Self-adaptive optimization method for processing error compensation of five-axis numerical control machine tool

Abstract

The invention discloses a multi-dimensional real-time coupling error compensation method for a five-axis numerical control machine tool. The method comprises the following steps: (1) acquiring geometric error, thermal error, dynamic error and coupling error data of the five-axis machine tool in real time by utilizing a sensor network arranged on key parts of the machine tool; (2) by constructing a five-axis linkage nonlinear error compensation model, performing traceability analysis on errors generated in the machining process by adopting a reverse error propagation algorithm, and dynamically optimizing the compensation amount according to error sources; (3) establishing a causal relationship and a probability distribution model of each error source based on a Bayesian network, updating joint distribution of the error sources by using real-time sensor data, predicting a comprehensive error through Bayesian reasoning, and adjusting motion parameters of each axis; through the real-time, multi-dimensional and intelligent error compensation technology, the machining precision, efficiency and adaptability of the five-axis numerical control machine tool are remarkably improved, the production cost is reduced, the product quality is improved, and the five-axis numerical control machine tool has higher competitiveness in the complex and high-end manufacturing field.

Description

Technical Field

[0001] The invention belongs to the technical field of five-axis CNC machine tool processing, and particularly relates to a self-adaptive optimization method for five-axis CNC machine tool processing error compensation. Background Art

[0002] Five-axis CNC machine tools, due to their multi-axis machining capabilities and high precision, are widely used in high-end manufacturing fields such as aerospace, automotive, mold processing, and medical devices. With the increasing demand for high-precision machining of complex parts in the manufacturing industry, the application of machine tools in machining complex surfaces, large parts, and high-hardness materials faces severe challenges. However, traditional error compensation methods have many shortcomings in terms of accuracy, efficiency, and adaptability.

[0003] Error sources are diverse and complex: Five-axis CNC machine tools are subject to a variety of errors during machining, including geometric errors (guideway straightness deviation, rotary axis angular error, etc.), thermal errors (thermal expansion of the spindle and key machine components), dynamic errors (vibration, cutting force fluctuations), and the coupled effects of these errors. Traditional error compensation methods typically use a single model to address various error types, making it difficult to effectively address multi-dimensional coupled errors, resulting in machining accuracy that fails to meet the requirements of high-end parts.

[0004] Lack of real-time performance: Traditional methods rely on offline error modeling and compensation, separating error analysis and compensation from the machining process. This makes them unable to respond in real time to dynamically changing errors during machining (such as temperature changes and dynamic vibration). This hysteresis-based compensation method is less adaptable to complex machining tasks (such as high-speed cutting or thin-walled part machining), easily leading to error accumulation and reduced machining quality.

[0005] Inadequate zoning and path optimization: Existing methods typically process the error distribution of the workspace based on a global static model, ignoring the heterogeneity of error distribution within different spatial regions. This lack of path optimization and dynamic compensation prevents full utilization of low-error regions, resulting in low machining path efficiency and limited error control effectiveness.

[0006] Limited effectiveness in controlling vibration and dynamic errors: Vibration is a common and significant source of dynamic error during machining. Traditional passive vibration control (such as damping devices) is difficult to effectively adapt to different working conditions, especially during high-speed machining or machining of thin-walled parts. Vibration-induced trajectory deviations often lead to reduced surface quality and part shape deviations, and existing methods have limited effectiveness in compensating for these deviations.

[0007] Insufficient feedback and optimization capabilities: Traditional error compensation methods are mostly open-loop control, unable to optimize the model based on actual error data after processing. Compensation accuracy decreases over time. Lacking intelligent learning and dynamic update capabilities, they are difficult to adapt to complex working conditions or multi-task processing requirements. Summary of the Invention

[0008] Therefore, the purpose of the present invention is to provide an adaptive optimization method for five-axis CNC machine tool processing error compensation. The present invention significantly improves the processing accuracy, efficiency and adaptability of five-axis CNC machine tools through real-time, multi-dimensional and intelligent error compensation technology, reduces production costs and improves product quality, making it more competitive and economical in complex and high-end manufacturing fields.

[0009] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0010] A multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool comprises the following steps:

[0011] (1) Using a sensor network placed on key machine tool components to collect real-time data on geometric errors, thermal errors, dynamic errors, and coupling errors of a five-axis machine tool;

[0012] (2) By constructing a five-axis linkage nonlinear error compensation model, the inverse error propagation algorithm is used to trace the errors generated during the machining process and dynamically optimize the compensation amount according to the error source;

[0013] (3) Based on the Bayesian network, a causal relationship and probability distribution model of each error source is established, the joint distribution of the error sources is updated using real-time sensor data, the comprehensive error is predicted through Bayesian reasoning, and the motion parameters of each axis are adjusted;

[0014] (4) Based on the disturbance characteristics of the machining trajectory, an adaptive learning algorithm is used to learn the trajectory disturbance pattern in real time and dynamically update the error compensation model;

[0015] (5) Partition the machine tool workspace, calibrate the error distribution of different areas, and combine the partition compensation model with the five-axis linkage kinematic model to call the corresponding compensation parameters for each motion area in the machining path in real time;

[0016] (6) The above error compensation strategy is integrated into the CNC system, and the tool posture, feed speed and processing path are adjusted through the machine tool controller to achieve dynamic compensation of multi-dimensional real-time coupling errors.

[0017] Wherein, the sensor network includes:

[0018] Laser interferometers and linear displacement sensors are respectively arranged on the machine tool guide rails and rotary axis support positions to collect geometric error data;

[0019] Thermocouples and infrared thermal imagers are respectively placed on the spindle, motor and machine tool bracket to monitor the temperature changes of key machine tool components and collect thermal error data;

[0020] Acceleration sensors and vibration sensors are installed on the tool holder and machine slide to detect dynamic disturbance errors;

[0021] Angle encoders are placed on the rotary table and the swing head axis to collect angle deviation data during multi-axis linkage motion.

