Wiring harness production line load balance and fatigue optimization method and system

Abstract

The invention provides a load balance and fatigue optimization method and system for a wire harness production line, and the method comprises the steps: collecting the operation parameters of all stations and the state data of workers in real time through a sensor and an equipment interface, calculating the load distribution of the stations through a time sequence analysis algorithm, and carrying out the dynamic task distribution in combination with a fatigue index of the workers. The method comprises the steps of firstly identifying overload and idle stations, then optimizing a task transfer path based on a linear programming algorithm, transferring tasks of the overload stations to the idle stations where low-fatigue workers are located, meanwhile, updating production takt time by adopting real-time data, triggering a dynamic matching algorithm to adjust an operation sequence when necessary, and finally, completing the task transfer process. And by comparing the new beat distribution with the fatigue index, generating a task execution sequence meeting load balance and fatigue optimization double constraints, and issuing the task execution sequence to a production line for execution. The production efficiency can be effectively improved, the fatigue degree of workers is reduced, and intelligent management of the wire harness production line is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for load balancing and fatigue optimization of a wiring harness production line. Background Art

[0002] Wire harness production, as a vital area in the manufacturing industry, is directly related to the stability of the electrical system and the overall performance of the product. Its efficiency improvement is not only the core manifestation of corporate competitiveness, but also the key to promoting the transformation of intelligent manufacturing. With the advancement of Industry 4.0, optimizing production processes and improving resource utilization have become an inevitable trend in the development of the industry. However, traditional wire harness production methods face many bottlenecks in improving efficiency, and there is an urgent need to break through the limitations of the existing technical framework to achieve a higher level of intelligence and refined management.

[0003] Currently, many solutions still rely on manual experience or simple automated equipment. While these solutions can improve production to a certain extent, they often neglect the dynamic balance of workload between workstations and the scientific management of worker fatigue, resulting in limited gains in production efficiency. More importantly, these approaches lack in-depth utilization of real-time data, making them unable to accurately respond to sudden changes in production, and thus, the overall optimization effects are difficult to sustain. In complex wiring harness production scenarios, the problems of uneven workstation load and accumulated worker fatigue are particularly prominent, becoming a persistent obstacle to efficiency improvement.

[0004] In this area, the core challenges mainly focus on three technical factors: real-time control of production rhythm, intelligent allocation of workstation loads, and precise matching of workers' fatigue curves. Since the production rhythm fails to dynamically match the actual operation complexity, some workstations may be in an overloaded state for a long time, while other workstations have idle resources, resulting in low overall production line efficiency. At the same time, the existing system makes it difficult to reasonably arrange tasks and rest time according to the workers' fatigue curve, which in turn affects product quality consistency and production stability. These technical factors have not been effectively resolved, directly leading to the unique problems of waste of production resources and efficiency bottlenecks.

[0005] Therefore, how to collect production data in real time and combine it with the status of workers to intelligently control the load and task distribution of each workstation, thereby maximizing overall efficiency while ensuring quality consistency, has become a key issue in improving wire harness production efficiency.

[0006] Solving this problem requires not only technological innovation, but also finding the best combination between dynamic balance and humanized management to provide the industry with a replicable optimization path. Summary of the Invention

[0007] In order to solve the technical problems raised in the above background technology, the first aspect of the present invention provides a load balancing and fatigue optimization method for a wiring harness production line, the method comprising: S1, through the sensor and equipment interface, obtains the operation time, complexity parameters and worker movement frequency data of each workstation in the wire harness production, and obtains the real-time distribution characteristics of production rhythm and load; S2, based on the real-time distribution characteristics, uses a time series analysis algorithm to calculate the load value and complexity deviation of each workstation at the current beat, and determines the dynamic change trend of the uneven load between workstations; S3, extracts the specific identification of overloaded and idle workstations and the load difference from the dynamic change trend, and determines which workstations need to adjust their tasks to achieve a preliminary balance in load distribution; S4, obtaining the heart rate, action repetition rate, and rest interval data from the workers' wearable devices, and calculating the real-time fatigue index of each worker based on a preset fatigue curve model to obtain a quantitative distribution of fatigue status; S5, based on the quantitative distribution and load difference, optimize the task distribution plan through the linear programming algorithm to determine the transfer path of overloaded workstation tasks to idle workstations where low-fatigue workers are located; S6, after the task transfer path is generated, the production rhythm parameters are updated using real-time data. If the rhythm deviation exceeds the rhythm threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new rhythm distribution; S7, through comparative analysis of the new beat distribution and fatigue index, it is determined whether the task distribution scheme meets the dual constraints of load balance and fatigue optimization, and the final task execution sequence is obtained; S8, based on the final task execution sequence, generates operation instructions and rest period arrangements for each workstation, sends them to the production line through the equipment terminal, and obtains real-time feedback data after execution to verify the stability of the optimization results.

[0008] Optionally, step S1 obtains the operation duration, complexity parameters, and worker movement frequency data of each workstation in the wire harness production through sensors and equipment interfaces to obtain real-time distribution characteristics of production rhythm and load, including: Step S11, obtaining original records of operation duration, complexity parameters, and action frequency of the workstation, wherein the original records are collected through the sensor interface and the equipment interface; Step S12, filtering the original records using a preset screening threshold, determining the correspondence between the action frequency and the operation duration, and obtaining preliminary features of the workstation data; Step S13, calculating the production tact time using a preset formula based on the complexity parameter and operation duration in the preliminary features, and determining the tact time distribution of each workstation; Step S14: if the beat distribution exceeds a preset range, the complexity parameter is adjusted using real-time data to obtain dynamic characteristics of the load distribution; Step S15, based on the load distribution and production rhythm in the dynamic characteristics, using the K-means clustering algorithm to divide the workstation load levels and determine the classification characteristics of the load distribution; Step S16, obtaining the duration characteristics of the workstation data by comparing the classification characteristics with the real-time data, and determining the change trend of the load distribution; Step S17: Based on the distribution characteristics in the change trend, a linear regression algorithm is used to predict the subsequent production cycle to obtain an optimized adjustment value of the load distribution.

[0009] Optionally, step S2, calculating the load value and complexity deviation of each workstation at the current beat using a time series analysis algorithm based on real-time distribution characteristics to determine the dynamic change trend of uneven load among workstations, includes: Step S21: Obtain the load value and complexity data of each workstation at the current beat from the real-time distribution characteristics, calculate the load value deviation using the ARIMA model, and obtain preliminary results of unevenness between workstations; Step S22: Process the load value and complexity deviation through the ARIMA model, extract the dynamic change characteristics, and obtain the time series data of the load of each workstation; Step S23: Based on the dynamic change characteristics, K-means clustering is used to group the time series data to determine the load unevenness trend among the workstations; Step S24: If the load unevenness trend exceeds the preset unevenness threshold, the fluctuation range of the unevenness is obtained through historical data comparison and analysis, and the key workstation is determined; Step S25: For key workstations, extract the correlation features between beat and complexity from the real-time distribution, calculate the adjusted load value using linear regression, and obtain the optimized load distribution; Step S26, based on the optimized load distribution, use the ARIMA model to predict future trends and determine the long-term changes in unevenness between workstations; Step S27: By comparing the prediction result with the current load distribution, a dynamic adjustment plan is determined to obtain the final load balancing strategy.

[0010] Optionally, step S3 extracts specific identifiers of overloaded workstations and idle workstations and load differences from the dynamic change trend, and determines which workstations need to adjust their tasks to achieve a preliminary balance in load distribution, including: Step S31, obtaining the load value of each workstation from the real-time data, and obtaining the load difference by calculating the difference between the load values ​​of each workstation; Step S32, comparing the load difference with a first load threshold value, where the first load threshold value is set based on historical data; Step S33: if the load difference exceeds the first load threshold, marking the overloaded workstation and the idle workstation to determine the workstation status; Step S34: extract the workstation ID through the workstation number in the workstation status, obtain the specific locations of the overloaded workstation and the idle workstation, and determine the task adjustment direction; Step S35, assigning tasks of idle workstations to overloaded workstations according to the task adjustment direction, and generating a task allocation table; Step S36, using a linear regression algorithm to predict the adjusted load distribution result; Step S37, extracting a new load difference from the adjusted load distribution result; Step S38: If the new load difference still exceeds the first load threshold, the task allocation table is repeatedly adjusted until the load difference is within the first load threshold, thereby obtaining a final load allocation plan. Step S39: Use the K-means clustering algorithm to group the workstation status and determine the stability of the load distribution within each group.

[0011] Optionally, step S4, obtaining heart rate, action repetition rate, and rest interval data from workers' wearable devices, and calculating each worker's real-time fatigue index using a preset fatigue curve model to obtain a quantitative distribution of fatigue status, includes: Step S41, collecting heart rate data, movement repetition rate, and rest interval duration through a wearable device, storing them in a time series format, and obtaining an original data set; Step S42, performing outlier detection on the original data set according to a preset heart rate threshold, marking data points where the heart rate data exceeds the heart rate threshold, and obtaining a filtered data set; Step S43, using a linear regression model to calculate the real-time fatigue value of each worker based on the filtered data set to obtain a fatigue index set; Step S44, calculating the ratio of the action repetition rate to the rest interval duration based on the fatigue index set. If the ratio is higher than a preset fatigue index threshold, the fatigue index is adjusted using a weighted method to obtain a revised fatigue index set. Step S45, using the K-means clustering algorithm to divide the worker states according to the modified fatigue index set to obtain preliminary grouping of fatigue states; Step S46, analyzing the fatigue index variation trend of each group using the moving average method to obtain the quantitative distribution of fatigue status; Step S47 , extracting the dynamic features of the worker status from the quantified distribution, and using a linear regression model to predict future changes in real-time fatigue to obtain a fatigue trend sequence.

