataset; and extracting the theoretical residual vector based on the perturbation-induced production scheduling state and the ideal production scheduling time series dataset; wherein, both the actual residual vector and the theoretical residual vector include time delay residual, resource consumption residual and energy consumption fluctuation residual.

[0079] This embodiment provides a dual-channel residual extraction step; specifically, after the aforementioned ideal benchmark and perturbation simulation have been obtained, the system does not directly compare the actual results with the simulation results, but first encodes the deviation of the two from the ideal state in a unified manner, thereby mapping data from different sources to the same comparable residual space;

[0080] In the specific implementation process, the actual residual vector is obtained by subtracting the ideal production scheduling time series dataset from the actual execution data; the theoretical residual vector is obtained by subtracting the ideal production scheduling time series dataset from the perturbation-induced production scheduling state; for ease of calculation, each work order can be expressed as a multi-dimensional residual unit, which includes at least time delay residual, resource consumption residual and energy consumption fluctuation residual.

[0081] Taking T1 as an example, if the ideal start time is 08:00:01, the ideal end time is 08:06:41, the ideal wire consumption is 50 meters, and the ideal energy consumption is 1.2 kWh; the actual record is 08:00:03 start, 08:07:19 end, 52 meters of wire consumption, and 1.35 kWh of energy consumption, then the actual residual can be expressed as [start delay of 2 seconds, end delay of 38 seconds, 2 meters more resource consumption, and 0.15 kWh more energy consumption]; similarly, if the simulation state gives the corresponding values ​​of 08:00:01 start, 08:07:21 end, 51.8 meters of wire consumption, and 1.33 kWh of energy consumption, then the theoretical residual can be expressed as [start delay of 0 seconds, end delay of 40 seconds, 1.8 meters more resource consumption, and 0.13 kWh more energy consumption];

[0082] When multiple work orders exist, the residual units of each work order can be concatenated in chronological order to form a residual vector sequence; for example, under three work orders T1, T2, and T3, the actual residual sequence can be denoted as:

[0083] ;

[0084] The theoretical residual sequence can be denoted as:

[0085] ;

[0086] Here, each quadruple corresponds in sequence to the start / end time deviation, resource consumption deviation, and energy consumption fluctuation deviation; subsequent coupled calculations are based on this unified expression.

[0087] Compared to simply looking at the delay, this embodiment incorporates resources and energy consumption into the residual, which can effectively identify problems where the time deviation does not exceed a preset threshold, but resources and energy efficiency have deteriorated; for example, if the equipment finishes only 5 seconds late, but the energy consumption is significantly higher than the ideal value, it may still point to tool wear or increased mechanical resistance, rather than business interruption.

[0088] As an anomaly handling mechanism, if the actual energy consumption record is temporarily unavailable, the energy consumption fluctuation residual can be estimated based on the equipment power curve and runtime, and marked as the source of the estimation in the record; if a work order is split into two equipment for parallel execution, the segmented residual can be extracted at the equipment level first, and then aggregated into a total residual according to the work order number; if the ideal data and the actual data cannot correspond one-to-one at the work order granularity, for example, if a temporary splitting and reorganization occurs, the system will first remap the task segments with the same process and similar delivery time, and then perform differential to avoid residual distortion.

[0089] Within the same automotive wiring harness order batch, the system extracted that M2-related work orders generally exhibited increased end-of-line delays, slight positive deviations in wire consumption, and increased energy consumption fluctuations, while M3-related work orders only showed time fluctuations below the tolerance threshold. After encoding, the former showed three-dimensional characteristics highly similar to theoretical residuals related to tool wear, while the latter resembled normal cycle fluctuations.

[0090] The purpose of this step is to unify the actual production scheduling results and theoretical damage results into a comparable deviation vector, laying the data foundation for subsequent time-series alignment and topology matching.

[0091] Furthermore, the coupling similarity between the actual residual vector and the theoretical residual vector is obtained in the following way: Based on dynamic time warping, the time delay residual, resource consumption residual and energy consumption fluctuation residual are first normalized using a pre-calibrated maximum tolerance threshold, and then the normalized data is time-series aligned; the aligned residual data is mapped to a topological structure representing the relationship between task nodes; the coupling similarity is obtained by weighted calculation based on the time alignment distance and the topological structure matching degree.