[0022] A real-time data acquisition system uses EtherCAT or CAN bus communication protocols to connect to the above sensors and transmit the collected data to the machine tool controller;

[0023] The data fusion module is embedded in the machine tool controller and uses the Kalman filter algorithm to reduce noise, correct, and classify and store the collected error data.

[0024] The error analysis module calculates the changing trends of the geometric error, thermal error, dynamic error and coupling error of the five-axis machine tool based on sensor data, and inputs the error results into the motion compensation module of the controller.

[0025] The implementation method of step (2) is as follows:

[0026] 1) Use Kalman filter algorithm to reduce noise and preprocess the collected data;

[0027] 2) Construct a full-dimensional nonlinear error model for the five-axis linkage, use sparse tensor decomposition technology to reduce the dimensionality of the high-dimensional error tensor, and extract the main error coupling relationships;

[0028] 3) By constructing a differentiable error propagation function and using the inverse error propagation algorithm to trace the source of the machining error, adaptive weight allocation is performed according to the contribution rate of the error source, and critical errors are compensated first;

[0029] 4) Using a variational autoencoder (VAE) to generate multiple sets of error compensation strategy samples, and optimizing the initial compensation value through deep learning of historical processing data;

[0030] 5) Use genetic algorithm (GA) to globally optimize the compensation strategy, take the machining accuracy after error compensation as the fitness function, and dynamically adjust the compensation amount;

[0031] 6) The optimized compensation amount is input into the machine tool controller in real time through the edge computing module to adjust the tool posture, feed speed and five-axis linkage motion parameters to achieve dynamic compensation of machining errors.

[0032] Wherein, the implementation method of step (3) is:

[0033] 1) Based on the collected data, the causal relationship between different error sources is identified through the Granger causality test method, and the initial topology of the Bayesian network is generated using a greedy algorithm;

[0034] 2) Using variational inference algorithms to optimize the joint probability distribution model of the Bayesian network to construct the causal relationship and joint probability distribution of each error source;

[0035] 3) Using real-time sensor data to update the node states in the Bayesian network, the comprehensive error of the five-axis CNC machine tool under specific machining conditions is calculated based on Bayesian reasoning;

[0036] 4) Based on the Bayesian inference results, the tool position, rotation angle and five-axis linkage motion parameters are dynamically adjusted to achieve real-time compensation of machining errors;

[0037] 5) The compensation results are updated to the Bayesian network model through a feedback loop embedded in the machine tool controller, and the causal relationship model and error compensation strategy are optimized by combining multi-task learning.

[0038] Wherein, the implementation method of step (4) is:

[0039] 1) The empirical mode decomposition (EMD) technique is used to decompose the disturbance signal into several intrinsic mode functions (IMFs), and the characteristic frequency and amplitude change trend of the disturbance signal are extracted in combination with wavelet transform;

[0040] 2) Build a trajectory disturbance pattern recognition model based on a long short-term memory network (LSTM), input the extracted disturbance features into the model, and dynamically update the model parameters through an online learning mechanism to adapt to changes in real-time processing conditions;

[0041] 3) Based on the disturbance pattern recognition results, the reinforcement learning algorithm is used to optimize and adjust the five-axis linkage motion parameters. The tool posture and motion path are optimized for compensable disturbances, and the active vibration suppression function is used to reduce the impact of difficult-to-compensate disturbances.

[0042] 4) Through the closed-loop feedback mechanism, the compensated processing trajectory data is input into the disturbance pattern recognition model in real time to further improve the model's adaptability to complex disturbance conditions.

[0043] Wherein, the implementation method of step (5) is:

[0044] 1) Cluster analysis of the error data in the workspace of the five-axis machine tool is performed based on the density clustering algorithm (DBSCAN), and the workspace is divided into multiple partitions with consistent error characteristics;

[0045] 2) Use the Voronoi diagram optimization algorithm to adjust the partition boundaries to make the error transition area between partitions smooth;

[0046] 3) Using the local polynomial regression method to construct an error distribution model in each partition, and dynamically updating the partition error model in combination with real-time sensor data;

[0047] 4) According to the current position of the tool and the machining path, the error compensation parameters of the corresponding partition are called, and the error compensation transition at the partition boundary is smoothed by the Bezier curve interpolation algorithm;

[0048] (5) For cross-partition processing paths, a dynamic path planning algorithm is used to optimize the processing path, and the error compensation strategy between partitions is dynamically adjusted in combination with a deep reinforcement learning algorithm.

[0049] Wherein, the implementation method of step (6) is:

[0050] 1) Embed a dynamic compensation module in the machine tool CNC system. This module implements real-time control of error compensation through a three-layer structure consisting of a perception layer, a decision layer, and an execution layer.

[0051] 2) Using the fuzzy logic control algorithm, fuzzy rules are generated according to the sources and magnitudes of geometric errors, thermal errors, dynamic errors, and coupling errors, and the final error compensation is calculated through a dynamic weighted fusion method;

[0052] 3) Dynamically optimize the machining path through reinforcement learning algorithms to avoid the machining path passing through areas with significant errors. Combined with the adaptive adjustment function of motion parameters, the feed speed and tool posture are adjusted in real time.