[0012] Optionally, step S5, optimizing the task distribution plan based on the quantitative distribution and the load difference using a linear programming algorithm to determine a transfer path for overloaded workstation tasks to idle workstations where low-fatigue workers are located, includes: Step S51, calculating the quantized distribution and the load difference by a linear programming algorithm to obtain a comparison result of the task load of the overloaded workstation and the idle workstation; Step S52, extracting overloaded workstation data from the load difference to determine the task load that needs to be transferred; Step S53: obtaining the distribution data of low-fatigue workers from a pre-established worker status database, and determining the receiving capacity of the corresponding idle workstations according to the worker fatigue levels; Step S54, using linear programming to construct a transfer path matrix to distribute the task load of the overloaded workstations to the idle workstations; Step S55, adjusting the transfer path matrix according to the task load comparison result to obtain an optimized task distribution result; Step S56, updating the worker distribution data according to the workstation status, and determining whether the task load is within a second load threshold range; Step S57: If the task load exceeds the second load threshold, the transfer path matrix is ​​recalculated through linear programming to obtain a final optimization solution.

[0013] Optionally, in step S6, after the task transfer path is generated, real-time data is used to update the production rhythm parameters. If the rhythm deviation exceeds the rhythm threshold, a dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new rhythm distribution, including: Step S61, acquiring real-time data through sensors, updating production rhythm parameters, and obtaining the current rhythm value; Step S62: If the difference between the current beat value and the target value exceeds the beat threshold, the beat deviation state is determined to determine whether the trigger condition is met; Step S63: Using the KNN algorithm to process the beat deviation, adjust the execution order of the workstation operations, and obtain an optimized operation sequence; Step S64, recalculating the load of each workstation through the optimized operation sequence to obtain a preliminary beat distribution; Step S65: After obtaining the preliminary beat distribution, compare it with the historical beat data to determine whether there is abnormal fluctuation and determine the stability of the adjusted sequence; Step S66, based on the stability judgment result, the beat distribution is smoothed using the exponential weighted average method to obtain the final beat distribution; Step S67: Update the execution plan of the workstation operation through the final beat distribution to complete the dynamic optimization of the production beat.

[0014] Optionally, step S7, determining whether the task distribution scheme satisfies the dual constraints of load balancing and fatigue optimization by comparing the new beat distribution with the fatigue index, and obtaining the final task execution sequence, includes: Step S71, using a greedy algorithm to generate an initial task distribution plan based on the collected new beat distribution and fatigue index data; Step S72, comparing the task allocation in the initial task distribution plan with the new beat distribution, and calculating the load difference of each task node; Step S73: If the load difference exceeds the third load threshold, the task distribution plan is adjusted, tasks are reallocated, and an updated task distribution result is generated; Step S74: Calculate the fatigue index of each task node based on the updated task allocation result and compare it with the preset fatigue threshold; Step S75: If the fatigue index exceeds the fatigue threshold, further adjust the task allocation to ensure that the fatigue optimization meets the constraint conditions; Step S76, generating a task sequence that meets the load balance according to the adjusted task allocation result; Step S77: Perform a secondary check on the task sequence and the fatigue index to ensure that the fatigue index of each task node is lower than the fatigue threshold, thereby obtaining an optimized execution sequence. Step S78: Integrate the new beat distribution with the optimized execution sequence to determine the final task execution sequence.

[0015] Optionally, step S8 generates operating instructions and rest period arrangements for each workstation based on the final task execution sequence, sends them to the production line via the equipment terminal, and obtains real-time feedback data after execution to verify the stability of the optimization results, including: Step S81, generating workstation operation instructions and rest period arrangements from the task sequence according to preset rules, and obtaining an operation instruction set and a period table; Step S82: Send the operation instruction set and time table to the production line through the equipment terminal to obtain execution status data; Step S83, pre-processing the execution status data, extracting the operation time, workstation status and equipment operating parameters, and generating a feedback data set; Step S84, using principal component analysis to extract features from the feedback data set and calculate process efficiency, workstation load, and equipment failure rate; Step S85, determining whether the process efficiency, workstation load, and equipment failure rate exceed the status threshold. If so, adjusting the task sequence and rest schedule to obtain an updated instruction set; Step S86: Send the updated instruction set to the production line via the device terminal to obtain new execution status data; Step S87: Repeat feature extraction and state threshold judgment on the new execution state data to determine whether the optimization result is stable.

[0016] A second aspect of the present invention provides a wire harness production line load balancing and fatigue optimization system, which uses the above-mentioned method to perform load balancing and fatigue optimization on the wire harness production line, and the system includes: The data acquisition module is used to obtain the operation time, complexity parameters and worker movement frequency data of each workstation in the wire harness production through sensors and equipment interfaces, and obtain the real-time distribution characteristics of production rhythm and load; The load analysis module is used to calculate the load value and complexity deviation of each workstation under the current beat based on the real-time distribution characteristics using a time series analysis algorithm, and to determine the dynamic trend of uneven load among workstations; The trend extraction module is used to extract the specific identification of overloaded and idle workstations and the load difference from the dynamic change trend, and determine which workstations need to adjust their tasks to achieve a preliminary balance in load distribution; The fatigue calculation module is used to obtain the heart rate, action repetition rate and rest interval duration data from the workers' wearable devices, and calculate the real-time fatigue index of each worker based on the preset fatigue curve model to obtain the quantitative distribution of fatigue status; The task optimization module is used to optimize the task distribution plan based on the quantitative distribution and load difference through a linear programming algorithm, and determine the transfer path of tasks from overloaded workstations to idle workstations where low-fatigue workers are located; The beat adjustment module is used to update the production beat parameters using real-time data after the task transfer path is generated. If the beat deviation exceeds the beat threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new beat distribution; The constraint judgment module is used to determine whether the task distribution plan meets the dual constraints of load balance and fatigue optimization by comparing the new beat distribution with the fatigue index, and obtain the final task execution sequence; The instruction issuing module is used to generate operation instructions and rest period arrangements for each workstation based on the final task execution sequence, and issue them to the production line through the equipment terminal to obtain real-time feedback data after execution to verify the stability of the optimization results.

[0017] The present invention provides a method and system for load balancing and fatigue optimization of a wire harness production line. The method collects the operating parameters and worker status data of each workstation in real time through sensors and equipment interfaces, calculates the workstation load distribution using a time series analysis algorithm, and dynamically allocates tasks in combination with the worker fatigue index. The present invention first identifies overloaded and idle workstations, then optimizes the task transfer path based on a linear programming algorithm, transfers the tasks of overloaded workstations to idle workstations where low-fatigue workers are located, and simultaneously uses real-time data to update the production rhythm, and triggers a dynamic matching algorithm to adjust the operation sequence when necessary. Finally, by comparing the new rhythm distribution with the fatigue index, a task execution sequence that meets the dual constraints of load balancing and fatigue optimization is generated and sent to the production line for execution. The present invention can effectively improve production efficiency, reduce worker fatigue, and realize intelligent management of wire harness production lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The present invention is a flow chart of a method for load balancing and fatigue optimization of a wiring harness production line.

[0019] Figure 2 This is a structural schematic diagram of a load balancing and fatigue optimization system for a wiring harness production line of the present invention. DETAILED DESCRIPTION

[0020] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, the first aspect of the present invention provides a method for load balancing and fatigue optimization of a wiring harness production line, comprising: S1, through the sensor and equipment interface, obtains the operation time, complexity parameters and worker movement frequency data of each workstation in the wire harness production, and obtains the real-time distribution characteristics of production rhythm and load.

[0022] Optionally, this step also includes: Step S11, obtaining original records of operation duration, complexity parameters, and action frequency of the workstation, wherein the original records are collected through the sensor interface and the equipment interface; Step S12, filtering the original records using a preset screening threshold, determining the correspondence between the action frequency and the operation duration, and obtaining preliminary features of the workstation data; Step S13, calculating the production tact time using a preset formula based on the complexity parameter and operation duration in the preliminary features, and determining the tact time distribution of each workstation; Step S14: if the beat distribution exceeds a preset range, the complexity parameter is adjusted using real-time data to obtain dynamic characteristics of the load distribution; Step S15, based on the load distribution and production rhythm in the dynamic characteristics, using the K-means clustering algorithm to divide the workstation load levels and determine the classification characteristics of the load distribution; Step S16, obtaining the duration characteristics of the workstation data by comparing the classification characteristics with the real-time data, and determining the change trend of the load distribution; Step S17: Based on the distribution characteristics in the change trend, a linear regression algorithm is used to predict the subsequent production cycle to obtain an optimized adjustment value of the load distribution.

[0023] Specifically, obtaining original records of the operation duration, complexity parameters, and action frequency of the workstation is usually achieved through sensors and equipment interfaces.

[0024] For example, on an automotive assembly line, sensors could be installed on power tools used to tighten screws, recording the duration and frequency of each operation. Complexity parameters might be determined by the diversity of tasks at the workstation, such as the involvement of multiple screw types or installation steps. This data collection method provides a real-time reflection of working conditions, providing a foundation for subsequent analysis. Preset filtering thresholds are used to filter the raw records, determining the correlation between action frequency and operation duration, and generating preliminary features.

[0025] Specifically, if the average operation duration at a certain workstation is 10 seconds, the frequency is 6 times per minute, and the screening threshold is set to an abnormal frequency of less than 5 times, then the workstation can be preliminarily judged to be operating smoothly. This step eliminates abnormal data, ensures the accuracy of feature extraction, and improves the reliability of analysis. Based on the complexity parameters and operation duration in the preliminary features, the production rhythm is calculated using a preset formula to determine the rhythm distribution.