[0092] This embodiment provides a coupling similarity calculation step; specifically, comparing the actual residual with the theoretical residual item by item is easily affected by the shift in the order of tasks within a preset time window; for example, a work order in an actual workshop may be shifted as a whole due to the delay of upstream materials, but its deviation shape is still similar to the theoretical damage state; therefore, this embodiment first performs time alignment, and then performs comprehensive matching in combination with the task association topology;

[0093] In the specific implementation process, dynamic time warping is performed on the actual residual sequence and the theoretical residual sequence. For ease of explanation, let the end delays of the actual sequence on the three work orders be [38 seconds, 5 seconds, 40 seconds], and the theoretical sequence be [40 seconds, 4 seconds, 39 seconds]. If the second work order in reality has a slight positional shift due to the insertion of the work order, dynamic time warping can allow the second point of the first sequence to be optimally aligned with the second or third point of the second sequence, thereby obtaining a smaller alignment distance.

[0094] To eliminate the impact of dimensional differences between different physical quantities such as time (seconds), resources (meters), and energy consumption (kilowatt-hours) on distance measurement, before performing multidimensional alignment calculation, the system first normalizes the residual components of each dimension using the pre-calibrated maximum tolerance threshold of each dimension, converting the absolute residuals with different dimensions into dimensionless relative residuals in the [0,1] interval.

[0095] For example, if the maximum tolerance threshold for the time dimension is 500 seconds, then the 40-second delay residual will be normalized to 0.08. The normalized resource consumption residual and energy consumption fluctuation residual are included in the multidimensional alignment. The multidimensional relative residual vector corresponding to each work order is regarded as a spatial point, and the local alignment cost is calculated based on the multidimensional Euclidean distance to obtain a unified time-series distance value. For example, after alignment, the average time dimension distance is 0.08, the resource dimension distance is 0.05, and the energy consumption dimension distance is 0.06. After weighting according to one-third equal weights, the comprehensive time-series alignment distance can be 0.063.

[0096] Project the aligned residuals into a task-related topology; this topology is not a geometric figure, but a structural graph representing the sequential dependencies, resource sharing, and waiting propagation relationships between task nodes; taking T1, T2, and T3 as an example, if T3 waits for the A-type cable, and the cable simultaneously affects the equipment used by T1, then an associated edge can be formed: T1—Resource Sharing—T3; if T2 runs independently, then only its own node is maintained; a simplified adjacency matrix can be further constructed: if T1 and T3 have resource coupling, then the positions (1,3) and (3,1) in the matrix are marked as 1, and the other irrelevant positions are marked as 0;

[0097] The matching degree between the real topology and the theoretical topology can be calculated by the edge overlap ratio. Specifically, the numerator is the number of completely identical associated edges in both, and the denominator is the total number of associated edges in the theoretical topology. For example, if two out of three edges in both are identical, the topology matching degree is 0.67; if all three edges are identical, the degree is 1.00.

[0098] The temporal alignment distance and topological matching degree are weighted and fused into a coupling similarity; the general transformation and fusion formula is: Temporal similarity = 1 - Comprehensive temporal alignment distance; Coupling similarity = Temporal similarity + Topological matching degree; where, For time-series correlation weights, For topological relevance weights, both satisfy the following conditions: ,and and All are greater than or equal to 0;

[0099] Since all underlying indicators have been normalized, the resulting coupling similarity is a dimensionless scalar within the interval [0,1]. For example, the temporal distance can be converted into a temporal similarity of 1-0.063=0.937, and then fused with the topological matching degree of 0.90 with weights of 0.6 and 0.4 to obtain a coupling similarity of 0.922. The higher this value, the closer the actual deviation is to a certain theoretical damage mode.

