Wire harness manufacturing management system and method based on artificial intelligence

Abstract

The invention discloses a wire harness manufacturing management system and method based on artificial intelligence, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: inputting a structured feature vector and a process constraint condition into a multi-target reinforcement learning model, and generating a process parameter instruction; the process parameter instruction is executed, the product yield is monitored in real time, and cross-process quality tracing is carried out on the abnormal product yield; historical data of abnormal batches in crimping, assembling and testing procedures are traced, the defect root cause probability of each procedure is calculated based on a Bayesian causal network, and primary and secondary root causes of quality defects are obtained and analyzed; and dynamically updating process constraint conditions according to primary and secondary root cause analysis results, adjusting a reward function of the multi-target reinforcement learning model, and re-optimizing a process parameter instruction. According to the invention, the operation state of the production equipment is accurately captured, the production process is optimized and adjusted in real time, and a closed-loop management process is formed, so that the production flexibility, efficiency and product quality are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and particularly to a wire harness manufacturing management system and method based on artificial intelligence. Background Art

[0002] The wire harness manufacturing process relies on manual experience and simple automated equipment to achieve the entire process from raw material cutting, crimping, assembly to final testing. However, with the introduction of the concept of Industry 4.0 and the progress of artificial intelligence technology, more and more intelligent means have been used in the wire harness manufacturing process. These technologies include but are not limited to using sensors to collect equipment operation status information, applying machine learning algorithms for data analysis, and realizing digital management of the production process through a cyber-physical system. Nevertheless, most of the current existing solutions still mainly focus on the technical improvement of a single aspect.

[0003] The existing technologies have significant deficiencies, including: First, most current ones lack an in-depth understanding of the operation status of production equipment. This results in limited capabilities in early warning and fault prevention, and it is unable to adjust production process parameters in a timely manner to avoid potential quality problems. Second, the existing quality management methods usually focus on post-event analysis rather than real-time monitoring and dynamic adjustment, which limits the ability to quickly respond to production line anomalies and may lead to a large amount of resources being wasted on rework and repair of defective products. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a wire harness manufacturing management method based on artificial intelligence to solve the problems of insufficient utilization of production equipment operation status information and lack of real-time monitoring and dynamic adjustment capabilities in the prior art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a wire harness manufacturing management method based on artificial intelligence, which includes,

[0008] Collecting vibration signals, thermal imaging data and production order information of wire harness manufacturing equipment, and generating a structured feature vector through preprocessing; inputting the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate a process parameter instruction; executing the process parameter instruction, monitoring the product yield in real time, and performing cross-process quality traceability on abnormal product yields; tracing the historical data of abnormal batches in the crimping, assembly and testing processes, calculating the probability of defect root causes for each process based on a Bayesian causal network, obtaining the primary and secondary root causes of quality defects, and analyzing them; dynamically updating the process constraint conditions according to the analysis results of the primary and secondary root causes, adjusting the reward function of the multi-objective reinforcement learning model and re-optimizing the process parameter instruction.

[0009] As a preferred solution of the artificial intelligence-based wire harness manufacturing management method described in the present invention, wherein: the preprocessing includes performing a fast Fourier transform on the original vibration signal, extracting spectral features, and calculating the vibration standard deviation;

[0010] Calculate the surface temperature gradient of the wire for the temperature distribution, and encode the work order priority as an urgency coefficient.

[0011] As a preferred solution of the artificial intelligence-based wire harness manufacturing management method described in the present invention, wherein: the generation of the structured feature vector is specifically carried out as follows.

[0012] Align the vibration standard deviation and the surface temperature gradient of the wire by timestamp, and combine them with the urgency coefficient to form a structured feature vector.

[0013] As a preferred solution of the artificial intelligence-based wire harness manufacturing management method described in the present invention, wherein: the input of the structured feature vector and the process constraint conditions into the multi-objective reinforcement learning model to generate process parameter instructions is specifically carried out as follows.

[0014] Construct the state space and action space of the multi-objective reinforcement learning model based on the structured feature vector;

[0015] Design a multi-objective reward function based on the state space, action space, and process constraint conditions;

[0016] Perform parallel simulation in the digital twin environment based on the state space, action space, and multi-objective reward function to generate process parameter instructions.

[0017] As a preferred solution of the artificial intelligence-based wire harness manufacturing management method described in the present invention, wherein: the execution of the process parameter instructions, the real-time monitoring of the product yield, and the cross-process quality traceability of the abnormal product yield are specifically carried out as follows.

[0018] Execute the process parameter instructions, collect the actual crimping force, actual cutting speed, and the number of qualified products, and calculate the real-time product yield;

[0019] Perform abnormal detection on the real-time product yield, calculate the abnormal risk index, and judge the production status according to the abnormal risk index. When the generation status is abnormal, trigger the quality traceability process.

