Cutting method for automobile wire harness processing

Abstract

The invention relates to the technical field of automobile wire harness processing, and discloses a cutting method for automobile wire harness processing. The method comprises the following steps: acquiring geometric feature data and material attribute data of a to-be-processed wire harness; constructing a dynamic cutting parameter mapping model based on the geometric feature data; generating a deformation compensation amount according to the material attribute data and inputting the deformation compensation amount into the model; cutting equipment is driven to execute cutting operation based on model output; physical response data in the cutting process are collected in real time, and whether a preset iteration condition is met or not is judged; if yes, a final cutting result is output, and if not, the model is updated, and related steps are repeated. According to the method, the dynamic parameter model is constructed, deformation compensation is introduced, cutting parameters are iteratively optimized in combination with real-time feedback, the defects that traditional fixed parameter cutting is poor in adaptability and cutting quality problems are prone to occurring are overcome, the cutting quality stability and the machining efficiency are improved, dependence on artificial experience is reduced, and the method is suitable for large-scale popularization and application. And the processing requirements of diversified automobile wire harnesses are met.

Description

Technical Field

[0001] This invention relates to the field of automotive wiring harness processing technology, specifically a cutting method for automotive wiring harness processing. Background Technology

[0002] In the automotive manufacturing industry, automotive wiring harnesses serve as a crucial carrier connecting various electronic components, and their processing precision directly impacts the overall operational performance and safety of the vehicle. The cutting process, a vital component of the automotive wiring harness manufacturing process, significantly influences subsequent assembly and connection stability. With the continuous development of automotive technology, the structure of automotive wiring harnesses has become increasingly complex. Not only are there diverse differences in the diameter and number of wires, but the materials used are also becoming more varied, ranging from traditional copper core wires to alloy wires with special properties. Some wiring harnesses are also wrapped with insulation layers of different materials, all of which place higher demands on the cutting process.

[0003] Most common automotive wiring harness cutting methods on the market employ fixed parameters. Before cutting, operators typically set parameters such as cutting speed and cutting force based on experience or simple specification manuals, and then drive the cutting equipment to uniformly cut batches of wiring harnesses. However, this method has significant limitations. Due to differences in the geometric characteristics of different wiring harnesses—for example, some harnesses have a larger diameter and more internal conductors, while others have a smaller diameter and relatively simple structure—fixed parameter cutting can easily lead to incomplete cutting for harnesses with unique geometric characteristics, meaning some conductors are not severed and require secondary processing, increasing processing steps and time costs. Conversely, for simple wiring harnesses, excessively high cutting parameters may cause deformation or damage to the harness ends, affecting subsequent connection results.

[0004] Different wire harness materials exhibit significantly different deformation characteristics during cutting. For example, copper core wires have good ductility, but excessive force during cutting can easily cause end deformation. Alloy wires, on the other hand, have higher hardness, requiring greater cutting force. Improper parameter settings can lead to low cutting efficiency and excessive wear on the cutting tool, shortening its lifespan. Existing cutting methods often fail to adequately consider the deformation caused by the wire harness material properties and lack targeted compensation measures. This makes it difficult to guarantee the flatness of the cut wire harness ends, and some wire harnesses even suffer from insulation damage and exposed wires, posing safety hazards for the subsequent use of automotive wiring harnesses.

[0005] Existing cutting methods lack an effective real-time feedback and adjustment mechanism during the cutting process. After the cutting equipment is started, it usually runs continuously according to the preset parameters. Even when situations such as wire harness position deviation and material property fluctuations occur during the cutting process, it is impossible to detect and adjust the cutting parameters in a timely manner. This "one-size-fits-all" mode results in poor stability of cutting quality. For wire harnesses in the same batch, some may be cut qualified while some are unqualified, increasing the defective rate of products. At the same time, for unqualified cutting products, screening, rework or scrapping is required, which not only wastes raw materials but also reduces the overall production efficiency, making it difficult to meet the current requirements of the automotive manufacturing industry for high-precision, high-efficiency and high-stability wire harness processing. Summary of the Invention

[0006] The purpose of the present invention is to provide a cutting method for automotive wire harness processing to solve the problems raised in the above background technology.

[0007] To achieve the above purpose, the present invention provides a cutting method for automotive wire harness processing, and the method includes: Step S1, obtaining geometric feature data and material property data of the wire harness to be processed; Step S2, constructing a dynamic cutting parameter mapping model based on the geometric feature data; Step S3, generating a deformation compensation amount according to the material property data and inputting it into the dynamic cutting parameter mapping model; Step S4, driving the cutting equipment to perform wire harness cutting operations based on the output of the dynamic cutting parameter mapping model; Step S5, real-time collecting physical response data during the cutting process and judging whether it meets the preset iteration conditions; Step S6, if the preset iteration conditions are met, outputting the final cutting result; if not, updating the dynamic cutting parameter mapping model and repeating steps S3 to S5.