[0100] As an anomaly handling mechanism, if the actual sequence length is lower than the preset length threshold, such as only one work order is collected, then the topological information is insufficient. In this case, the weight of temporal similarity can be temporarily increased to reduce the influence of the topological part. If the coupling similarity of multiple theoretical patterns is close, such as the difference is no more than 0.03, then a definite conclusion will not be output directly, but the review process will be initiated. If abnormal alignment paths occur in dynamic time warping, such as a segment of actual residual being repeatedly mapped to theoretical points exceeding the preset limit, unreasonable matching can be prevented by limiting the maximum distortion window width.

[0101] In shifts of continuous processing of type A wire harnesses, the system found that the actual residuals were highly similar to the theoretical pattern of tool wear + type A wire spool contention in terms of time, resources and energy consumption. Moreover, the task association graph also showed that M1 and M2 shared the competition for wire of the same specification. Therefore, the obtained coupling similarity was significantly higher than that of other candidate patterns.

[0102] The purpose of this step is not only to compare the magnitude of production scheduling deviations at the numerical level, but also to objectively analyze their distribution similarity in multidimensional features and their topological path matching degree between task nodes, thereby improving the robustness and reliability of the cause identification model.

[0103] Furthermore, determining the cause of production scheduling deviation based on coupling similarity includes: extracting a decision boundary based on the distribution of historical abnormal production scheduling samples, which serves as a first threshold and a second threshold, and setting the first threshold to be greater than the second threshold; when the coupling similarity is greater than or equal to the first threshold, determining that the current deviation corresponds to a latent loss factor and / or a collaborative conflict factor;

[0104] If the actual residual vector has the highest coupling similarity with the theoretical residual vector corresponding to the latent loss factor among multiple theoretical residual vectors, the current deviation is determined to correspond to the latent loss factor, and a maintenance command is output; if the actual residual vector has the highest coupling similarity with the theoretical residual vector corresponding to the cooperative conflict factor among multiple theoretical residual vectors, the current deviation is determined to correspond to the cooperative conflict factor, and a rescheduling command is output.

[0105] When the current deviation is determined to correspond to both the implicit loss factor and the collaborative conflict factor, maintenance instructions and rescheduling instructions are output in parallel. When the coupling similarity is less than or equal to the second threshold, the current deviation is determined to correspond to a business disturbance, and a hold instruction is output. The business disturbance includes new work orders, work order priority changes, delivery date adjustments, or order cancellations. When the coupling similarity is greater than the second threshold but less than the first threshold, a review instruction is output, and differential extraction and coupling decision are re-executed based on the updated equipment status timing data.

[0106] This embodiment provides a decision and instruction output mechanism based on dual thresholds. Specifically, after obtaining the coupling similarity in the previous embodiment, if hierarchical decision rules are lacking, the system is prone to two problems: if a single judgment benchmark is used, if the threshold is set too small, it is easy to trigger maintenance and cause misjudgment; if the threshold is set too large, it is easy to cause latent abnormal features to be not effectively identified. Therefore, this embodiment introduces a high threshold, a low threshold, and an intermediate verification area.

[0107] In the specific implementation process, the first threshold can be preset to 0.85 and the second threshold to 0.55. The first and second thresholds are dynamic decision boundaries extracted by the system in advance by using clustering algorithms or statistical quantile methods based on the known abnormal sample distribution in the historical benchmark production scheduling dataset. If the coupling similarity is not less than 0.85, it indicates that the actual deviation is highly consistent with a certain theoretical pattern, and it can be determined that the current deviation mainly originates from implicit losses and / or cooperative conflicts.

[0108] Furthermore, if the similarity to the theoretical pattern of sub-health of equipment is the highest, then a maintenance instruction is output, such as generating an M2 tool inspection and replacement work order; if the similarity to the theoretical pattern of material delivery delay or resource preemption is the highest, then a rescheduling instruction is output, such as adjusting T3 from M1 to the idle M2 or postponing low-priority work orders to release coil resources for high-priority A-type wire harnesses.

[0109] If the coupling similarity is not higher than 0.55, it indicates that the actual residual differs significantly from the theoretical model of the known latent problem, and is more likely to be a normal production disturbance, such as new order insertions, early delivery by customers, or order cancellation and rescheduling. In this case, the system outputs a hold instruction, that is, to maintain the current production scheduling master topology unchanged and only record this business disturbance for the upper-level planning system to refer to. This can avoid frequent disruptions to the production rhythm due to normal order changes.