[0020] As a preferred solution of the artificial intelligence-based wire harness manufacturing management method described in the present invention, wherein: trace the historical data of the abnormal batch in the crimping, assembly, and testing processes, calculate the probability of the root cause of defects in each process based on the Bayesian causal network, obtain the primary and secondary root causes of the quality defects, and conduct analysis, specifically as follows.

[0021] Collect the historical data of abnormal batches in the crimping, assembly, and testing processes, and process them through data alignment and feature standardization to generate standardized features;

[0022] Construct a dynamic Bayesian causal network based on the processed standardized features and the physical causal relationships between processes, and calculate the defect root cause probability;

[0023] Conduct primary and secondary root cause analysis based on the defect root cause probability.

[0024] As a preferred solution of the wire harness manufacturing management method based on artificial intelligence according to the present invention, wherein: dynamically update the process constraint conditions according to the results of the primary and secondary root cause analysis, adjust the reward function of the multi-objective reinforcement learning model, and re-optimize the process parameter instructions. The specific steps are as follows:

[0025] According to the results of the primary and secondary root cause analysis, perform primary cause constraint update and secondary cause constraint update;

[0026] According to the primary cause constraint update and the secondary cause constraint update, adjust the reward function of the multi-objective reinforcement learning model, and re-generate the optimized process parameter instructions.

[0027] In a second aspect, the present invention provides a wire harness manufacturing management system based on artificial intelligence, including a data processing module for collecting vibration signals, thermal imaging data, and production order information of wire harness manufacturing equipment, and generating a structured feature vector through preprocessing; an intelligent optimization module for inputting the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate process parameter instructions; a real-time monitoring module for executing the process parameter instructions, real-time monitoring the product yield, and performing cross-process quality traceability on the abnormal product yield; a quality analysis module for tracing the historical data of abnormal batches in the crimping, assembly, and testing processes, calculating the defect root cause probability of each process based on the Bayesian causal network, obtaining the primary and secondary root causes of quality defects, and performing analysis; a process adjustment module for dynamically updating the process constraint conditions according to the results of the primary and secondary root cause analysis, adjusting the reward function of the multi-objective reinforcement learning model, and re-optimizing the process parameter instructions.

[0028] In a third aspect, the present invention provides a computer device, including a memory and a processor, wherein: when the computer program is executed by the processor, it implements any step of the wire harness manufacturing management method based on artificial intelligence as described in the first aspect of the present invention.

[0029] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, it implements any step of the wire harness manufacturing management method based on artificial intelligence as described in the first aspect of the present invention.

[0030] The beneficial effects of the present invention are as follows: By using a multi-objective reinforcement learning model to generate process parameter instructions based on these feature vectors and process constraint conditions, the accurate capture of the operating state of production equipment and the real-time optimization and adjustment of production processes are achieved. Further, by monitoring the real-time product yield and detecting abnormalities, combining Bayesian causal network analysis to identify the root causes of defects, dynamically updating process constraint conditions and re-optimizing process parameters, a closed-loop management process is formed, thereby significantly improving the flexibility, efficiency, and product quality of production. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0032] Figure 1 It is a flowchart of the wire harness manufacturing management method based on artificial intelligence in Embodiment 1.

[0033] Figure 2 It is a flowchart of preprocessing to generate structured feature vectors in Embodiment 1.

[0034] Figure 3 It is a flowchart of a multi-objective reinforcement learning model to generate process parameter instructions in Embodiment 1.

[0035] Figure 4 It is a flowchart of cross-process quality traceability and root cause analysis in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification.

[0037] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0038] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that can be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that are mutually exclusive of other embodiments.

[0039] Embodiment 1, refer to Figures 1 to 4, which is the first embodiment of the present invention. This embodiment provides an artificial intelligence-based wire harness manufacturing management method, including the following steps:

[0040] S1: Collect the vibration signals, thermal imaging data, and production order information of the wire harness manufacturing equipment, and generate structured feature vectors through preprocessing;

[0041] S1.1: Collect the vibration signals, thermal imaging data, and production order information of the wire harness manufacturing equipment;

[0042] Furthermore, install a three-axis vibration sensor on the main shaft of the crimping machine to collect vibration signals in real time;

[0043] Deploy an infrared thermal imager at the wire guide wheel to collect the surface temperature distribution of the wire in real time;

[0044] Read the priority and remaining delivery time of the current work order in real time through the API interface.

[0045] S1.2: Preprocess the vibration signals, thermal imaging data, and production order information collected from the wire harness manufacturing equipment;

[0046] Furthermore, perform a fast Fourier transform on the original vibration signal, extract spectral features, and calculate the vibration standard deviation;

[0047] Calculate the surface temperature gradient of the wire for the temperature distribution;

[0048] Encode the work order priority as an urgency coefficient, and normalize the remaining delivery time to a 0-1 scale.