[0053] 4) Integrate active vibration suppression function and use predictive control algorithm to adjust the feed rhythm and reduce dynamic error superposition;

[0054] 5) Through the closed-loop feedback mechanism, the actual error value after processing is fed back to the dynamic compensation module, the fuzzy rules and compensation strategy are updated, and the compensation accuracy is optimized through the real-time evaluation module.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] The present invention is aimed at multi-dimensional real-time coupling error compensation for five-axis CNC machine tools. Compared with traditional error processing methods, it demonstrates significant technical, operational and economic advantages:

[0057] 1.Technological advantages

[0058] (1) Real-time dynamic compensation: Traditional error compensation usually relies on static error models and offline data analysis. Error compensation lags behind the processing process and is difficult to cope with dynamic errors (such as vibration, thermal deformation) and real-time working condition changes. The present invention collects a variety of error data through a real-time sensor network, and combines dynamic error compensation models (such as nonlinear error compensation, Bayesian reasoning, and disturbance pattern recognition) to adjust the tool posture, feed speed, and processing path in real time during the processing. The error response speed is significantly improved, and the compensation time is shortened from the traditional minute level to millisecond level. The processing accuracy is improved by 20%-50%, meeting the processing requirements of high-precision complex parts.

[0059] (2) Multi-dimensional error coupling processing: Traditional methods usually process geometric errors, thermal errors, and dynamic errors separately, without fully considering the coupling relationship between multiple errors, resulting in limited compensation effects. The present invention uses error tracing and back propagation algorithms to analyze the error coupling relationship, combined with Bayesian networks and partition modeling methods, to dynamically adjust the compensation amount of multiple error sources and solve the complex problems caused by error superposition. It accurately compensates for the combined effects of multi-dimensional error superposition, reducing error residuals by more than 30%. It reduces the mutual amplification effect of error sources and improves overall compensation efficiency.

[0060] (3) Vibration suppression and trajectory disturbance optimization: In traditional methods, vibration problems are usually passively addressed by reducing cutting speed or adding vibration damping devices, but these measures may sacrifice processing efficiency or increase costs. The present invention dynamically adjusts the processing rhythm and tool path through real-time vibration pattern recognition (LSTM model) and active vibration suppression functions, and optimizes the compensation strategy in combination with reinforcement learning algorithms. It effectively reduces the impact of vibration on processing accuracy and surface quality, and improves surface roughness by 15%-30%. Processing stability is significantly improved under high-speed cutting, reducing the probability of deformation of thin-walled parts.

[0061] (4) Closed-loop feedback and model optimization: Traditional methods mostly use open-loop control, and the compensation results cannot be fed back to the error model, making it difficult to achieve continuous optimization of accuracy. This invention introduces a closed-loop feedback mechanism, uses the actual error data after processing for model optimization, and combines it with a deep learning algorithm to continuously improve the compensation strategy. This allows the processing accuracy to be gradually improved in repeated tasks, and the error compensation model is more robust. The system stability and adaptability are enhanced in long-term operation.

[0062] 2. Operational advantages

[0063] (1) High degree of intelligence and automation: Traditional methods rely on manual intervention to complete error correction and process parameter adjustment, which requires high operator skills and is prone to problems such as untimely adjustments or insufficient error compensation. The present invention uses sensor networks, automatic modeling, and real-time dynamic compensation technology to achieve full process automation from data acquisition to compensation execution. This significantly reduces operational complexity and reduces operator skill requirements by more than 30%. It improves the consistency and repeatability of the processing process and is suitable for mass production.

[0064] (2) Partition modeling and path optimization: Traditional methods use a single error model to process the entire workspace, ignoring the regional differences in error distribution within the space, resulting in local under-compensation or over-compensation. The present invention uses a density clustering algorithm to partition the error distribution, combines it with dynamic path planning to avoid high-error areas, and smoothes the partition boundaries. This makes the processing path more intelligent, the compensation parameters change continuously, and the error mutations at the path junctions are eliminated. Processing efficiency is improved by 15%-30%, and path optimization saves cutting time.

[0065] 3. Application and economic advantages

[0066] (1) Adaptability to complex working conditions and diverse tasks: Traditional methods are only applicable to single or fixed working conditions and cannot quickly adapt to diverse processing tasks such as complex curved surfaces, large-sized parts, and high-hardness materials. The present invention automatically adjusts the compensation strategy to adapt to diverse working conditions and task requirements through a dynamic compensation model and adaptive learning algorithm. It meets the needs of multiple fields such as complex curved surface processing (such as turbine blades), high-hardness mold cavity processing, and large-sized frame parts processing. The company's order coverage is expanded and its market competitiveness is improved.

[0067] (2) Improved production efficiency: Traditional error compensation methods require pausing processing to adjust the tool or path, which is time-consuming and labor-intensive, and reduces processing efficiency. The present invention eliminates the need to interrupt processing during dynamic real-time compensation, shortening process time through automated and intelligent operations. The processing cycle is shortened by 15%-40%, significantly improving production efficiency. Equipment utilization is increased by more than 20%.

[0068] (3) Reduced production costs: Traditional methods have high rework and scrap rates, increasing material waste and additional process costs. This invention significantly reduces rework and scrap rates through precise error compensation, reducing material waste and labor costs. The rework rate is reduced by 30%-50%, saving material and processing costs. This reduces equipment maintenance costs during long-term operation and extends the service life of machine tools by 15%-20%.

[0069] (4) Improved product quality: Traditional methods limit product quality due to their error control capabilities, making it difficult to meet the stringent requirements of the high-end market. The precise multi-dimensional error compensation of this invention enables products to meet high standards in dimensional accuracy, surface roughness, and consistency. This significantly improves product quality in high-precision fields such as aerospace and medical devices. Part consistency is enhanced during mass production, reducing the defective rate by 50%.