[0008] Preferably, the geometric feature data includes the shape distribution vector of the wire harness cross-section, the curvature change sequence in the wire harness length direction, and the thickness gradient value of the wire harness outer sheath.

[0009] Preferably, the dynamic cutting parameter mapping model includes a feature encoding layer, a parameterized operation layer, and an action decoding layer; The feature encoding layer decomposes the geometric feature data into basis function weight coefficients; The parameterized operation layer performs a non-linear mapping on the basis function weight coefficients according to the deformation compensation amount; The action decoding layer reconstructs the mapped basis function weight coefficients into a cutting speed control sequence and a cutting pressure control sequence.

[0010] Preferably, the deformation compensation amount is generated through the following steps: Extract the elastic modulus distribution and thermal expansion coefficient matrix from the material property data; Based on the temperature fluctuation value of the wire harness processing environment, the coupling offset between the elastic modulus distribution and the thermal expansion coefficient matrix is ​​calculated; Based on the coupling offset, axial deformation compensation components and radial deformation compensation components are generated.

[0011] Preferably, the preset iteration conditions include the cumulative number of cutting operations reaching a maximum threshold, the absolute change in physical response data generated by two adjacent cutting operations being less than a convergence threshold, or the relative change rate of physical response data generated by two adjacent cutting operations being less than a convergence threshold.

[0012] Preferably, the physical response data includes the flatness index of the wire harness cut morphology, the distribution density value of the burrs on the wire harness end face, and the offset distance of the conductor inside the wire harness.

[0013] Preferably, updating the dynamic cutting parameter mapping model includes: Calculate the deviation between the objective function value and the constraint function value of the current cutting result; Based on the deviation, the gradient projection algorithm is used to adjust the internal weight matrix of the dynamic cutting parameter mapping model.

[0014] Preferably, the constraint function values ​​include the maximum acceleration threshold of the cutting equipment, the wear safety boundary of the cutting tool, and the minimum yield strength of the wire harness material.

[0015] Preferably, step S4 further includes: Based on the spatial layout topology of the wire harness processing equipment, the cutting speed control sequence and the cutting pressure control sequence are assigned to the main processing node and the auxiliary processing node; The main processing node and the auxiliary processing node synchronously execute the cutting action based on a real-time data exchange protocol.

[0016] Preferably, the real-time data exchange protocol includes: When the instruction transmission delay of the main processing node exceeds the latency threshold, the local cached instruction sequence of the auxiliary processing node is activated; Based on the feedback value of wire harness processing progress, the instruction distribution frequency of the main processing node and the auxiliary processing node is dynamically adjusted.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This cutting method for automotive wire harness processing, starting from the actual needs of wire harness manufacturing, provides a more scientific and efficient solution for the wire harness cutting process through multi-stage optimization design. Before cutting, the geometric feature data and material property data of the wire harness to be processed are obtained, so that subsequent cutting parameter settings no longer rely on experience but are based on a precise understanding of the specific situation of the harness. A dynamic cutting parameter mapping model is constructed based on the geometric feature data, enabling a correspondence between cutting parameters and the geometric characteristics of the wire harness. Wire harnesses of different diameters and structures can obtain suitable initial cutting parameters, avoiding the problems of incomplete cutting or harness damage caused by the mismatch between parameters and the geometric characteristics of the wire harness in traditional fixed-parameter cutting, thus improving the rationality of cutting parameter settings from the source.

[0018] Based on geometric features, this method also generates deformation compensation based on material property data and inputs it into a dynamic cutting parameter mapping model. This design fully considers the deformation differences of wire harnesses made of different materials during the cutting process. By introducing deformation compensation, the cutting parameters can be adjusted in a targeted manner. For example, for highly ductile and easily deformable materials, the cutting force and speed can be adjusted by the compensation to reduce the extrusion deformation of the wire harness during cutting. For materials with high hardness, the cutting parameters can be optimized by the compensation to ensure cutting efficiency while reducing tool wear. This effectively solves the cutting quality problems caused by neglecting material properties in traditional cutting methods, improves the flatness of the wire harness ends after cutting, and reduces the occurrence of defects such as insulation layer damage and exposed wires.