[0110] When the coupling similarity is between 0.55 and 0.85, it falls within the review interval. This review interval is not directly considered an anomaly, nor is it directly regarded as a normal production disturbance; instead, a review instruction is output. After review, more recent equipment status time-series data can be updated, and residual extraction and coupling decision are re-executed. For example, if the similarity of a work order at 08:20 is 0.72, the system does not immediately issue a maintenance conclusion but continues to collect M2 cycle time, tool vibration, and energy consumption changes over the next 10 minutes. If the similarity rises to 0.88 after review, it is converted to maintenance or rescheduling; if it drops to 0.50, it is classified as a business disturbance.

[0111] As an anomaly handling mechanism, if two theoretical patterns have a similarity score higher than the first threshold at the same time, such as equipment sub-health and resource preemption both being above 0.87, then maintenance instructions and rescheduling instructions can be output in parallel. If, after outputting rescheduling, it is found that the target equipment no longer has the capability in the latest window, then the current production schedule is maintained and a manual review mark is added. If the number of reviews reaches the preset limit and still cannot converge, then the system should classify the deviation into the unknown pattern library for absorption during subsequent factor updates, rather than infinitely looping the review.

[0112] During the morning shift, the system calculated a coupling similarity of 0.91 for the M2 related deviation, corresponding to the highest matching mode of tool wear, and therefore automatically generated a maintenance work order. At the same time, the similarity calculation for the A-type wire spool waiting chain was 0.88, corresponding to material coordination conflict, so the system reassigned some A-type work orders to subsequent idle windows. Another batch of C-type wire harness work orders added by customers at the last minute had a coupling similarity of only 0.32, and the system maintained the original equipment health judgment based on this to avoid triggering maintenance unnecessarily.

[0113] The purpose of this step is to clearly distinguish between three scenarios—identified abnormalities, normal business changes, and uncertainties—by using tiered thresholds, thereby reducing the risk of mishandling and omissions.

[0114] Furthermore, when the method is applied to a distributed wire cutting machine cluster scenario, the task instruction data includes work order data, wire type data, delivery weight data, wire harness length data, and quantity data; the equipment status timing data includes cutting cycle time data, tool wear data, wire spool remaining data, and standby idle time data; and the actual execution timing data includes actual start time, actual completion time, actual resource usage records, and actual energy consumption records.

[0115] This embodiment provides a field-based data organization method for distributed wire cutting machine groups; specifically, in the aforementioned automotive wiring harness factory scenario, in order to connect production scheduling modeling, disturbance injection, residual extraction, and cause determination, it is necessary to impose stable constraints on the business meaning of various types of data.

[0116] In practice, task instruction data can be organized around the work order granularity; for example, work order O101 corresponds to type A wire harness, with a single piece length of 0.5 meters, a quantity of 100 pieces, and a delivery weight of 0.9; work order O102 corresponds to type B wire harness, with a single piece length of 0.6 meters, a quantity of 80 pieces, and a delivery weight of 0.7; wire type data is used to describe the wire specifications, colors, terminal combinations, or switching rules corresponding to different work orders; delivery weight is used to reflect the urgency of orders in ideal production scheduling and rescheduling; wire harness length and quantity directly determine processing time and wire resource consumption;

[0117] The equipment status timing data is organized around the equipment and time slice; for example, M1's cutting cycle time is stable at 4.8 to 5.0 seconds / piece within the window from 08:00 to 08:05, the tool wear coefficient increases from 0.18 to 0.22, the spool balance decreases from 120 meters to 65 meters, and the idle time is 0; M2's idle time is lower than the first time threshold within the same window, but the tool wear rate increases more than the preset rate threshold; the spool balance here can help determine whether resource contention will occur, and the idle time can help identify whether the equipment is idling due to waiting for materials or instructions;