[0049] It should be noted that the vibration spectrum in the 0.5-5 kHz frequency band is obtained through a 1024-point fast Fourier transform (FFT) weighted by a Hanning window, the probability density function of the frequency-domain energy distribution is calculated, and the root mean square value of the time-domain signal is extracted as the vibration standard deviation; an infrared thermal imager (resolution 320x240 pixels) is used to obtain the surface temperature field distribution matrix of the wire, the temperature gradient per millimeter is calculated along the crimping direction, and the standard deviation of the temperature field is statistically analyzed; the priority label is mapped to the urgency coefficient according to the work order attribute coding rule, and the normalized urgency is calculated by the ratio of the remaining delivery time to the total planned time.

[0050] It should be noted that the specific process of calculating the probability density function of the frequency-domain energy distribution is as follows: After the time-domain vibration signal is preprocessed, it is converted into a frequency-domain complex sequence through fast Fourier transform, and the squared amplitude value of each frequency point is extracted as the instantaneous energy distribution; based on a sliding time window, the energy distribution in continuous time periods is integrated to generate a frequency-domain energy time series matrix; the kernel density estimation algorithm is used to perform non-parametric probability modeling on the energy distribution, where the bandwidth parameter is adaptively determined by the Silverman criterion, and the Gaussian kernel function acts on the energy amplitude dimension, and finally the probability density distribution curve of the frequency-domain energy value is output, and its integral area satisfies the normalization condition. The effectiveness of the calculation is verified through the goodness-of-fit test of the frequency-domain energy distribution histogram and the probability density curve.

[0051] The priority label is based on the customer level (strategic customers are automatically of high priority) and the delivery time buffer rate formula for ordinary work orders;

[0052] S1.3: Align the timestamps of the preprocessed vibration signal, thermal imaging data, and production order information, and generate a feature vector;

[0053] Perform linear interpolation on the preprocessed vibration signal and align it with the preprocessed temperature distribution in time;

[0054] It should be noted that to solve the sampling rate difference, linear interpolation is performed on the vibration signal: According to the sampling moments of the temperature sequence, two adjacent sampling points of the vibration signal are selected, and the amplitude of the intermediate point is calculated according to the time ratio to generate vibration data that is strictly aligned with the time stamps of the temperature sequence.

[0055] The aligned vibration signal, temperature distribution, and urgency coefficient are combined by splicing to form a structured feature vector.

[0056] S2: Input the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate process parameter instructions;

[0057] S2.1: Construct the state space and action space of the multi-objective reinforcement learning model based on the structured feature vector;

[0058] Perform vibration feature enhancement, temperature feature expansion, and urgency feature optimization based on the structured feature vector, and generate the state space through combination;

[0059] It should be noted that vibration feature enhancement is to obtain the smoothness index of the vibration signal by calculating the standard deviation of the first derivative of the vibration signal;

[0060] Temperature feature expansion is to obtain the uniformity index of the temperature distribution by calculating the coefficient of variation of the temperature distribution;

[0061] The urgent feature optimization calculates the dynamic weight of the production task through the exponential decay function of the urgency coefficient and the remaining delivery time.

[0062] Discretize based on the pressing force and cutting speed in the process parameters and combine them into an action space;

[0063] It should be noted that the pressing force is discretized into 10 levels, with each level spaced 4 kN apart, covering a range from 10 kN to 50 kN;

[0064] The cutting speed is discretized into 10 levels, with each level spaced 1.5 m / min apart, covering a range from 5 m / min to 20 m / min.

[0065] S2.2: Design a multi-objective reward function based on the state space, action space, and process constraint conditions;

[0066] Furthermore, collect the defect rate, energy consumption ratio, and delay risk of the product, obtain the quality score, energy consumption score, and delivery date score, and perform weight allocation. Generate a multi-objective reward function through fixed-weight integration;

[0067] It should be noted that real-time collect the defect rate data of the current batch of products, the energy consumption data per unit time of the production equipment, and the remaining delivery time data of the work order. The defect rate is calculated by the ratio of the number of defective products to the total output. The energy consumption data is quantified by the ratio of the current energy consumption value to the historical benchmark energy consumption value. The delay risk is calculated according to the reciprocal of the remaining delivery time and the historical average delivery time;

[0068] Convert the defect rate to a quality score through a non-linear mapping function. Based on the ratio of the current energy consumption to the benchmark energy consumption, use a piecewise function to convert it to an energy consumption score. The delay risk is mapped to a delivery date score through an exponential function;

