A high-power laser cutting optical path system, a laser cutting head, and its usage method

Abstract

This invention relates to the field of laser cutting technology, and discloses a high-power laser cutting optical path system, a laser cutting head, and a method for using the same. The system includes modules for dynamic cutting path calculation, multi-dimensional cutting difference analysis, abnormal region correlation and location, and adaptive cutting strategy generation. The dynamic cutting path calculation module builds a model based on historical laser cutting data and collects real-time data on laser power fluctuations, beam quality distribution, and cutting speed sequence. The multi-dimensional cutting difference analysis module receives this information, performs a three-dimensional difference analysis with the path measured by the optical sensor, and generates a workpiece-level difference coefficient matrix. The abnormal region correlation and location module inputs the matrix into a spatial geometric analysis network to generate an abnormal probability heat map of the cutting defect propagation path, locating abnormal cutting areas. The adaptive cutting strategy generation module configures parameters according to the heat map and applies multi-band impedance perturbations to the adjacent nodes of the cutting head with the largest changes in abnormal probability gradient.

Description

Technical Field

[0001] This invention relates to the field of laser cutting technology, specifically to a high-power laser cutting optical path system, a laser cutting head, and a method of using the same. Background Technology

[0002] In the field of high-power laser cutting, as industrial production demands increasing precision, efficiency, and material adaptability, traditional optical path systems are gradually revealing numerous limitations. Traditional systems often employ preset, fixed cutting paths, making it difficult to cope with the effects of laser power fluctuations, beam quality variations, and differences in material properties during the cutting process. For example, when laser power fluctuates instantaneously due to equipment operating conditions, the cutting depth and width under a fixed path will deviate. This deviation can lead to defects such as burrs and cracks in the workpiece, especially when processing high-hardness alloys or composite materials.

[0003] Existing methods for analyzing cutting differences are mostly limited to detecting positional deviations within a two-dimensional plane, neglecting the impact of uneven energy distribution along the depth direction. Comparisons between the measured and theoretical paths acquired by optical sensors often focus only on the offset of planar coordinates, lacking effective analysis of energy density differences caused by variations in beam focusing depth. This makes it difficult to detect some cutting defects hidden within the material in a timely manner.

[0004] Abnormal area location relies on manual experience or simple threshold judgments, failing to establish a correlation between cutting defects and material properties or cutting head position. When local defects appear on the workpiece, it is difficult to quickly trace the propagation path and cause of the defect, leading to increased rework rates and limited production efficiency. In terms of adaptive adjustment, traditional systems often modify parameters based on a single feedback signal, such as adjusting the cutting head speed only based on positional deviations. This lacks a comprehensive consideration of the coupled effects of multiple factors, making it difficult to achieve accurate adaptation to complex cutting scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a high-power laser cutting optical path system, a laser cutting head, and a method of using the same, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a high-power laser cutting optical path system, the system comprising:

[0007] The dynamic cutting path calculation module constructs a dynamic cutting path calculation model based on historical laser cutting operation data. It collects the actual laser power fluctuation parameters, beam quality distribution index, and cutting speed sequence data in real time during the current cutting process, and outputs theoretical cutting path information through the dynamic cutting path calculation model.

[0008] The multi-dimensional cutting difference analysis module receives the theoretical cutting path information and performs a three-dimensional difference analysis with the actual cutting path information measured by the optical sensor. The three-dimensional difference analysis includes the cumulative position deviation, depth energy offset, and velocity sequence similarity index, generating a workpiece-level difference coefficient matrix.

[0009] The abnormal region association and positioning module inputs the difference coefficient matrix into the spatial geometric analysis network, and combines the material property characteristic parameters and the cutting head position coordinate information to generate an abnormal probability thermal distribution map of the cutting defect propagation path, which is used to locate the abnormal cutting physical region.

[0010] The adaptive cutting strategy generation module configures cutting adjustment parameters based on the anomaly probability heat map, including applying multi-band impedance perturbation operations to the adjacent node cutting heads with the largest anomaly probability gradient change in the anomaly probability heat map.

[0011] Preferably, the dynamic cutting path calculation module specifically includes a historical cutting feature mining component and a dynamic cutting path calculation model construction component;

[0012] The historical cutting feature mining component processes historical laser cutting operation data, performs multi-dimensional decomposition operations, including extracting the energy ratio values ​​of steady-state and transient components of power fluctuation parameters, establishing a correlation matrix between beam quality distribution index and cutting load rate through beam coupling analysis, and adjusting the speed sequence mode under different working conditions using a dynamic sequence alignment algorithm.

[0013] The dynamic cutting path calculation model building component transmits the processed historical running data to the hybrid prediction network. The hybrid prediction network has a time-series prediction unit based on the cutting head aging curve to generate basic path prediction values, a fully connected network with embedded spatial attention mechanism to correct the prediction deviation caused by beam distortion, and a sequence feature compensator to dynamically adjust the prediction weight coefficients according to the real-time acquired velocity sequence. The heterogeneous data acquisition unit deployed through optical sensors synchronously captures the zero-crossing distortion rate and phase jitter characteristics of power fluctuation parameters, the fractional content and angular distribution range of beam quality distribution indicators, the amplitude abrupt change gradient and time interval entropy parameters of the velocity sequence. The theoretical path calculation component inputs the real-time acquired data into the dynamic cutting path calculation model to obtain theoretical cutting path information.

[0014] Preferably, the theoretical cutting path information calculation process includes adaptive noise cancellation processing based on the cutting stage to eliminate the measurement noise caused by environmental vibration interference; integrating the associated feature vectors of power fluctuation parameters, beam quality distribution index, and velocity sequence through a spatiotemporal feature fusion algorithm; and outputting theoretical cutting path information including the fluctuation range under normal operating conditions, which is dynamically updated with the aging state of the cutting head.

[0015] Preferably, the multidimensional cutting difference analysis module specifically includes a position cumulative deviation calculation unit, a depth energy shift detection unit, and a velocity sequence similarity evaluation unit;

[0016] The position cumulative deviation calculation unit performs a sliding comparison between the theoretical cutting path information and the measured cutting path information using a preset distance window, and uses a dynamic sequence alignment algorithm to align the asynchronously sampled path sequence data, calculates the cumulative deviation within each window, and generates a position deviation vector.

[0017] The depth energy offset detection unit performs frequency domain decomposition on the depth data of theoretical and measured cutting path information, calculates the energy spectral density ratio, extracts the energy offset index of each depth layer, and constructs the depth offset vector.

[0018] The velocity sequence similarity evaluation unit matches the velocity sequence distance distribution between theoretical cutting path information and measured cutting path information based on the structure matching algorithm, calculates the phase synchronization error value of velocity mutation points, quantifies the difference index of velocity interval distribution, and generates a velocity similarity vector. The difference coefficient matrix generation unit performs tensor concatenation operation on the position deviation vector, depth offset vector, and velocity similarity vector, and eliminates the influence of dimensional differences through feature importance weighted normalization processing, outputting a third-order difference coefficient matrix with dimensions of workpiece number multiplied by timestamp multiplied by difference type.

[0019] Preferably, in the frequency domain analysis stage, the depth energy offset detection unit first extracts the corresponding depth data from the theoretical cutting path information and the measured cutting path information, and uses the spectral decomposition method to perform multi-scale frequency band analysis on each group of depth signals to extract the energy distribution characteristic parameters within the preset sensitive frequency band interval. The sensitive frequency band interval covers the cutting frequency range of typical metal materials. After extraction, the energy density of the theoretical cutting path information and the measured cutting path information within the sensitive frequency band interval is quantified and calculated, and the offset index data of each order of depth layer is extracted based on the relative offset degree of the two. The energy offset results of all orders are summarized to construct a depth offset vector.

