Intelligent assembly platform and assembly method for wiring harness terminal

Abstract

The invention belongs to the technical field of automobile wire harness manufacturing, and discloses a wire harness terminal intelligent assembly platform and method, and the method comprises the steps: carrying out the multi-mode feature collection and analysis of a wire harness terminal, and constructing a terminal feature knowledge base; intelligent terminal classification and matching are carried out based on the terminal feature knowledge base, and an assembly association network is generated; performing assembly path optimization and multi-robot collaborative planning according to the assembly association network to obtain a dynamic assembly strategy; performing real-time force feedback adaptive control on the assembly process of the wire harness terminal based on a dynamic assembly strategy, performing multi-sensor fusion monitoring and abnormal state recognition on the assembly process, and constructing an assembly quality evaluation model; performing assembly result verification and process parameter self-optimization based on the assembly quality evaluation model to obtain a wiring harness assembly comprehensive evaluation result; and the assembling quality and reliability of the wire harness terminal are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive wiring harness manufacturing, and more specifically, to an intelligent assembly platform and assembly method for wiring harness terminals. Background Art

[0002] A patent with the application publication number CN119126726A discloses an intelligent monitoring and management system for a wiring harness production line, including a wiring harness terminal crimping process database module, a wiring harness terminal crimping text data acquisition module, a wiring harness terminal shape text data processing module, a wiring harness terminal pressure text data processing module, a wiring harness terminal crimping process analysis module, an error correction analysis and adjustment module, a wiring harness terminal crimping management judgment module, and a wiring harness terminal crimping human-computer interaction module; by collecting and preprocessing text data in the wiring harness terminal crimping process, obtaining the first monitoring index and the second monitoring index through analyzing the wiring harness terminal shape and the wiring harness terminal crimping pressure, and by real-time monitoring the shape change and pressure change of the terminal during the crimping process, defective products can be found in time and the alarm can be triggered to stop the machine, reducing the time and labor costs of manual inspection and significantly improving the production efficiency.

[0003] Modern automobiles use various types of terminals, and each type of terminal has differences in shape, size, material, electrical characteristics, etc. Traditional terminal classification relies on rule matching or manual identification, which is difficult to meet the intelligent requirements of large-scale assembly, resulting in terminal misalignment and assembly failure, affecting the stability of subsequent electrical connections. There are many types of terminals in the wiring harness, and it is difficult for manual assembly or simple robot vision recognition to accurately distinguish the terminal models, especially prone to errors in identifying similar terminals; the wiring harness has a complex layout inside the automobile, and the terminal connection involves multiple branches. Traditional path planning adopts a static sequence and lacks the ability of dynamic adjustment, resulting in low robot assembly efficiency and even assembly failure; terminal assembly requires high-precision alignment. If there are small pose errors during the assembly process, it will lead to insufficient or excessive insertion force, affecting the assembly quality and terminal life; traditional robot assembly methods rely on fixed trajectories, lack compliant control and error compensation, and cannot adapt to small deformations or position offsets of the terminals.

[0004] In view of this, the present invention proposes an intelligent assembly platform and assembly method for wiring harness terminals to solve the above problems. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: An intelligent assembly method for wiring harness terminals, the method includes: Performing multi-modal feature acquisition and analysis on the wiring harness terminals to construct a terminal feature knowledge base; Based on the terminal feature knowledge base, performing intelligent classification and matching of the terminals to generate an assembly association network; Optimize the assembly path and perform multi-robot collaborative planning according to the described assembly association network to obtain a dynamic assembly strategy; Perform real-time force feedback adaptive control on the assembly process of the wire harness terminals based on the dynamic assembly strategy; Perform multi-sensor fusion monitoring and abnormal state recognition on the assembly process, and construct an assembly quality evaluation model; Verify the assembly result and perform self-optimization of process parameters based on the assembly quality evaluation model to obtain a comprehensive evaluation result of wire harness assembly.

[0006] Furthermore, perform multi-modal feature acquisition and analysis on the wire harness terminals, and construct a terminal feature knowledge base, including: Use a multi-spectral imaging system to scan the surface of the terminal, obtain multi-dimensional feature data of the terminal surface, and perform image enhancement processing on the multi-dimensional feature data to obtain an enhanced feature map; Perform deep learning feature extraction on the enhanced feature map, obtain the geometric feature vector and material feature vector of the terminal, and perform dimensionality reduction processing on the geometric feature vector and material feature vector of the terminal to obtain a feature subspace; Use 3D laser scanning to measure the shape of the terminal, obtain accurate 3D point cloud data of the terminal, and perform registration and denoising on the 3D point cloud data to obtain a digital model of the terminal; Perform topological analysis on the digital model of the terminal, identify key connection surfaces and contact areas, and extract accurate dimensional parameters to obtain a terminal parameter set; Fuse the feature subspace and the terminal parameter set through a multi-layer graph structure to construct a terminal feature association graph, and establish a self-updating terminal feature knowledge base based on the terminal feature association graph.

[0007] Furthermore, perform intelligent classification and matching of terminals based on the terminal feature knowledge base to generate an assembly association network, including: Perform clustering analysis on the terminal feature knowledge base to obtain terminal category groups, and construct a terminal classification decision tree based on the terminal category groups to obtain an intelligent classification model; Use the intelligent classification model to perform real-time identification and classification of the terminals to be assembled, obtain terminal category labels, and perform analysis of terminal assembly requirements based on the terminal category labels to obtain an assembly requirement matrix; Perform topological analysis on the wire harness structure, identify terminal connection nodes and circuit directions, obtain a wire harness topology diagram, and perform electrical connection constraint analysis based on the wire harness topology diagram to obtain connection constraint conditions; According to the assembly requirement matrix and the connection constraint conditions, perform optimal calculation of terminal matching to obtain an optimal terminal pairing scheme; Perform graph structure modeling on the optimal terminal pairing scheme, construct the multi-dimensional relationships between terminal nodes, and generate an assembly association network.

[0008] Further, perform assembly path optimization and multi-robot collaborative planning based on the assembly association network to obtain a dynamic assembly strategy, including: Perform topological sorting on the assembly association network to obtain an initial assembly sequence, and perform assembly timing simulation according to the initial assembly sequence to identify timing conflict points and obtain a conflict constraint set; Construct an assembly path optimization model based on the conflict constraint set, and use a heuristic algorithm to perform path search optimization to obtain a multi-objective optimized assembly path; Decompose the multi-objective optimized assembly path into a set of subtasks, and perform assembly resource allocation according to the set of subtasks to obtain an initial collaborative assembly plan; Use the multi-agent reinforcement learning method to perform simulation training on the initial collaborative assembly plan to obtain a robot collaborative strategy network, and perform collaborative behavior optimization based on the robot collaborative strategy network to obtain an optimized collaborative plan; Perform robustness analysis on the optimized collaborative plan, establish an abnormal response strategy library, and combine the abnormal response strategy library and the optimized collaborative plan to form a dynamic assembly strategy.