[0069] Statistically analyze the contribution degrees of the quality score, energy consumption score, and delivery date score to the comprehensive production performance in historical data. The quality contribution degree is calculated by the ratio of the historical average improvement amplitude of the yield to the total improvement amplitude of the three objectives. The energy consumption contribution degree is calculated by the ratio of the historical average reduction amplitude of the energy consumption to the total reduction amplitude of the three objectives. The delivery date contribution degree is calculated by the ratio of the historical average reduction amplitude of the delay risk to the total reduction amplitude of the three objectives;

[0070] Normalize the quality, energy consumption, and delivery date contribution degrees and use them as the quality score weight coefficient, energy consumption score weight coefficient, and delivery date score weight coefficient respectively. The sum of the weight coefficients is constantly 1;

[0071] Based on the temperature and humidity thresholds in the process constraint conditions, calculate the temperature overrun penalty factor and humidity overrun penalty factor respectively, and calculate the penalty item weight coefficient in combination with the historical overrun event frequency;

[0072] Specifically, the specific process of calculating the temperature overrun penalty factor and the humidity overrun penalty factor based on the temperature threshold and humidity threshold defined in the process constraint conditions is as follows: Real-time collect the surface temperature of the wire and the ambient humidity data. When the measured temperature value exceeds the preset temperature threshold, the ratio of the overrun part to the temperature threshold is used as the basic ratio, and the positive value of this ratio is taken and raised to the power of 1.5 to generate the temperature overrun penalty factor; when the measured humidity value exceeds the preset humidity threshold, the same method is used to calculate the positive value of the overrun ratio and raise it to the power of 1.2 to generate the humidity overrun penalty factor; the historical overrun event frequency is obtained by statistically calculating the percentage of the cumulative duration of temperature and humidity overruns in the past 30 production batches in the total production duration; the penalty term weight coefficient is composed of the current penalty factor multiplied by a dynamic adjustment factor containing the natural logarithm of the historical overrun frequency. Finally, the temperature and humidity penalty terms are incorporated into the multi-objective reward function according to the weighted results to form a non-linear reinforcement constraint mechanism for high-frequency and high-amplitude overrun behaviors.

[0073] Generate the target reward function by linearly weighted summation of the quality score, energy consumption score, delivery date score, and penalty term with fixed weights.

[0074] S2.3: Perform parallel simulation in the digital twin environment based on the state space, action space, and multi-objective reward function to generate the optimal process parameters;

[0075] Furthermore, based on Monte Carlo sampling, randomly select 500 groups of parameter combinations from the action space, call the dynamic equation of the hydraulic cylinder of the press and the plastic deformation model of the wire, and initialize the simulation environment;

[0076] It should be noted that the Monte Carlo sampling method generates 500 groups of combinations of pressing force and cutting speed in the discretized action space with the pressing force parameter range from 10 kN to 50 kN and the cutting speed parameter range from 5 m / min to 20 m / min according to the uniform distribution rule with a pressing force interval of 0.8 kN and a cutting speed interval of 0.3 m / min;

[0077] Call the dynamic equation of the hydraulic cylinder of the press (including the force transfer relationship of the effective area of the piston, oil pressure, and damping coefficient) and the elastic-plastic constitutive model of the wire (describing the stress-strain relationship of the material's yield strength and strain hardening characteristics), and set the initial simulation conditions including the zero calibration of the piston displacement, the initial strain state of the wire, and the reference values of the ambient temperature and humidity to complete the initialization of the digital twin simulation environment.

[0078] Calculate the piston displacement through the pressing force, synchronously record the oil pressure fluctuation and the servo motor torque, calculate the terminal pressing deformation amount based on the stress-strain model of the wire, count the number of qualified products, collect vibration data, temperature data, and production beat data, obtain the spectral entropy, temperature standard deviation, and delay risk, and assign weights;

[0079] It should be noted that based on the input pressing force, the piston displacement is calculated by integrating the effective area of the piston and the real-time hydraulic pressure in the dynamic equation of the hydraulic cylinder of the press, and the hydraulic pressure fluctuation curve and the real-time torque value converted from the driving current of the servo motor are recorded synchronously;

[0080] The specific process of calculating the piston displacement based on the input pressing force and synchronously recording the hydraulic pressure fluctuation and the servo motor torque is as follows: The pressing force calculates the real-time hydraulic pressure value through the force balance relationship between the effective area of the piston and the real-time hydraulic pressure in the dynamic equation of the hydraulic cylinder of the press. Considering the influence of the damping coefficient of the hydraulic cylinder on the piston movement speed, the time integral of the piston acceleration is performed to obtain the displacement; The hydraulic pressure fluctuation curve records the instantaneous pressure change at a millisecond sampling rate by the pressure sensor to generate a time-domain pressure fluctuation waveform; The driving current of the servo motor is collected in real time by a high-precision current transformer, and the current signal is converted into a torque value based on the motor torque conversion coefficient (calibrated value of the nameplate parameter), and the torque time series is recorded synchronously; The above process fully realizes the mapping relationship between the pressing force and the displacement, the dynamic characteristics of the hydraulic pressure, and the synchronous monitoring of the driving torque, providing a multi-physical field coupling data basis for the subsequent optimization of process parameters.