[0020] Preferably, the structure matching algorithm in the velocity sequence similarity evaluation unit adopts the edit distance algorithm to perform position matching operation on the set of mutation points in the two sequences and identify the phase synchronization error value.

[0021] Preferably, the abnormal region association and positioning module specifically includes a workpiece topology modeling unit, an abnormal propagation simulation unit, and a probability distribution generation unit;

[0022] The workpiece topology modeling unit constructs a topology diagram of the workpiece node connection relationship based on the position coordinate information of the cutting head, marks the material impedance parameters between each node, and superimposes the reverse heat conduction constraint conditions of the heat-affected zone access point on the topology diagram to generate a geometric topology model including the impedance matrix and the node admittance matrix.

[0023] The anomaly propagation simulation unit maps the difference coefficient matrix to the corresponding nodes of the workpiece topology model, performs anomaly propagation inference operation based on graph neural network, and the anomaly propagation inference calculation includes calculating the attenuation factor of abnormal heat flow based on node impedance parameters, capturing cross-regional anomaly correlation feature vector through multi-head attention mechanism, and simulating the diffusion path trajectory of abnormal heat flow in the topology network using Monte Carlo method.

[0024] The probability distribution generation unit counts the frequency of abnormal heat flow occurrences in simulated propagation for each path, calculates the abnormal heat flow residence probability value in combination with material impedance parameters, generates an abnormal probability thermal distribution map covering the entire workpiece, and marks the set of suspicious paths whose probability values ​​exceed the preset abnormal residence probability threshold; the physical region positioning unit performs spatial clustering analysis on the abnormal probability thermal distribution map, identifies abnormal probability clustering areas, and delineates the abnormal cutting physical boundary range based on the cutting head position coordinates and the topological connection relationship of the workpiece.

[0025] Preferably, the abnormal area association and positioning module further includes outputting suspicious device identification information and abnormal propagation main path information;

[0026] The suspicious device identification information is based on the workpiece nodes connected by the suspicious path set. The workpiece nodes are bound to the actual cutting head to form a suspicious device identification set, indicating the potential abnormal cutting source or affected terminal.

[0027] The anomaly propagation main path information is obtained by recording the node paths and their sequence sequence experienced in each round of propagation during the anomaly diffusion process in Monte Carlo simulation. The frequency of occurrence of each path is counted in all simulated paths, and the path sequence with the highest cumulative frequency is selected as the anomaly propagation main path. The output anomaly propagation main path sequence is a structured ordered node list, which reflects the main propagation trajectory of the anomaly information in the workpiece.

[0028] Preferably, the present invention further includes a high-power laser cutting head, suitable for the high-power laser cutting optical path system described above. The laser cutting head includes an integrated optical path transmission component, a real-time feedback component, and a dynamic adjustment component. The integrated optical path transmission component includes a collimating lens unit, a focusing lens group unit, and a beam-shaping galvanometer unit. The collimating lens unit receives the original beam output from the laser generator and performs collimation processing. The focusing lens group unit focuses the collimated beam onto the workpiece surface to form a cutting spot. The beam-shaping galvanometer unit dynamically adjusts the beam deflection angle according to the cutting adjustment parameters issued by the adaptive cutting strategy generation module. The real-time feedback component is connected to the multi-dimensional cutting difference analysis module and includes an embedded optical sensor array. It captures the actual position coordinates of the cutting spot, energy distribution parameters, and temperature gradient data of the heat-affected zone in real time, and feeds the captured data back to the dynamic cutting path calculation module. The dynamic adjustment component is linked with the abnormal area correlation and positioning module. Based on the set of suspicious paths output by the abnormal probability thermal distribution map, it automatically adjusts the focal length offset of the cutting head and the protective gas pressure value. The focal length offset is compensated for at the nanometer level by a piezoelectric ceramic actuator, and the protective gas pressure value is adjusted at the millisecond level by a proportional valve.

[0029] Preferably, the present invention also includes a method of using a high-power laser cutting optical path system, applied to the high-power laser cutting optical path system as described above, the method comprising the following steps:

[0030] Step S1: Load the material property feature parameters of the target workpiece and the preset cutting trajectory data through the dynamic cutting path calculation module to construct an initial dynamic cutting path calculation model;

[0031] Step S2: Activate the real-time feedback component to capture laser power fluctuation parameters, beam quality distribution index and cutting speed sequence data during the actual cutting process, and synchronously input the data into the multi-dimensional cutting difference analysis module;

[0032] Step S3: The multidimensional cutting difference analysis module performs a three-dimensional difference analysis operation, generates a workpiece-level difference coefficient matrix, and transmits it to the abnormal area association and positioning module;

[0033] Step S4: The abnormal area association and positioning module maps the difference coefficient matrix according to the topology map of the workpiece node connection relationship, and outputs the abnormal probability heat distribution map and suspicious equipment identification information;

[0034] Step S5: When the detected abnormal probability value exceeds the preset abnormal dwell probability threshold, the adaptive cutting strategy generation module sends a high-frequency laser monitoring command and a multi-band impedance disturbance command to the laser cutting head.

[0035] Step S6: The dynamic adjustment component of the laser cutting head responds to the command, performs focal length offset compensation operation and protective gas pressure gradient adjustment operation, and feeds back the adjusted data to the dynamic cutting path calculation module to update the theoretical cutting path information.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] The dynamic cutting path calculation module constructs a computational model based on historical operational data, and collects parameters such as laser power fluctuations, beam quality distribution, and cutting speed sequences in real time to generate a theoretical cutting path. This process eliminates the dependence on fixed paths and can dynamically adjust the path planning according to the real-time status during the cutting process, making the theoretical path more closely match the actual cutting needs and reducing path deviations caused by changes in environmental and equipment factors.

[0038] The multidimensional cutting difference analysis module introduces three-dimensional difference analysis, covering cumulative positional deviation, depth energy offset, and velocity sequence similarity index, generating a workpiece-level difference coefficient matrix. Compared to traditional two-dimensional analysis, this comprehensive comparison method can more fully capture the differences between theoretical and measured paths, focusing not only on surface position offsets but also reflecting changes in energy distribution and velocity matching in the depth direction, making the evaluation of the cutting process more three-dimensional and accurate.

[0039] The anomaly region correlation and localization module inputs the difference coefficient matrix into the spatial geometric analysis network, and combines it with material property characteristic parameters and cutting head position coordinate information to generate an anomaly probability heat map. This method breaks through the limitations of traditional methods that rely on manual or simple threshold judgments. By establishing correlations between data, it can clearly present the propagation path of cutting defects and possible anomaly regions, making localization more targeted and facilitating the rapid identification of the physical area where the problem occurs.

[0040] The adaptive cutting strategy generation module applies multi-band impedance perturbation to the adjacent cutting heads with the largest changes in anomaly probability gradients based on the anomaly probability heat map. This adjustment method is not a simple modification of a single parameter, but a precise operation based on comprehensive analysis of multiple factors. It can intervene in anomaly areas in a timely manner, and cope with complex changes in cutting state through multi-band impedance perturbation. This allows the cutting process to be corrected in a timely manner when deviation trends appear, thereby improving the stability and adaptability of the overall cutting process. Attached Figure Description

[0041] Figure 1 This is a schematic diagram illustrating the working principle of the high-power laser cutting optical path system described in this invention.