[0009] Further, perform real-time force feedback adaptive control on the assembly process of the wire harness terminals, including: Install a micro force sensor array on the end effector of the robot, collect multi-dimensional force feedback data during the assembly process in real time, and perform filtering processing on the multi-dimensional force feedback data to obtain force signal characteristics; Construct a mechanical state observer based on the force signal characteristics, estimate the terminal contact state in real time, and perform assembly stage identification according to the terminal contact state to obtain the current assembly stage identifier; According to the current assembly stage identifier, automatically adjust the force control parameters, construct a stage-based adaptive impedance controller, and perform fine-tuning of the robot trajectory based on the stage-based adaptive impedance controller; Use visual servo technology to perform real-time tracking of the terminal position, obtain the terminal pose error, and perform multi-modal fusion of the terminal pose error and the force feedback data to obtain a comprehensive error vector; Perform adaptive compensation control based on the comprehensive error vector to dynamically adjust the pose of the end effector of the robot.

[0010] Further, perform multi-sensor fusion monitoring and abnormal state identification on the assembly process, and construct an assembly quality evaluation model, including: Deploy acoustic, visual, force - tactile, and electrical property sensors to build an all - around perception network, synchronously collect and pre - process the data of each sensor, and obtain a multi - modal perception data stream; Extract features and perform time - frequency domain analysis on the multi - modal perception data stream to obtain an assembly process feature spectrum, and establish a normal assembly mode library based on the assembly process feature spectrum; Use a deep anomaly detection algorithm to compare the real - time assembly process with the normal assembly mode library, identify potential abnormal patterns, and obtain abnormal feature vectors; Perform clustering analysis on the abnormal feature vectors, construct an abnormal type knowledge graph, and perform abnormal cause reasoning based on the abnormal type knowledge graph to obtain an abnormal diagnosis result; Perform correlation analysis on the assembly process data and the abnormal diagnosis result with the assembly quality standard to establish an assembly quality assessment model.

[0011] Further, perform assembly result verification and process parameter self - optimization based on the assembly quality assessment model to obtain a comprehensive evaluation result of wire harness assembly, including: Conduct electrical performance tests and mechanical strength detections on the assembled terminals, obtain assembly result measurement data, and compare the assembly result measurement data with the standard requirements to obtain a quality compliance index; Based on the quality compliance index and the assembly process data, use Bayesian network modeling to analyze the influencing factors of assembly quality and obtain a process parameter sensitivity matrix; According to the process parameter sensitivity matrix, construct an optimization objective function for process parameters, and use an evolutionary algorithm for multi - objective optimization calculation to obtain an optimized process parameter set; Conduct small - batch verification experiments on the optimized process parameter set, evaluate the parameter optimization effect, and perform incremental updates on the process knowledge base according to the verification results; Integrate assembly quality data, process parameter optimization effect, and production efficiency indicators to construct a multi - level evaluation system, generate a comprehensive evaluation result of wire harness assembly, and feedback the evaluation result to the assembly system for continuous improvement.

[0012] A smart wire harness terminal assembly platform, including: An acquisition module: used to perform multi - modal feature acquisition and analysis on wire harness terminals and construct a terminal feature knowledge base; A classification module: based on the terminal feature knowledge base, perform intelligent classification and matching of terminals to generate an assembly association network; An optimization module: according to the assembly association network, perform assembly path optimization and multi - robot collaborative planning to obtain a dynamic assembly strategy; A control module: based on the dynamic assembly strategy, perform real - time force feedback adaptive control to achieve precise assembly positioning; Evaluation module: Conduct multi-sensor fusion monitoring and abnormal state recognition for the assembly process, and construct an assembly quality evaluation model; Feedback module: Verify the assembly result and self-optimize the process parameters based on the assembly quality evaluation model to obtain a comprehensive evaluation result of the wire harness assembly.

[0013] A wire harness terminal intelligent assembly device includes: a memory and at least one processor, and instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the wire harness terminal intelligent assembly device to execute the above-mentioned wire harness terminal intelligent assembly method.

[0014] A computer-readable storage medium stores instructions, and when it runs on a computer, it enables the computer to execute the above-mentioned wire harness terminal intelligent assembly method.

[0015] The technical effects and advantages of a wire harness terminal intelligent assembly platform and an assembly method of the present invention: By collecting and analyzing multi-modal features of wire harness terminals, the present invention constructs a self-updating terminal feature knowledge base, realizes the comprehensive encoding and characterization of terminal geometric and material features, and improves the accuracy and comprehensiveness of feature recognition. Using the multi-layer graph structure fusion technology, the construction of the terminal feature association graph is realized, and the richness and relevance of feature expression are enhanced. Through the intelligent classification model and assembly requirement analysis, the accurate classification and optimal matching of terminals are realized, and the accuracy and efficiency of assembly are improved. By adopting topological analysis and multi-objective optimization algorithms, the intelligent planning and optimization of the assembly path are realized, and assembly conflicts and resource waste are reduced. Based on the multi-agent reinforcement learning method, the collaborative assembly planning of multiple robots is realized, and the overall cooperation efficiency and robustness of the system are improved. Through the real-time force feedback adaptive control technology, the precise positioning and compliant control during the terminal assembly process are realized. Using the multi-sensor fusion monitoring technology, the omni-directional perception of the assembly process and the real-time recognition of abnormal states are realized, and the comprehensiveness and timeliness of quality monitoring are improved. By adopting the deep anomaly detection algorithm and the abnormal type knowledge graph, the root cause and propagation path of assembly anomalies are deeply explored, and the accuracy of fault diagnosis is enhanced. Through the Bayesian network and evolutionary algorithm, the sensitivity analysis and self-optimization of process parameters are realized, and an adaptive process optimization closed-loop is formed to continuously improve the assembly quality and efficiency. By constructing a multi-level evaluation system, a complete comprehensive evaluation mechanism for wire harness assembly is formed, which provides strong support for the continuous improvement of the assembly system and helps to realize the intelligent, high-quality and high-efficiency management of the wire harness terminal assembly process. Description of the Drawings

[0016] Figure 1 It is a schematic diagram of a wire harness terminal intelligent assembly method of the present invention; Figure 2 Schematic diagram of an intelligent assembly platform for wire harness terminals of the present invention. Specific embodiments

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1

[0019] Please refer to Figure 1 As shown, an intelligent assembly method for wire harness terminals in this embodiment includes: Step S1: Collect and analyze multi-modal features of wire harness terminals to construct a terminal feature knowledge base; Step S2: Based on the terminal feature knowledge base, perform intelligent classification and matching of terminals to generate an assembly association network; Step S3: Optimize the assembly path and plan multi-robot collaboration according to the assembly association network to obtain a dynamic assembly strategy; Step S4: Perform real-time force feedback adaptive control on the assembly process of wire harness terminals based on the dynamic assembly strategy; Step S5: Perform multi-sensor fusion monitoring and abnormal state recognition on the assembly process to construct an assembly quality evaluation model; Step S6: Based on the assembly quality evaluation model, verify the assembly result and self-optimize the process parameters to obtain a comprehensive evaluation result of wire harness assembly.