[0118] The actual execution timeline data is organized around the execution result; for example, the actual start time of O101 on M2 is 08:00:03, the actual completion time is 08:07:19, the actual resource usage record includes the use of type A cable reel number L07, the actual consumption of 52 meters, and the actual energy consumption record of 1.35 kWh; this type of data is directly used to construct the actual residual; if there are only start and completion times without resource usage details, the system will have difficulty distinguishing between two types of problems: slow time but normal resources and slow time and abnormal resources. Therefore, this embodiment emphasizes the synchronous collection of resource and energy consumption records;

[0119] As an anomaly handling mechanism, if some older equipment does not yet support direct uploading of tool wear data, cycle time fluctuations and maintenance intervals can be used instead; if the energy consumption record sampling frequency is low and cannot be accurately mapped to a single work order, it can be allocated according to the work order execution time window; if the harness length of some work orders is the theoretical length in the manufacturing execution system, but there are redundant cutting segments in the field, the actual cumulative cutting length should be used as the standard, and the theoretical length should be retained as a reference item.

[0120] In the same automotive door wiring harness production chain, the system uses work orders, line type, delivery date, length, and quantity as production scheduling inputs, cutting cycle time, tool wear, wire reel margin, and standby idle time as status inputs, and start-up, completion, resource usage, and energy consumption as execution feedback inputs; these three types of data together form the basis for closed-loop analysis throughout the shift.

[0121] The purpose of this step is to clarify the collection boundaries and usage locations of each data item in the distributed wire cutting machine cluster scenario, ensuring that each step in the entire method chain has corresponding data support.

[0122] Furthermore, the collaborative conflict factors include material delivery delay factors and node resource preemption factors; the node resource preemption factor is used to characterize the competitive relationship between multiple execution nodes for materials of the same specification; the queue blocking factor is used to characterize the waiting time changes caused by the material delivery delay factor and the node resource preemption factor.

[0123] This embodiment provides a refined construction method for collaborative conflict factors. Specifically, in the perturbation simulation of the previous embodiment, if the collaborative problem is handled only with the general concept of variable waiting time, although the delay can be simulated, it is difficult to distinguish whether the delay is due to logistics and distribution or due to multiple nodes competing for the same specification of material at the same time. This embodiment decomposes collaborative conflict into two types of basic factors and further synthesizes them into queue blocking factors.

[0124] In practice, the material delivery delay factor mainly describes the time delay in the delivery of wire reels from the warehouse, buffer area, or front-end station to the equipment, and is quantified as a dimensionless normalized index. For example, the delivery time of type A wire reels increases from 20 seconds to 50 seconds, a delay of 30 seconds. Based on the maximum allowable delay threshold, such as 50 seconds, after normalization, the material delivery delay factor can be obtained as 0.6. The node resource preemption factor describes the degree of overlap in the demand for wire of the same specification by multiple devices within the same time window.

[0125] The factor is calculated as follows: it is based on the ratio of the total number of devices requesting the same specification of material within the current time window to the number of available inventory reels of that specification of material. For example, when the number of devices requesting the same specification of material is greater than the number of inventory reels, such as when M1 and M2 simultaneously request type A wire from 08:00 to 08:10, and there is only one main wire reel available for immediate use, the system determines that the resource is insufficient, and the corresponding node resource preemption factor can be rated as 0.8. If there are two reels of the same specification of wire in the inventory, and the number of devices requesting the same specification of material is equal to the number of inventory reels, the system determines that the resource is in a tight balance, and the corresponding node resource preemption factor can be rated as 0.2.

[0126] In practical calculations, the system can combine the two into a queue blocking factor; for example, a linear weighted formula can be used for calculation:

[0127] Queue blocking factor = Material delivery delay factor + Node resource preemption factor

[0128] in, and These are weighting coefficients; and they must satisfy mathematical constraints. This is to ensure that the calculated queue blocking factor is effectively limited to the range of [0,1], consistent with the dimensions and data format of each sub-factor; assuming and Both are 0.5. If the material delivery delay factor is 0.6 and the node resource preemption factor is 0.8, then the calculated queue blocking factor is 0.7.