[0042] Figure 2 A flowchart for calculating theoretical cutting path information;

[0043] Figure 3 A flowchart for depth energy migration detection;

[0044] Figure 4 A flowchart for the correlation and localization of abnormal regions;

[0045] Figure 5 A flowchart illustrating the operation of a high-power laser cutting head. Detailed Implementation

[0046] 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.

[0047] Please see Figure 1 This invention provides a high-power laser cutting optical path system, the system comprising:

[0048] The system includes a dynamic cutting path calculation module, a multi-dimensional cutting difference analysis module, an abnormal region correlation and localization module, and an adaptive cutting strategy generation module. The specific implementation methods are as follows:

[0049] The dynamic cutting path calculation module constructs a dynamic cutting path calculation model based on historical laser cutting operation data. It collects the actual laser power fluctuation parameters, beam quality distribution index, and cutting speed sequence data in real time during the current cutting process, and outputs theoretical cutting path information through the dynamic cutting path calculation model.

[0050] The multidimensional cutting difference analysis module receives theoretical cutting path information and performs three-dimensional difference analysis with the actual cutting path information measured by optical sensors, including cumulative position deviation, depth energy offset and velocity sequence similarity index, to generate a workpiece-level difference coefficient matrix.

[0051] The abnormal region correlation and localization module inputs the difference coefficient matrix into the spatial geometric analysis network, and combines the material property characteristic parameters and the cutting head position coordinate information to generate an abnormal probability thermal distribution map of the cutting defect propagation path, which is used to locate the abnormal cutting physical region.

[0052] The adaptive cutting strategy generation module configures cutting adjustment parameters based on the anomaly probability heat map, including applying multi-band impedance perturbation operations to the cutting heads of adjacent nodes with the largest anomaly probability gradient changes in the anomaly probability heat map.

[0053] Example 1: See Figure 2The dynamic cutting path calculation module calculates theoretical cutting path information through a historical cutting feature mining component and a dynamic cutting path calculation model construction component.

[0054] The historical cutting feature mining component decomposes historical laser cutting operation data into multiple dimensions, extracting the energy ratio of steady-state and transient components of power fluctuation parameters. The steady-state component reflects the long-term stable trend of laser power, while the transient component captures short-term fluctuation characteristics. Through beam coupling analysis, a correlation matrix between beam quality distribution indicators and cutting load rate is established. This matrix is ​​used to quantify the variation law of beam quality under different load conditions. The dynamic sequence alignment algorithm adjusts the velocity sequence pattern under different operating conditions, eliminating time sequence deviations caused by changes in equipment status or material differences, and ensuring the comparability of historical data under different operating conditions.

[0055] The dynamic cutting path calculation model building component transmits processed historical data to a hybrid prediction network. This network includes a time-series prediction unit based on the cutting head aging curve, used to generate the base path prediction values. The cutting head aging curve is modeled by long-term monitoring of key performance parameters of the cutting head, such as the optical attenuation rate of the focusing lens group and the mechanical wear of the galvanometer system, to establish an aging trend model. A fully connected network with an embedded spatial attention mechanism corrects the prediction bias caused by beam distortion. This mechanism analyzes the spatial distribution characteristics of the beam along the transmission path, identifies distortion regions that may affect cutting accuracy, and dynamically adjusts the weights during the prediction process. A sequence feature compensator dynamically adjusts the prediction weight coefficients based on the real-time acquired velocity sequence, ensuring that the model can adapt to real-time changes in cutting speed.

[0056] A heterogeneous data acquisition unit deployed with optical sensors synchronously captures the zero-crossing distortion rate and phase jitter characteristics of power fluctuation parameters. The zero-crossing distortion rate reflects the degree of distortion in the laser power waveform, while the phase jitter characteristics are used to evaluate the temporal stability of the laser output. The fractional content and angular distribution range of the beam quality distribution index are measured by an optical sensor array. The fractional content describes the distribution of beam energy in different spatial frequency bands, while the angular distribution range reflects the beam divergence characteristics. The amplitude abrupt change gradient and time interval entropy parameters of the velocity sequence are used to quantify the dynamic changes in cutting speed. The amplitude abrupt change gradient captures the instantaneous magnitude of speed changes, while the time interval entropy assesses the degree of randomness in the velocity sequence.

[0057] The theoretical path calculation component inputs real-time collected data into the dynamic cutting path calculation model to calculate the theoretical cutting path information. The calculation process first performs adaptive noise cancellation processing based on the cutting stage to eliminate the measurement noise caused by environmental vibration interference. This processing employs a multi-scale filtering algorithm, dynamically adjusting filtering parameters according to different cutting stages to ensure noise suppression while preserving effective signal characteristics. A spatiotemporal feature fusion algorithm integrates the associated feature vectors of power fluctuation parameters, beam quality distribution indices, and velocity sequences to construct a multi-dimensional feature space. This algorithm analyzes the temporal and spatial correlations of different features, extracts key factors affecting the cutting path, and establishes dynamic mapping relationships between features. The final output theoretical cutting path information includes the normal operating condition fluctuation range, which is dynamically updated according to the aging state of the cutting head, ensuring that the theoretical path reflects the actual operating status of the equipment.

[0058] The training process of the dynamic path cutting calculation model employs a joint optimization strategy using historical cutting data and real-time running data. Historical data is used to establish initial model parameters, while real-time data continuously adjusts model weights through an online learning mechanism. During model training, a sliding window mechanism is used to segment the historical data, ensuring the model can adapt to data changes at different time scales. The training objective function comprehensively considers path prediction accuracy, computational efficiency, and model stability, balancing different performance metrics through a multi-objective optimization algorithm. After model deployment, a periodic self-checking mechanism evaluates prediction performance, triggering a model parameter update process when prediction deviation exceeds a threshold.

[0059] The theoretical cutting path information is output in a structured data format, including key parameters such as path coordinate sequence, cutting energy distribution, and velocity variation curve. The path coordinate sequence describes the trajectory of the cutting head on the workpiece surface, while the cutting energy distribution reflects the dynamic changes in laser energy along the path. The velocity variation curve records the real-time adjustment process of the cutting speed, providing a benchmark reference for subsequent difference analysis. The theoretical path information is transmitted to the multi-dimensional cutting difference analysis module via a high-speed data interface, ensuring data synchronization and real-time performance.

[0060] Example 2: See Figure 3 The multidimensional cutting difference analysis module realizes the three-dimensional difference analysis between theoretical cutting path information and measured cutting path information through the position cumulative deviation calculation unit, depth energy offset detection unit, and velocity sequence similarity evaluation unit.

[0061] The cumulative position deviation calculation unit performs a sliding comparison between the theoretical and measured paths using a preset distance window. The window size is dynamically adjusted according to the cutting accuracy requirements. A dynamic sequence alignment algorithm processes asynchronously sampled path sequence data, eliminating timing misalignments caused by differences in sensor sampling rates or communication delays. The aligned path data calculates the cumulative deviation within each window, generating a position deviation vector. This vector records the path's offset in three-dimensional space, including deviation components in the X, Y, and Z directions.

[0062] The depth energy shift detection unit performs frequency domain decomposition on the depth data of the theoretical and measured paths. The depth data reflects the energy deposition of the laser within the material, and energy distribution characteristics in different frequency bands are extracted using spectral analysis. Preset sensitive frequency bands cover the cutting frequency range of typical metallic materials, such as the molten pool oscillation frequency of low-carbon steel and the thermal conductivity characteristic frequency of aluminum alloys. The energy spectral density ratio is used to calculate the difference in energy distribution between the theoretical and measured paths in each frequency band, quantifying the degree of energy shift. The energy shift index of each depth layer constitutes a depth shift vector, which reflects anomalies in the distribution of cutting energy along the material thickness direction.