[0070] Summarize

[0071] Compared with traditional methods, the present invention significantly improves the machining accuracy, efficiency and adaptability of five-axis CNC machine tools through real-time, multi-dimensional and intelligent error compensation technology, reduces production costs and improves product quality, making it more competitive and economically efficient in complex and high-end manufacturing fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them:

[0073] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0074] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0075] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0076] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0077] See also Figure 1 , a multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool, comprising the following steps:

[0078] (1) Using a sensor network placed on key machine tool components to collect real-time data on geometric errors, thermal errors, dynamic errors, and coupling errors of a five-axis machine tool;

[0079] The sensor network comprises:

[0080] Laser interferometer and linear displacement sensor, among which the laser interferometer is installed along the linear guide rail, rotary axis support structure and key positioning points of the machine tool, and is used for high-precision measurement of machine tool guide rail motion errors (such as straightness, perpendicularity, parallelism, etc.) and the positioning accuracy deviation of the rotary axis. The displacement and geometric deviation measurement with nanometer-level precision is achieved through the principle of laser interference. High-precision laser interferometers provided by Shanghai Precision Instruments, Wuhan Zhongke Tianqi, Shenzhen Hyde Optics and other companies can be directly used; linear displacement sensors are arranged at key positions between the moving parts of the machine tool (such as slides, spindle boxes) and the support structure, and the displacement deviation data generated by each axis of the machine tool during movement is collected. Micron-level displacement error data is obtained in real time to provide basic data for geometric error modeling. High-precision grating scales, magnetic scales and other displacement sensors produced by Xi'an Sente, Beijing Times and other companies can be directly used.

[0081] Thermocouples and infrared thermal imagers. Among them, thermocouples are installed in key components that are easily affected by temperature, such as the spindle, motor, and machine tool bracket. They monitor the temperature changes of the machine tool and the structural deformation caused by heat diffusion, thereby providing thermal error data. They are highly accurate, low-latency, and have the ability to detect temperature gradients. There are many mature products on the market; the infrared thermal imager is fixed at an appropriate viewing angle position to monitor the temperature distribution and local hotspots of the entire machine tool, provide overall and local thermal distribution maps of the machine tool, identify the temperature field distribution and its changing trends, non-contact measurement, high resolution, and real-time output of thermal imaging data. Industrial-grade infrared thermal imagers can be provided by companies such as Hikvision, Shanghai Tektronix, and Shenzhen Ruichuang.

[0082] Acceleration sensors and vibration sensors are installed on the tool holding device and machine tool slide to detect dynamic disturbance errors. Acceleration sensors are installed on the tool holding device, machine tool slide and other moving parts to collect acceleration data during the operation of the machine tool, reflecting the vibration characteristics and dynamic disturbance errors of the machine tool components. They have high dynamic response and can capture vibration signals caused by high-speed machining. High-sensitivity acceleration sensors produced by Beijing Aerospace Measurement and Control, Suzhou Xunpeng, Nantong Zhenhua and other companies can be used. Vibration sensors are installed at key positions of the spindle and tool holding device to monitor vibration signals during machine tool operation and provide dynamic error data. They have high resolution and can identify low-frequency and high-frequency vibration components. Piezoelectric vibration sensors and MEMS vibration sensors produced by Shenyang Xinsong, Guangzhou Xike and other companies can be used.

[0083] Angle encoders are installed in multi-axis linkage parts such as rotary worktables and swing head shafts to collect angular deviation data of rotating and swinging axes, analyze angular errors in five-axis linkage motion, and provide high-resolution and high-precision real-time angular data. High-precision angle encoders provided by Guangzhou Zhongwang Longteng, Beijing Huaxia Tianxin, and Shenyang Machine Tool Supporting Company can be used.

[0084] The real-time data acquisition system uses EtherCAT or CAN bus communication protocols to connect the above sensors and transmit the collected data to the machine tool controller, achieving high real-time and multi-channel data synchronous acquisition. It supports millisecond-level data transmission rates, high bandwidth, low latency, and strong anti-interference capabilities. It is suitable for complex multi-sensor systems and uses EtherCAT or CAN bus communication modules provided by Beijing Advantech, Weikong Technology, Hollysys and other companies.

[0085] The data fusion module, embedded in the machine tool controller, uses the Kalman filter algorithm to reduce noise, correct, and classify and store various error data collected, improving data reliability and accuracy, filtering out noise interference, updating data status in real time, and maintaining high-precision data output. It can use domestic embedded systems (such as ARM architecture) equipped with open source algorithm libraries (such as filter packages in Python and MATLAB) to perform multi-sensor data fusion. The hardware uses embedded controllers such as Advantech IPC and North China Industrial Control.

[0086] The error analysis module calculates the changing trends of the geometric error, thermal error, dynamic error and coupling error of the five-axis machine tool based on sensor data, inputs the error results into the motion compensation module of the controller, extracts the error development law through time series analysis, and inputs the error results into the motion compensation module in the controller. The CNC system provided by companies such as Huazhong CNC, Guangzhou CNC (GSK), and Dazzle Laser supports customized error compensation modules. It only needs to provide a data input interface to embed the error compensation strategy into the machine tool controller.

[0087] (2) By constructing a five-axis linkage nonlinear error compensation model, the inverse error propagation algorithm is used to trace the errors generated during the machining process and dynamically optimize the compensation amount according to the error source;

[0088] The implementation method of step (2) is:

[0089] 1) The Kalman filter algorithm is used to reduce noise and preprocess the collected data. Based on the real-time acquisition of geometric error, thermal error, dynamic error, and coupling error data, noise filtering and smoothing are performed. Through iterative optimization of state estimation and observation data, noise is filtered out and high-confidence error data is obtained. Data standardization methods are used to normalize all types of error data to ensure a unified data format, facilitating subsequent error model construction and analysis.

[0090] 2) Construct a full-dimensional nonlinear error model for the five-axis linkage. Using CAE software platforms such as Huayun Data SimuWorks or CASIC's Tiangong Simulation, sparse tensor decomposition technology is used to reduce the dimensionality of the high-dimensional error tensor and extract the main error coupling relationships. The high-dimensional error data is represented as a multi-order tensor (geometric, thermal, dynamic, coupling, and other error coupling relationships). Sparse tensor decomposition is used for dimensionality reduction, extracting key error coupling features from the high-dimensional data, eliminating redundant error information, and improving data computation efficiency. Using the secondary development capabilities of the Guangzhou CNC (GSK) CNC system, a nonlinear error propagation model is used to construct the coupling relationships between error sources. Error trends are dynamically updated using real-time data to accurately reflect the error conditions of the five-axis machine tool.