[0026] In one possible implementation, if a workstation is complex, such as requiring five different actions in 50 seconds, the formula can be simply set as "total time / number of actions," resulting in a single cycle of 10 seconds. Comparing the cycle times of each workstation provides a visual reflection of the balance of production rhythm, providing a basis for optimization. If the cycle distribution exceeds the preset range, the complexity parameters are adjusted based on real-time data to capture the dynamic characteristics of the load distribution.

[0027] For example, if a workstation's cycle time reaches 15 seconds, exceeding the standard range by 10-12 seconds, real-time data analysis can reveal that its complexity has increased due to the temporary addition of tasks. Adjustments and recalculations can dynamically reflect load changes and facilitate timely intervention. Based on the load distribution and production cycle time in the dynamic characteristics, the K-means clustering algorithm is used to classify workstation load levels and determine classification characteristics.

[0028] In one example, assuming there are 10 workstations with a cycle time distribution between 8 and 15 seconds, K-means clustering can categorize them into three load categories: low, medium, and high. For example, 8-10 seconds is considered low load. This classification clearly distinguishes workstation status and facilitates resource allocation. By comparing the classification features with real-time data, the duration characteristics of the workstation data are obtained and the changing trend of load distribution is determined.

[0029] If a workstation's load increases from low to medium, and the duration increases from 9 seconds to 11 seconds, it can be inferred that its workload has increased. This trend analysis helps identify bottlenecks in advance and improve production stability. Based on the distribution characteristics of the changing trend, a linear regression algorithm is used to predict the subsequent production cycle and obtain the optimized adjustment value.

[0030] It's understandable that if a workstation's cycle time for the past five days was 10, 11, 12, 13, and 14 seconds, linear regression can predict it will be 15 seconds on the sixth day. Adjusting personnel or equipment accordingly can optimize load distribution, reduce production fluctuations, and improve overall efficiency.

[0031] It's important to note that each technical step is closely linked, forming a closed-loop logic from data collection to predictive optimization. This process not only enhances the scientific nature of production management but also reduces costs and increases output through dynamic adjustments, making it an indispensable technical support for intelligent manufacturing.

[0032] Optionally, step S16, obtaining the duration characteristics of the workstation data by comparing the classification results with the real-time data and determining the change trend of the load distribution, further includes: Step S161 : Obtain the fluctuation range of the action frequency by comparing the real-time data with the workstation data, and determine the initial state of the distribution feature.

[0033] Step S162 , based on the distribution characteristics in the initial state, a preset frequency threshold is used to filter the action frequency to obtain a preliminary adjustment value of the complexity parameter.

[0034] Step S163: Calculate the deviation range of the production rhythm based on the preliminary adjustment value and the real-time data, and determine the dynamic offset of the load distribution.

[0035] Step S164: If the dynamic offset exceeds the preset range, a correction value of the load distribution is obtained by adjusting the complexity parameter.

[0036] In step S165 , based on the correspondence between the correction value and the action frequency, the K-means clustering algorithm is used to divide the load levels of the workstation data and determine the classification boundaries of the duration features.

[0037] Step S166 , by matching the classification boundary with the real-time data, obtain the predicted value of the change trend and determine the subsequent fluctuation of the distribution characteristics.

[0038] Step S167: Based on the correlation between subsequent fluctuations and production rhythm, a linear regression algorithm is used to adjust the complexity parameter to obtain an optimized value of the load distribution.

[0039] Specifically, the fluctuation range of the action frequency is obtained by comparing the real-time data with the workstation data, and the initial state of the distribution characteristics is determined.

[0040] It can be understood that the action frequency reflects the number of operations completed by workers in a unit of time, and the fluctuation range reveals the preliminary characteristics of the operation stability.

[0041] For example, in wire harness production, the number of times a worker at a certain workstation tightens screws per minute may fluctuate between 20 and 25 times. By recording this data with sensors and comparing it with the historical average, it is possible to determine whether the current working conditions are stable.

[0042] In a possible implementation, based on the distribution characteristics in the initial state, a preset frequency threshold is used to filter the action frequency to obtain a preliminary adjustment value of the complexity parameter.

[0043] Specifically, if the normal range of action frequency is set to 18 to 26 times per minute, when the frequency of a certain workstation drops to 15 times, it indicates that the operation may be hindered and the complexity parameter needs to be adjusted upward.

[0044] For example, if a workstation involves tying multiple wire harnesses, the frequency reduction may be due to wire entanglement. In this case, the complexity parameter is adjusted from 1.2 to 1.5 to reflect the actual difficulty. This adjustment lays the foundation for subsequent analysis. Based on the initial adjustment value and real-time data, the deviation range of the production cycle is calculated to determine the dynamic shift in load distribution.

[0045] For example, suppose a workstation's standard cycle time is 30 seconds per piece, but real-time data shows it as 35 seconds, a deviation of 5 seconds. By analyzing the frequency and complexity of movements, it can be determined that the deviation is due to workers frequently adjusting tool positions. This identification helps identify potential bottlenecks in a timely manner. If the dynamic deviation exceeds the preset range, the complexity parameter is adjusted to obtain a corrected load distribution.

[0046] For example, when the deviation exceeds 10%, the complexity parameter is further adjusted from 1.5 to 1.8, while observing whether the load is balanced. This dynamic correction can effectively cope with sudden changes in production. Based on the correspondence between the correction value and the action frequency, the K-means clustering algorithm is used to classify the load level of the workstation data and determine the classification boundaries of the duration feature.

[0047] In one example, assume that the motion frequencies of five workstations in a certain section are 20, 22, 25, 18, and 23 times per minute, respectively. After clustering, these workstations are classified into three load levels: high, medium, and low: 25, 22-23, and 18-20 times per minute, respectively. This classification clearly defines the load status of different workstations and facilitates resource allocation. By matching the classification boundaries with real-time data, a predicted value for the change trend is obtained, and subsequent fluctuations in the distribution characteristics can be determined.

[0048] If the frequency of high-load workstations continues to rise to 26 times per minute, this may indicate worker fatigue or equipment failure. By comparing this with historical trends, we can predict increased load within the next hour. This prediction provides a basis for early intervention. Based on the correlation between subsequent fluctuations and production cycle time, a linear regression algorithm is used to adjust the complexity parameter to obtain the optimal value for the load distribution.

[0049] For example, if frequency fluctuations are positively correlated with tempo, regression analysis might suggest reducing complexity from 1.8 to 1.6 to reduce workstation stress. This optimization ensures a balance between productivity and worker comfort.

[0050] It should be noted that the implementation of each step depends on the accuracy of real-time data.

[0051] For example, if the sensor records a low frequency due to dust interference, the equipment needs to be calibrated in time. This meticulous process improves the reliability of analysis and helps optimize the overall production process.

[0052] Optionally, the step S17, using a linear regression algorithm to predict the subsequent production cycle based on the distribution characteristics in the change trend to obtain the optimized adjustment value of the load distribution, further includes: Step S171 , based on the distribution characteristics in the change trend, linear regression is used to predict the subsequent production rhythm to obtain a preliminary adjustment value.

[0053] Step S172 : According to the correspondence between the preliminary adjustment value and the distribution characteristic, the standard deviation of the load distribution is calculated to obtain the offset range of the load distribution and determine the boundary of the dynamic change.

[0054] Step S173: If the dynamically changing boundary exceeds the preset boundary threshold, the production cycle is recalculated through linear regression to obtain a corrected value of the load distribution.

[0055] In step S174 , based on the matching between the correction value and the change trend, K-means clustering is used to classify the load distribution levels to obtain an optimized and adjusted classification result.

[0056] Step S175 , based on the association between the classification result and the subsequent prediction, the variance of the load distribution is calculated to determine the fluctuation range of the load distribution and the stability of the adjustment value.

[0057] In step S176, based on the stability of the adjustment value and the deviation of the production cycle, the distribution characteristics are filtered through the distribution threshold to obtain the optimized value of the load distribution.

[0058] Step S177 , based on the corresponding relationship between the optimized value and the change trend, K-means clustering is used to readjust the boundary of the load distribution to obtain the final adjustment result.

[0059] Specifically, based on the distribution characteristics in the changing trend, linear regression is used to predict the subsequent production rhythm and obtain the preliminary adjustment value.

[0060] It can be understood that linear regression uses historical data and current trends to predict the changes in rhythm over a period of time in the future.

[0061] For example, in wire harness production, the historical cycle time for a certain workstation was 30 seconds per piece. Recent trends show a gradual increase to 32 seconds. Linear regression predicts that the cycle time could reach 33 seconds within the next hour. This prediction provides a basis for subsequent adjustments. Based on the correspondence between the preliminary adjustment value and the distribution characteristics, the standard deviation of the load distribution is calculated to determine the offset range and determine the boundaries of dynamic changes.

[0062] Specifically, the standard deviation reflects the degree of dispersion of loads between workstations.

[0063] For example, if the tact times of five workstations are 33, 31, 34, 30, and 32 seconds, respectively, and the calculated standard deviation is 1.58 seconds, the deviation range can be set to ±1.6 seconds, with the boundaries between 28.4 and 35.6 seconds. This method clearly defines the fluctuation range. If the dynamically changing boundaries exceed the preset boundary threshold, the production tact time is recalculated using linear regression to obtain a corrected value.

[0064] In one possible implementation, assuming a threshold of ±2 seconds, if the threshold is exceeded, real-time data is used for regression, resulting in a correction to 31 seconds. This step ensures that the load distribution more closely matches actual operating conditions. To match the correction value with the change trend, K-means clustering is used to classify the load distribution and obtain the classification results for optimization and adjustment.