[0020] In a specific embodiment, the process of executing step S1 may specifically include the following steps: (1) Use a multi-spectral imaging system to scan the surface of the terminal, obtain multi-dimensional feature data of the terminal surface, and perform image enhancement processing on the multi-dimensional feature data to obtain an enhanced feature map; (2) Perform deep learning feature extraction on the enhanced feature map, obtain the geometric feature vector and material feature vector of the terminal, and perform dimensionality reduction processing on the geometric feature vector and material feature vector of the terminal to obtain a feature subspace; (3) Use three-dimensional laser scanning to measure the shape of the terminal, obtain accurate three-dimensional point cloud data of the terminal, and perform registration and denoising on the three-dimensional point cloud data to obtain a digital model of the terminal; (4) Perform topological analysis on the digital model of the terminal, identify key connection surfaces and contact areas, and extract accurate dimensional parameters to obtain a terminal parameter set; (5) Fuse the feature subspace and the terminal parameter set through a multi-layer graph structure to construct a terminal feature association map, and establish an automatically updated terminal feature knowledge base based on the terminal feature association map; Specifically, use a multi-spectral imaging system to scan the terminal surface to obtain the surface multi-dimensional feature data of the terminal. The surface multi-dimensional feature data includes image information in the visible light, infrared, and ultraviolet spectral ranges. For example, visible light images (400 - 700 nm) are used to capture the appearance features of the terminal, near-infrared images (700 - 1100 nm) are used to detect the thickness and uniformity of the surface treatment layer, and ultraviolet images (200 - 400 nm) are used to discover surface micro-defects and contaminants. Perform image enhancement processing on the obtained multi-dimensional feature data, including denoising, contrast enhancement, and edge sharpening operations, to obtain an enhanced feature map. Gaussian filtering is used for denoising, histogram equalization is used for contrast enhancement, and the Sobel operator is used for edge sharpening. Perform deep learning feature extraction on the enhanced feature map. Use a pre-trained convolutional neural network or other suitable deep learning models, and through transfer learning methods, fine-tune the pre-trained model to make it suitable for the terminal feature extraction task. Automatically extract the geometric features and material features of the terminal from the enhanced feature map through the fine-tuned pre-trained model. Geometric features include shape, contour, surface curvature, etc., and material features include metal type, surface treatment process, etc. These features together constitute the geometric feature vector and material feature vector of the terminal. Perform dimensionality reduction processing on the geometric feature vector and material feature vector of the terminal to obtain a feature subspace. Common dimensionality reduction methods include principal component analysis and linear discriminant analysis. Through dimensionality reduction processing, while retaining the main original information, the redundancy of the features is significantly reduced, and a more compact feature subspace is obtained. Use three-dimensional laser scanning to measure the shape of the terminal to obtain accurate three-dimensional point cloud data of the terminal. A laser triangulation system with an accuracy of up to 5 microns is used to measure the shape of the terminal, and the terminal is scanned in all directions. During the scanning process, the terminal is placed on a precision rotary table, the laser line sweeps across the terminal surface, and the deformation of the laser line is captured by a camera to calculate the three-dimensional coordinates of the surface points. Through multi-angle scanning, dense point cloud data containing hundreds of thousands to millions of points is obtained to completely describe the three-dimensional geometry of the terminal. Three-dimensional laser scanning can capture the accurate shape and size information of the terminal with micron-level accuracy, providing basic data for subsequent geometric analysis. Perform registration and denoising processing on the obtained three-dimensional point cloud data to eliminate the errors and noises generated during the scanning process, and obtain a clean and accurate digital model of the terminal. The iterative closest point algorithm is used in the registration process to align the point cloud data obtained from multiple scans to a unified coordinate system. Suppose there are two sets of point cloud data and , the iterative closest point algorithm realizes registration by iteratively minimizing the adaptation objective function. The formula of the adaptation objective function is: , where, Represents the point cloud data within Represents the point cloud data within represents the rotation matrix represents the translation vector. The denoising process uses statistical outlier filtering and bilateral filtering methods to remove outliers and smooth the noise. The registration process ensures that the point cloud data from multiple scans can be correctly aligned, while the denoising process eliminates outliers and sampling errors. Topological analysis is performed on the terminal digital model to identify key connection surfaces and contact areas. The topological analysis uses region growing algorithms and RANSAC algorithms. Topological analysis can identify the structural features of the terminal, including geometric forms such as planes, curved surfaces, grooves, and protrusions, especially the key contact surfaces and insertion areas responsible for electrical connections. For example, for pin-type terminals, the cylindrical surface and conical lead-in surface of the insertion part are identified; for socket-type terminals, the internal contact spring pieces and lead-in inclined surfaces are identified. By analyzing the geometric characteristics of these areas, precise dimensional parameters are extracted, including key parameters such as contact area, insertion depth, and contact angle, to obtain the terminal parameter set. These parameters are crucial for ensuring the accuracy of assembly and the reliability of electrical connections. The feature subspace and the terminal parameter set are fused into a multi-layer graph structure to construct a terminal feature association graph. The association graph adopts a three-layer structure: the bottom-layer nodes represent the original features and parameters, the middle-layer nodes represent feature combinations and parameter associations, and the top-layer nodes represent terminal types and functional attributes. In the association graph, nodes represent different features and parameters, edges represent the association relationships between them, and weights represent the association strength, which are calculated through correlation analysis functions. Common correlation analysis functions include the Pearson coefficient function and the Spearman rank correlation coefficient function. Through the multi-layer graph structure, the hierarchical relationships and complex dependencies between different features can be expressed. Based on the terminal feature association graph, a self-updating terminal feature knowledge base is established. This knowledge base can be continuously expanded and optimized as new terminal samples are added, improving the ability to identify and analyze new types of terminals.