[0019] This method does not operate statically according to initial parameters during the cutting process. Instead, it collects physical response data in real time and determines whether preset iteration conditions are met. This real-time feedback mechanism can promptly capture anomalies during the cutting process, such as changes in cutting force caused by wire harness positional shifts or differences in cutting response caused by fluctuations in local material properties. When the preset iteration conditions are not met, the dynamic cutting parameter mapping model is updated promptly, and the relevant steps are repeated, achieving dynamic adjustment of cutting parameters. This iterative optimization mode breaks the limitations of the traditional "one-cut" cutting method, allowing the cutting process to flexibly adapt to actual conditions, effectively improving the stability of cutting quality, reducing the number of defective products in the same batch of wire harnesses, and lowering the costs of subsequent screening and rework. Through precise parameter setting and dynamic adjustment, this method can optimize cutting efficiency while ensuring cutting quality. It avoids secondary processing caused by improper parameters, reducing ineffective processing time; reasonable parameter settings reduce excessive tool wear, extend tool life, reduce production interruptions caused by tool replacement, and indirectly improve overall production efficiency. Furthermore, this method requires no complex manual intervention. From data acquisition to parameter adjustment and cutting execution, it forms a relatively complete automated process, reducing reliance on operator experience, minimizing the impact of human error, improving the consistency and reliability of wire harness processing, and better adapting to the large-scale, high-precision production needs of the modern automotive manufacturing industry. Attached Figure Description

[0020] Figure 1 This is a timing diagram of the cutting method for automotive wiring harness processing according to the present invention; Figure 2 A flowchart illustrating the geometric feature data of a cutting method used in automotive wiring harness manufacturing; Figure 3 A flowchart for generating deformation compensation amount for cutting methods used in automotive wiring harness processing; Figure 4 A flowchart for acquiring physical response data of a cutting method used in automotive wiring harness processing. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 The present invention provides a cutting method for automotive wiring harness processing, the method comprising:

[0023] Obtain the geometric feature data and material property data of the wire harness to be processed. The geometric feature data includes the shape distribution vector of the wire harness cross-section, the curvature change sequence, and the outer sheath thickness gradient value; the material property data includes the elastic modulus distribution and the thermal expansion coefficient matrix. A dynamic cutting parameter mapping model is constructed based on the geometric feature data. This model consists of a feature encoding layer, a parameterized computation layer, and an action decoding layer, used to convert the input data into a cutting control sequence. Deformation compensation is generated based on the material property data and input into the dynamic cutting parameter mapping model to correct the cutting parameters. The cutting equipment executes the cutting operation according to the control sequence output by the model, while simultaneously acquiring physical response data in real time, including the cut morphology smoothness, burr distribution density, and conductor offset distance. If the iteration conditions are met (e.g., the number of cuts reaches a threshold or the change in physical response data is less than a convergence threshold), the final result is output; otherwise, the model parameters are updated and the adjustment and cutting process is repeated.

[0024] Example 1: See Figure 2 The acquisition of geometric feature data is achieved through a specific combination of equipment. A 3D laser scanner performs a circumferential scan along the axis of the wire harness to be processed at 0.5mm intervals to acquire point cloud data of the cross-sectional contour. After preprocessing, the point cloud of each cross-section is used to generate a closed polygon using a contour fitting algorithm. This polygon is divided into 36 sectors at equal angles, and the radial distance from the contour point of each sector to the center of the cross-section is calculated to form a 36-dimensional vector as the basis for the shape distribution vector. The amplitude of the first 12 harmonic components is further extracted through discrete Fourier transform, and finally compressed into a 12-dimensional shape distribution vector. The measurement of the curvature change sequence is completed by a contact probe arranged on the feeding track. The probe records the spatial coordinates of the wire harness centerline at a sampling frequency of 10Hz, and calculates the curvature value at 5mm intervals using the three-point difference method to form a curvature change sequence along the length direction. The thickness gradient of the outer layer was obtained using a non-contact ultrasonic thickness gauge. Eight circumferential measurement points were selected within a 10cm length range. After recording the thickness value at each point, a thickness distribution field was constructed using a bilinear interpolation algorithm. Finally, the rate of change of thickness along the circumferential and axial directions was extracted as the gradient value.

[0025] The feature encoding layer receives three types of geometric feature data: a 12-dimensional shape distribution vector, a curvature variation sequence (200 sampling points along the length direction), and an 8×8 thickness gradient matrix. First, the curvature sequence is decomposed using wavelet packet decomposition, employing the db4 wavelet basis function for a 5-layer decomposition, extracting the level 3 detail coefficients as curvature feature components. The thickness gradient matrix is ​​transformed to the frequency domain using a 2D discrete cosine transform, retaining the first 16 low-frequency coefficients. The shape distribution vector is directly used as the input component. These three feature components are input in parallel to a fully connected network containing three hidden layers (64, 32, and 16 neurons respectively), which outputs 32-dimensional basis function weight coefficients through the ReLU activation function.