[0065] For example, the cycle time data for five workstations was clustered into three categories: high load (34 seconds), medium load (32-33 seconds), and low load (30-31 seconds). This classification intuitively reflects the status of each workstation and facilitates targeted optimization. Based on the correlation between the classification results and subsequent predictions, the variance of the load distribution was calculated to determine the fluctuation range and the stability of the adjustment value.

[0066] Preferably, if the variance is small, such as 0.5 seconds, the adjustment value is stable; if the variance reaches 2 seconds, the data needs to be re-examined. This analysis ensures the reliability of the adjustment. Based on the stability of the adjustment value and the beat deviation, the distribution characteristics are screened using a preset deviation threshold to obtain the optimized value of the load distribution.

[0067] In one embodiment, if the deviation threshold is 3 seconds and the cycle time deviation at a particular workstation is 4 seconds, the optimized value after screening is 31 seconds. This screening improves load balancing. Based on the correspondence between the optimized value and the change trend, K-means clustering is used to readjust the load distribution boundaries to obtain the final adjustment result.

[0068] For example, after adjustment, the boundary may be narrowed to 29 seconds to 34 seconds, and the clustering results are more accurate. This approach makes resource allocation more reasonable and improves production stability.

[0069] It should be noted that each step depends on the real-time and accuracy of the data.

[0070] For example, if a sensor's beat recording is too high due to environmental interference, it needs to be calibrated in a timely manner. This meticulous processing ensures that the analysis results are more practical.

[0071] S2, based on the real-time distribution characteristics, uses the time series analysis algorithm to calculate the load value and complexity deviation of each workstation under the current beat, and determines the dynamic change trend of the uneven load among workstations.

[0072] Optionally, this step also includes: In step S21, the load value and complexity data of each workstation at the current beat are obtained from the real-time distribution characteristics, and the load value deviation is calculated using the ARIMA model to obtain a preliminary result of the unevenness between workstations.

[0073] Step S22: Process the load value and complexity deviation through the ARIMA model, extract the dynamic change characteristics, and obtain the time series data of the load of each workstation.

[0074] In step S23 , K-means clustering is used to group the time series data according to the dynamic change characteristics to determine the load unevenness trend among the workstations.

[0075] In step S24, if the load unevenness trend exceeds the preset unevenness threshold, the fluctuation range of the unevenness is obtained through historical data comparison and analysis, and the key workstations are determined.

[0076] In step S25 , for key workstations, the beat and complexity correlation features are extracted from the real-time distribution, and the adjusted load value is calculated using linear regression to obtain the optimized load distribution.

[0077] In step S26 , based on the optimized load distribution, the ARIMA model is used to predict future trends and determine the long-term changes in the unevenness between workstations.

[0078] Step S27: By comparing the prediction result with the current load distribution, a dynamic adjustment plan is determined to obtain the final load balancing strategy.

[0079] Specifically, the load value and complexity data of each workstation at the current beat are obtained from the real-time distribution characteristics.

[0080] Understandably, this requires reliance on information collected by sensors in real time.

[0081] For example, on an automotive wiring harness production line, a workstation is responsible for connecting multiple wires. Sensors record a 12-second cycle, and the load value reflects 500 operations per day. The complexity data is determined by the number of wire types connected, such as three different wire sizes. When using the ARIMA model to calculate the load value deviation.

[0082] Specifically, load fluctuations are analyzed based on historical data.

[0083] For example, the load values ​​for the past five days were 480, 490, 500, 510, and 495, respectively. By calculating the deviation using the model, we find that the current 500 values ​​are higher than the average, initially suggesting possible unevenness among workstations. When dealing with load value and complexity deviations, the ARIMA model can extract dynamic characteristics.

[0084] For example, if the load value of a certain workstation increases from 490 to 520 times in one day, and the complexity increases from 2 to 4 due to the addition of new wire types, the time series data will show a trend of gradually increasing load over time. This dynamic feature provides a basis for subsequent grouping.

[0085] When K-means clustering is used to group time series data, preferably, the loads of the 10 workstations can be divided into three categories: low, medium, and high.

[0086] For example, a load value of 400-450 times is low, 450-500 times is medium, and 500 times or more is high. This indicates that high-load workstations are concentrated at the front end of the production line, reflecting an uneven load trend. If the load unevenness trend exceeds the preset unevenness threshold, such as the proportion of high-load workstations exceeding 30%, it is necessary to compare and analyze the fluctuation range through historical data.

[0087] In one possible implementation, the load value of a certain workstation fluctuated between 490 and 520 times in the past week, while the load value of another workstation remained stable at 450 times. The former is determined to be a critical workstation.

[0088] When extracting features related to cycle time and complexity for key workstations, for example, where the cycle time is 12 seconds and the complexity is high due to the large number of wire types, the load value after linear regression calculation may be reduced to 480 times, forming an optimized distribution. Based on this optimized load distribution, the ARIMA model can predict future trends.

[0089] In one embodiment, if the load value stabilizes at 480 after adjustment, it is predicted that the load value in the next three days will be 482, 485, and 488, respectively, indicating that the imbalance is gradually decreasing.

[0090] It should be noted that by comparing the prediction results with the current load, a dynamic adjustment plan can be determined.

[0091] For example, if a load increase at a certain workstation is found to be related to a temporary increase in tasks, a final balancing strategy can be developed by reducing its complexity, such as reducing the number of wire types to two. This approach can identify problems in a timely manner and optimize resource allocation.

[0092] Specifically, from multiple perspectives, sensor data provides the foundation for analysis, ARIMA models capture dynamic changes, K-means clustering clarifies classification, and linear regression refines the adjustment plan. These steps support each other, forming a complete logic that ensures balanced workstation loads and improves production stability.

[0093] It is understandable that the combination of real-time data and historical data can not only reflect the current status, but also predict future trends and provide support for management decisions.

[0094] Preferably, the identification and adjustment of key workstations can also avoid bottlenecks and ensure production continuity.

[0095] S3, extracts the specific identification and load difference of overloaded workstations and idle workstations from the dynamic change trend, and determines which workstations need to adjust their tasks to achieve a preliminary balance in load distribution.

[0096] Optionally, this step also includes: Step S31, obtaining the load value of each workstation from real-time data, and obtaining the load difference by calculating the difference between the load values ​​of each workstation.

[0097] Step S32 : comparing the load difference with a first load threshold value, where the first load threshold value is set based on historical data.

[0098] Step S33: If the load difference exceeds the first load threshold, mark the overloaded workstation and the idle workstation and determine the workstation status.

[0099] Step S34: extract the workstation identification through the workstation number in the workstation status, obtain the specific locations of the overloaded workstation and the idle workstation, and determine the task adjustment direction.

[0100] Step S35: Allocate tasks of idle workstations to overloaded workstations according to the task adjustment direction, and generate a task allocation table.

[0101] Step S36: Use a linear regression algorithm to predict the adjusted load distribution result.

[0102] Step S37: extracting a new load difference from the adjusted load distribution result.

[0103] Step S38: If the new load difference still exceeds the first load threshold, the task allocation table is repeatedly adjusted until the load difference is within the first load threshold, and a final load allocation plan is obtained.

[0104] Step S39: Use the K-means clustering algorithm to group the workstation status and determine the stability of the load distribution within each group.

[0105] Specifically, when obtaining the load value of each workstation from real-time data, it can be understood that this relies on the sensor system on the production line.

[0106] For example, on an automotive wiring harness production line, one workstation completes 500 operations per day, while another performs 450. By calculating the difference, a load difference of 50 is calculated. This load difference reflects the disparity in task distribution between workstations and provides a basis for subsequent adjustments.

[0107] For example, if the load values ​​of 10 workstations range from 400 to 550 times, and the difference ranges from 20 to 150 times, it is initially indicated that there is an imbalance. The load difference is compared with a preset first load threshold, which is usually set based on historical data.

[0108] In a possible implementation, if historical data shows that the load difference is 30 times on average and fluctuates between 20 and 40 times, the first load threshold may be set to 50 times.

[0109] Specifically, the current difference value of 50 times is equal to the first load threshold, which requires further processing when exceeded.

[0110] It should be noted that the first load threshold is set to take into account the normal fluctuation range of the production line to ensure the rationality of the adjustment. If the load difference exceeds the first load threshold, the overloaded and idle workstations are marked.

[0111] For example, a workstation with a load of 550 times is marked as overloaded, and one with a load of 400 times is marked as idle. This marking clearly distinguishes the workstation status.

[0112] Preferably, after extracting the identification by the workstation number, the overloaded workstation can be located in the front section and the idle workstation in the back section, and the task should be transferred to the front section. After the task adjustment direction is determined, the idle workstation task is assigned to the overloaded workstation.

[0113] In one embodiment, an idle workstation has 50 tasks per day, which are transferred to the overloaded workstation to generate an allocation table, and the load of the overloaded workstation is reduced to 500 times.

[0114] It should be noted that the allocation table needs to record the workstation number and task quantity for easy execution.

[0115] For example, when linear regression is used to predict the adjusted load result, based on the data of the past five days, which are 510, 500, 490, 495, and 500 times, respectively, the predicted adjusted load stabilizes at 490 times, and the new load difference drops from 50 times to 40 times, which is close to the first load threshold.

[0116] It can be understood that the linear regression predicts the future distribution through the historical trend, and if the new difference still exceeds the first load threshold, the adjustment is repeated.

[0117] For example, the secondary adjustment reduces the load to 480 times, and the difference is reduced to 30 times, which falls within the first load threshold range, and the final solution is formed.