[0021] In a specific embodiment, the process of executing step S2 may specifically include the following steps: (1) Perform clustering analysis on the terminal feature knowledge base to obtain terminal category groups, and construct a terminal classification decision tree based on the terminal category groups to obtain an intelligent classification model; (2) Use the intelligent classification model to perform real-time identification and classification on the terminals to be assembled, obtain terminal category labels, and perform terminal assembly requirement analysis based on the terminal category labels to obtain an assembly requirement matrix; (3) Perform topological analysis on the wire harness structure to identify terminal connection nodes and wire routes, obtain a wire harness topology diagram, and perform electrical connection constraint analysis based on the wire harness topology diagram to obtain connection constraint conditions; (4) According to the assembly requirement matrix and connection constraint conditions, perform terminal matching optimization calculation to obtain the optimal terminal pairing scheme; (5) Conduct graph structure modeling on the optimal terminal pairing scheme, construct the multi-dimensional relationship between terminal nodes, and generate an assembly association network; Specifically, perform clustering analysis on the terminal feature knowledge base. Apply the K-means++ or hierarchical clustering algorithm to perform unsupervised clustering on the terminal feature data in the knowledge base to form natural terminal category groups. The clustering process is based on the Euclidean distance or Mahalanobis distance in the feature space, and terminals with similar features are grouped into the same group. The number of groups is adaptively determined by the silhouette coefficient evaluation index to ensure the minimization of intra-group differences and the maximization of inter-group differences. The clustering result forms terminal category groups, and each group represents a terminal type with similar features. Based on the terminal category groups, construct a terminal classification decision tree. Adopt the CART or C4.5 decision tree algorithm, and use information gain or Gini coefficient as the node splitting criterion. Each non-leaf node of the decision tree represents a judgment condition for features, the path represents the classification rule, and the leaf node represents the final terminal category. Optimize the decision tree structure through pruning to avoid overfitting and improve the generalization ability, and finally obtain an efficient and accurate intelligent classification model. Use the intelligent classification model to perform real-time identification and classification on the terminals to be assembled. First, obtain the image of the terminals to be assembled through the vision system, and use the same feature extraction process to obtain the feature vector F_new. Input F_new into the classification decision tree model, and obtain the terminal category label through feature judgment. The classification result includes key information such as the model, polarity, and applicable scenario of the terminal, providing an accurate classification basis for subsequent assembly. Perform terminal assembly requirement analysis according to the terminal category label, considering the physical size matching degree, electrical performance matching degree, and mechanical performance matching degree of the terminals, and calculate the assembly requirement matrix; The calculation formula for the physical size matching degree is: , where and respectively represent class and class terminal feature sizes; The calculation formula for the electrical performance matching degree is: , where and respectively represent class and class terminal voltage parameters, and respectively represent class and class terminal current parameters; The calculation formula for the mechanical performance matching degree is: , where represents class terminal insertion and extraction force, represents class terminal insertion and extraction force, It is a standardized parameter with a value range of [1, 10]. The assembly requirement matrix comprehensively describes the matching and adaptation relationships between different types of terminals, providing data support for subsequent optimization. Perform topological analysis on the wiring harness structure, representing the wiring harness as an undirected graph. Through graph traversal algorithms such as breadth-first search (BFS) or depth-first search (DFS), identify terminal connection nodes and line directions. For complex wiring harnesses, use community discovery algorithms to identify modular structures. Obtain the wiring harness topology graph through topological analysis, clearly showing the connection relationships and hierarchical structures between terminals. Based on the wiring harness topology graph, conduct electrical connection constraint analysis, considering electrical parameters such as current capacity, voltage level, signal type, as well as EMC compatibility requirements, to obtain a set of connection constraint conditions. The set of constraint conditions provides a criterion for judging the legality of terminal connections, ensuring the safety and reliability of electrical connections. According to the assembly requirement matrix and connection constraint conditions, perform terminal matching optimization calculations, and use the Hungarian algorithm to solve the maximum weight matching problem of a bipartite graph. Represent the two sets of terminals to be matched as a bipartite graph , where and respectively represent the two sets of terminal nodes, and the weight of the edge set represents the matching and adaptation degree in the assembly requirement matrix (such as electrical performance matching degree). The Hungarian algorithm solves the optimal matching within the time complexity of O(n³) through marking operations and alternating path searches. When the algorithm terminates, an optimal terminal pairing scheme is obtained, ensuring the maximum total matching and adaptation degree under the premise of meeting the connection constraint conditions. Through optimization calculations, an optimal terminal pairing scheme is obtained, ensuring the best performance and reliability of terminal matching. Perform graph structure modeling on the optimal terminal pairing scheme to construct a multi-layer directed graph. The multi-layer directed graph contains three layers of information: the physical connection layer represents the physical plugging relationship of terminals, the electrical connection layer represents the current, voltage, and signal transmission paths, and the assembly sequence layer represents the timing dependency relationship of terminal installation; the edges in the multi-layer directed graph not only represent physical connection relationships but also include multi-dimensional relationships such as electrical connections, mechanical fits, and assembly sequences. Analyze the centrality, connectivity, and community structure of terminal nodes through graph algorithms to evaluate the robustness and reliability of the assembly network, and finally generate a complete assembly association network. This network provides a basic data structure support for subsequent assembly path planning and optimization.

[0022] In a specific embodiment, the process of executing step S3 may specifically include the following steps: (1) Perform topological sorting on the assembly association network to obtain the initial assembly sequence, and perform assembly timing simulation according to the initial assembly sequence to identify timing conflict points and obtain a conflict constraint set; (2) Build an assembly path optimization model based on the conflict constraint set and use a heuristic algorithm to perform path search optimization to obtain a multi-objective optimized assembly path; (3) Decompose the multi-objective optimized assembly path to obtain a set of subtasks, and perform assembly resource allocation according to the set of subtasks to obtain an initial collaborative assembly plan; (4) Use the multi-agent reinforcement learning method to perform simulation training on the initial collaborative assembly plan to obtain a robot collaborative policy network, and optimize the collaborative behavior based on the robot collaborative policy network to obtain an optimized collaborative plan; (5) Conduct a robustness analysis on the optimized collaborative plan, establish an exception response strategy library, and combine the exception response strategy library and the optimized collaborative plan to form a dynamic assembly strategy; Specifically, perform a topological sorting on the assembly association network to determine the initial assembly order that meets the constraint conditions; conduct a timing simulation based on the initial assembly order to identify possible timing conflict points; construct an optimization model and use a heuristic algorithm to optimize the assembly path; decompose the optimized assembly path and perform resource allocation; use the multi-agent reinforcement learning method to optimize the collaborative strategy; conduct a robustness analysis and establish an exception response strategy library, and finally form a dynamic assembly strategy. Perform a topological sorting on the assembly association network, and represent the pre-order relationship of terminal assembly through a directed graph, where nodes represent terminal assembly tasks and edges represent assembly dependency relationships. Use the Kahn algorithm to perform a topological sorting on the assembly association network. This algorithm first identifies the nodes with an in-degree of 0 (assembly tasks without pre-dependencies), adds them to the sorting result, and then removes these nodes and the outgoing edges connected to them. Repeat this process until the graph is empty or a loop is found. Through the topological sorting, obtain the initial assembly order to ensure that all assembly pre-order relationships are satisfied. Conduct an assembly timing simulation according to the initial assembly order, and use the discrete event simulation method to construct a timing model of the assembly process.