[0026] The parameterized computation layer consists of a nonlinear mapping module. This module receives 32-dimensional basis function weight coefficients from the feature encoding layer and a 4-dimensional compensation vector (containing 2-dimensional axial and 2-dimensional radial components) from the deformation compensation module. The compensation vector is extended to 32 dimensions through concatenation and then subjected to a Hadamard product with the original weight coefficients. The result is input to a gated recurrent unit containing 128 hidden states. The features are nonlinearly reorganized using the tanh activation function, and the updated 32-dimensional weight coefficients are output.

[0027] The updated 32-dimensional weight coefficients are split into three sets of sub-vectors: the first set of 12-dimensional vectors is reconstructed into cross-sectional shape control parameters through inverse Fourier transform; the second set of 8-dimensional vectors is reconstructed into curvature compensation parameters through inverse wavelet transform; and the third set of 12-dimensional vectors is reconstructed into thickness control parameters through inverse discrete cosine transform. These three sets of parameters are input to the motion planner, which generates the toolpath trajectory based on the cross-sectional shape control parameters, adjusts the feed rate curve based on the curvature compensation parameters, and calculates the pressure distribution based on the thickness control parameters. The final output is a discretized cutting speed control sequence (time-speed value pairs, sampling interval 10ms) and a cutting pressure control sequence (spatial position-pressure value pairs, resolution 1mm).

[0028] During execution, the wire harness is fixed to a six-degree-of-freedom positioning platform. A 3D laser scanner is installed at the feeding inlet, simultaneously acquiring geometric data during feeding. The thickness gauge employs a rotating scanning head design, completing thickness measurement during feeding pauses. The calculation of the feature encoding layer is performed on an industrial control computer, with a processing cycle controlled within 50ms. The control sequence output from the motion decoding layer is transmitted to the cutting execution unit via real-time Ethernet. The execution unit includes a servo motor-driven rotating cutter head and a pneumatic pressure regulation system, achieving precise synchronous control of speed and pressure.

[0029] Example 2: See Figure 3 The process of generating deformation compensation begins with the accurate acquisition of material property data. In the pre-processing stage, sample segments of the wire harness to be processed are placed in a constant temperature chamber and subjected to segmental tensile testing using an electronic universal testing machine. The test environment temperature varies in 10℃ increments within the range of -10℃ to 50℃, and elastic modulus values ​​at three axial positions are collected at each temperature point. The elastic modulus distribution data is stored in the form of a three-dimensional tensor; the first dimension represents the temperature nodes, the second dimension represents the axial position coordinates, and the third dimension records the measured modulus value. The thermal expansion coefficient matrix is ​​measured using a dynamic thermomechanical analyzer. The wire harness sample is fixed in a quartz fixture and heated and cooled at a rate of 2℃ / min within the same temperature range, simultaneously recording the radial and axial expansion coefficient change curves, ultimately forming a three-dimensional coefficient matrix of temperature-position-direction.

[0030] Eight high-precision temperature sensors are evenly distributed around the cutting station, located in the equipment frame, feeding track, pneumatic circuit, and tool cooling area, respectively. The sensors collect ambient temperature data at a frequency of 10Hz, and the data from each node is fused using a Kalman filter algorithm to output the spatiotemporal distribution of the current temperature fluctuation value in real time. This distribution data is then transformed into dynamic boundary conditions for the finite element model, including changes in the air convection coefficient in the processing area, the thermal conductivity gradient of the equipment's metal components, and the radiation exchange parameters of the wire harness surface.

[0031] A refined 3D model of the wire harness to be cut is established. Each segment of the harness is discretized into multiple layers of shell elements: the innermost conductor layer is rigidly constrained, the insulation layer is set as an orthotropic material, and the outermost layer is defined as a viscoelastic body. The elastic modulus distribution tensor is mapped to the element attribute field of the finite element model, and the thermal expansion coefficient matrix is ​​input as the temperature load function. The solver performs a step-by-step coupled calculation: first, the temperature field transmission process in the time dimension is calculated, and then the temperature field result is used as the initial condition input to the stress field module. After solving, the displacement deviation of each element node is extracted, and after Gaussian filtering and noise reduction, a coupled offset dataset is formed.

[0032] One hundred calculation sections are divided along the length of the wire harness, and the axial displacement deviation of all elements within each section is averaged. A continuous curve of axial offset is constructed using cubic spline interpolation. Combined with the time function of the wire harness feeding speed, the spatial curve is converted into a time-series compensation component. This component is output in real time by a digital signal processor as a two-dimensional array of timestamps and offsets, with a time resolution of 1 ms.