[0081] The pressing force and the displacement are input into the wire stress-strain model, and the terminal pressing deformation amount is calculated according to the material yield strength and the hardening index. The number of qualified products is counted by comparing the deformation tolerance range (±0.05 mm); The time-domain signal of the spindle acceleration is collected by a three-axis vibration sensor (sampling rate 10 kHz), and the energy distribution in the frequency band of 0.5 - 5 kHz is extracted by fast Fourier transform, and the power spectral entropy value is calculated as the spectral entropy; The infrared thermal imager is used to record the temperature field distribution data on the wire surface, and the temperature standard deviation is counted; The delay risk coefficient is quantified according to the ratio of the work order plan cycle timestamp to the simulated pressing-cutting cycle time; The weight coefficients are allocated according to the historical contributions of the quality score, the energy consumption score, and the delivery date score (quality 60%, energy consumption 25%, delivery date 15%) to complete the parameter configuration of the multi-objective reward function.

[0082] It should also be noted that the specific process of calculating the terminal pressing deformation amount according to the material yield strength and the hardening index is as follows: Based on the elastic-plastic constitutive model of the wire, the stress state of the terminal material under the action of the pressing force is decomposed into the elastic stage and the plastic stage. In the elastic stage, the strain is calculated according to Hooke's law. When the stress exceeds the yield strength, the plastic strain increment is corrected according to the strain hardening law described by the hardening index; The balance equation of the pressing force and the displacement is solved by the iterative method, and combined with the geometric parameters of the terminal (such as the initial cross-sectional area, the size of the pressing die cavity) and the power-law relationship of the hardening index, the cumulative strain in the plastic deformation stage is calculated by integration, and finally the deformation amount of the terminal after pressing is output; The residual analysis is performed on the calculation result of the deformation amount and the measured data of the high-precision laser displacement sensor. After verifying the model accuracy, it is applied to the determination of qualified products.

[0083] Couple the spectral entropy and the temperature standard deviation through a non - linear function as a stability penalty term to dynamically adjust the reward value. If the temperature or humidity exceeds the limit, calculate the exceeding ratio and generate a penalty term to reduce the reward value;

[0084] It should be noted that based on the spectral entropy of the vibration signal and the data of the wire surface temperature standard deviation, a dynamic stability evaluation index is generated through a non - linear function to adjust the calculation weight of the multi - objective reward function; when the real - time temperature exceeds 90 °C, calculate the temperature exceeding ratio, and when the ambient humidity exceeds 75% RH, calculate the humidity exceeding ratio. Multiply the sum of the absolute values of the two by the penalty coefficient 0.5 to generate a constraint penalty term, which is superimposed on the reward function to reduce the comprehensive score.

[0085] Screen the parameter combinations that are at least one - dimension strictly better than other solutions and have no significant deterioration in other dimensions among the four dimensions of quality, energy consumption, delivery time, and stability. Calculate the comprehensive score, screen the solution set with a comprehensive score higher than 0.7, apply ±5% parameter perturbations to the first 10 groups of solutions, and after simulation, screen the stable parameters with a yield rate (the yield rate refers to the percentage of the number of qualified products in the total production quantity of this batch) fluctuation less than 2% and a temperature standard deviation change less than 15%, and generate process parameter instructions;

[0086] It should be noted that by traversing all parameter combinations, screen the non - dominated solution set in which at least one of the quality score, energy consumption score, delivery time score, or stability score is strictly better than other solutions (the index value is more than 5% higher), and the remaining indexes do not show a deterioration amplitude exceeding 5%;

[0087] Use the weighted sum method to calculate the comprehensive score, and screen the parameter combinations with a comprehensive score higher than 0.7 as candidate solutions; apply ±5% amplitude perturbations to the crimping force and cutting speed of the first 10 groups of parameters in the candidate solution set (such as perturbing 45 kN to 42.75 kN and 47.25 kN, and 15 m / min to 14.25 m / min and 15.75 m / min). Perform crimping deformation simulation and production beat simulation of the perturbed parameter combinations in the digital twin simulation environment, and screen the stable parameter combinations with a yield rate fluctuation amplitude less than 2 percentage points (such as the reference yield rate of 98% is not less than 96% after perturbation) and a temperature standard deviation change rate less than 15% (such as the reference standard deviation of 5 °C does not exceed 5.75 °C after perturbation); generate target parameter instructions according to the mean and standard deviation of the crimping force (such as mean 45 kN, standard deviation 0.5 kN) and the mean and standard deviation of the cutting speed (such as mean 15 m / min, standard deviation 0.2 m / min) of the stable parameter set.