[0091] 3) By constructing a differentiable error propagation function, using the Baidu PaddlePaddle or Alibaba Damo Academy MNN framework, the reverse error propagation algorithm is used to trace the source of the processing error, and adaptive weight allocation is performed according to the contribution rate of the error source, with priority given to compensating for key errors; through the chain derivation of the differentiable function, the overall error is back-propagated to each error source, and the contribution rate of each error source to the overall error is calculated.

[0092] 4) Use variational autoencoder (VAE) to generate multiple sets of error compensation strategy samples, optimize the initial compensation value through deep learning of historical processing data, use historical processing data to train variational autoencoder to generate multiple sets of error compensation strategy samples, and VAE generates initial error compensation value to shorten the model adaptation time.

[0093] 5) Use genetic algorithms (GA) to globally optimize the compensation strategy, use the machining accuracy after error compensation as the fitness function, and dynamically adjust the compensation amount; use the Huazhong CNC secondary development platform or the optimization library in the Python environment to iteratively optimize the compensation strategy to ensure optimal accuracy.

[0094] 6) The optimized compensation amount is input into the machine tool controller in real time through the edge computing module, adjusting the tool posture, feed rate, and five-axis linkage motion parameters to achieve dynamic compensation of machining errors. The controller applies the compensation amount to the machine tool execution unit in real time and verifies the compensation effect through a feedback loop. The compensation calculation and execution are integrated to achieve high response speed and real-time compensation functions. The actual machining error data after compensation is fed back to the error analysis module, dynamically updating the nonlinear error model and error compensation strategy. Combined with multi-task learning, it further optimizes error tracing and compensation accuracy, realizes closed-loop feedback optimization, and improves the long-term compensation accuracy and stability of the five-axis machine tool.

[0095] (3) Based on the Bayesian network, a causal relationship and probability distribution model of each error source is established, the joint distribution of the error sources is updated using real-time sensor data, the comprehensive error is predicted through Bayesian reasoning, and the motion parameters of each axis are adjusted;

[0096] The implementation method of step (3) is:

[0097] 1) Based on the collected data, Granger causality testing is used to identify the causal relationships between different error sources. Historical error data is used to analyze the causal relationships between geometric, thermal, dynamic, and coupled errors, identifying the primary error sources and their impact paths. Specifically, the causal relationships between primary and secondary error sources are identified by analyzing the time lag effect of a particular error source on other error changes in the historical error data. For example, an increase in spindle temperature can cause thermal deformation, which in turn affects both geometric and dynamic errors. The identified causal relationships are represented in the form of a directed acyclic graph (DAG), with nodes representing various error sources (e.g., thermal and geometric errors) and edges representing causal relationships. A greedy algorithm is then used to generate the initial topology of a Bayesian network. Using the causal testing results, the initial topology of the error sources is generated based on the Bayesian network model to describe the correlations among geometric, thermal, and dynamic errors. Specifically, based on the Bayesian network, a joint probability distribution of the error sources is defined to reflect the dependencies and variation patterns of the error sources. For example, the state distribution of a thermal error node may be determined by data from a temperature sensor and thermocouples, while the state of a dynamic error node is affected by both geometric error and vibration.

[0098] 2) A variational inference algorithm is used to optimize the joint probability distribution model of the Bayesian network to construct the causal relationship and joint probability distribution of each error source: the joint probability distribution of each node in the Bayesian network is iteratively updated through real-time error data, the main coupling characteristics between errors are extracted, the model accuracy is dynamically corrected, and the problem of high computational complexity is solved. Variational inference converts complex probability distributions into simple distribution approximations, thereby efficiently estimating the probability state of each error source.

[0099] 3) Use real-time sensor data to update the node status in the Bayesian network, and calculate the comprehensive error of the five-axis CNC machine tool under specific machining conditions based on Bayesian reasoning: Use the Kalman filter algorithm to fuse and denoise the sensor data, and update the status of each node in the Bayesian network in real time to ensure that the dynamic changes of the error source can be captured by the model.

[0100] 4) Based on the results of Bayesian reasoning, the tool position, rotation angle, and five-axis linkage motion parameters are dynamically adjusted to achieve real-time compensation of machining errors: Through real-time sensor data input, the comprehensive error of the five-axis machine tool under the current machining conditions is calculated, and the tool position, rotation angle, and motion parameters are dynamically adjusted. The machine tool's motion trajectory, feed speed, etc. are optimized in real time based on the Bayesian reasoning results to ensure that the machining path error is minimized.

[0101] 5) The compensation results are updated to the Bayesian network model through a feedback loop embedded in the machine tool controller, and the causal relationship model and error compensation strategy are optimized by combining multi-task learning. By integrating real-time data and historical processing data, deep learning technology is used to continuously improve the accuracy of the causal relationship model and the compensation effect.

[0102] (4) Based on the disturbance characteristics of the machining trajectory, an adaptive learning algorithm is used to learn the trajectory disturbance pattern in real time and dynamically update the error compensation model;

[0103] The implementation method of step (4) is:

[0104] 1) Empirical Mode Decomposition (EMD) is used to decompose the disturbance signal into several intrinsic mode functions (IMFs). Wavelet transform is then used to extract the characteristic frequencies and amplitude trends of the disturbance signal. Machining trajectory disturbance signals typically consist of low-frequency structural vibrations and high-frequency dynamic noise. EMD is used to separate these signals and obtain a clearer trajectory disturbance pattern. Wavelet transform converts the time domain signal to the frequency domain, extracting the characteristic frequencies and amplitude trends of the disturbance signal, thereby identifying the primary disturbance source of the machining trajectory.