[0033] Within each cross-section, a polar coordinate system is established with the center of the wire harness as the origin. The unit displacement vector is decomposed into radial and tangential components. Data points with radial components greater than a threshold are selected for cluster analysis to identify weak areas in the outer layer. The thickness gradient data and radial displacement data are used to calculate the Hadamard product in the polar coordinate grid, and the result is input into a convolutional neural network for feature extraction. The network outputs two compensation parameters: a circumferential non-uniform shrinkage adjustment factor and a local pressure intensity coefficient. The former is used to correct the radius of the toolpath trajectory, and the latter is converted into the control voltage curve of the pneumatic servo valve.

[0034] The axial compensation component drives the phase compensator of the feeding servo motor, superimposing micro-displacement pulses onto the base feeding speed. The radial compensation component is converted into four independent control signals: the first signal controls the radial feed cylinder pressure of the rotary cutter, the second adjusts the circumferential clamping force of the auxiliary support roller, the third changes the local power density of the laser heater, and the fourth controls the wind speed distribution of the cooling air curtain. All compensation signals are synchronously transmitted to the execution terminal through a fiber optic ring network, with a transmission delay controlled within 0.5ms and a timing error with the cutting action not exceeding ±10μs.

[0035] During the process control cycle, the temperature monitoring system updates environmental parameters every 200ms. The finite element solver uses an incremental calculation method, reloading only the mesh of regions where the change exceeds 5% in each calculation, compressing the calculation time to within 80ms. The compensation output module has a dual buffering mechanism: the compensation parameters calculated in the previous cycle are executed in the current cycle, while the new generation of compensation data is updated in the background to avoid control flow interruption.

[0036] Example 3: See Figure 4 The preset iteration condition judgment mechanism is based on the real-time acquisition and analysis of multi-dimensional physical response data. The physical response data acquisition system consists of three independent detection modules, which simultaneously monitor the cut morphology, burr distribution, and conductor position, respectively. The cut morphology detection module uses a confocal laser microscope, with its optical probe installed at a 45-degree angle 5mm behind the cutting tool, and the scanning frequency is synchronously matched with the cutting speed. After each cut, the probe performs a layer scan along the Z-axis of the cut cross-section to acquire three-dimensional surface morphology data. After median filtering, five feature parameters are extracted from this data: the waviness amplitude of the cut edge, the peak-valley height difference of the cutting surface, the slope change rate of the transition region, the material accumulation volume, and the width of the heat-affected zone. The flatness index is calculated using a composite weighted method.

[0037] in: Represents wave amplitude. Indicates the difference in height between peaks and valleys. The rate of change of the slope It is the volume of the material stacked. The width of the heat-affected zone. to These are the normalized weighting coefficients for each parameter. These coefficients are dynamically adjusted based on the type of wire harness material. For example, for PVC-coated wire harnesses, the weight of the slope change rate is set to 0.35, while for silicone-insulated wire harnesses, the weight of the heat-affected zone width is increased to 0.28.

[0038] The burr distribution detection module is integrated into the unloading area of ​​the cutting station, comprising a high-speed industrial camera and a ring LED light source. The camera captures images of the wire harness end face at a rate of 200 frames per second. The image processing algorithm first extracts the edge contour using the Canny operator, and then uses morphological opening operations to separate adhered burrs. Each identified burr feature is quantified into three parameters: projected area, aspect ratio, and tip curvature radius. The distribution density value is defined as the number of standardized burrs per square millimeter. The standardization process considers burr size weights, with a counting coefficient of 1.5 for large burrs and 0.8 for small burrs. The change in burr distribution between two adjacent cuts is evaluated by calculating the cosine similarity of the feature vectors. When the similarity exceeds 0.98, it is considered to have reached a convergence state.

[0039] The X-ray tube operates in pulse mode, with each pulse lasting 50 μs, and the pulse interval synchronized with the cutting cycle. Projection data acquired by the detector array is used to reconstruct cross-sectional images using a filtered back-projection algorithm, achieving an image resolution of 5 μm. Conductor position identification employs an improved Hough transform algorithm. First, the elliptical features of the conductor region are extracted, and then the offset vector of each conductor center relative to a standard position is calculated. The statistical measure of the offset distance is the average L2 norm of all conductor offset vectors, and compensation is made for the rigid displacement component of the beam during detection.

[0040] The first level of judgment is based on the cumulative number of cutting operations. When the counter reaches a preset threshold (typically 8 times), the iteration process is forcibly terminated. The second level of judgment monitors the absolute change in physical response data. A convergence threshold of 0.02 mm is set for the flatness index, 1.2 burr density changes / mm², and 15 μm for the conductor offset distance. The third level of judgment focuses on the relative change rate, calculating the percentage difference between two adjacent measurements. The relative change rate thresholds are 1.8% for the flatness index, 2.5% for burr density, and 3% for conductor offset. These three judgment levels form an OR logical relationship; if any condition is met, the iteration terminates.