[0118] In one embodiment, repeated adjustments ensure maximum resource utilization.Finally, K-means clustering is used to group the workstation status.

[0119] For example, loads were categorized into three groups: low, medium, and high. Low loads ranged from 400 to 450 times, medium loads from 450 to 500 times, and high loads above 500 times. After grouping, it was found that medium-load workstations accounted for the highest proportion, confirming their stability.

[0120] Specifically, this classification helps identify potential bottlenecks.

[0121] For example, after the load of a certain workstation was adjusted from 550 times to 480 times, production continuity was improved and stagnation caused by overload was avoided.

[0122] In one possible implementation, after tasks at idle workstations are reallocated, overall efficiency is improved and resource waste is reduced.

[0123] Understandably, this approach optimizes the production line layout through dynamic adjustments.

[0124] S4, obtains the heart rate, action repetition rate and rest interval duration data from the workers' wearable devices, combines them with the preset fatigue curve model to calculate the real-time fatigue index of each worker, and obtains the quantitative distribution of fatigue status.

[0125] Optionally, this step also includes: Step S41: collect heart rate data, action repetition rate, and rest interval duration through the wearable device, store them in a time series format, and obtain an original data set.

[0126] Step S42: performing outlier detection on the original data set according to a preset heart rate threshold, marking data points where the heart rate data exceeds the heart rate threshold, and obtaining a filtered data set.

[0127] Step S43 : Using a linear regression model, the real-time fatigue value of each worker is calculated based on the filtered data set to obtain a fatigue index set.

[0128] Step S44, based on the fatigue index set, calculate the ratio of the action repetition rate to the rest interval duration. If the ratio is higher than the preset fatigue index threshold, adjust the fatigue index by a weighted method to obtain a revised fatigue index set.

[0129] Step S45 , using the K-means clustering algorithm, the worker states are divided according to the modified fatigue index set to obtain preliminary grouping of fatigue states.

[0130] Step S46 , for the preliminary grouping, using the moving average method to analyze the fatigue index variation trend of each group to obtain the quantitative distribution of fatigue status.

[0131] Step S47 , extracting the dynamic features of the worker status from the quantified distribution, and using a linear regression model to predict future changes in real-time fatigue to obtain a fatigue trend sequence.

[0132] Specifically, collecting heart rate data, movement repetition rate, and rest interval duration through wearable devices can be understood as a way to use smart hardware to monitor workers' physiological and behavioral status in real time.

[0133] For example, on an electronics assembly line, workers wear wristbands that record their heart rate every minute. Their motion repetition rate is calculated by counting the number of times a component is installed, and their rest intervals are determined by the device detecting periods of inactivity. For example, if a worker has a heart rate of 80 beats per minute, a motion repetition rate of 200 times per hour, and a rest interval of 10 minutes per hour, this data can be stored chronologically to form a continuous time series, facilitating subsequent analysis.

[0134] In a possible implementation, when detecting abnormal points according to a preset heart rate threshold, the normal heart rate may be set to 60-100 beats / minute.

[0135] For example, if a worker's heart rate spikes to 120 beats per minute, exceeding the heart rate threshold, this is marked as an outlier. After filtering, data within the normal range, such as 80 or 85 beats per minute, is retained, while outliers like 120 beats per minute are removed, resulting in a more reliable dataset. This approach ensures the accuracy of subsequent analysis.

[0136] Specifically, when using a linear regression model to calculate real-time fatigue values, trends can be established based on filtered data.

[0137] For example, a worker's heart rate is 80, 90, and 95 beats per minute for three consecutive hours, and the repetition rate of the movements gradually increases. The model calculates the fatigue index through these variables, such as 2.0 initially and then rising to 3.5.

[0138] It should be noted that the fatigue index reflects the degree of physical exhaustion of workers and provides a basis for management.

[0139] Preferably, when calculating the ratio of action repetition rate to rest interval length, assuming that a worker's action repetition rate is 200 times per hour and the rest interval is 10 minutes, the ratio is 20. If the fatigue index threshold is 15 and the ratio exceeds the threshold, the fatigue index is adjusted by weighting.

[0140] For example, an original index of 3.5 may be adjusted to 4.0 after weighting. This adjustment takes into account the additional effect of fatigue caused by high repetitions.

[0141] In one embodiment, when K-means clustering is used to divide worker states, fatigue indexes may be divided into three categories: low, medium, and high.

[0142] For example, an index below 2.0 is considered low, 2.0-4.0 is considered medium, and 4.0 and above is considered high. A worker with an index of 4.0 is assigned to the high fatigue group, while another worker with an index of 2.5 is assigned to the medium fatigue group. This grouping intuitively reflects the distribution of worker status.

[0143] It's understandable that when analyzing fatigue index trends using the moving average method, assuming a group of workers have five-hour indices of 3.0, 3.2, 3.5, 3.4, and 3.3, the moving average shows a slowly rising trend. This method smooths out short-term fluctuations and reveals patterns of fatigue accumulation.

[0144] For example, after extracting dynamic features from the quantitative distribution, linear regression is used to predict future fatigue changes. For example, if a worker's fatigue index over the past five hours was 3.0, 3.1, 3.3, 3.5, and 3.6, it's predicted to reach 3.8 in the next hour. This prediction enables early intervention to prevent excessive fatigue. This approach optimizes worker status management through dynamic monitoring and prediction.

[0145] S5, based on the quantitative distribution and load difference, the task distribution plan is optimized through the linear programming algorithm to determine the transfer path of overloaded workstation tasks to idle workstations where low-fatigue workers are located.

[0146] Optionally, this step also includes: Step S51 , calculating the quantized distribution and the load difference through a linear programming algorithm, and obtaining a comparison result of the task loads of the overloaded workstations and the idle workstations.

[0147] Step S52: extract overloaded workstation data from the load difference to determine the task load that needs to be transferred.

[0148] Step S53: obtaining the distribution data of low-fatigue workers from a pre-established worker status database, and determining the receiving capacity of the corresponding idle workstations according to the worker fatigue levels.

[0149] Step S54: linear programming is used to construct a transfer path matrix to distribute the task load of the overloaded workstations to the idle workstations.

[0150] Step S55 , adjusting the transfer path matrix according to the task load comparison result to obtain an optimized task distribution result.

[0151] Step S56: update the worker distribution data according to the workstation status and determine whether the task load is within the second load threshold range.

[0152] Step S57: If the task load exceeds the second load threshold, the transfer path matrix is ​​recalculated through linear programming to obtain a final optimization solution.

[0153] Specifically, calculating the quantitative distribution and load difference through a linear programming algorithm can be understood as a way to balance the distribution of workstation tasks using mathematical optimization methods.

[0154] For example, consider an electronics assembly line with five workstations that must complete 500 products per hour. Suppose the quantitative distribution for a certain period shows that workstation 1 is loaded with 150 pieces and workstation 2 with 50 pieces. The algorithm calculates a load difference of 100 pieces, indicating that workstation 1 is overloaded and workstation 2 is idle. This comparison provides a basis for subsequent adjustments.

[0155] In one possible implementation, when extracting overloaded workstation data from the load difference, the amount of tasks that need to be transferred is determined. Specifically, if workstation 1 is overloaded by 100 items, 50 of the tasks can be transferred.

[0156] It should be noted that the transfer amount needs to be determined based on the actual situation of the production line, such as the distance between workstations or the skills of workers.

[0157] For example, if workstation 1 is responsible for screen installation and is overloaded by 50 pieces, this part of the task can be transferred to an adjacent idle workstation. When obtaining the distribution data of low-fatigue workers from the pre-established worker status database, the receiving capacity can be determined through historical records.

[0158] For example, the database shows that the fatigue index of the worker at workstation 2 has remained stable at 1.5 for the past two hours, far below the high fatigue threshold of 4.0, indicating that he has enough energy to accept tasks.

[0159] Preferably, if the worker at station 2 is good at screen installation, then the receiving ability is stronger. This matching improves the feasibility of task transfer. When using linear programming to construct the transfer path matrix, the task allocation logic between stations must be considered.

[0160] In one embodiment, a matrix of five workstations shows that workstation 1 transfers 50 pieces to workstation 2, and workstation 3 transfers 20 pieces to workstation 4. The routing must be designed to ensure that the overall load is balanced.

[0161] For example, after the transfer, the load at workstation 1 drops to 100 pieces, while that at workstation 2 increases to 100 pieces. This approach avoids excessive pressure on a single workstation. By adjusting the transfer path matrix based on the task load comparison results, allocation efficiency can be optimized.

[0162] For example, if the load of workstation 2 approaches the upper limit after receiving 50 pieces, the matrix is ​​adjusted to transfer part of the tasks to workstation 5, so that the load of each workstation is finally about 100 pieces.

[0163] It is understandable that this dynamic adjustment reduces the bottleneck phenomenon. When updating the worker distribution data according to the workstation status, it is necessary to determine whether the load is within the second load threshold range.

[0164] Specifically, if the preset second load threshold for each workstation is 120 pieces, the adjusted load for all workstations is 100 pieces, which meets the requirement. This update provides data support for real-time monitoring. If the task load exceeds the second load threshold, the transfer path matrix is ​​recalculated.

[0165] For example, during a certain period, the load at workstation 1 rose to 130 pieces, exceeding the second load threshold of 120 pieces. Through linear programming, 10 tasks were transferred to workstation 3, ultimately returning the load to a normal range. This iterative optimization ensures the stability of the production line.

[0166] S6, after the task transfer path is generated, the production rhythm parameters are updated using real-time data. If the rhythm deviation exceeds the rhythm threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new rhythm distribution.