[0105] 2) A trajectory perturbation pattern recognition model based on a long short-term memory (LSTM) network was constructed. Extracted perturbation features were input into the model, and the model parameters were dynamically updated through an online learning mechanism to adapt to changes in real-time processing conditions. The LSTM can effectively process the time series characteristics of trajectory perturbation signals and identify the changing patterns of trajectory perturbations by memorizing both short-term and long-term signal features. The online learning mechanism continuously updates the LSTM model parameters to adapt to real-time trajectory perturbation changes during the processing process, such as new perturbation patterns caused by changes in processing materials or speed.

[0106] 3) Based on the disturbance pattern recognition results, a reinforcement learning algorithm is used to optimize and adjust the five-axis motion parameters. Tool posture and motion path are optimized for compensable disturbances, while active vibration suppression is used to reduce the impact of difficult-to-compensate disturbances. Reinforcement learning dynamically adjusts the five-axis motion parameters by learning how different motion parameter combinations compensate for trajectory disturbances by defining a reward mechanism (e.g., reduction in machining error). For high-frequency disturbances that are difficult to compensate directly, active vibration suppression adjusts the feed rate to avoid resonance zones and reduce dynamic errors.

[0107] 4) Through a closed-loop feedback mechanism, compensated machining trajectory data is fed into the disturbance pattern recognition model in real time, further improving the model's adaptability to complex disturbance conditions. The compensated trajectory data contains trajectory deviations not captured by the disturbance pattern recognition model. This feedback corrects the model, further improving its recognition accuracy. Offline training combines real-time and historical data, optimizing the robustness and generalization capabilities of the disturbance pattern recognition model through reinforcement learning.

[0108] (5) Partition the machine tool workspace, calibrate the error distribution of different areas, and combine the partition compensation model with the five-axis linkage kinematic model to call the corresponding compensation parameters for each motion area in the machining path in real time;

[0109] The implementation method of step (5) is:

[0110] 1) Based on the density clustering algorithm (DBSCAN), the error data within the workspace of the five-axis machine tool is clustered and analyzed, dividing the workspace into multiple partitions with consistent error characteristics. Different areas of the machine tool workspace will produce different error distributions due to structural characteristics, thermal deformation, and machining trajectory distribution. For example, the area near the edge of the machine tool may cause increased geometric errors due to insufficient support stiffness, while the area near the spindle may cause thermal errors due to heat diffusion. Using the density clustering algorithm (DBSCAN), the spatial error data is grouped according to error characteristics and divided into multiple partitions with consistent error characteristics. The density characteristics of the error data are used to identify high-error areas and low-error areas, and the initial boundaries of the areas are established. Combining physical characteristics and error distribution, points with similar error density are divided into the same area, eliminating the problem of overlapping partitions caused by continuous error changes.

[0111] 2) A Voronoi diagram optimization algorithm is used to adjust partition boundaries to smooth the error transition regions between partitions. Based on the initial error partition boundaries, a Voronoi diagram optimization algorithm is used to generate smooth partition boundaries for the error transition regions. The Voronoi diagram is generated based on the center point of each error partition, ensuring the smoothest possible transition between boundaries and avoiding discontinuities in compensation parameters caused by sudden changes in partition boundaries. During partition optimization, priority is given to the coverage of the machining path, and more refined partition adjustments are performed on areas frequently traversed by the path to ensure consistent error distribution within the partition.

[0112] 3) A local polynomial regression method is used to construct an error distribution model within each partition, and the partition error model is dynamically updated based on real-time sensor data. Within each partition, a local polynomial regression method is used to construct an error distribution model to describe how the error within the partition changes with the machining path. The regression model inputs include error source data (such as geometric error and thermal error) and machining path parameters (such as position and speed). Based on real-time sensor data, the partition error model is dynamically corrected to reflect the real-time changes in error during the machining process. For example, when the machining path approaches a partition boundary, the model is interpolated and adjusted based on the data from adjacent partitions to reduce error fluctuations at the boundary.

[0113] 4) Based on the tool's current position and machining path, the error compensation parameters for the corresponding partition are called to ensure that the compensation amount accurately matches the error distribution in the current area. A Bezier curve interpolation algorithm is used to smooth the error compensation transition at the partition boundaries, preventing sudden changes in the error compensation amount at these boundaries. For complex paths, the compensation parameters of adjacent partitions are combined to dynamically and smoothly compensate for the error effects in the transition area.

[0114] 5) For cross-partition machining paths, a dynamic path planning algorithm is used to optimize the machining path, and a deep reinforcement learning algorithm is combined to dynamically adjust the error compensation strategy between partitions. When the machining path covers multiple partitions, the dynamic path planning algorithm optimizes path selection to avoid areas with large errors as much as possible. For areas that cannot be avoided, a combination of path optimization and parameter adjustment strategies is used to minimize the impact of errors. Through the reinforcement learning algorithm, the impact of different path selection and compensation parameter combinations on machining accuracy is continuously studied during the path planning process, optimizing the compensation strategy for cross-partition paths.

[0115] (6) The above error compensation strategy is integrated into the CNC system, and the tool posture, feed speed and processing path are adjusted through the machine tool controller to achieve dynamic compensation of multi-dimensional real-time coupling errors.

[0116] The implementation method of step (6) is:

[0117] 1) Embed a dynamic compensation module in the machine tool CNC system. This module implements real-time control of error compensation through a three-layer structure of perception layer, decision layer and execution layer. The three-layer structure of the dynamic compensation module is as follows:

[0118] Perception layer: Collects geometric error, thermal error, dynamic error and coupling error data through real-time sensors.