[0023] For each task, its execution time consists of a deterministic base time and a stochastic variation time. During the simulation process, the start time and completion time of the task are statistically recorded. The start time of the task satisfies the condition that the task can only start after all its preceding tasks are completed. By performing Z Monte Carlo simulations, where Z is an integer greater than zero, the time distribution and resource occupancy of the tasks are statistically analyzed to identify timing conflict points. The conflict types include: resource conflict: two tasks simultaneously require the same resource; space conflict: there is an overlap in the operation areas of two tasks; path conflict: there is an intersection in the movement path of the robot. For each identified conflict point, the conflict task pair, conflict type, and conflict probability are recorded. The conflict probability is calculated by dividing the number of simulation times when the conflict occurs by the total number of simulation times. Finally, a conflict constraint set is formed. These conflict constraints provide key constraint conditions for subsequent path optimization. Based on the conflict constraint set, an assembly path optimization model is constructed. This model is a multi-objective optimization problem, and the objectives include: minimizing the total assembly time, minimizing the conflict probability, and maximizing the resource utilization rate; the constraint conditions include: task precedence relationship constraints, resource capacity constraints, and spatial interference constraints. An improved genetic algorithm is used for path search optimization. The algorithm process includes: initializing the population, where each individual represents a feasible assembly path; calculating the fitness of each individual; generating a new population through crossover, mutation, and selection operations. The crossover uses the partially mapped crossover operator, the mutation uses the insertion mutation operator, and the selection uses the roulette wheel and elitist retention strategy; repeat the iteration until convergence or the maximum number of iterations is reached. Through this algorithm, a multi-objective optimized assembly path is obtained, achieving a balance among time efficiency, conflict avoidance, and resource utilization. The multi-objective optimized assembly path is decomposed into tasks using a hierarchical decomposition method, and the overall assembly path is decomposed into multiple subtask sets. The decomposition process considers factors such as the spatio-temporal correlation of the tasks, similarity of resource requirements, and execution complexity. The specific algorithm is as follows: first, classify the tasks according to the assembly type (such as insertion, fixing, detection, etc.); then perform spatio-temporal clustering on each type of task using the K-means algorithm. The clustering features include task execution time, spatial position, and resource requirements, etc.; finally, form subtask sets according to the clustering results. Assembly resource allocation is performed according to the subtask sets. The resources include robots, end effectors, tooling fixtures, etc. The resource allocation uses a mixed integer linear programming (MILP) model. The decision variable represents whether to allocate the resource Res to the subtask Work. The objective function is to minimize the total assembly time and balance the resource load. By solving the MILP model, an initial collaborative assembly plan is obtained, specifying the subtasks and execution order for each robot. A multi-agent reinforcement learning method is used to simulate and train the initial collaborative assembly plan, and a multi-robot collaborative environment model is constructed. The environmental state includes the positions of each robot, task completion status, and resource occupancy status; the actions include the movement, grasping, and assembly of the robot; the reward function considers the task completion time, resource utilization rate, and collaborative effect.The Multi-Agent Proximal Policy Optimization (MAPPO) algorithm is used for training, and each robot agent learns its own policy function and value function. During the training process, collaborative learning among agents is promoted through shared observations and an experience pool. After thousands of simulation iterations of training, a robot collaborative policy network is obtained, which can output optimal collaborative actions according to the environmental state. Based on the robot collaborative policy network, collaborative behavior optimization is carried out, including path planning optimization, task switching optimization, and conflict avoidance optimization, to obtain an optimized collaborative plan. Robustness analysis is performed on the optimized collaborative plan, using methods of perturbation injection and sensitivity analysis. Perturbations include common anomalies such as execution time fluctuations, positioning errors, and grasping failures. For each perturbation, the change in the plan performance is calculated to obtain robustness metrics. Based on the results of the robustness analysis, an abnormal situation response strategy library is established, and recovery strategies are designed for different types of anomalies. For example, for the grasping failure anomaly, a three-level recovery strategy of retry-adjust-replace is designed; for the task conflict anomaly, strategies for timing adjustment and path replanning are designed; for the resource failure anomaly, strategies for task reassignment and degraded execution are designed. Combining the optimized collaborative plan and the abnormal situation response strategy library forms a dynamic assembly strategy.

[0024] In a specific embodiment, the process of executing step S4 may specifically include the following steps: (1) Install a micro force sensor array on the end effector of the robot to collect multi-dimensional force feedback data during the assembly process in real time, and filter the multi-dimensional force feedback data to obtain force signal features; (2) Construct a mechanical state observer based on the force signal features to estimate the terminal contact state in real time, and identify the assembly stage according to the terminal contact state to obtain the current assembly stage identifier; (3) Automatically adjust the force control parameters according to the current assembly stage identifier, construct a stage-adaptive impedance controller, and perform fine-tuning of the robot trajectory based on the stage-adaptive impedance controller to achieve compliant assembly control; (4) Use visual servo technology to track the terminal position in real time, obtain the terminal pose error, and perform multi-modal fusion of the terminal pose error and the force feedback data to obtain a comprehensive error vector; (5) Perform adaptive compensation control based on the comprehensive error vector to dynamically adjust the pose of the robot end effector and achieve sub-millimeter-level precise assembly positioning.

[0025] Specifically, a micro force sensor array is installed at the end effector of the robot to form a distributed force feedback network, which can collect multi-dimensional force feedback data during the assembly process in real time, including three-axis force and three-axis torque information, and comprehensively sense physical quantities such as contact force, friction force, and insertion resistance during the terminal assembly process. The collected multi-dimensional force feedback data is digitally filtered, including operations such as Kalman filtering, wavelet denoising, and low-pass filtering, to eliminate high-frequency noise and environmental interference in the sensor signal and extract stable and reliable force signal features. Through the filtering process, the signal-to-noise ratio of the force signal is effectively improved, providing a high-quality perception basis for subsequent state estimation and control decision-making. Based on the processed force signal features, a non-linear mechanical state observer is constructed to realize the real-time estimation of the terminal contact state. The mechanical state observer uses the recursive Bayesian estimation method to model the assembly process as a hidden Markov process. By analyzing the temporal and spatial distribution features of the force signal, it infers the current contact state between the terminal and the socket, including states such as non-contact, initial contact, partial insertion, and full insertion. According to the estimated terminal contact state, an assembly stage recognition algorithm based on a finite state machine is used to divide the assembly process into an approach stage, a search stage, an insertion stage, and a locking stage, and output the current assembly stage identifier, providing a decision basis for subsequent adaptive control. According to the identified current assembly stage identifier, the force control parameters are automatically adjusted, including the impedance stiffness matrix, damping coefficient, and force control gain, etc., to construct a stage-based adaptive impedance controller. In the approach stage, a high position stiffness is adopted to ensure accurate approach; in the search stage, the lateral stiffness is reduced to adapt to the position error; in the insertion stage, the longitudinal damping is adjusted to control the insertion speed; in the locking stage, the force control gain is increased to ensure full insertion. Based on the constructed stage-based adaptive impedance controller, the robot trajectory correction amount is calculated in real time, and the pre-planned trajectory is fine-tuned to achieve compliant assembly control, effectively coping with position errors and uncertainties during the assembly process. Visual servo technology is used to track the position of the terminal in real time. The terminal image is captured by a high-frame-rate industrial camera, and combined with a deep learning object detection algorithm, the position and pose of the terminal and the socket are accurately identified. The vision system adopts an eye-in-hand or eye-out-of-hand configuration to ensure continuous observation of the assembly area. Through visual measurement, the pose error of the terminal relative to the target position is obtained, including the position error vector and the attitude error matrix. The pose error of the terminal obtained by vision and the force feedback data provided by the force sensor are fused in a multi-modal manner, and a Kalman filter or a particle filter is used to achieve complementary fusion to obtain a more accurate and reliable comprehensive error vector, which comprehensively reflects the state deviation during the assembly process. Based on the calculated comprehensive error vector, an adaptive compensation controller is designed to dynamically adjust the position and pose of the robot end effector. The compensation controller adopts a hybrid control strategy, flexibly switching between position control and force control modes in different directions to achieve precise regulation of the assembly process.The compensation control algorithm adaptively adjusts the compensation gain and compensation direction according to the error magnitude and change trend, ensuring the stability and convergence of the compensation process. Through continuous closed-loop control and error compensation, high-precision adjustment of the pose of the robot end effector is achieved, ultimately reaching sub-millimeter-level assembly accuracy and ensuring the reliability and consistency of terminal insertion.