[0167] Optionally, this step also includes: Step S61: Acquire real-time data through sensors, update production rhythm parameters, and obtain the current rhythm value.

[0168] Step S62: If the difference between the current beat value and the target value exceeds the beat threshold, the beat deviation state is determined to determine whether the trigger condition is met.

[0169] In step S63, the KNN algorithm is used to process the beat deviation, adjust the execution order of the workstation operations, and obtain an optimized operation sequence.

[0170] Step S64: recalculate the load of each workstation through the optimized operation sequence to obtain a preliminary beat distribution.

[0171] Step S65: After obtaining the preliminary beat distribution, compare it with the historical beat data to determine whether there is abnormal fluctuation and determine the stability of the adjusted sequence.

[0172] Step S66: Based on the stability judgment result, the beat distribution is smoothed using the exponential weighted average method to obtain the final beat distribution.

[0173] Optionally, the beat distribution is calculated using the following formula: in, represents the smoothed beat distribution, α represents the smoothing coefficient, n represents the length of historical data involved in the calculation, ω represents the original beat distribution, and t represents the current moment.

[0174] Step S67: Update the execution plan of the workstation operation through the final beat distribution to complete the dynamic optimization of the production beat.

[0175] Specifically, obtaining real-time data through sensors is the basis for optimizing production rhythm.

[0176] For example, on an electronic product assembly line, sensors collect the number of products completed at each workstation and the time interval every minute. The data reflects that the current pace is 10 pieces per minute.

[0177] It should be noted that this data directly affects subsequent parameter updates. For example, the beat value will be adjusted to a value closer to the actual value based on real-time data collection. If the beat value differs significantly from the target value, a deviation status must be determined.

[0178] In a possible implementation, the target tempo is 12 pieces per minute, while the current tempo is 10 pieces. The difference of 2 pieces exceeds the preset tempo threshold of 1 piece, triggering an adjustment condition.

[0179] Specifically, this judgment can be based on a comparison of data over several consecutive minutes. For example, if the value is consistently below the target value for three minutes, this indicates that the deviation is not a random fluctuation. When using the KNN algorithm to process beat deviations, similar scenarios are searched for based on historical data.

[0180] Preferably, the algorithm extracts the operation sequence when the takt time was 10 pieces in the past from the database and finds the adjustment solution that is closest to the current working condition.

[0181] For example, historical records show that by scheduling part preparation steps in advance, the cycle time was increased to 11 pieces. This approach quickly generates an optimized sequence based on adjacent cases. When the load is recalculated based on the optimized operation sequence, the workload is redistributed among the workstations.

[0182] It's understandable that if one workstation's load decreases from 100 pieces per hour to 90 pieces due to earlier part preparation and reduced waiting time, while another workstation's load increases to 110 pieces per hour due to taking on additional steps. This preliminary distribution provides a basis for subsequent analysis. When comparing historical cycle time data to identify unusual fluctuations, the cycle time trend over the past hour is examined.

[0183] In one embodiment, if the historical tact rate has been stable between 11-12 pieces and suddenly drops to 10 pieces, it may indicate equipment delays or worker fatigue.

[0184] It should be noted that this comparison can identify potential problems in a timely manner and ensure the reliability of sequence adjustment. When using the exponentially weighted average method to smooth the beat distribution, recent and historical data are integrated.

[0185] For example, the current beat is 10 pieces, and the previous ones were 11 pieces and 12 pieces respectively. Through smoothing, a stable value close to 11 pieces is obtained.

[0186] Specifically, a smoothing coefficient of 0.3 can be set to prioritize recent data. This approach reduces the impact of short-term fluctuations and makes the beat more consistent. When the execution plan is updated using the final beat distribution, workstation tasks are rescheduled based on the smoothed results.

[0187] In one possible implementation, a workstation was originally scheduled to complete 100 pieces per hour, but after adjustment it was reduced to 105 pieces, while another workstation was reduced from 110 pieces per hour to 100. This dynamic optimization ensures a more stable production rhythm and avoids excessive local pressure.

[0188] For example, if workers at a certain workstation reduce their need for overtime, overall efficiency can also be improved.

[0189] S7, through comparative analysis of the new beat distribution and fatigue index, it is judged whether the task distribution scheme meets the dual constraints of load balance and fatigue optimization, and the final task execution sequence is obtained.

[0190] Optionally, this step also includes: Step S71: Generate an initial task distribution plan using a greedy algorithm based on the collected new beat distribution and fatigue index data.

[0191] Step S72 : Compare the task allocation in the initial task distribution plan with the new beat distribution, and calculate the load difference of each task node.

[0192] Step S73: If the load difference exceeds a preset third load threshold, the task distribution plan is adjusted, tasks are reallocated, and an updated task distribution result is generated.

[0193] Step S74: Calculate the fatigue index of each task node based on the updated task allocation result and compare it with the preset fatigue threshold.

[0194] Step S75: If the fatigue index exceeds the fatigue threshold, the task allocation is further adjusted to ensure that the fatigue optimization satisfies the constraint conditions.

[0195] Step S76: Generate a task sequence that meets the load balance requirement based on the adjusted task allocation result.

[0196] In step S77 , the task sequence and the fatigue index are verified twice to ensure that the fatigue index of each task node is lower than the fatigue threshold, thereby obtaining an optimized execution sequence.

[0197] Step S78: Integrate the new beat distribution with the optimized execution sequence to determine the final task execution sequence.

[0198] Specifically, when generating the initial task distribution plan based on the collected data, a greedy algorithm can be used to make quick decisions.

[0199] For example, on an electronics assembly line, suppose the new tempo distribution shows that workstation A completes 8 pieces per minute and workstation B completes 12 pieces per minute, while fatigue index data indicates that the current fatigue value of the worker at workstation A is 60 and that of the worker at workstation B is 40. The greedy algorithm will prioritize assigning tasks to workstation B, which has a higher tempo and lower fatigue, in order to maximize efficiency.

[0200] Specifically, we might assign all tasks for a batch of 50 products to workstation B, while temporarily keeping workstation A from adding any new tasks. This approach is simple and direct, allowing for a quick initial plan. When comparing the task assignments with the new cycle time distribution, we need to calculate the load difference.

[0201] In one possible implementation, the target load of workstation B is 700 pieces per hour, but due to the addition of new tasks, the actual load increases to 750 pieces, a difference of 50 pieces. If the preset third load threshold is 30 pieces, then the standard is exceeded.

[0202] It should be noted that this comparison can intuitively reflect whether the distribution is reasonable.

[0203] For example, workstation A's load is 450 pieces per hour, which is within expectations. However, an overall imbalance in the distribution could affect subsequent production. If the load variance exceeds the standard, adjusting the task distribution plan is a key step.

[0204] Preferably, part of the tasks of workstation B can be transferred to workstation A.

[0205] For example, after reallocating 20 tasks, the load at workstation B dropped to 730 pieces, and the load at workstation A increased to 470 pieces. The differences were both controlled within the third load threshold.

[0206] Understandably, this adjustment needs to take into account workstation capacity to avoid new problems caused by insufficient capacity. Calculating the fatigue index is crucial for the updated allocation results.

[0207] In one example, after adjustment, the fatigue index of workstation A rises to 65, while the preset third load threshold is 60, indicating an overload. The fatigue index may be based on a comprehensive assessment of work hours and task intensity, for example, increasing by 5 points for every additional 10 tasks. This assessment can quantify the worker's condition and provide a basis for optimization. If the fatigue index exceeds the limit, further adjustment is necessary.

[0208] Specifically, the workload at workstation A can be reduced to 460 pieces, lowering the fatigue index to 58, while tasks can be distributed to other workstations. This approach ensures that workers are not overloaded and maintains long-term production stability. When generating a task sequence that meets the load balance, the results are comprehensively adjusted.

[0209] For example, workstation A produces 460 pieces per hour and workstation B produces 730 pieces per hour, forming a preliminary sequence.

[0210] It should be noted that this sequence needs to take into account process connections to avoid production interruptions due to adjustments. When rechecking the fatigue index, ensure that all workstations meet the requirements.

[0211] In one possible implementation, the fatigue index for workstation B is 45, within the standard, and the fatigue index for workstation A is 58, also within the fatigue threshold. This verification can identify potential risks and improve solution reliability. Finally, integrating the new beat distribution with the execution sequence creates the final plan.

[0212] For example, the task volume of workstation A is 8 pieces per minute and 460 pieces per hour, which is combined with the task volume of workstation B at 12 pieces and 730 pieces per hour to ensure the overall rhythm is coordinated.

[0213] For example, this integration can reduce waiting time between workstations and improve assembly line efficiency.

[0214] S8, based on the final task execution sequence, generates operation instructions and rest period arrangements for each workstation, sends them to the production line through the equipment terminal, and obtains real-time feedback data after execution to verify the stability of the optimization results.

[0215] Optionally, this step also includes: Step S81: Generate workstation operation instructions and rest period arrangements from the task sequence according to preset rules to obtain an operation instruction set and a period table.

[0216] Step S82: Send the operation instruction set and time table to the production line through the equipment terminal to obtain execution status data.

[0217] Step S83: pre-process the execution status data, extract the operation time, workstation status, and equipment operation parameters, and generate a feedback data set.

[0218] Step S84: Use principal component analysis to extract features from the feedback data set and calculate process efficiency, workstation load, and equipment failure rate.

[0219] Step S85, determine whether the process efficiency, workstation load, and equipment failure rate exceed the status threshold. If so, adjust the task sequence and rest arrangement to obtain an updated instruction set.

[0220] Step S86: Send the updated instruction set to the production line through the device terminal to obtain new execution status data.