[0129] The queue congestion factor can be mapped to a waiting time correction amount; the specific calculation model is: corrected waiting time = ideal waiting time + (queue congestion factor × maximum magnification window duration); where the ideal waiting time and the maximum magnification window duration are both in units of time, while the queue congestion factor is a dimensionless proportionality coefficient, thus ensuring the consistency of dimensions on both sides of the equation; taking a task with an ideal waiting time of 10 seconds as an example, if the maximum magnification window duration is set to 50 seconds, then the corrected waiting time can be 10 + 0.7 × 50 = 45 seconds; in this way, the material delivery delay and node resource competition are jointly converted into waiting time correction parameters that can be called by the simulation model;

[0130] Furthermore, node resource preemption can also be expressed as a graph structure; if M1, M2, and M3 correspond to three execution nodes respectively, where M1 and M2 both request type A cables, and M3 requests type B cables, then there is a competing edge between M1 and M2 in the resource contention graph, and no competing edge between M3 and the other nodes; this contention graph can directly participate in subsequent topology matching, so that the waiting caused by preemption is reflected not only in the duration, but also in the relational structure;

[0131] As an exception handling mechanism, if the delivery system temporarily switches to manual pallet delivery, making it impossible to directly obtain the standard delivery time, it can be deduced from the outbound scanning time and the equipment pallet confirmation time; if there are substitute reels for the same specifications of materials, substitutability can be introduced into the preemption calculation to reduce the intensity of competition; if the waiting time corresponding to the queue blocking factor calculation exceeds the upper limit of the production window, the waiting time will not be simply increased, but the production scheduling split or task transfer logic will be triggered.

[0132] During the concentrated production of type A wire harnesses in the morning, M1 and M2 simultaneously compete for type A wire reels, while the delivery from warehouse to machine is delayed due to forklifts blocking the aisle. Based on this, the system calculates material delivery delay factors and node resource preemption factors that are greater than preset thresholds, and then synthesizes a high queue blocking factor. In the simulation, the start time of T3 is shifted back by more than the time tolerance. If the actual deviation also shows the same waiting propagation structure, it can be more accurately determined that the deviation comes from coordination conflict rather than equipment failure.

[0133] The purpose of this step is to break down collaborative conflicts into two traceable sources, so that waiting anomalies can be both quantified and further subdivided and located during causal analysis.

[0134] Furthermore, after outputting rescheduling instructions, maintenance instructions, hold instructions, or review instructions, the process also includes: writing the execution results back to the baseline production scheduling dataset; updating the implicit loss factor and collaborative conflict factor based on the written-back execution results according to the preset recursive correction rules or sliding window statistical rules; and regenerating the perturbation-induced production scheduling state based on the updated factors to form a closed-loop adaptive processing flow.

[0135] This embodiment provides a closed-loop adaptive update mechanism. Specifically, if the system stops tracking the consequences after outputting maintenance, rescheduling, holding, or review instructions, the aforementioned analysis can only remain at the one-time judgment level and cannot gradually correct implicit loss factors and collaborative conflict factors. Therefore, this embodiment continuously writes the execution results back to the baseline production scheduling dataset to form a rolling learning closed loop.

[0136] In the specific implementation process, when the system outputs a certain type of instruction, the corresponding results will be collected in subsequent production. For example, if a maintenance instruction has been issued to M2 and the tool has been replaced, the cycle time, energy consumption, and completion time after maintenance will be re-entered into the dataset. If a rescheduling has been performed, and T3 has been reassigned from M1 to M2, the waiting time, spool occupancy, and delivery performance after the reassignment will also be written back. Based on these written-back data, the system updates the implicit loss factor and the coordination conflict factor.

[0137] One update method is a recursive correction rule; for example, if the original cycle decay factor of M2 was 1.10, and the actual residual of three consecutive work orders has been significantly reduced after maintenance, it can be recursively reduced to 1.05 by the formula: new factor = 0.7 × old factor + 0.3 × new observation factor value; where the new observation factor value refers to the current dimensionless decay coefficient recalculated by the system according to the aforementioned mapping rule based on the latest observed actual execution data, so as to ensure the uniformity of dimensions in the calculation process and gradually approach the real state;

[0138] Another update method is the sliding window statistical rule; for example, if the average delay of A-type cable reel delivery is statistically analyzed in the last 30 minutes, and the delivery delay factor is adjusted accordingly if the average delay decreases from 35 seconds to 18 seconds; if the frequency of overlapping requests for A-type cable by M1 and M2 in the most recent window decreases, the node resource preemption factor is reduced synchronously.