[0026] In a specific embodiment, the process of executing step S5 may specifically include the following steps: (1) Deploy acoustic, visual, force-tactile, and electrical property sensors to construct an all-round perception network, and synchronously collect and preprocess the data of each sensor to obtain a multi-modal perception data stream; (2) Extract features and perform time-frequency domain analysis on the multi-modal perception data stream to obtain the assembly process feature spectrum, and establish a normal assembly mode library based on the assembly process feature spectrum; (3) Use a deep anomaly detection algorithm to compare the real-time assembly process with the normal assembly mode library, identify potential abnormal modes, and obtain abnormal feature vectors; (4) Conduct clustering analysis on the abnormal feature vectors, construct an abnormal type knowledge graph, and perform abnormal cause reasoning based on the abnormal type knowledge graph to obtain an abnormal diagnosis result; (5) Correlate and analyze the assembly process data and the abnormal diagnosis result with the assembly quality standard to establish an assembly quality assessment model; Specifically, acoustic, visual, force / touch, and electrical property sensors are deployed to build an all-round perception network to achieve multi-dimensional monitoring of the assembly process. The acoustic sensor captures the sound characteristics during the assembly process, including insertion sound, locking sound, and abnormal friction sound; the acoustic sensor uses a high-precision microphone array (sampling rate 48 kHz) and is installed around the assembly station; the visual sensor includes a high-speed industrial camera (resolution 1920×1080, frame rate 60 fps) and a 3D depth camera; the force / touch sensor uses a six-axis force / torque sensor (accuracy ±0.1 N) and is installed at the end of the assembly mechanism; the electrical property sensor includes a four-wire resistance measuring instrument (accuracy 0.1 mΩ) and an insulation tester. The visual sensor monitors the assembly actions and the change of terminal position in real time; the force / touch sensor records the insertion force curve and the peak locking force; the electrical property sensor measures the contact resistance and insulation performance. The data of each sensor are synchronously collected and preprocessed, including time alignment, filtering and noise reduction, and outlier processing, to obtain a multi-modal perception data stream. Time alignment ensures that the timestamps of different sensor data are accurately matched, filtering and noise reduction eliminate environmental interference, and outlier processing removes obvious error data points. Feature extraction and time-frequency domain analysis are performed on the multi-modal perception data stream. Wavelet transform and short-time Fourier transform are used to perform time-frequency decomposition on the acoustic signal to extract frequency features and energy distribution features. A curve feature extraction algorithm is applied to the force / touch data to obtain key points, slopes, and area features of the force curve. Motion features and position features are extracted from the visual data through optical flow method and target tracking algorithm. The stability features of electrical connections are extracted from the electrical property data through analysis of the resistance change rate and volatility. The extracted multi-dimensional features are combined to form an assembly process feature spectrum to describe the dynamic characteristics of the entire assembly process. Based on the feature spectra of a large number of normal assembly samples, density estimation and boundary learning methods are applied to establish a normal assembly pattern library, which contains the feature distribution range and typical patterns of the normal assembly process. A deep anomaly detection algorithm is used to compare the real-time assembly process with the normal assembly pattern library. A five-layer autoencoder network (input layer - encoding layer - latent representation layer - decoding layer - output layer) is designed. The feature spectrum is input into the network, and the network parameters are trained by minimizing the reconstruction error. The autoencoder or variational autoencoder is used to encode and reconstruct the real-time feature spectrum, and the reconstruction error is calculated , where is the original feature, It is a reconstruction feature. When the reconstruction error exceeds the preset reconstruction threshold, it is determined as a potential anomaly. At the same time, the one-class SVM or isolation forest algorithm is used to calculate the anomaly score of the feature spectrum. The higher the anomaly score, the greater the likelihood of anomaly. Combining the reconstruction error and the anomaly score, a two-dimensional anomaly criterion is formed, and a decision boundary is set. The decision boundary includes: the reconstruction error threshold and the anomaly score threshold. When the two-dimensional anomaly criterion crosses the boundary, that is, the reconstruction error is greater than the reconstruction error threshold and the anomaly score is greater than the anomaly score threshold, it is determined as an anomaly and the corresponding anomaly feature vector is extracted. The anomaly feature vector records the multi-dimensional feature state when the anomaly occurs, providing a basis for subsequent analysis. Cluster analysis is performed on the anomaly feature vectors, using hierarchical clustering or density clustering algorithms to group similar anomaly feature vectors to form anomaly clusters. Based on the clustering results, an anomaly type knowledge graph is constructed. The nodes of the graph represent anomaly types, the edges represent the association relationships between anomalies, and the node attributes include anomaly feature descriptions, severity levels, and occurrence frequencies. Causal reasoning and rule reasoning methods are applied to establish a mapping relationship from anomaly features to potential causes , where represents the anomaly feature, represents the possible cause. Through the maximum a posteriori probability criterion, the most likely anomaly cause is determined to obtain the anomaly diagnosis result. The assembly process data and the anomaly diagnosis result are associated and analyzed with the assembly quality standard to establish a quality index mapping function , where is the normal assembly feature, is the anomaly feature, is the diagnosis result. The index mapping function is approximated by multiple regression or neural network methods to construct an assembly quality evaluation model. This model can predict the final assembly quality score based on the assembly process characteristics and anomaly situations and give improvement suggestions. The model evaluation dimensions include mechanical connection reliability, electrical contact stability, and service life expectancy, which are combined through weighting to form a comprehensive quality index. The model is continuously optimized in an online learning manner, and the prediction accuracy is continuously improved as data accumulates.