[0041] The laser microscope's scanning trigger signal lags behind the tool position signal by 200μs to ensure stable kerf capture. The industrial camera's exposure time is synchronized with the end-of-cut signal, with an additional 5ms mechanical vibration attenuation period. The X-ray pulse is triggered during the tool return phase to avoid electromagnetic interference generated during cutting. All detection data is input into the judgment module after being timestamped. The module uses a sliding window mechanism to handle timing deviations, with a window width set to 10ms.

[0042] The raw detection data retains complete waveforms and images, is compressed and stored in a high-speed cache, and records of the three most recent iterations are retained. Feature extraction results are stored in the form of a structured array, including timestamps, tool position coordinates, and various physical response parameters. The execution log of the judgment logic records in detail the intermediate variables and trigger states of each evaluation for subsequent process analysis. When the iteration terminates, the system automatically generates a process report containing all key parameters, and the report data format is compatible with the MES system interface standard.

[0043] The anomaly handling mechanism for physical response data comprises a three-level response: Level 1 addresses transient interference, such as detection noise or brief environmental disturbances, filtering outomas through digital filtering and majority voting algorithms. Level 2 handles persistent deviations; when three consecutive measurements exceed the expected fluctuation range, the sensor calibration process is triggered. Level 3 addresses systemic failures, such as X-ray tube aging or optical lens contamination, automatically switching to redundant detection channels and issuing alarms to prompt maintenance.

[0044] Example 4: The update process of the dynamic cutting parameter mapping model is implemented based on an optimization algorithm. The objective function value is calculated using a multi-index fusion method. The input parameters include the real-time acquired cut smoothness index, burr distribution density value, and conductor offset distance. The smoothness index comes from the scanning data of a laser microscope, the burr density is based on image analysis results, and the conductor offset distance is taken from X-ray reconstruction data. The objective function value is generated by a weighted summation formula, and the weight coefficients are dynamically configured according to the wire harness type: for multi-core shielded wire harnesses, the weight of the conductor offset distance is increased to 0.6; for single-core power wire harnesses, the weight of the burr density is set to 0.55. The calculation process is completed on a dedicated hardware acceleration card, and the processing latency is controlled within 2ms.

[0045] Equipment constraints are collected by vibration sensors and encoders mounted on the cutting equipment spindle, monitoring acceleration values ​​and positional deviations in real time. Tool constraints utilize a tool condition monitoring module, including a contact wear probe and an infrared temperature sensor. Material constraints rely on an online material testing unit to perform micro-tensile tests on wire harness sample segments during cutting intervals. Threshold settings for each constraint parameter are shown in Table 1.

[0046] Table 1: Constraint Function Threshold Setting Table.

[0047] Constraint Type Monitoring parameters Threshold setting Equipment constraints Maximum acceleration ≤5.0m / s² vibration amplitude ≤0.05mm (RMS) Tool constraints Edge wear width ≤0.12mm tool temperature ≤85℃ Material constraints Minimum yield strength ≥80% of the nominal value of the material Elongation at break ≥75% of the nominal value of the material The deviation calculation module receives the objective function value and constraint function value and performs standardization processing. The objective function value is normalized to the [0,1] interval, and the constraint function value is converted into a violation degree coefficient (0 indicates no violation, 1 indicates severe violation). The deviation is defined as the Euclidean distance between the normalized objective function value and the constraint violation coefficient. The calculation process is implemented using a fixed-point arithmetic unit, with an output precision of 0.001.

[0048] The gradient projection algorithm consists of four stages: The first stage calculates the partial derivatives of the objective function with respect to the model weight matrix, achieved through perturbation analysis: applying a small change of ±0.1% to each element of the weight matrix and observing the change in the objective function value. The second stage calculates the gradient direction of the constraint functions, sampling near the constraint boundaries using the finite difference method. The third stage performs projection operations, projecting the gradient of the objective function onto the constrained feasible region, eliminating components in the same direction as the constraint boundary normal vector. The fourth stage determines the update step size, dynamically adjusting it based on the current deviation and historical convergence trend, with the maximum step size limited to 0.5% of the weight matrix norm.

[0049] The weights of the feature encoding layer are updated directly using the gradient projection results; the weights of the parameterized computation layer are updated using a forgetting factor, which is set to 0.7; the weights of the action decoding layer are updated through orthogonalization to maintain the linear independence of the output space. The updated weight matrix is ​​then validated by singular value decomposition to remove ill-conditioned solutions with a condition number greater than 1000.