[0139] After the factor update is completed, the system regenerates the production scheduling state after the disturbance based on the new factors. In this way, the theoretical residual is no longer a fixed template, but evolves dynamically with the changes in the actual state of the workshop. If the maintenance is effective, the subsequent theoretical residual will shrink. If the resource conflict is still not relieved after rescheduling, the collaborative conflict theoretical mode will remain active, indicating that further optimization of material distribution or production scheduling strategies is needed.

[0140] As an anomaly handling mechanism, if the data deteriorates after a maintenance, the system will not directly overwrite the old factors, but will archive the result as an anomaly sample separately to avoid an abnormal bias in long-term statistical parameters caused by a single accidental failure. If the number of samples in the sliding window is lower than the minimum effective sample size, such as only 1 to 2 work orders, the system can keep the cause unchanged until the sample reaches the minimum update size. If the written-back data is inconsistent with the previous judgment for a long period of time, such as multiple windows showing no improvement after maintenance and a decrease in similarity, the system can reduce the credibility weight of the corresponding theoretical model and trigger the review process first.

[0141] After two consecutive shifts of operation in the automotive wiring harness factory, the system found that after the tool was changed, the end delay and energy consumption fluctuation of M2 were significantly restored. Therefore, the cycle time decay factor corresponding to the sub-health of the equipment was gradually reduced. At the same time, after the warehouse route was adjusted, the delivery of A-type wire reels was more stable, and the queue blockage factor also decreased with the sliding window statistics. The updated disturbance simulation is closer to the new workshop conditions, and the accuracy of subsequent decisions is improved accordingly.

[0142] The purpose of this step is to transform the entire method from identifying a single anomaly into a closed-loop process of continuously correcting the system parameter model, thereby maintaining adaptability and stability during the long-term operation of the distributed wire cutting machine cluster.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-objective scheduling data processing method for distributed wire cutting machine clusters, characterized in that, include: Collect task instruction data, equipment status timing data, and actual execution timing data including energy consumption records, and perform time alignment and data integration to generate a baseline production scheduling dataset; An ideal constraint model is constructed based on the benchmark production scheduling dataset to generate an ideal production scheduling time series dataset. The ideal production scheduling time series dataset includes task start time, task end time and resource consumption trajectory, as well as associated cycle time parameters, waiting parameters and resource usage parameters. Based on preset implicit loss factors, collaborative conflict factors, and combinations thereof, the cycle time parameters, waiting parameters, and resource occupancy parameters in the ideal production scheduling time series dataset are corrected. Various perturbation-induced production scheduling states are generated through simulation. The implicit loss factor is used to characterize equipment cycle time decay and energy efficiency decline, while the collaborative conflict factor is used to characterize the waiting impact caused by material delivery delays and resource contention. Based on the actual execution time series data and the ideal production scheduling time series dataset, the actual residual vector is extracted, and based on the production scheduling state after various disturbances and the ideal production scheduling time series dataset, various theoretical residual vectors are extracted. Among them, the actual residual vector and the various theoretical residual vectors include at least the time delay residual and the resource consumption residual. Based on the coupling similarity between the actual residual vector and various theoretical residual vectors, the causes of production scheduling deviations are determined, and rescheduling instructions, maintenance instructions, hold instructions, or review instructions are output according to the determination results.

2. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 1, characterized in that, Before integrating task instruction data, device status timing data, and actual execution timing data, the process also includes: The task instruction data, device status timing data, and actual execution timing data are preprocessed to obtain preprocessed data. The preprocessing includes missing value imputation, outlier detection, timestamp alignment, and data standardization.

3. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 1, characterized in that, An ideal constraint model is constructed based on the benchmark production scheduling dataset, generating an ideal production scheduling time series dataset, including: Set the equipment health status to correspond to the rated processing cycle time; Set the network transmission delay to a pre-calibrated baseline transmission delay; Set the material changeover time to a pre-defined standard changeover time. Based on mixed-integer linear programming, the integrated data is solved to generate an ideal production scheduling time series dataset.

4. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 1, characterized in that, The production scheduling status after disturbance is generated through simulation, including: Based on the preset implicit loss factor and collaborative conflict factor, the cycle time parameter, waiting parameter and resource usage parameter in the ideal production scheduling time series dataset are corrected; The equipment health score is determined based on tool wear data, fault interval duration, or cycle time fluctuation index, and the equipment health score is mapped to a cycle time decay factor. Map material delivery delays and resource contention states to queue blocking factors; Based on discrete event simulation, the corrected production scheduling process is forward-engineered to generate the production scheduling state after disturbance.

5. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 1, characterized in that, Extract the actual residual vector and the theoretical residual vector, including: Based on the collected actual execution time-series data and ideal production scheduling time-series dataset, extract the actual residual vector; Based on the perturbed production scheduling status and the ideal production scheduling time series dataset, the theoretical residual vector is extracted; The actual residual vector and the theoretical residual vector both include time delay residual, resource consumption residual and energy consumption fluctuation residual.

6. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 5, characterized in that, The coupling similarity between the actual residual vector and the theoretical residual vector is obtained in the following way: Based on dynamic time warping, the time delay residual, resource consumption residual and energy consumption fluctuation residual are first normalized by using the pre-calibrated maximum tolerance threshold, and then the normalized data are time-series aligned. The aligned residual data is mapped into a topological structure representing the relationship between task nodes; The coupling similarity is calculated by weighting the temporal alignment distance and the topological matching degree.

7. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 6, characterized in that, Determining the causes of production scheduling deviations based on coupling similarity includes: The decision boundary is extracted based on the distribution of historical production scheduling anomalies, and used as the first threshold and the second threshold, with the first threshold set to be greater than the second threshold. When the coupling similarity is greater than or equal to the first threshold, the current deviation is determined to correspond to the latent loss factor and / or the cooperative conflict factor; if the coupling similarity between the actual residual vector and the theoretical residual vector corresponding to the latent loss factor among multiple theoretical residual vectors is the highest, the current deviation is determined to correspond to the latent loss factor, and a maintenance command is output. If the coupling similarity between the actual residual vector and the theoretical residual vector corresponding to the cooperative conflict factor among multiple theoretical residual vectors is the highest, the current deviation is determined to correspond to the cooperative conflict factor, and a rescheduling instruction is output; when the current deviation is determined to correspond to both the implicit loss factor and the cooperative conflict factor, maintenance instructions and rescheduling instructions are output in parallel. When the coupling similarity is less than or equal to the second threshold, the current deviation is determined to correspond to a business disturbance, and a hold instruction is output. The business disturbance includes new work orders, changes in work order priority, delivery date adjustments, or order cancellations. When the coupling similarity is greater than the second threshold and less than the first threshold, a verification instruction is output, and differential extraction and coupling decision are re-executed based on the updated device state time series data.

8. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 1, characterized in that, When the method is applied to a distributed wire cutting machine cluster scenario, the task instruction data includes work order data, wire type data, delivery weight data, wire harness length data, and quantity data. Equipment status timing data includes cutting cycle time data, tool wear data, spool balance data, and standby idle time data; The actual execution timeline data includes the actual start time, actual completion time, actual resource usage records, and actual energy consumption records.

9. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to claim 4, characterized in that, Collaboration conflict factors include material delivery delay factors and node resource preemption factors; The node resource preemption factor is used to characterize the competitive relationship among multiple execution nodes for materials of the same specification; The queue blocking factor is used to characterize the changes in waiting time caused by material delivery delay factor and node resource preemption factor.

10. The multi-objective scheduling data processing method for distributed wire cutting machine groups according to any one of claims 1 to 9, characterized in that, After outputting rescheduling instructions, maintenance instructions, hold instructions, or review instructions, the following are also included: Write the execution results back to the baseline production scheduling dataset; Based on the execution results after writing back, the implicit loss factor and the collaborative conflict factor are updated according to the preset recursive correction rules or sliding window statistical rules. The perturbation-induced production scheduling status is regenerated based on the updated factors to form a closed-loop adaptive processing flow.

Images