[0088] S3: Execute the process parameter instructions, monitor the product yield rate in real - time, and conduct cross - process quality traceability for abnormal product yield rates;

[0089] S3.1: Execute the process parameter instructions, collect the actual pressing force, actual cutting speed, and the number of qualified products, and calculate the real-time product yield.

[0090] Furthermore, write the pressing force and cutting speed instructions into the device PLC through the edge controller, set the rotational speed of the servo motor and the opening degree of the hydraulic valve, and start the production process.

[0091] The pressing force sensor (with an accuracy of ±0.3 kN) collects the actual pressing force, the optical encoder (with a resolution of ±0.1 m / min) collects the actual cutting speed. For every 50 wire harnesses produced, the AOI optical detector counts the number of qualified products and calculates the real-time product yield.

[0092] It should be noted that based on the edge controller, write the optimal pressing force instruction value (such as 45 kN ± 0.5 kN) and the cutting speed instruction value (such as 15 m / min ± 0.2 m / min) into the programmable logic controller (PLC) of the wire harness manufacturing equipment, drive the hydraulic servo system to adjust the opening degree of the oil pressure valve and the rotational speed of the servo motor; collect the actual pressing force value in real time through the pressing force sensor (range 0 - 100 kN, accuracy ±0.3 kN), and synchronously record the actual cutting speed value through the optical encoder (resolution ±0.1 m / min); after every 50 wire harnesses are produced, the automatic optical detector (AOI) scans and counts the number of qualified products based on the preset terminal deformation tolerance range (±0.05 mm) and the insulation sheath integrity standard, and calculates the real-time yield (number of qualified products / 50).

[0093] S3.2: Perform anomaly detection on the real-time product yield, and judge the production status according to the anomaly detection result. The expression is:

[0094]

[0095] where Ψ is the anomaly risk index, ΔF is the absolute deviation between the actual pressing force and the target value, ΔV is the absolute deviation between the actual cutting speed and the target value, Q is the real-time product yield, T is the wire surface temperature, H is the ambient humidity, F * is the target pressing force, V * is the target cutting speed;

[0096] Set the anomaly threshold according to historical data and perform anomaly determination;

[0097] When the anomaly risk index is greater than the anomaly threshold, it is determined that the production status is abnormal, and the quality traceability process is triggered;

[0098] When the anomaly risk index is less than the anomaly threshold, it is determined that the production status is normal, and production continues.

[0099] S4: Trace the historical data of the abnormal batch in the crimping, assembly, and testing processes, calculate the probability of the root cause of defects in each process based on the Bayesian causal network, obtain the primary and secondary root causes of quality defects, and conduct analysis;

[0100] S4.1: Collect the historical data of the abnormal batch in the crimping, assembly, and testing processes, and process them through data alignment and feature standardization to generate standardized features;

[0101] It should be noted that in the crimping process, the time-domain vibration signal of the abnormal batch is sampled by a three-axis vibration sensor deployed on the main shaft of the equipment. After band-pass filtering, the vibration spectrum entropy feature is calculated;

[0102] In the assembly process, a humidity sensor is used to record the ambient humidity, and a pressure sensor is used to collect the fixture pressure;

[0103] In the testing process, an insulation resistance tester is used to measure the resistance between the wire harness terminals;

[0104] Through the RFID tag of the abnormal batch, the timestamps of each process are associated to align the time windows of the crimping vibration data, the assembly humidity data, and the testing resistance data; perform standardization on the vibration spectrum entropy, humidity standardization, and insulation resistance standardization to generate standardized features.

[0105] S4.2: According to the processed standardized features and the physical causal relationships between processes, construct a dynamic Bayesian causal network, and calculate the probability of the root cause of defects. The expression is:

[0106]

[0107] where P(Z i ) is the probability of the root cause of the defect of the i-th historical data of the standardized feature Z i , Z i is the standardized feature of the i-th historical data, Z k is the standardized feature of the k-th historical data in the root cause combination, is the dynamic weight of the i-th historical data, is the dynamic weight of the k-th historical data in all possible root cause combinations, ρ ij is the correlation coefficient of the j-th historical data that is correlated with the target historical data point for the i-th historical data, ρ kj is the correlation coefficient of the k-th historical data in the root cause combination and the j-th historical data that is correlated with the target historical data point, Φ is the cumulative distribution function of the standard normal distribution, mapping the parameter fluctuation range to a probability value, λ is the time decay coefficient, Δt i is the interval between the occurrence time of the i-th historical data and the defect detection time, Δt kis the interval between the occurrence time of the k-th historical data in the root cause combination and the defect detection time, τ is the time window, K is the total number of possible root cause combinations, and m is the number of other parameters correlated with the i-th historical data;

[0108] It should be noted that is the time decay factor, which controls the timeliness of historical data. The longer the time interval, the lower the weight; ρ i is the correlation coefficient between the i-th historical data and the historical defect.