[0119] Decision layer: integrated error compensation models and algorithms, including nonlinear error compensation models, Bayesian reasoning models and disturbance pattern recognition models.

[0120] Execution layer: outputs compensation instructions in real time to control tool posture, feed speed and five-axis linkage motion parameters.

[0121] The dynamic compensation module is embedded in the controller through the interface of the CNC system to ensure synchronization with the CNC program, process error data in real time and update compensation instructions.

[0122] Integration method: The dynamic compensation module can communicate with the control system through an open interface (such as EtherCAT or CAN bus protocol), receive sensor data in real time and output compensation parameters.

[0123] 2) Using the fuzzy logic control algorithm, fuzzy rules are generated according to the sources and magnitudes of geometric errors, thermal errors, dynamic errors, and coupling errors, and the final error compensation is calculated through a dynamic weighted fusion method:

[0124] Define the error sources (such as position error, temperature error) and magnitudes as input variables.

[0125] The fuzzy rule base is applied to calculate the compensation amount and generate a dynamic mapping relationship between error type and compensation priority.

[0126] Output the dynamic compensation amount after fusion.

[0127] The compensation amounts for multiple error sources are weighted and integrated, prioritizing key error sources to ensure optimal overall compensation. For example, in areas where thermal and dynamic errors are coupled, priority is given to compensating for short-term, high-amplitude changes in dynamic errors.

[0128] 3) Dynamically optimize the machining path through reinforcement learning algorithms to avoid the machining path passing through areas with significant errors. Combined with the adaptive adjustment function of motion parameters, the feed speed and tool posture are adjusted in real time.

[0129] A path optimization model is established using a reinforcement learning algorithm, with improved machining path accuracy defined as the reward objective. Different path combinations are continuously tried to learn the optimal path selection strategy, avoiding paths passing through areas with significant errors.

[0130] Based on the compensation amount calculated by the error compensation module, the machining path is dynamically modified to reduce the areas affected by high errors. Combined with the dynamic programming algorithm, machining areas with smaller errors are prioritized and the path smoothness and speed distribution are adjusted in real time.

[0131] 4) Integrated active vibration suppression utilizes a predictive control algorithm to adjust the feed cadence and reduce dynamic error accumulation. Real-time vibration data is used to predict the next vibration trend, allowing the machining cadence to be adjusted in advance to avoid resonance zones. The tool's feed rate and motion trajectory frequency are adjusted to minimize machining vibration amplitude. Vibration sensor data, combined with a control algorithm, generates vibration suppression parameters, adjusting the machine tool's motion cadence in real time to offset vibration. In high-dynamic error zones, the machining speed is reduced to minimize vibration accumulation.

[0132] 5) Through a closed-loop feedback mechanism, the actual error value after processing is fed back to the dynamic compensation module. The compensation amount is updated through data comparison, gradually improving model accuracy. The fuzzy rules and compensation strategy are updated, and compensation accuracy is optimized through the real-time evaluation module. The compensation model is optimized based on feedback data, and multi-task learning is performed in conjunction with historical error data to improve the model's adaptability to complex processing conditions. During the processing, the evaluation module performs real-time quantitative analysis of the compensation effect (such as residual error distribution) and dynamically adjusts the compensation strategy.

[0133] The following are five representative examples in actual working process:

[0134] Example 1: Machining a turbine blade with complex surfaces

[0135] Task description: Process the complex curved surface of turbine blades, with a surface roughness of Ra0.4μm and a profile deviation of less than ±5μm.

[0136] Working process:

[0137] Data Acquisition: A laser interferometer monitors the linearity error of the tool path, while thermocouples and infrared thermal imagers capture temperature changes in the spindle and machining area in real time. A vibration sensor records vibration data generated by high cutting speeds, and an angle encoder detects slight deviations in the tool angle.

[0138] Modeling and Analysis: The compensation module identified thermal error as the primary source of error and also discovered that dynamic vibration had a significant impact on trajectory in localized areas. Based on the data, a coupled model of thermal and dynamic errors was established.

[0139] Partition calibration: The area near the spindle is designated as a high thermal error zone, while areas further away from the spindle have relatively lower errors. By modeling the zone errors, the compensation module generates independent compensation parameters for each zone.

[0140] Trajectory Perturbation Optimization: An LSTM model identifies vibration patterns, showing that spindle vibration is most pronounced at high cutting speeds. The system reduces cutting speeds and optimizes the tool path using a dynamic compensation algorithm.

[0141] Real-time compensation: Adjusts tool angle and path based on real-time sensor data to offset expansion caused by temperature changes and dynamic vibration.

[0142] Feedback and optimization: After processing is completed, a laser interferometer is used to measure the surface error of the blade, and the data is fed back to the compensation module to optimize the next processing strategy.

[0143] Example 2: Processing large-size aviation frame parts

[0144] Task Description: Process a 3-meter-long aviation frame part, ensuring the straightness error is less than 0.1mm.

[0145] Working process:

[0146] Data Collection: Infrared thermal imagers detect the temperature distribution in the processing area, linear displacement sensors monitor the linear displacement deviation of the guide rails, and vibration sensors collect vibration data caused by the large size of the parts.

[0147] Modeling and Analysis: The system identified thermal expansion as the primary source of linearity errors in long guide rails. Furthermore, large-scale machining resulted in insufficient support stiffness at the rail ends. Thermal and geometric error compensation models were established.

[0148] Partition modeling:

[0149] The workspace is divided into three partitions:

[0150] Central area: thermal error is significant.

[0151] End region: Insufficient stiffness leading to geometric errors.

[0152] Boundary area: mixed errors.

[0153] Path Optimization and Compensation: Dynamic path planning prioritizes avoiding areas of high error, such as reducing cutting speed in the center to minimize thermal expansion. At the end of the process, geometric errors are compensated by adjusting the position of the guide rails in real time.