[0063] The velocity sequence similarity evaluation unit analyzes the velocity sequence differences between theoretical and measured paths based on a structure matching algorithm. An edit distance algorithm matches the set of abrupt change points in the two sequences, identifying key time nodes of velocity changes. The phase synchronization error value calculates the alignment of abrupt change points on the time axis, reflecting the synchronicity differences in velocity adjustments. The velocity interval distribution difference index quantifies the dispersion of velocity fluctuations using statistical methods, generating a velocity similarity vector. This vector contains evaluation results across three dimensions: velocity matching degree, phase error, and distribution difference.

[0064] The difference coefficient matrix generation unit performs tensor concatenation on the position deviation vector, depth offset vector, and velocity similarity vector. Before concatenation, all vectors are normalized to eliminate the influence of different physical dimensions on the analysis results. A feature importance weighting mechanism dynamically adjusts the weight coefficients of each vector according to the cutting process requirements; for example, position deviation has a higher weight in high-precision cutting scenarios, while depth energy offset has a larger weight in thick plate cutting scenarios. The normalized vectors are organized into a three-dimensional tensor structure, forming a third-order difference coefficient matrix with three dimensions: workpiece number, timestamp, and difference type. This matrix structure supports rapid retrieval and batch analysis, facilitating subsequent processing by the anomaly detection module.

[0065] The position cumulative deviation calculation unit employs an adaptive window adjustment strategy to optimize computational efficiency. The initial window size is set based on material thickness and cutting speed, and the window size is dynamically adjusted during calculation according to the deviation change rate. Larger windows are used in areas with gradual deviation changes to reduce computational load, while smaller windows are used in areas with frequent abrupt changes to improve analysis accuracy. The window sliding step size is proportional to the cutting head movement speed to ensure that the sampling density matches the cutting accuracy requirements. The position deviation vector is stored using a circular buffer structure, optimizing memory usage efficiency while ensuring data integrity.

[0066] The frequency domain analysis of the depth energy migration detection unit employs a multi-resolution processing method. Coarse-grained analysis rapidly scans the energy distribution across the entire frequency band, identifying key frequency bands where anomalies may exist. Fine-grained analysis performs high-precision decomposition of suspicious frequency bands, extracting subtle energy difference features. The frequency band division scheme pre-configures multiple templates based on material type; for example, the stainless steel template focuses on characteristic frequency bands of the heat-affected zone, while the copper alloy template focuses on energy rebound frequency bands caused by high reflectivity. The calculation of the energy spectral density ratio incorporates a moving average filter to suppress the interference of random noise on the analysis results. The update cycle of the depth migration vector is synchronized with the change in cutting layer depth, ensuring that the energy analysis results are consistent with the actual cutting progress.

[0067] The structural matching algorithm for the velocity sequence similarity evaluation unit employs a bidirectional comparison strategy. Forward matching identifies abrupt changes starting from the sequence's beginning, while backward matching verifies the matching results by tracing back from the end point. This bidirectional verification mechanism improves the accuracy of abrupt change identification and avoids false matches caused by local fluctuations. A dynamic time warping algorithm is introduced to calculate the phase synchronization error, compensating for nonlinear time distortions between the theoretical and measured paths. The velocity similarity vector generation process includes an outlier filtering step, eliminating instantaneous velocity sampling points that significantly exceed reasonable limits, thus improving the reliability of the evaluation results.

[0068] The difference coefficient matrix employs a hierarchical storage structure. The bottom layer contains the raw sampled data, preserving complete time-series information; the middle layer stores the calculated values ​​of various vectors in the windowed statistical results; and the top layer provides a matrix view, offering a data interface for analytical tasks. This structure balances data processing efficiency and flexibility, supporting both real-time monitoring and historical backtracking applications. The matrix update mechanism utilizes a transaction processing model to ensure data consistency in a multi-threaded environment. The difference coefficients are visualized using a heatmap, intuitively presenting the distribution of anomalies at different locations and time points.

[0069] The real-time processing capability of the multidimensional segmentation difference analysis module is achieved through a pipelined architecture. Data acquisition, preprocessing, feature calculation, and matrix generation form a parallel processing pipeline, with each stage connected via a high-speed data bus. The computational task scheduler dynamically allocates computing resources based on system load, prioritizing analysis tasks with high real-time requirements. The module's internal status monitoring system continuously tracks the computational latency and resource usage of each unit, automatically triggering load balancing strategies when performance bottlenecks are detected.

[0070] The analysis results of the difference coefficient matrix are output in a standardized data format, including matrix dimension descriptions, timestamp sequences, and coefficient value arrays. The output interface supports multiple data transmission protocols to meet the access requirements of different downstream modules. The compression storage algorithm for matrix data selects the optimal compression strategy based on data type characteristics; for example, differential encoding is used for positional deviation data, and frequency domain coefficient truncation is used for depth energy data. Historical matrix archiving management employs a time-sharded storage strategy to balance storage space usage and query efficiency.

[0071] The module's calibration and maintenance process includes two parts: sensor calibration and algorithm parameter adjustment. Sensor calibration uses standard test pieces to obtain baseline data and establishes a correction relationship between measured values ​​and true values. Algorithm parameter adjustment optimizes key parameters such as window size and frequency band division by analyzing the characteristics of the difference matrix under typical operating conditions. Calibration records are stored in association with the difference analysis results, forming a complete quality traceability chain. The maintenance diagnostic tool provides matrix anomaly pattern analysis capabilities to assist technicians in quickly locating the source of system deviations.

[0072] Example 3: The velocity sequence similarity evaluation unit uses an improved edit distance algorithm to handle the velocity sequence comparison problem between theoretical paths and measured paths.

[0073] This algorithm introduces a speed change rate weighting factor into the traditional edit distance calculation, making it more sensitive to matching key speed segments in the cutting process. For a theoretical speed sequence T and a measured speed sequence R of length n, the structure matching cost function is defined as:

[0074]

[0075] in: This represents the total matching cost. This represents the rate of change of velocity at the i-th sampling point in the theoretical sequence. This represents the rate of change of velocity at the j-th point in the measured sequence. Indicates the speed range weighting factor. Represents the timing decay coefficient. Represents the absolute value of sequence position differences

[0076] The rate of change of velocity in this formula is calculated by the velocity difference between adjacent sampling points, with weighting factors... According to the cutting process specifications, the value is larger in the acceleration-sensitive region and smaller in the constant-speed range. (Time-series decay coefficient) By controlling the impact of the matching point spacing on the total cost, the algorithm is made more inclined to match velocity abrupt change points with similar timing.

[0077] The workpiece topology modeling unit of the anomaly region correlation and localization module constructs a weighted graph structure based on the cutting head position coordinates. Nodes represent discretized detection areas on the workpiece surface, and edge weights include material impedance parameters and thermal conductivity characteristics. Impedance parameters are obtained from a material database and include physical properties such as thermal conductivity and specific heat capacity. The heat-affected zone constraints are based on the inverse heat conduction equation, and the main heat flow paths are marked on the topology graph. The node admittance matrix Y of the geometric topology model is calculated through node connectivity and material parameters and is used for subsequent anomaly propagation simulation.

[0078] The anomaly propagation simulation unit employs a graph neural network to handle the mapping relationship between the difference coefficient matrix and the topology model. The difference coefficients are spatially assigned to corresponding topology nodes, forming the initial anomaly excitation signal. The aggregation function of the graph neural network is designed considering the attenuation characteristics of material impedance, and the information transfer weights between adjacent nodes are proportional to the elements of the admittance matrix. A multi-head attention mechanism captures cross-regional anomaly correlations, with each attention head focusing on propagation modes with different physical characteristics, such as heat conduction-dominated and stress wave-dominated modes. Temperature gradient constraints are introduced into the attention weight calculation to ensure that the anomaly propagation direction conforms to the second law of thermodynamics.