[0221] Step S87: Repeat feature extraction and state threshold judgment on the new execution state data to determine whether the optimization result is stable.

[0222] Specifically, when generating workstation operation instructions and rest period arrangements from task sequences through preset rules, a specific plan can be formulated based on the workload of the workstation and the status of the workers.

[0223] For example, on an electronics assembly line, suppose workstation A completes 460 units per hour and workstation B 730. A pre-set rule might be to schedule a 5-minute break after every 200 units. For workstation A, instructions could be generated to operate four cycles per hour, with a 5-minute break after completing 115 units per cycle, for a total of 20 minutes. Workstation B, on the other hand, would operate three cycles per hour, with a break after approximately 243 units per cycle, for a total of 15 minutes. This arrangement breaks tasks into manageable units while preventing workers from working continuously for extended periods.

[0224] In one possible implementation, after the operation instruction set and time table are sent to the production line through the equipment terminal, the execution status data will be fed back in real time.

[0225] For example, the terminal at workstation A shows that the actual completion time for a cycle is 15 minutes, while that at workstation B is 12 minutes, exceeding the expected time by 10 minutes. This data can reflect the actual execution status.

[0226] It should be noted that feedback data may also include whether the equipment is functioning properly and whether workers have taken breaks on time, which facilitates subsequent analysis. When preprocessing the execution status data, key indicators can be extracted.

[0227] In one example, workstation A lasted 15 minutes, and the status indicated that the worker had taken a 2-minute break. Equipment operating parameters indicated normal motor speed. Workstation B lasted 12 minutes, and the status was normal, with slightly elevated equipment temperature. This information is integrated to generate a feedback dataset that intuitively reflects the production line's operating status. When extracting features from this feedback dataset using principal component analysis, we can focus on process efficiency, workstation load, and equipment failure rate.

[0228] For example, process efficiency is measured by the ratio of operating time to planned time. Workstation A's ratio is 15 to 10, indicating low efficiency. Workstation load is based on the comparison of completed tasks against capacity. Workstation B's 730 pieces are near the upper limit. Equipment failure rate is assessed based on the number of abnormal temperature and speed events. High temperatures at workstation B may indicate risk. This type of analysis can help identify core issues.

[0229] When judging whether these indicators exceed the status threshold, it is preferred to set a process efficiency below 90% as unqualified, a workstation load exceeding 95% as overload, and a failure rate exceeding 5% as requiring an early warning.

[0230] Specifically, the efficiency of workstation A is 66.7%, which exceeds the standard; the load of workstation B is close to the upper limit but does not exceed it, and the failure rate is 3% due to temperature abnormality, which does not reach the status threshold.

[0231] It is understandable that in response to exceeding standards, it is imperative to adjust the task sequence.

[0232] For example, workstation A's workload could be reduced to 430 pieces, with a 10-minute break added. Workstation B's workload would remain unchanged, but equipment cooling would be enhanced. After the updated instruction set is delivered via the device terminal, the new execution status data will reflect the adjustment results.

[0233] In one example, the operating time at workstation A dropped to 12 minutes, and efficiency rose to 83%, still slightly below the target, but a significant improvement. The temperature at workstation B returned to normal, and the failure rate dropped to 1%. This feedback validated the feasibility of the adjustments. Repeated analysis and assessment of the new execution status data confirms the stability of the optimization results.

[0234] For example, if the efficiency of workstation A remained above 85% for three consecutive hours, and the failure rate of workstation B remained below 1%, the adjusted plan would have balanced the load and reduced risk. This iterative process ensures long-term stable operation of the production line.

[0235] See also Figure 2 In a second aspect of the present invention, a load balancing and fatigue optimization system for a wiring harness production line is provided. The load balancing and fatigue optimization system for the wiring harness production line is performed using the above method. The system mainly comprises: The data acquisition module is used to obtain the operation time, complexity parameters and worker movement frequency data of each workstation in the wire harness production through sensors and equipment interfaces, and obtain the real-time distribution characteristics of production rhythm and load; The load analysis module is used to calculate the load value and complexity deviation of each workstation under the current beat based on the real-time distribution characteristics using a time series analysis algorithm, and to determine the dynamic trend of uneven load among workstations; The trend extraction module is used to extract the specific identification of overloaded and idle workstations and the load difference from the dynamic change trend, and determine which workstations need to adjust their tasks to achieve a preliminary balance in load distribution; The fatigue calculation module is used to obtain the heart rate, action repetition rate and rest interval duration data from the workers' wearable devices, and calculate the real-time fatigue index of each worker based on the preset fatigue curve model to obtain the quantitative distribution of fatigue status; The task optimization module is used to optimize the task distribution plan based on the quantitative distribution and load difference through a linear programming algorithm, and determine the transfer path of tasks from overloaded workstations to idle workstations where low-fatigue workers are located; The beat adjustment module is used to update the production beat parameters using real-time data after the task transfer path is generated. If the beat deviation exceeds the beat threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new beat distribution; The constraint judgment module is used to determine whether the task distribution plan meets the dual constraints of load balance and fatigue optimization by comparing the new beat distribution with the fatigue index, and obtain the final task execution sequence; The instruction issuing module is used to generate operation instructions and rest period arrangements for each workstation based on the final task execution sequence, and issue them to the production line through the equipment terminal to obtain real-time feedback data after execution to verify the stability of the optimization results.

[0236] The present invention provides a method and system for load balancing and fatigue optimization of a wire harness production line. The method collects the operating parameters and worker status data of each workstation in real time through sensors and equipment interfaces, calculates the workstation load distribution using a time series analysis algorithm, and dynamically allocates tasks in combination with the worker fatigue index. The present invention first identifies overloaded and idle workstations, then optimizes the task transfer path based on a linear programming algorithm, transfers the tasks of overloaded workstations to idle workstations where low-fatigue workers are located, and simultaneously uses real-time data to update the production rhythm, and triggers a dynamic matching algorithm to adjust the operation sequence when necessary. Finally, by comparing the new rhythm distribution with the fatigue index, a task execution sequence that meets the dual constraints of load balancing and fatigue optimization is generated and sent to the production line for execution. The present invention can effectively improve production efficiency, reduce worker fatigue, and realize intelligent management of wire harness production lines.

[0237] Although the present invention has been described in detail above using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made thereto. Therefore, such modifications and improvements, without departing from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A method for load balancing and fatigue optimization of a wiring harness production line, characterized in that: The method comprises: S1, through the sensor and equipment interface, obtains the operation time, complexity parameters and worker movement frequency data of each workstation in the wire harness production, and obtains the real-time distribution characteristics of production rhythm and load; S2, based on the real-time distribution characteristics, uses a time series analysis algorithm to calculate the load value and complexity deviation of each workstation at the current beat, and determines the dynamic change trend of the uneven load between workstations; S3, extracts the specific identification of overloaded and idle workstations and the load difference from the dynamic change trend, and determines which workstations need to adjust their tasks to achieve a preliminary balance in load distribution; S4, obtaining the heart rate, action repetition rate, and rest interval data from the workers' wearable devices, and calculating the real-time fatigue index of each worker based on a preset fatigue curve model to obtain a quantitative distribution of fatigue status; S5, based on the quantitative distribution and load difference, optimize the task distribution plan through the linear programming algorithm to determine the transfer path of overloaded workstation tasks to idle workstations where low-fatigue workers are located; S6, after the task transfer path is generated, the production rhythm parameters are updated using real-time data. If the rhythm deviation exceeds the rhythm threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new rhythm distribution; S7, through comparative analysis of the new beat distribution and fatigue index, it is determined whether the task distribution scheme meets the dual constraints of load balance and fatigue optimization, and the final task execution sequence is obtained; S8, based on the final task execution sequence, generates operation instructions and rest period arrangements for each workstation, sends them to the production line through the equipment terminal, and obtains real-time feedback data after execution to verify the stability of the optimization results.

2. The method according to claim 1, characterized in that Step S1, through sensors and equipment interfaces, obtains the operation duration, complexity parameters, and worker motion frequency data of each workstation in the wiring harness production process, and obtains the real-time distribution characteristics of the production rhythm and load, including: Step S11, obtaining original records of operation duration, complexity parameters, and action frequency of the workstation, wherein the original records are collected through the sensor interface and the equipment interface; Step S12, filtering the original records using a preset screening threshold, determining the correspondence between the action frequency and the operation duration, and obtaining preliminary features of the workstation data; Step S13, calculating the production tact time using a preset formula based on the complexity parameter and operation duration in the preliminary features, and determining the tact time distribution of each workstation; Step S14: if the beat distribution exceeds a preset range, the complexity parameter is adjusted using real-time data to obtain dynamic characteristics of the load distribution; Step S15, based on the load distribution and production rhythm in the dynamic characteristics, using the K-means clustering algorithm to divide the workstation load levels and determine the classification characteristics of the load distribution; Step S16, obtaining the duration characteristics of the workstation data by comparing the classification characteristics with the real-time data, and determining the change trend of the load distribution; Step S17: Based on the distribution characteristics in the change trend, a linear regression algorithm is used to predict the subsequent production cycle to obtain an optimized adjustment value of the load distribution.