[0109] S4.3: Conduct primary and secondary root cause analysis based on the defect root cause probability;

[0110] Furthermore, sort the defect root cause probabilities from high to low and output the list of primary and secondary root causes;

[0111] Set a determination threshold according to historical data. When the defect root cause probability is greater than the determination threshold, it is determined as the primary cause; when the defect root cause probability is less than the determination threshold, it is determined as the secondary cause.

[0112] S5: Dynamically update the process constraint conditions according to the results of the primary and secondary root cause analysis, adjust the reward function of the multi-objective reinforcement learning model, and re-optimize the process parameter instructions;

[0113] S5.1: According to the results of the primary and secondary root cause analysis, perform primary cause constraint update and secondary cause constraint update;

[0114] It should be noted that based on the results of the primary and secondary root cause analysis, first, dynamically update the process constraint conditions for the primary cause parameters (such as excessive crimping temperature or crimping force deviation), adjust the upper temperature limit from 90°C to 85°C, and tighten the allowable deviation of the crimping force from ±0.5 kN to ±0.3 kN, and design a non-linear penalty function to strengthen the fluctuation effect; for the secondary cause parameters (such as insufficient insulation resistance or cutting speed deviation), update them to increase the lower resistance limit to 95 MΩ and compress the speed deviation limit to ±0.1 m / min, and use a penalty term that couples environmental factors; through weighted integration of the primary and secondary penalty terms into the multi-objective reward function (primary cause weight 0.7, secondary cause weight 0.4), trigger a parameter re-optimization mechanism with priority grading (primary cause immediately interrupts optimization, secondary cause is optimized after batches), and form an autonomous decision-making process of "root cause analysis - constraint update - dynamic penalty - closed-loop iteration".

[0115] S5.2: According to the primary cause constraint update and secondary cause constraint update, adjust the reward function of the multi-objective reinforcement learning model, and re-generate the optimized process parameter instructions;

[0116] It should be noted that based on the update of the primary cause constraints (such as the upper limit of the crimping temperature being adjusted from 90 °C to 85 °C, and the allowable deviation of the crimping force being tightened from ±0.5 kN to ±0.3 kN) and the update of the secondary cause constraints (such as the lower limit of the insulation resistance being increased to 95 MΩ, and the limit value of the cutting speed deviation being compressed to ±0.1 m / min), a non-linear penalty term is added to the reward function of the multi-objective reinforcement learning model; 500 groups of parameter combinations are generated by performing Monte Carlo sampling (the step size of the crimping force is 0.8 kN, and the step size of the cutting speed is 0.3 m / min) in the digital twin environment, and a new generation of process parameter instructions are generated based on the Pareto front screening.

[0117] This embodiment also provides an artificial intelligence-based wire harness manufacturing management system, including: a data processing module, configured to collect vibration signals, thermal imaging data, and production order information of wire harness manufacturing equipment, and generate a structured feature vector through preprocessing; an intelligent optimization module, configured to input the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate process parameter instructions; a real-time monitoring module, configured to execute the process parameter instructions, monitor the product yield in real time, and perform cross-process quality traceability on abnormal product yields; a quality analysis module, configured to trace the historical data of abnormal batches in the crimping, assembly, and testing processes, calculate the probability of the root cause of defects in each process based on the Bayesian causal network, obtain the primary and secondary root causes of quality defects, and perform analysis; a process adjustment module, configured to dynamically update the process constraint conditions according to the results of the primary and secondary root cause analysis, adjust the reward function of the multi-objective reinforcement learning model, and re-optimize the process parameter instructions.

[0118] This embodiment also provides a computer device applicable to the case of the artificial intelligence-based wire harness manufacturing management method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the artificial intelligence-based wire harness manufacturing management method as proposed in the above embodiment.

[0119] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads provided on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0120] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for managing wire harness manufacturing based on artificial intelligence as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disc.