[0079] The Monte Carlo simulation process employs a random walk strategy on the topological network. Each walk starts from a node with a large difference coefficient, and the transition probability is proportional to the product of the admittance matrix elements and the attention weights. The walk path records the sequence of nodes traversed and their dwell times. Termination conditions for a single simulation include reaching the workpiece boundary or the cumulative attenuation exceeding a threshold. After large-scale repeated simulations, the arrival frequency of anomalous heat flux at each node is statistically analyzed, and the anomalous dwell probability is calculated in conjunction with material heat capacity parameters. During the simulation, the walk step size is dynamically adjusted, with a smaller step size used in anomalous accumulation regions to improve resolution.

[0080] The probability distribution generation unit performs kernel density estimation on the Monte Carlo simulation results, generating a continuous and smooth anomaly probability heatmap. The kernel function bandwidth is adaptively selected based on the workpiece size and node density, automatically reducing the bandwidth in boundary regions to avoid estimation bias. The color mapping scheme of the heatmap uses nonlinear transformation to highlight regions where the probability values ​​are within the critical range. The extraction of the suspicious path set is based on the change in probability gradient, connecting gradient maxima points to form the hypothesis of the main path of anomaly propagation.

[0081] The physical region positioning unit applies an improved DBSCAN algorithm to perform spatial clustering of the thermal distribution map. The algorithm parameters are dynamically adjusted based on the workpiece's geometric features, maintaining a proportional relationship between the minimum neighborhood radius and the cutting spot diameter. The clustering results undergo morphological closing operations to eliminate small noise areas and fill boundary depressions. The final output of abnormal cutting physical boundaries is approximated using polygons, with vertex coordinates aligned with the coordinate system of the cutting head motion control system.

[0082] The maximum spanning tree algorithm is used to generate the main propagation path information for anomalies. Nodes in the topology network are used as vertices, and the frequency of anomalous heat flow is used as edge weights to construct a fully connected graph. The Prim algorithm grows from the node with the highest probability, adding the edge with the largest weight outside the current cut set in each iteration until all significant anomalous nodes are covered. The generated ordered list of nodes is sorted by access time, and the material properties and anomalous characteristics of each node are labeled to form a structured propagation path description.

[0083] The process of binding suspicious equipment identification information establishes a mapping relationship between workpiece nodes and the cutting head actuator. Each node is associated with the position coordinates of its nearest neighbor cutting head. When a node is marked as suspicious, its associated cutting head enters a detailed diagnostic mode. The identification information includes the cutting head number, optical component status code, and mechanical positioning deviation value, used to guide subsequent maintenance operations. The binding relationship is established using a spatial nearest neighbor algorithm, taking into account the cutting head's working radius and kinematic constraints.

[0084] The spatial geometry analysis network employs a hierarchical feature extraction architecture. Bottom-level convolutions process local geometric features, mid-level graph convolutions aggregate regional characteristics, and the top-level fully connected network synthesizes global anomaly indicators. Network training uses a semi-supervised learning method, with a small amount of labeled data used to adjust classification boundaries and a large amount of unlabeled data used for feature representation learning. The loss function design incorporates topology-preserving constraints to ensure that the learned feature space maintains the original workpiece geometric relationships. The network inference process uses a sliding window strategy, where large-sized workpieces are segmented into overlapping blocks, processed separately, and then the results are fused.

[0085] The anomaly probability heatmap update mechanism employs an incremental recalculation strategy. Newly collected differential data triggers a resimulation of the local area, updating only the probability values ​​of affected nodes. The scope of incremental calculation is determined based on an anomaly propagation speed estimation, ensuring the inclusion of potentially changing areas. Historical probability distributions are cached using a sliding time window, supporting probability trend analysis and short-term prediction. Version management of the distribution map records the timestamp and modified area of ​​each update, facilitating backtracking of the anomaly development process.

[0086] The parameter configuration for dynamic impedance perturbation is based on probabilistic gradient analysis. For adjacent node pairs with the largest gradient changes, the correlation coefficient between their material impedance differences and probability changes is calculated. The perturbation frequency is selected to avoid the material's inherent resonant frequency band, and the amplitude is dynamically adjusted according to the current cutting depth. Multi-band perturbation is achieved by superimposing Gaussian pulses with different center frequencies, and the energy distribution of each band is matched with the abnormal characteristic spectrum. The perturbation application timing is synchronized with the cutting head's movement trajectory to avoid interfering with the normal cutting process.

[0087] The physical region boundary coordinate transformation module handles the mapping relationship between the workpiece coordinate system and the machine coordinate system. The coordinate transformation matrix is ​​obtained through a calibration process and includes rotation, translation, and scaling parameters. The vertices of the boundary polygon are interpolated and refined according to machining accuracy requirements, and the generated path point sequence is added to the speed planning module. The coordinate transformation process considers workpiece clamping deviation compensation and corrects positioning errors through reference point alignment. Collision detection is performed before the transformed machine coordinates are sent to the motion controller to ensure the safety of the abnormal area handling process.

[0088] The anomaly localization results are verified using a multi-sensor data fusion method. Thermal imaging data verifies the temperature distribution of the anomaly area, acoustic emission signals analyze the internal defect characteristics of the material, and visual inspection confirms the surface morphology anomaly. Data from each sensor is synchronized via timestamps and spatially registered to the same coordinate system. The verification results are fed back to adjust parameters of the probabilistic model, such as the transition probabilities from Monte Carlo simulations or the attention weights of a graph neural network. The iterative verification process gradually improves the localization accuracy until the evidence from all sensors reaches a consensus.

[0089] The maintenance decision support system analyzes the correlation between anomaly location results and equipment status data. A spatiotemporal correlation model is established between cutting head maintenance records and anomaly area locations to identify recurring failure modes. The decision rule base includes multi-dimensional conditions such as material properties, process parameters, and equipment status, outputting maintenance priority suggestions. The support system provides interactive causal analysis tools, allowing technicians to explore the predicted effects of different maintenance schemes. The execution results of maintenance actions are recorded and used to optimize decision rules.

[0090] Example 4: See Figure 4 The process of generating suspicious equipment identification information in the abnormal area association and positioning module is achieved through the binding operation between the workpiece node and the actual cutting head.

[0091] In the example of cutting stainless steel sheets, when node N-17 in the workpiece topology model is marked as the starting point of a suspicious path, the system retrieves the spatial coordinates of this node (x=245mm, y=178mm) and queries the cutting head position log. According to the motion control system record, cutting head H-03 was within ±2mm of this coordinate at timestamp T-428, and its optical lens temperature sensor showed abnormal temperature rise. The system establishes a binding relationship between cutting head H-03 and node N-17, generating a record item in the suspicious device identifier set, including fields such as cutting head serial number, abnormal start time, and temperature deviation value. For multi-cutting head collaborative operation scenarios, the binding operation considers the weight allocation of the overlapping area of ​​the working radius. When a coordinate point is located at the intersection of the working ranges of multiple cutting heads, the system will simultaneously mark all potentially affected cutting heads and label them with confidence scores.