[0050] The current weight matrix is ​​stored in DDR4 memory block A, and the next-generation weight matrix is ​​updated and calculated in block B. Once the update is complete, a hardware switch enables seamless switching within milliseconds. An anomaly detection mechanism continuously monitors the numerical stability of the weight matrix; when a NaN value or numerical overflow is detected, it automatically rolls back to the previous valid version.

[0051] The tool wear threshold decreases linearly with the cumulative cutting length; for every 100 meters of additional cutting length, the threshold decreases by 0.01 mm. The material yield strength threshold is compensated for by ambient temperature; for every 10°C increase in temperature, the threshold decreases by 2% of the material's nominal value. The equipment acceleration threshold is automatically relaxed by 20% in high-speed cutting mode, but the vibration amplitude threshold is simultaneously tightened by 30%.

[0052] The model update verification process consists of three steps: First, a virtual cut is performed in the simulation environment to predict the physical response data; second, a short-stroke trial cut is performed with an actual cut length of 5cm; finally, the difference between the predicted and measured values ​​is compared, and a second update is triggered when the error of key parameters exceeds 5%. The verification data is input into the historical database for optimizing the algorithm's parameter self-tuning.

[0053] Example 5: The allocation and execution of the cutting control sequence are implemented based on a distributed device architecture. The main processing node adopts a high-precision five-axis linkage servo cutting machine, equipped with a rotary cutter head and a pressure feedback system. The auxiliary processing nodes include four sets of pneumatic support fixtures, each with an independent radial clamping force adjustment unit and an axial position servo mechanism. The spatial layout topology between the devices is established by measuring with a laser tracker, recording the position matrix of each node in the global coordinate system. This matrix includes the relative distance, azimuth angle, and connection path length between the main node and each auxiliary node, used to calculate the transmission priority of the control sequence.

[0054] The cutting speed control sequence and cutting pressure control sequence output by the dynamic cutting parameter mapping model are decomposed into tasks using an allocation algorithm. The speed control sequence is fully allocated to the main processing node, and this sequence includes the correspondence between timestamps and speed values, with a time resolution of 10 milliseconds. The pressure control sequence is weighted and allocated according to the spatial layout topology: first, the topological distance weight coefficient from each auxiliary node to the main node is calculated, with closer nodes having higher weights; then, the spatial location information of the pressure sequence is mapped to the action area of ​​the auxiliary nodes, generating a pressure-location subsequence specific to each node. The allocation results are encapsulated into an instruction data packet, containing the target node address, execution time window, and action parameters.

[0055] The master processing node acts as the time reference source, sending a synchronization clock signal every 100 milliseconds. Instruction data packets are broadcast to all nodes within 5 milliseconds after the rising edge of the clock signal. Data packet transmission uses a deterministic Ethernet protocol, with maximum jitter controlled within 50 microseconds. Upon receiving the instruction, each node synchronously executes the action at the start of the next clock cycle. Execution status feedback data is returned to the master node 20 milliseconds before the end of the cycle, forming a closed-loop control.

[0056] The first stage sets a hardware timestamp at the transport layer. When the master node's instruction transmission delay exceeds a 50-millisecond threshold, it triggers the auxiliary node's local cached instruction sequence. The cached sequence is taken from the previous valid cycle data, and the current cycle parameters are predicted using a linear extrapolation algorithm. The second stage establishes a delay prediction model, training an LSTM network based on historical transmission delay data to predict nodes that may exceed the limit in advance and proactively activate the cache. The third stage sets a safety fallback mode. When three consecutive cycles of transmission fail, the auxiliary node switches to a preset constant-voltage clamping mode and issues an alarm.

[0057] The progress monitoring system consists of an absolute encoder mounted on the feed roller, which measures the feed length of the wire harness in real time. The progress feedback value is calculated as the percentage deviation between the actual feed length and the planned length. The frequency adjustment algorithm inputs this deviation value into a proportional-integral controller and outputs a frequency adjustment coefficient. The base distribution frequency is 10 Hz, which can be dynamically increased to 1 kHz. When the progress lags behind by more than 5%, the frequency is increased in steps; when the progress is ahead, the frequency is decreased according to an inverse proportional function.

[0058] The coordinated operation of the main machining node and auxiliary nodes is achieved through kinematic compensation. When the main node performs a cutting action, it calculates the tool position coordinates in real time and broadcasts them to the auxiliary nodes. The auxiliary nodes calculate the following trajectory of their supporting fixtures based on the tool position, and the trajectory generation takes into account the real-time deformation compensation data of the wiring harness. The pressure adjustment unit of each fixture receives a position-pressure mapping table, increases the clamping force in advance in the area where the tool approaches, and reduces the pressure in the area where the tool moves away. The coordination error is monitored by a laser rangefinder to detect the relative position of the fixture and the tool. When the error exceeds 0.1 mm, online recalibration is triggered.