[0027] In a specific embodiment, the process of executing step S6 may specifically include the following steps: (1) Conduct electrical performance tests and mechanical strength detections on the assembled terminals, obtain the measurement data of the assembly results, and compare the measurement data of the assembly results with the standard requirements to obtain the quality compliance index; (2) Based on the quality compliance index and the assembly process data, use Bayesian network modeling to analyze the influencing factors of assembly quality and obtain the process parameter sensitivity matrix; (3) According to the process parameter sensitivity matrix, construct an optimization objective function for process parameters, and use an evolutionary algorithm for multi-objective optimization calculation to obtain the optimized process parameter set; (4) Conduct small-batch verification experiments on the optimized process parameter set, evaluate the parameter optimization effect, and incrementally update the process knowledge base according to the verification results to form an adaptive process optimization closed loop; (5) Construct a multi-level evaluation system by integrating assembly quality data, process parameter optimization effect, and production efficiency indicators, generate the comprehensive evaluation result of wire harness assembly, and feedback the evaluation result to the assembly system for continuous improvement; Specifically, electrical performance tests and mechanical strength inspections are carried out on the assembled terminals to obtain measurement data of the assembly results. The electrical performance tests include contact resistance measurement, insulation resistance measurement, and withstand voltage test to comprehensively evaluate the electrical reliability and insulation safety of the terminal connection. The mechanical strength inspections include pull-off force test, swing strength test, and vibration durability test to verify the firmness and stability of the mechanical connection of the terminal. The four-wire method is used to measure the contact resistance in the electrical performance test, with a measurement current of 100 mA and an accuracy of 0.1 mΩ; a high-voltage insulation tester is used to measure the insulation resistance, with a test voltage of 500 V, and the insulation resistance is required to be greater than 100 MΩ; a withstand voltage tester is used for the withstand voltage test, applying an AC voltage of 1000 V for 1 minute, and no breakdown and flashover phenomena are required. A microcomputer-controlled electronic universal testing machine is used for the pull-off force test in the mechanical strength inspection, with a pulling speed of 50 mm / min, and the maximum force value when the terminal is pulled out of the wire harness is recorded; a swing tester is used for the swing strength test, swinging at an angle of ±30° and a frequency of 2 Hz for 1000 times, and no loosening and damage of the terminal are required; a vibration table is used for the vibration durability test, with a vibration frequency of 5 - 500 Hz, an acceleration of 2 g, and a duration of 8 hours, and the terminal is required to maintain a stable connection. The measured performance data are quantitatively compared with the industry standard requirements and product specifications, and the deviation rate and compliance rate of each index are calculated to generate a quality compliance index. The quality compliance index comprehensively represents the assembly quality level through weighted average, providing a basis for subsequent process optimization. Based on the quality compliance index and assembly process data, a Bayesian network is used to model and analyze the factors affecting assembly quality. The assembly process data include multi-dimensional parameters such as the crimping force curve, insertion depth, crimping temperature, and operation time. The constructed Bayesian network model represents the causal relationship between process parameters and quality indicators through conditional probability distribution, quantifying the influence degree of each process parameter on the final quality. Through the sensitivity analysis method, the sensitivity coefficient of each process parameter to the quality indicator is calculated to form a process parameter sensitivity matrix. The sensitivity matrix intuitively shows the importance ranking of each process parameter, indicating the key direction of optimization. According to the process parameter sensitivity matrix, an optimization objective function for process parameters is constructed, and an evolutionary algorithm is used for multi-objective optimization calculation. The optimization objective function comprehensively considers multiple dimensions such as quality improvement, production efficiency, and resource consumption, forming a multi-objective optimization problem. The non-dominated sorting genetic algorithm (NSGA-II) or multi-objective particle swarm optimization algorithm is used for optimization calculation to search for the optimal solution set among a large number of parameter combinations. Through Pareto front analysis, an optimized process parameter set that balances each objective is selected to provide an optimization plan for actual production. A small-batch verification experiment is carried out on the optimized process parameter set to evaluate the effect of parameter optimization. Under controlled conditions, small-scale production tests are carried out using the optimized parameter set, and assembly quality and efficiency data are collected. Through comparative experiments, the actual effect of the optimized parameters is verified, and the quality improvement rate and efficiency improvement rate are calculated. The formula for the quality improvement rate is: , where represents the quality improvement rate, represents the quality compliance index under the new parameters, represents the quality compliance index under the original parameters. The calculation formula for the efficiency improvement rate is: , where, represents the efficiency improvement rate, represents the production efficiency under the new parameters and represents the production efficiency under the original parameters. According to the verification results, the process knowledge base is incrementally updated, and effective process optimization experiences and knowledge are precipitated into the knowledge base to form an adaptive process optimization closed-loop, realizing the continuous accumulation of process knowledge and the continuous improvement of the optimization ability. Combining the comprehensive assembly quality data, the effects of process parameter optimization, and the production efficiency indicators, a multi-level evaluation system is constructed to generate the comprehensive evaluation results of wire harness assembly. The evaluation system comprehensively evaluates from multiple dimensions such as product quality, production efficiency, resource consumption, and reliability. The weights of each indicator are determined by the Analytic Hierarchy Process (AHP), and the comprehensive evaluation score is calculated. The evaluation results are fed back to the assembly system in the form of visual charts to provide a basis for management decision-making and process improvement, realizing the continuous improvement and optimization of the assembly system.

[0028] In this embodiment, through multi-modal feature acquisition and analysis of wire harness terminals, a self-updating terminal feature knowledge base is constructed, realizing the comprehensive coding and characterization of terminal geometric and material features, and improving the accuracy and comprehensiveness of feature recognition. Using the multi-layer graph structure fusion technology, the construction of the terminal feature association map is realized, enhancing the richness and relevance of feature expression. Through the intelligent classification model and assembly requirement analysis, the accurate classification and optimal matching of terminals are realized, improving the accuracy and efficiency of assembly. By using topological analysis and multi-objective optimization algorithms, the intelligent planning and optimization of the assembly path are realized, reducing assembly conflicts and resource waste. Based on the multi-agent reinforcement learning method, the collaborative assembly planning of multiple robots is realized, improving the overall cooperation efficiency and robustness of the system. Through the real-time force feedback adaptive control technology, the precise positioning and compliant control during the terminal assembly process are realized. Using the multi-sensor fusion monitoring technology, the full-range perception of the assembly process and the real-time identification of abnormal states are realized, improving the comprehensiveness and timeliness of quality monitoring. By using the deep anomaly detection algorithm and the anomaly type knowledge graph, the root causes and propagation paths of assembly anomalies are deeply explored, enhancing the accuracy of fault diagnosis. Through the Bayesian network and evolutionary algorithm, the sensitivity analysis and self-optimization of process parameters are realized, forming an adaptive process optimization closed-loop, and continuously improving the assembly quality and efficiency. By constructing a multi-level evaluation system, a complete comprehensive evaluation mechanism for wire harness assembly is formed, providing strong support for the continuous improvement of the assembly system, and contributing to the intelligent, high-quality, and high-efficiency management of the wire harness terminal assembly process.

[0029] Embodiment 2

[0030] Please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description in Embodiment 1. A smart assembly platform for wire harness terminals is provided, including: Acquisition module: used to collect and analyze multi-modal features of wire harness terminals, and construct a terminal feature knowledge base; Classification module: based on the terminal feature knowledge base, perform intelligent classification and matching of terminals, and generate an assembly association network; Optimization module: according to the assembly association network, optimize the assembly path and plan multi-robot collaboration to obtain a dynamic assembly strategy; Control module: based on the dynamic assembly strategy, perform real-time force feedback adaptive control on the assembly process of wire harness terminals; Evaluation module: perform multi-sensor fusion monitoring and abnormal state recognition on the assembly process, and construct an assembly quality evaluation model; Feedback module: based on the assembly quality evaluation model, verify the assembly result and self-optimize the process parameters to obtain a comprehensive evaluation result of wire harness assembly.

[0031] This embodiment also provides a smart assembly device for wire harness terminals. The smart assembly device for wire harness terminals includes a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the processor executes the steps of the smart assembly method for wire harness terminals in the above embodiments.

[0032] This embodiment also provides a computer-readable storage medium. The computer-readable storage medium can be a non-volatile computer-readable storage medium, or can also be a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer executes the steps of the smart assembly method for wire harness terminals.

[0033] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to get a formula closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0034] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An intelligent assembly method for wire harness terminals, characterized in that, The method includes: Performing multi-modal feature acquisition and analysis on the wire harness terminals to construct a terminal feature knowledge base; Based on the terminal feature knowledge base, performing intelligent classification and matching of terminals to generate an assembly association network; According to the assembly association network, performing assembly path optimization and multi-robot collaborative planning to obtain a dynamic assembly strategy; Based on the dynamic assembly strategy, performing real-time force feedback adaptive control on the assembly process of the wire harness terminals; Performing multi-sensor fusion monitoring and abnormal state recognition on the assembly process to construct an assembly quality evaluation model; Based on the assembly quality evaluation model, performing assembly result verification and process parameter self-optimization to obtain a comprehensive evaluation result of the wire harness assembly.