[0121] In summary, through the present invention: by using a multi-objective reinforcement learning model to generate process parameter instructions based on these feature vectors and process constraint conditions, the accurate capture of the operating state of production equipment and the real-time optimization and adjustment of production processes are achieved. Further, by monitoring the real-time product yield and performing anomaly detection, combining Bayesian causal network analysis to identify the root causes of defects, dynamically updating the process constraint conditions and re-optimizing the process parameters, a closed-loop management process is formed, thereby significantly improving the flexibility, efficiency, and product quality of production.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based wire harness manufacturing management method, characterized in that: Including, Collect the vibration signals, thermal imaging data, and production order information of the harness manufacturing equipment, and generate a structured feature vector through preprocessing; Input the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate process parameter instructions; Execute the process parameter instructions, monitor the product yield in real time, and conduct cross-process quality traceability for abnormal product yields; Trace the historical data of the abnormal batch in the crimping, assembly, and testing processes, calculate the probability of the root cause of defects in each process based on the Bayesian causal network, obtain the primary and secondary root causes of quality defects, and conduct analysis; Dynamically update the process constraint conditions according to the analysis results of the primary and secondary root causes, adjust the reward function of the multi-objective reinforcement learning model, and optimize the process parameter instructions.

2. The method for harness manufacturing management based on artificial intelligence according to claim 1, wherein: The preprocessing includes performing a fast Fourier transform on the original vibration signal, extracting spectral features, and calculating the vibration standard deviation; Calculate the surface temperature gradient of the wire for the temperature distribution, and encode the work order priority as an urgency coefficient.

3. The method for harness manufacturing management based on artificial intelligence according to claim 2, wherein: The specific steps for generating the structured feature vector are as follows: Align the vibration standard deviation and the surface temperature gradient of the wire by timestamp, and combine them with the urgency coefficient into a structured feature vector.

4. The method for harness manufacturing management based on artificial intelligence according to claim 3, wherein: The specific steps for inputting the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate process parameter instructions are as follows: Construct the state space and action space of the multi-objective reinforcement learning model based on the structured feature vector; Design a multi-objective reward function based on the state space, action space, and process constraint conditions; Execute parallel simulation in the digital twin environment based on the state space, action space, and multi-objective reward function to generate process parameter instructions.

5. The method for managing the manufacturing of wire harnesses based on artificial intelligence according to claim 4, wherein: The specific steps for executing the process parameter instructions, monitoring the product yield in real time, and conducting cross-process quality traceability for abnormal product yields are as follows: Execute the process parameter instructions, collect the actual crimping force, actual cutting speed, and the number of qualified products, and calculate the real-time product yield; Conduct abnormal detection on the real-time product yield, calculate the abnormal risk index, and judge the production status based on the abnormal risk index. When the production status is abnormal, trigger the quality traceability process.

6. The method for harness manufacturing management based on artificial intelligence according to claim 5, wherein: The specific steps for tracing the historical data of the abnormal batch in the crimping, assembly, and testing processes, calculating the probability of the root cause of defects in each process based on the Bayesian causal network, obtaining the primary and secondary root causes of quality defects, and conducting analysis are as follows: Collect the historical data of the abnormal batch in the crimping, assembly, and testing processes, and process them through data alignment and feature standardization to generate standardized features; Construct a dynamic Bayesian causal network based on the processed standardized features and the physical causal relationship between processes, and calculate the probability of the root cause of defects; Conduct primary and secondary root cause analysis based on the probability of the root cause of defects.

7. The method for harness manufacturing management based on artificial intelligence according to claim 6, characterized in that: The specific steps for dynamically updating the process constraint conditions according to the analysis results of the primary and secondary root causes, adjusting the reward function of the multi-objective reinforcement learning model, and re-optimizing the process parameter instructions are as follows: Conduct primary cause constraint update and secondary cause constraint update according to the analysis results of the primary and secondary root causes; Adjust the reward function of the multi-objective reinforcement learning model according to the primary cause constraint update and secondary cause constraint update, and re-generate optimized process parameter instructions.

8. An artificial intelligence-based wire harness manufacturing management system, based on the artificial intelligence-based wire harness manufacturing management method according to any one of claims 1 to 7, characterized in that: It includes a data processing module, an intelligent optimization module, a real-time monitoring module, a quality analysis module, a process adjustment module, and a closed-loop management module. The data processing module is used to collect the vibration signals, thermal imaging data, and production order information of the wire harness manufacturing equipment, and generate a structured feature vector through preprocessing. The intelligent optimization module is used to input the structured feature vector and process constraint conditions into a multi-objective reinforcement learning model to generate process parameter instructions. The real-time monitoring module is used to execute the process parameter instructions, monitor the product yield in real time, and perform cross-process quality traceability for abnormal product yields. The quality analysis module is used to trace the historical data of abnormal batches in the crimping, assembly, and testing processes, calculate the probability of the root cause of defects in each process based on the Bayesian causal network, obtain the primary and secondary root causes of quality defects, and conduct analysis. The process adjustment module is used to dynamically update the process constraint conditions according to the analysis results of the primary and secondary root causes, adjust the reward function of the multi-objective reinforcement learning model, and re-optimize the process parameter instructions.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based wire harness manufacturing management method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based wire harness manufacturing management method according to any one of claims 1 to 7.

Images