3. The method according to claim 1, characterized in that Step S2, which uses a time series analysis algorithm to calculate the load value and complexity deviation of each workstation at the current beat based on the real-time distribution characteristics, and determines the dynamic change trend of the uneven load between workstations, includes: Step S21: Obtain the load value and complexity data of each workstation at the current beat from the real-time distribution characteristics, calculate the load value deviation using the ARIMA model, and obtain preliminary results of unevenness between workstations; Step S22: Process the load value and complexity deviation through the ARIMA model, extract the dynamic change characteristics, and obtain the time series data of the load of each workstation; Step S23: Based on the dynamic change characteristics, K-means clustering is used to group the time series data to determine the load unevenness trend among the workstations; Step S24: If the load unevenness trend exceeds the preset unevenness threshold, the fluctuation range of the unevenness is obtained through historical data comparison and analysis, and the key workstation is determined; Step S25: For key workstations, extract the correlation features between beat and complexity from the real-time distribution, calculate the adjusted load value using linear regression, and obtain the optimized load distribution; Step S26, based on the optimized load distribution, use the ARIMA model to predict future trends and determine the long-term changes in unevenness between workstations; Step S27: By comparing the prediction result with the current load distribution, a dynamic adjustment plan is determined to obtain the final load balancing strategy.

4. The method according to claim 1, wherein Step S3 extracts the specific identifiers and load differences of overloaded and idle workstations from the dynamic change trend, and determines which workstations need to adjust their tasks to achieve a preliminary balance in load distribution, including: Step S31, obtaining the load value of each workstation from the real-time data, and obtaining the load difference by calculating the difference between the load values ​​of each workstation; Step S32, comparing the load difference with a first load threshold value, where the first load threshold value is set based on historical data; Step S33: if the load difference exceeds the first load threshold, marking the overloaded workstation and the idle workstation to determine the workstation status; Step S34: extract the workstation ID through the workstation number in the workstation status, obtain the specific locations of the overloaded workstation and the idle workstation, and determine the task adjustment direction; Step S35, assigning tasks of idle workstations to overloaded workstations according to the task adjustment direction, and generating a task allocation table; Step S36, using a linear regression algorithm to predict the adjusted load distribution result; Step S37, extracting a new load difference from the adjusted load distribution result; Step S38: If the new load difference still exceeds the first load threshold, the task allocation table is repeatedly adjusted until the load difference is within the first load threshold, thereby obtaining a final load allocation plan. Step S39: Use the K-means clustering algorithm to group the workstation status and determine the stability of the load distribution within each group.

5. The method according to claim 1, wherein Step S4, obtaining the heart rate, action repetition rate, and rest interval data from the workers' wearable devices, and calculating the real-time fatigue index of each worker in combination with a preset fatigue curve model to obtain a quantitative distribution of fatigue status, includes: Step S41, collecting heart rate data, movement repetition rate, and rest interval duration through a wearable device, storing them in a time series format, and obtaining an original data set; Step S42, performing outlier detection on the original data set according to a preset heart rate threshold, marking data points where the heart rate data exceeds the heart rate threshold, and obtaining a filtered data set; Step S43, using a linear regression model to calculate the real-time fatigue value of each worker based on the filtered data set to obtain a fatigue index set; Step S44, calculating the ratio of the action repetition rate to the rest interval duration based on the fatigue index set. If the ratio is higher than a preset fatigue index threshold, the fatigue index is adjusted using a weighted method to obtain a revised fatigue index set. Step S45, using the K-means clustering algorithm to divide the worker states according to the modified fatigue index set to obtain preliminary grouping of fatigue states; Step S46, analyzing the fatigue index variation trend of each group using the moving average method to obtain the quantitative distribution of fatigue status; Step S47 , extracting the dynamic features of the worker status from the quantified distribution, and using a linear regression model to predict future changes in real-time fatigue to obtain a fatigue trend sequence.

6. The method according to claim 1, characterized in that The step S5, optimizing the task distribution plan based on the quantitative distribution and the load difference through a linear programming algorithm, and determining the transfer path of the overloaded workstation tasks to the idle workstations where the low-fatigue workers are located, includes: Step S51, calculating the quantized distribution and the load difference by a linear programming algorithm to obtain a comparison result of the task load of the overloaded workstation and the idle workstation; Step S52, extracting overloaded workstation data from the load difference to determine the task load that needs to be transferred; Step S53: obtaining the distribution data of low-fatigue workers from a pre-established worker status database, and determining the receiving capacity of the corresponding idle workstations according to the worker fatigue levels; Step S54, using linear programming to construct a transfer path matrix to distribute the task load of the overloaded workstations to the idle workstations; Step S55, adjusting the transfer path matrix according to the task load comparison result to obtain an optimized task distribution result; Step S56, updating the worker distribution data according to the workstation status, and determining whether the task load is within a second load threshold range; Step S57: If the task load exceeds the second load threshold, the transfer path matrix is ​​recalculated through linear programming to obtain a final optimization solution.

7. The method according to claim 1, characterized in that The step S6, after the task transfer path is generated, uses real-time data to update the production rhythm parameters. If the rhythm deviation exceeds the rhythm threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new rhythm distribution, including: Step S61, acquiring real-time data through sensors, updating production rhythm parameters, and obtaining the current rhythm value; Step S62: If the difference between the current beat value and the target value exceeds the beat threshold, the beat deviation state is determined to determine whether the trigger condition is met; Step S63: Using the KNN algorithm to process the beat deviation, adjust the execution order of the workstation operations, and obtain an optimized operation sequence; Step S64, recalculating the load of each workstation through the optimized operation sequence to obtain a preliminary beat distribution; Step S65: After obtaining the preliminary beat distribution, compare it with the historical beat data to determine whether there is abnormal fluctuation and determine the stability of the adjusted sequence; Step S66, based on the stability judgment result, the beat distribution is smoothed using the exponential weighted average method to obtain the final beat distribution; Step S67: Update the execution plan of the workstation operation through the final beat distribution to complete the dynamic optimization of the production beat.

8. The method according to claim 1, characterized in that The step S7, determining whether the task distribution scheme satisfies the dual constraints of load balancing and fatigue optimization by comparing the new beat distribution with the fatigue index, and obtaining the final task execution sequence, includes: Step S71, using a greedy algorithm to generate an initial task distribution plan based on the collected new beat distribution and fatigue index data; Step S72, comparing the task allocation in the initial task distribution plan with the new beat distribution, and calculating the load difference of each task node; Step S73: If the load difference exceeds a preset third load threshold, the task distribution plan is adjusted, tasks are reallocated, and an updated task distribution result is generated; Step S74: Calculate the fatigue index of each task node based on the updated task allocation result and compare it with the preset fatigue threshold; Step S75: If the fatigue index exceeds the fatigue threshold, further adjust the task allocation to ensure that the fatigue optimization meets the constraint conditions; Step S76, generating a task sequence that meets the load balance according to the adjusted task allocation result; Step S77: Perform a secondary check on the task sequence and the fatigue index to ensure that the fatigue index of each task node is lower than the fatigue threshold, thereby obtaining an optimized execution sequence. Step S78: Integrate the new beat distribution with the optimized execution sequence to determine the final task execution sequence.

9. The method according to claim 1, characterized in that Step S8 generates operating instructions and rest period arrangements for each workstation based on the final task execution sequence, sends them to the production line via the equipment terminal, and obtains real-time feedback data after execution to verify the stability of the optimization results, including: Step S81, generating workstation operation instructions and rest period arrangements from the task sequence according to preset rules, and obtaining an operation instruction set and a period table; Step S82: Send the operation instruction set and time table to the production line through the equipment terminal to obtain execution status data; Step S83, pre-processing the execution status data, extracting the operation time, workstation status and equipment operating parameters, and generating a feedback data set; Step S84, using principal component analysis to extract features from the feedback data set and calculate process efficiency, workstation load, and equipment failure rate; Step S85, determining whether the process efficiency, workstation load, and equipment failure rate exceed the status threshold. If so, adjusting the task sequence and rest schedule to obtain an updated instruction set; Step S86: Send the updated instruction set to the production line via the device terminal to obtain new execution status data; Step S87: Repeat feature extraction and state threshold judgment on the new execution state data to determine whether the optimization result is stable.

10. A wire harness production line load balancing and fatigue optimization system, characterized in that: The method according to any one of claims 1 to 9 is used to perform load balancing and fatigue optimization on a wiring harness production line, the system comprising: The data acquisition module is used to obtain the operation time, complexity parameters and worker movement frequency data of each workstation in the wire harness production through sensors and equipment interfaces, and obtain the real-time distribution characteristics of production rhythm and load; The load analysis module is used to calculate the load value and complexity deviation of each workstation under the current beat based on the real-time distribution characteristics using a time series analysis algorithm, and to determine the dynamic trend of uneven load among workstations; The trend extraction module is used to extract the specific identification of overloaded and idle workstations and the load difference from the dynamic change trend, and determine which workstations need to adjust their tasks to achieve a preliminary balance in load distribution; The fatigue calculation module is used to obtain the heart rate, action repetition rate and rest interval duration data from the workers' wearable devices, and calculate the real-time fatigue index of each worker based on the preset fatigue curve model to obtain the quantitative distribution of fatigue status; The task optimization module is used to optimize the task distribution plan based on the quantitative distribution and load difference through a linear programming algorithm, and determine the transfer path of tasks from overloaded workstations to idle workstations where low-fatigue workers are located; The beat adjustment module is used to update the production beat parameters using real-time data after the task transfer path is generated. If the beat deviation exceeds the beat threshold, the dynamic matching algorithm is triggered to adjust the operation sequence of each workstation to obtain a new beat distribution; The constraint judgment module is used to determine whether the task distribution plan meets the dual constraints of load balance and fatigue optimization by comparing the new beat distribution with the fatigue index, and obtain the final task execution sequence; The instruction issuing module is used to generate operation instructions and rest period arrangements for each workstation based on the final task execution sequence, and issue them to the production line through the equipment terminal to obtain real-time feedback data after execution to verify the stability of the optimization results.

Images