2. The intelligent assembly method of the wire harness terminal according to claim 1, wherein The performing multi-modal feature acquisition and analysis on the wire harness terminals to construct a terminal feature knowledge base includes: Using a multi-spectral imaging system to scan the surface of the terminals, obtaining multi-dimensional feature data of the terminal surface, and performing image enhancement processing on the multi-dimensional feature data to obtain an enhanced feature map; Performing deep learning feature extraction on the enhanced feature map, obtaining geometric feature vectors and material feature vectors of the terminals, and performing dimensionality reduction processing on the geometric feature vectors and material feature vectors of the terminals to obtain a feature subspace; Using 3D laser scanning to measure the shape of the terminals, obtaining accurate 3D point cloud data of the terminals, and performing registration and denoising on the 3D point cloud data to obtain a digital model of the terminals; Performing topological analysis on the digital model of the terminals, identifying key connection surfaces and contact areas, and extracting accurate dimensional parameters to obtain a set of terminal parameters; Performing multi-layer graph structure fusion on the feature subspace and the set of terminal parameters to construct a terminal feature association graph, and establishing a self-updating terminal feature knowledge base based on the terminal feature association graph.

3. The intelligent assembly method of the wire harness terminal according to claim 1, wherein The based on the terminal feature knowledge base, performing intelligent classification and matching of terminals to generate an assembly association network includes: Performing clustering analysis on the terminal feature knowledge base to obtain terminal category groups, and constructing a terminal classification decision tree based on the terminal category groups to obtain an intelligent classification model; Using the intelligent classification model to perform real-time identification and classification of the terminals to be assembled, obtaining terminal category labels, and performing analysis of the assembly requirements of the terminals according to the terminal category labels to obtain an assembly requirement matrix; Performing topological analysis on the wire harness structure, identifying terminal connection nodes and line directions to obtain a wire harness topology graph, and performing electrical connection constraint analysis based on the wire harness topology graph to obtain connection constraint conditions; According to the assembly requirement matrix and the connection constraint conditions, performing optimal terminal pairing calculation to obtain an optimal terminal pairing scheme; Performing graph structure modeling on the optimal terminal pairing scheme, constructing multi-dimensional relationships between terminal nodes, and generating an assembly association network.

4. The intelligent assembly method of the wire harness terminal according to claim 1, wherein, The according to the assembly association network, performing assembly path optimization and multi-robot collaborative planning to obtain a dynamic assembly strategy includes: Performing topological sorting on the assembly association network to obtain an initial assembly sequence, and performing assembly timing simulation according to the initial assembly sequence to identify timing conflict points to obtain a conflict constraint set; Construct an assembly path optimization model based on the conflict constraint set, and use a heuristic algorithm to perform path search optimization to obtain a multi-objective optimized assembly path; Decompose the multi-objective optimized assembly path to obtain a set of subtasks, and perform assembly resource allocation according to the set of subtasks to obtain an initial collaborative assembly plan; Use a multi-agent reinforcement learning method to perform simulation training on the initial collaborative assembly plan to obtain a robot collaborative policy network, and perform collaborative behavior optimization based on the robot collaborative policy network to obtain an optimized collaborative plan; Conduct a robustness analysis on the optimized collaborative plan, establish an abnormal response strategy library, and combine the abnormal response strategy library and the optimized collaborative plan to form a dynamic assembly strategy.

5. The intelligent assembly method of the wire harness terminal according to claim 1, wherein, The real-time force feedback adaptive control of the assembly process of the wire harness terminals based on the dynamic assembly strategy includes: Install a micro force sensor array on the end effector of the robot, collect multi-dimensional force feedback data during the assembly process in real time, and filter the multi-dimensional force feedback data to obtain force signal features; Construct a mechanical state observer based on the force signal features, estimate the terminal contact state in real time, and perform assembly stage identification according to the terminal contact state to obtain the current assembly stage identifier; According to the current assembly stage identifier, automatically adjust the force control parameters, construct a stage-based adaptive impedance controller, and perform fine-tuning of the robot trajectory based on the stage-based adaptive impedance controller; Use visual servo technology to track the terminal position in real time, obtain the terminal pose error, and perform multi-modal fusion of the terminal pose error and the force feedback data to obtain a comprehensive error vector; Perform adaptive compensation control based on the comprehensive error vector to dynamically adjust the pose of the end effector of the robot.

6. The intelligent assembly method of the wire harness terminal according to claim 1, wherein, The multi-sensor fusion monitoring and abnormal state identification of the assembly process, and the construction of an assembly quality evaluation model include: Deploy acoustic, visual, force-tactile and electrical property sensors, construct an omni-directional perception network, and synchronously collect and preprocess the data of each sensor to obtain a multi-modal perception data stream; Perform feature extraction and time-frequency domain analysis on the multi-modal perception data stream to obtain an assembly process feature spectrum, and establish a normal assembly mode library based on the assembly process feature spectrum; Use a deep anomaly detection algorithm to compare the real-time assembly process with the normal assembly mode library, identify potential abnormal modes, and obtain an abnormal feature vector; Perform clustering analysis on the abnormal feature vector, construct an abnormal type knowledge graph, and perform abnormal cause reasoning based on the abnormal type knowledge graph to obtain an abnormal diagnosis result; Perform correlation analysis on the assembly process data and the abnormal diagnosis result with the assembly quality standard to establish an assembly quality evaluation model.

7. The intelligent assembly method of the wire harness terminal according to claim 1, wherein The verification of the assembly result and the self-optimization of the process parameters based on the assembly quality evaluation model to obtain a comprehensive evaluation result of the wire harness assembly includes: Perform electrical performance testing and mechanical strength detection on the assembled terminals, obtain the measurement data of the assembly result, and compare the measurement data of the assembly result with the standard requirements to obtain a quality compliance index; Based on the quality compliance index and assembly process data, a Bayesian network is used for modeling and analyzing the factors affecting assembly quality to obtain a process parameter sensitivity matrix; According to the process parameter sensitivity matrix, an optimization objective function for process parameters is constructed, and an evolutionary algorithm is used for multi-objective optimization calculation to obtain an optimized process parameter set; A small-batch verification experiment is carried out on the optimized process parameter set to evaluate the parameter optimization effect, and the process knowledge base is incrementally updated according to the verification results; Based on the assembly quality data, the process parameter optimization effect, and the production efficiency index, a multi-level evaluation system is constructed to generate a comprehensive evaluation result of the wire harness assembly, and the evaluation result is fed back to the assembly system for continuous improvement.

8. A smart assembly platform for wire harness terminals, which is used to execute the smart assembly method of wire harness terminals described in any one of claims 1-7, characterized in that It includes: A collection module: used for multi-modal feature collection and analysis of wire harness terminals to construct a terminal feature knowledge base; A classification module: based on the terminal feature knowledge base, intelligent classification and matching of terminals are carried out to generate an assembly association network; An optimization module: according to the assembly association network, assembly path optimization and multi-robot collaborative planning are carried out to obtain a dynamic assembly strategy; A control module: based on the dynamic assembly strategy, real-time force feedback adaptive control is carried out to achieve accurate assembly positioning; An evaluation module: multi-sensor fusion monitoring and abnormal state recognition of the assembly process are carried out to construct an assembly quality evaluation model; A feedback module: based on the assembly quality evaluation model, assembly result verification and process parameter self-optimization are carried out to obtain a comprehensive evaluation result of the wire harness assembly.

9. An intelligent assembly device for wire harness terminals, characterized in that, The wire harness terminal intelligent assembly device includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the wire harness terminal intelligent assembly device executes the wire harness terminal intelligent assembly method according to any one of claims 1-7.

10. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by the processor, the wire harness terminal intelligent assembly method according to any one of claims 1-7 is realized.

Images