[0092] The update mechanism for the suspicious device identifier set adopts an event-driven model. When laser cutting carbon fiber composite material, the optical sensor detects a sudden 30% increase in beam reflectivity at node N-09, triggering a real-time binding process. The system immediately scans all cutting head position data within a 5mm radius of that node over the past 3 minutes and finds an abnormal offset of 0.15mm in the Z-axis coordinate of the focusing lens of cutting head H-05. This information is added to the suspicious device identifier set and displayed in conjunction with historical maintenance records, indicating that the lens assembly calibration of this cutting head has been in operation for 98 hours, approaching the maintenance cycle threshold. In multi-cutting head systems, this type of real-time binding operation takes an average of 12ms to complete, ensuring rapid correlation between abnormal signals and device status.

[0093] The verification process for suspicious equipment identification information incorporates multi-dimensional cross-checking. In a stainless steel pipe cutting operation, the system initially identified cutting head H-02 as a suspicious device. However, motion trajectory playback showed that this cutting head was in standby mode during an abnormal period. The system automatically initiated a secondary verification process: retrieving thermal imager data confirmed that the temperature of the H-02 lens barrel did indeed fluctuate abnormally by 3°C; checking the work log of the adjacent cutting head H-01 revealed that its protective gas flow rate decreased by 25% during the abnormal period; ultimately, it was determined that the abnormality of H-01 caused thermal interference with H-02. The verification process in such complex scenarios typically lasts 45-60 seconds. The system will temporarily lock the suspicious equipment identification status until verification is complete to avoid false alarms interfering with production.

[0094] The visualization of the main path of anomaly propagation supports spatiotemporal analysis. When cutting multi-layer composite armor plates, the main path identified by the system exhibits periodic fluctuations along the Z-axis, corresponding to interface reflections between different material layers. The 3D visualization tool unfolds the path nodes according to a time series, showing that the propagation speed of anomalous heat flow in the ceramic layer (Z=2.1mm) is 1.7 times faster than in the metal layer (Z=3.4mm). The color gradient of the nodes on the time axis reflects the temperature change process, such as the gradual change from orange-red (320℃) at node N-28 to dark red (285℃) at node N-31. Operators can observe the dynamic process of anomaly propagation by sliding along the interactive time axis, and pause at any time to view the material properties and cutting parameters of the nodes.

[0095] The maintenance decision support function for suspicious equipment identification is specifically manifested in priority ranking. On a car chassis cutting line, the system simultaneously flagged two suspicious items: focusing lens contamination on cutting head H-04 and guide rail vibration on cutting head H-07. Analysis based on the abnormal thermal distribution map showed that the abnormal area associated with H-04 accounted for 18% of the total workpiece area, while H-07 only affected 6%. When the system automatically generated maintenance work orders, it classified H-04 as priority A (requiring immediate processing) and H-07 as priority B (to be processed within 8 hours). Each work order entry included a screenshot of the associated abnormal path, a table of equipment operating parameter deviations, and recommended maintenance measures. For example, work order H-04 suggested "cleaning the focusing lens and checking the cooling pipes." Historical data shows that this prioritization can reduce unplanned downtime by 22%.

[0096] The prediction function for the main path of anomaly propagation is achieved through a machine learning model. When cutting 10 aluminum alloy components of the same specifications consecutively, the system detected that the anomaly path consistently originated from the upper left corner of the workpiece. When the sixth workpiece began cutting, the system pre-deployed enhanced monitoring around the predicted path [N-05→N-08→N-11], and indeed captured the initial anomaly signal at node N-05 when the cutting reached 38%. The prediction model was trained based on the path patterns of the first five workpieces, considering over 30 feature parameters such as material batch, ambient temperature and humidity. When the overlap between the actual anomaly path and the predicted path exceeds 75%, the system automatically optimizes the cutting parameters, such as reducing the laser power in the upper left corner by 8% when cutting the seventh workpiece.

[0097] The storage of suspicious equipment identification information adopts a hierarchical data structure. Each identification entry contains three dimensions: a base layer (equipment ID, binding time), an evidence layer (sensor readings, video snapshots), and an analysis layer (anomaly type classification, impact assessment). When the cutting head H-09 is repeatedly marked as having a loose fiber optic interface, the system automatically creates a dedicated fault mode profile for that equipment, recording data such as temperature change curves and vibration spectrum characteristics of each anomaly. This structured information is transmitted to the factory's MES system via the OPC UA interface and stored in association with business data such as equipment maintenance history and spare parts replacement records, forming a complete equipment health management database.

[0098] The application of process optimization based on the main path of abnormal propagation is reflected in adaptive parameter adjustment. In a copper alloy cutting operation, the system detected that the main path extended along the material rolling direction, and the width of the heat-affected zone (HAZ) at the corresponding node exceeded the standard. The control system immediately initiated a compensation strategy: increasing the cutting speed by 12% at the upstream node and increasing the auxiliary gas pressure by 25 kPa at the downstream node. This local parameter adjustment based on the physical propagation path, compared to global parameter modification, maintains processing efficiency in non-abnormal areas. After compensation, subsequent quality inspection showed that the HAZ width was controlled within ±0.05 mm of the process requirements, and the overall cutting time increased by only 3.7%.

[0099] The closed-loop management of suspicious equipment identification is reflected in the automatic verification process. When the system marks the collimator of the H-12 cutting head as misaligned, it automatically performs a three-point calibration test upon the next startup of the equipment: first, a standard pattern is cut on the test board; then, the deviation between the actual cutting trajectory and the theoretical path is measured using a high-precision camera; finally, the calibration result is fed back to the identification system. If the verification passes (deviation <0.02mm), the suspicious identification is cleared; if it fails, it is escalated to a fault alarm and the equipment is locked. The entire verification process is completed automatically during the equipment warm-up phase, taking approximately 2 minutes and not affecting normal production schedules. This mechanism effectively prevents the problem of "over-maintenance," and historical data shows that approximately 31% of suspicious identifications can be cleared from alarms during the automatic verification process.

[0100] The cross-workpiece analysis function for the main path of anomaly propagation supports batch quality traceability. During the continuous processing of 20 wind turbine bearing housing components, the system detected that anomaly paths were concentrated around the flange holes. By retrieving cutting records of similar workpieces from the past three months, statistics showed that 86% of the anomaly paths had a similar spatial distribution pattern to the current workpiece. This cluster analysis triggered a process review, tracing back to the recently changed protective gas supplier. Quality engineers confirmed through path pattern comparison that insufficient purity of the new gas source, leading to cut oxidation, was the root cause of the path anomalies. The gas composition was subsequently adjusted to resolve this batch-specific issue. The system's path similarity analysis tool supports multi-condition filtering and comparison by time, material batch, equipment combination, and other criteria.

[0101] The suspicious device identification system employs a distributed hardware architecture. Each cutting head is equipped with an independent edge computing unit that caches motion parameters and sensor data from the past 8 hours in real time. When the central system initiates binding, the edge unit immediately uploads detailed operational logs for the specified time period, such as the servo motor current fluctuation curve of the H-15 cutting head during abnormal periods. This design reduces the network transmission burden, keeping the average response time for binding queries below 50ms. A data synchronization mechanism ensures that even if the network is temporarily interrupted, the edge unit can maintain a 72-hour data cache, automatically re-uploading critical event records once the connection is restored. All transmitted data is encrypted using AES-256 to prevent leakage of device status information.

[0102] Example 5: See Figure 5 The integrated optical path transmission component of the high-power laser cutting head adopts a modular design architecture, and the collimating lens unit is equipped with an adaptive zoom mechanism, which can automatically adjust the focal length position according to the original divergence angle of the laser generator output beam.