[0059] The data exchange network adopts a ring topology, with the master node acting as the network controller, connecting all auxiliary nodes via dual-channel fiber optic cables to form a closed loop. Data transmission employs a dual-backup mechanism: the master channel transmits real-time control commands, while the backup channel transmits status monitoring data. Network bandwidth allocation uses a time-division multiple access (TDMA) strategy, allocating 60% of the time slots to the master node and 10% to each auxiliary node. When a node is detected as offline, its allocated time slots are automatically redistributed to neighboring nodes to extend bandwidth.

[0060] The anomaly handling process comprises two levels of response: device-level and system-level. Device-level responses are executed by the local controllers of each node, automatically switching to safe torque mode when actuator over-limit or sensor malfunction is detected. System-level responses are coordinated by the master node; when critical nodes fail, topology weights are recalculated, and the pressure distribution scheme is dynamically adjusted. All anomaly events are logged with timestamps, and log data is stored synchronously with process parameters for subsequent diagnostic analysis.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cutting method for automotive wiring harness processing, characterized in that, Includes the following steps: Step S1: Obtain the geometric feature data and material property data of the wire harness to be processed; Step S2: Construct a dynamic cutting parameter mapping model based on the geometric feature data; Step S3: Generate deformation compensation amount based on the material property data and input it into the dynamic cutting parameter mapping model; Step S4: Based on the output of the dynamic cutting parameter mapping model, drive the cutting device to perform wire harness cutting operation; Step S5: Collect physical response data during the cutting process in real time and determine whether the preset iteration conditions are met; Step S6: If the preset iteration condition is met, the final cutting result is output; if not, the dynamic cutting parameter mapping model is updated and steps S3 to S5 are repeated. The dynamic cutting parameter mapping model includes a feature encoding layer, a parameterized operation layer, and an action decoding layer; The feature encoding layer decomposes the geometric feature data into basis function weight coefficients; The parameterized computation layer performs a nonlinear mapping on the basis function weight coefficients based on the deformation compensation amount; The action decoding layer reconstructs the mapped basis function weight coefficients into a cutting speed control sequence and a cutting pressure control sequence; The deformation compensation amount is generated through the following steps: Extract the elastic modulus distribution and thermal expansion coefficient matrix from the material property data; Based on the temperature fluctuation value of the wire harness processing environment, the coupling offset between the elastic modulus distribution and the thermal expansion coefficient matrix is ​​calculated; Based on the coupling offset, axial deformation compensation components and radial deformation compensation components are generated.

2. The cutting method for automotive wire harness processing according to claim 1, characterized in that, The geometric feature data includes the shape distribution vector of the wire harness cross section, the curvature change sequence along the wire harness length direction, and the thickness gradient value of the outer sheath of the wire harness.

3. The cutting method for automotive wire harness processing according to claim 1, characterized in that, The preset iteration conditions include the cumulative number of cutting operations reaching the maximum threshold, the absolute change in physical response data generated by two adjacent cutting operations being less than the convergence threshold, or the relative change rate of physical response data generated by two adjacent cutting operations being less than the convergence threshold.

4. The cutting method for automotive wire harness processing according to claim 3, characterized in that, The physical response data includes the flatness index of the wire harness cut morphology, the distribution density value of the burrs on the wire harness end face, and the offset distance of the conductor inside the wire harness.

5. The cutting method for automotive wiring harness processing according to claim 1, characterized in that, The updating of the dynamic cutting parameter mapping model includes: Calculate the deviation between the objective function value and the constraint function value of the current cutting result; Based on the deviation, the gradient projection algorithm is used to adjust the internal weight matrix of the dynamic cutting parameter mapping model.

6. The cutting method for automotive wiring harness processing according to claim 5, characterized in that, The constraint function values ​​include the maximum acceleration threshold of the cutting equipment, the wear safety boundary of the cutting tool, and the minimum yield strength of the wire harness material.

7. The cutting method for automotive wire harness processing according to claim 1, characterized in that, Step S4 also includes: Based on the spatial layout topology of the wire harness processing equipment, the cutting speed control sequence and the cutting pressure control sequence are assigned to the main processing node and the auxiliary processing node; The main processing node and the auxiliary processing node synchronously execute the cutting action based on a real-time data exchange protocol.

8. A cutting method for automotive wiring harness processing according to claim 7, characterized in that, The real-time data exchange protocol includes: When the instruction transmission delay of the main processing node exceeds the latency threshold, the local cached instruction sequence of the auxiliary processing node is activated; Based on the feedback value of wire harness processing progress, the instruction distribution frequency of the main processing node and the auxiliary processing node is dynamically adjusted.

Images