[0103] When the cutting system switches between laser sources of different power levels, the servo drive mechanism of the collimating lens completes position fine-tuning within 150 milliseconds, ensuring that the collimation error of the output beam is controlled within 0.05 mrad. The focusing lens unit adopts a multi-layer coating process and designs a special anti-reflection coating system for the 1064nm laser beam, ensuring that the lens temperature rise does not exceed 8°C above the ambient temperature during continuous operation. The beam-shaping galvanometer unit incorporates a high-speed position feedback sensor, achieving a galvanometer deflection angle resolution of 0.001 degrees. Combined with the control commands of the dynamic adjustment component, it can achieve precise tracking of complex curved trajectories.

[0104] The embedded optical sensor array of the real-time feedback component adopts a ring-shaped distribution design, with 12 high-sensitivity detection units arranged circumferentially around the cutting spot. Each detection unit includes a visible light CCD, an infrared thermal imager, and a spectral analysis module, synchronously acquiring spot position coordinates, energy distribution parameters, and heat-affected zone temperature gradient data at a sampling frequency of 2000Hz. Position coordinate measurement is based on the double-cross laser positioning principle, using image processing algorithms to identify the center position of the spot, achieving a spatial positioning accuracy of ±5μm. Energy distribution parameters are acquired through a beam-splitting prism array, measuring the asymmetry of energy distribution along the X / Y axes. Heat-affected zone temperature gradient data is measured non-contactly, with the infrared sensor array scanning the temperature field changes at the cutting edge with a spatial resolution of 50μm.

[0105] The dynamic adjustment component achieves real-time compensation of cutting parameters through a multi-axis linkage mechanism. A piezoelectric ceramic actuator controls the axial displacement of the focusing lens assembly, with each nanometer-level displacement corresponding to a 0.3μm adjustment of the focal position, and a compensation command response time of less than 1 millisecond. The protective gas pressure regulation system consists of a high-speed proportional valve and a miniature eddy current sensor. The proportional valve's opening resolution reaches 0.1%, and it can establish a new pressure balance within 10 milliseconds. The gas nozzle is designed with a rotatable structure, automatically adjusting the airflow injection angle to the optimal position based on the location of the abnormal area indicated by the anomaly probability thermal distribution map. All actuators in the adjustment component are equipped with redundant sensors to monitor the deviation between the actual execution status and the command value in real time. When the deviation exceeds a safety threshold, a protective shutdown is immediately triggered.

[0106] The data interaction between the laser cutting head and the optical path system adopts a dual-channel architecture of fiber optic communication and wireless transmission. Critical control commands are transmitted in real-time via gigabit fiber optic cable, ensuring a transmission latency of less than 0.5 milliseconds. Status monitoring data is transmitted wirelessly via the 5G millimeter-wave band, with each cutting head equipped with an independent network slice, achieving a stable data transmission rate of over 800Mbps. The communication protocol uses a time-sensitive network standard, giving the highest priority to critical data packets, ensuring timely delivery of control commands even under network congestion. A data encryption module performs end-to-end encryption on all transmitted commands and status information to prevent external interference and malicious attacks.

[0107] The thermal management system of the cutting head employs phase change material cooling technology. A micro heat pipe array is arranged around the collimating lens unit and focusing lens assembly, utilizing the latent heat absorption characteristics of the phase change material to control the operating temperature fluctuations of the optical components within ±0.5℃. The cooling system of the galvanometer drive motor uses eddy current cooling technology, employing a micro fan to generate rotating airflow to remove heat, ensuring that the motor winding temperature does not exceed the rated operating range. All cooling circuits are equipped with flow and temperature monitoring sensors, automatically adjusting pump power or switching to a backup cooling path when a decrease in cooling efficiency is detected. The status data of the thermal management system is uploaded to a central monitoring platform in real time, forming a complete heat load change trend chart.

[0108] The contamination protection system for optical components comprises multiple layers of protection. The outermost air curtain employs a ring-shaped airflow design, forming a stable air barrier around the optical window and effectively blocking over 99% of splash particles. The middle layer's self-cleaning coating possesses oleophobic and hydrophobic properties, making it difficult for contaminants to adhere to the optical surface. The inner layer's electrostatic dust removal device automatically activates during cutting intervals, using a high-voltage electrostatic field to adsorb residual microparticles. The operational status of the protection system is assessed in real-time by a particle counter; when a drop in transmittance exceeding 3% is detected, an automatic cleaning program is triggered or manual maintenance is prompted. Records of all protection actions form a complete maintenance log, used for analyzing contamination sources and optimizing protection strategies.

[0109] The cutting head's mechanical structure employs a lightweight magnesium alloy design, reducing its overall weight by 35% compared to traditional structures while maintaining sufficient rigidity and stability. The bearing system for moving parts utilizes magnetic levitation technology, eliminating positioning errors caused by mechanical friction. A quick-release interface design allows for optical module replacement within 3 minutes, and the optical path calibration process after module docking is fully automated, requiring no manual intervention. The structural health monitoring system uses a vibration sensor array to capture changes in mechanical resonant frequency in real time, prompting preventative maintenance when a decrease in structural stiffness is detected. Three-dimensional models of all mechanical components are stored in a digital twin system for simulation analysis and lifespan prediction.

[0110] 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.

[0111] 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 high-power laser cutting optical path system, characterized in that, include: The dynamic cutting path calculation module constructs a dynamic cutting path calculation model based on historical laser cutting operation data. It collects the actual laser power fluctuation parameters, beam quality distribution index, and cutting speed sequence data in real time during the current cutting process, and outputs theoretical cutting path information through the dynamic cutting path calculation model. The multi-dimensional cutting difference analysis module receives the theoretical cutting path information and performs a three-dimensional difference analysis with the actual cutting path information measured by the optical sensor. The three-dimensional difference analysis includes the cumulative position deviation, depth energy offset, and velocity sequence similarity index, generating a workpiece-level difference coefficient matrix. The abnormal region association and positioning module inputs the difference coefficient matrix into the spatial geometric analysis network, and combines the material property characteristic parameters and the cutting head position coordinate information to generate an abnormal probability thermal distribution map of the cutting defect propagation path, which is used to locate the abnormal cutting physical region. The adaptive cutting strategy generation module configures cutting adjustment parameters according to the anomaly probability heat map, including applying multi-band impedance perturbation operations to the adjacent node cutting heads with the largest anomaly probability gradient change in the anomaly probability heat map. The dynamic cutting path calculation module specifically includes a historical cutting feature mining component and a dynamic cutting path calculation model construction component; The historical cutting feature mining component processes historical laser cutting operation data, performs multi-dimensional decomposition operations, including extracting the energy ratio values ​​of steady-state and transient components of power fluctuation parameters, establishing a correlation matrix between beam quality distribution index and cutting load rate through beam coupling analysis, and adjusting the speed sequence mode under different working conditions using a dynamic sequence alignment algorithm. The dynamic cutting path calculation model building component transmits the processed historical running data to the hybrid prediction network. The hybrid prediction network has a time-series prediction unit based on the cutting head aging curve to generate basic path prediction values, a fully connected network with embedded spatial attention mechanism to correct the prediction deviation caused by beam distortion, and a sequence feature compensator to dynamically adjust the prediction weight coefficients according to the real-time acquired velocity sequence. The heterogeneous data acquisition unit deployed through optical sensors synchronously captures the zero-crossing distortion rate and phase jitter characteristics of power fluctuation parameters, the fractional content and angular distribution range of beam quality distribution indicators, the amplitude abrupt change gradient and time interval entropy parameters of the velocity sequence. The theoretical path calculation component inputs the real-time acquired data into the dynamic cutting path calculation model to obtain theoretical cutting path information.

2. The high-power laser cutting optical path system according to claim 1, characterized in that, The theoretical cutting path information calculation process includes adaptive noise cancellation processing based on the cutting stage to eliminate the measurement noise caused by environmental vibration interference; integrating the associated feature vectors of power fluctuation parameters, beam quality distribution index and velocity sequence through a spatiotemporal feature fusion algorithm; and outputting theoretical cutting path information including the fluctuation range under normal operating conditions, which is dynamically updated according to the aging state of the cutting head.

3. The high-power laser cutting optical path system according to claim 1, characterized in that, The multidimensional cutting difference analysis module specifically includes a position cumulative deviation calculation unit, a depth energy shift detection unit, and a velocity sequence similarity evaluation unit; The position cumulative deviation calculation unit performs a sliding comparison between the theoretical cutting path information and the measured cutting path information using a preset distance window, and uses a dynamic sequence alignment algorithm to align the asynchronously sampled path sequence data, calculates the cumulative deviation within each window, and generates a position deviation vector. The depth energy offset detection unit performs frequency domain decomposition on the depth data of theoretical and measured cutting path information, calculates the energy spectral density ratio, extracts the energy offset index of each depth layer, and constructs the depth offset vector. The velocity sequence similarity evaluation unit matches the velocity sequence distance distribution between theoretical cutting path information and measured cutting path information based on the structure matching algorithm, calculates the phase synchronization error value of velocity mutation points, quantifies the difference index of velocity interval distribution, and generates a velocity similarity vector. The difference coefficient matrix generation unit performs tensor concatenation operation on the position deviation vector, depth offset vector, and velocity similarity vector, and eliminates the influence of dimensional differences through feature importance weighted normalization processing, outputting a third-order difference coefficient matrix with dimensions of workpiece number multiplied by timestamp multiplied by difference type.

4. A high-power laser cutting optical path system according to claim 3, characterized in that, In the frequency domain analysis stage, the depth energy offset detection unit first extracts the corresponding depth data from the theoretical cutting path information and the measured cutting path information. It then uses the spectral decomposition method to perform multi-scale frequency band analysis on each group of depth signals, extracting the energy distribution characteristic parameters within a preset sensitive frequency band interval. The sensitive frequency band interval covers the cutting frequency range of typical metal materials. After extraction, the energy density of the theoretical cutting path information and the measured cutting path information within the sensitive frequency band interval is quantified and calculated. Based on the relative offset between the two, the offset index data of each order of depth layer is extracted. The energy offset results of all orders are summarized to construct a depth offset vector.

5. A high-power laser cutting optical path system according to claim 3, characterized in that, The structure matching algorithm in the velocity sequence similarity evaluation unit uses the edit distance algorithm to perform position matching operations on the set of mutation points in the two sequences and identify the phase synchronization error value.

6. A high-power laser cutting optical path system according to claim 1, characterized in that, The abnormal region correlation and localization module specifically includes a workpiece topology modeling unit, an abnormal propagation simulation unit, and a probability distribution generation unit; The workpiece topology modeling unit constructs a topology diagram of the workpiece node connection relationship based on the position coordinate information of the cutting head, marks the material impedance parameters between each node, and superimposes the reverse heat conduction constraint conditions of the heat-affected zone access point on the topology diagram to generate a geometric topology model including the impedance matrix and the node admittance matrix. The anomaly propagation simulation unit maps the difference coefficient matrix to the corresponding nodes of the workpiece topology model, performs anomaly propagation inference operation based on graph neural network, and the anomaly propagation inference calculation includes calculating the attenuation factor of abnormal heat flow based on node impedance parameters, capturing cross-regional anomaly correlation feature vector through multi-head attention mechanism, and simulating the diffusion path trajectory of abnormal heat flow in the topology network using Monte Carlo method. The probability distribution generation unit counts the frequency of abnormal heat flow occurrences in simulated propagation for each path, calculates the abnormal heat flow residence probability value in combination with material impedance parameters, generates an abnormal probability thermal distribution map covering the entire workpiece, and marks the set of suspicious paths whose probability values ​​exceed the preset abnormal residence probability threshold; the physical region positioning unit performs spatial clustering analysis on the abnormal probability thermal distribution map, identifies abnormal probability clustering areas, and delineates the abnormal cutting physical boundary range based on the cutting head position coordinates and the topological connection relationship of the workpiece.

7. A high-power laser cutting optical path system according to claim 6, characterized in that, The abnormal area association and location module also includes outputting suspicious device identification information and abnormal propagation main path information; The suspicious device identification information is based on the workpiece nodes connected by the suspicious path set. The workpiece nodes are bound to the actual cutting head to form a suspicious device identification set, indicating the potential abnormal cutting source or affected terminal. The anomaly propagation main path information is obtained by recording the node paths and their sequence sequence experienced in each round of propagation during the anomaly diffusion process in Monte Carlo simulation. The frequency of occurrence of each path is counted in all simulated paths, and the path sequence with the highest cumulative frequency is selected as the anomaly propagation main path. The output anomaly propagation main path sequence is a structured ordered node list, which reflects the main propagation trajectory of the anomaly information in the workpiece.

8. A high-power laser cutting head, suitable for the high-power laser cutting optical path system according to any one of claims 1 to 7, characterized in that, The laser cutting head includes an integrated optical path transmission component, a real-time feedback component, and a dynamic adjustment component. The integrated optical path transmission component includes a collimating lens unit, a focusing lens group unit, and a beam-shaping galvanometer unit. The collimating lens unit receives the original beam output from the laser generator and performs collimation processing. The focusing lens group unit focuses the collimated beam onto the workpiece surface to form a cutting spot. The beam-shaping galvanometer unit dynamically adjusts the beam deflection angle according to the cutting adjustment parameters issued by the adaptive cutting strategy generation module. The real-time feedback component is data-connected to the multi-dimensional cutting difference analysis module and includes an embedded optical sensor array. It captures the actual position coordinates of the cutting spot, energy distribution parameters, and temperature gradient data of the heat-affected zone in real time and feeds the captured data back to the dynamic cutting path calculation module. The dynamic adjustment component is linked with the abnormal area correlation and positioning module. Based on the suspicious path set output by the abnormal probability thermal distribution map, it automatically adjusts the focal length offset and protective gas pressure value of the cutting head. The focal length offset is compensated at the nanometer level through a piezoelectric ceramic actuator, and the protective gas pressure value is adjusted at the millisecond level through a proportional valve.

9. A method of using a high-power laser cutting optical path system, applied to the high-power laser cutting optical path system according to any one of claims 1 to 7, characterized in that, Includes the following steps: Step S1: Load the material property feature parameters of the target workpiece and the preset cutting trajectory data through the dynamic cutting path calculation module to construct an initial dynamic cutting path calculation model; Step S2: Activate the real-time feedback component to capture laser power fluctuation parameters, beam quality distribution index and cutting speed sequence data during the actual cutting process, and synchronously input the data into the multi-dimensional cutting difference analysis module; Step S3: The multidimensional cutting difference analysis module performs a three-dimensional difference analysis operation, generates a workpiece-level difference coefficient matrix, and transmits it to the abnormal area association and positioning module; Step S4: The abnormal area association and positioning module maps the difference coefficient matrix according to the topology map of the workpiece node connection relationship, and outputs the abnormal probability heat distribution map and suspicious equipment identification information; Step S5: When the detected abnormal probability value exceeds the preset abnormal dwell probability threshold, the adaptive cutting strategy generation module sends a high-frequency laser monitoring command and a multi-band impedance disturbance command to the laser cutting head. Step S6: The dynamic adjustment component of the laser cutting head responds to the command, performs focal length offset compensation operation and protective gas pressure gradient adjustment operation, and feeds back the adjusted data to the dynamic cutting path calculation module to update the theoretical cutting path information.

Images