[0154] Real-time execution and feedback: The compensation module adjusts the tool feed rate and path based on real-time sensor data. After machining, laser interferometer verification of straightness revealed that the error had been reduced to 0.08 mm.

[0155] Example 3: High-speed machining of thin-walled parts

[0156] Task description: Process a thin-walled part with a wall thickness of only 0.5mm. The vibration effect needs to be controlled and the surface roughness needs to reach Ra0.8μm.

[0157] Working process:

[0158] Data acquisition: The vibration sensor records the vibration amplitude of the tool in real time when machining thin walls, and the angle encoder monitors the tool's posture changes.

[0159] Modeling and Analysis: The system identifies deformation of thin-walled parts due to vibration as the primary source of error and isolates high-frequency vibration modes through disturbance signal analysis.

[0160] Disturbance pattern recognition: The LSTM model learns the characteristics of vibration signals in real time and predicts the next vibration trend.

[0161] Real-time compensation: The system reduces feed rate and adjusts cutting depth to minimize vibration amplitude. In vibration peak areas, active vibration suppression reduces dynamic errors by adjusting the machining rhythm.

[0162] Feedback and verification: After processing, the surface roughness was checked using an optical microscope, and the actual value was Ra0.75μm, which met the requirements.

[0163] Example 4: Processing a mold cavity made of high-hardness material

[0164] Task description: Processing mold cavities with a hardness of HRC60 requires controlling the impact of cutting force fluctuations on the trajectory, and the trajectory deviation must be less than ±10μm.

[0165] Working process:

[0166] Data acquisition: Laser interferometers monitor minute linear displacement changes of the guide rails, and vibration sensors capture vibrations caused by fluctuations in cutting forces.

[0167] Modeling and Analysis: A dynamic error model was developed to identify the coupling effects of cutting force fluctuations and vibrations. Thermal errors were identified as a minor error source that is only significant during high-speed machining.

[0168] Partition modeling: The mold cavity is divided into a deep cavity area (significant dynamic error) and a shallow cavity area (significant thermal error), and independent compensation models are established for each area.

[0169] Real-time compensation: In deep cavities, the system adjusts the tool feed rate to control cutting forces and optimizes tool posture to reduce vibration. In shallow cavities, the system uses temperature sensor data to adjust the path in real time to avoid dimensional deviations caused by thermal expansion.

[0170] Feedback optimization: After processing, the trajectory deviation of the mold cavity was verified by CMM (coordinate measuring machine). The actual error was ±8μm, which met the design requirements.

[0171] Example 5: Processing medical device parts

[0172] Task Description: Processing medical device parts with multiple curved surfaces requires ensuring surface continuity and profile deviation less than ±3μm.

[0173] Working process:

[0174] Data Collection: Thermocouples monitor temperature changes in the processing area, while infrared thermal imagers detect uneven heat distribution on complex curved surfaces. Vibration sensors capture dynamic vibrations caused by curvature changes in real time.

[0175] Modeling and Partitioning: Analyze the impact of thermal and dynamic errors on surface continuity. Divide the surface into high curvature areas (mainly caused by thermal errors) and low curvature areas (mainly caused by dynamic errors).

[0176] Path Optimization and Compensation: The system uses interpolation algorithms to smooth compensation parameters across regions, ensuring compensation continuity. It dynamically adjusts tool paths and reduces cutting speeds in areas of high error.

[0177] Real-time compensation: Real-time adjustment of tool posture to optimize surface curvature, combined with active vibration suppression to reduce dynamic errors.

[0178] Effect verification: Using a laser scanner to check the surface profile of the part, the actual deviation is ±2.5μm, which meets the strict requirements of medical devices.

[0179] This invention significantly improves the machining accuracy, stability, and adaptability of five-axis CNC machine tools, particularly in the high-precision machining of complex surfaces, high-hardness materials, large parts, and thin-walled parts. Its real-time compensation capabilities and intelligent operation reduce production costs, improve production efficiency, and increase product quality. It also has broad market application potential in high-end manufacturing, possessing significant economic value and industrial expansion prospects.

[0180] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool, characterized in that: The following steps are involved: (1) Using a sensor network placed on key machine tool components to collect real-time data on geometric errors, thermal errors, dynamic errors, and coupling errors of a five-axis machine tool; (2) By constructing a five-axis linkage nonlinear error compensation model, the inverse error propagation algorithm is used to trace the errors generated during the machining process and dynamically optimize the compensation amount according to the error source; (3) Based on the Bayesian network, a causal relationship and probability distribution model of each error source is established, the joint distribution of the error sources is updated using real-time sensor data, the comprehensive error is predicted through Bayesian reasoning, and the motion parameters of each axis are adjusted; (4) Based on the disturbance characteristics of the machining trajectory, an adaptive learning algorithm is used to learn the trajectory disturbance pattern in real time and dynamically update the error compensation model; (5) Partition the machine tool workspace, calibrate the error distribution of different areas, and combine the partition compensation model with the five-axis linkage kinematic model to call the corresponding compensation parameters for each motion area in the machining path in real time; (6) The above error compensation strategy is integrated into the CNC system, and the tool posture, feed speed and processing path are adjusted through the machine tool controller to achieve dynamic compensation of multi-dimensional real-time coupling errors.

2. The multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool according to claim 1, characterized in that: The sensor network comprises: Laser interferometers and linear displacement sensors are respectively arranged on the machine tool guide rails and rotary axis support positions to collect geometric error data; Thermocouples and infrared thermal imagers are respectively placed on the spindle, motor and machine tool bracket to monitor the temperature changes of key machine tool components and collect thermal error data; Acceleration sensors and vibration sensors are installed on the tool holder and machine slide to detect dynamic disturbance errors; Angle encoders are placed on the rotary table and the swing head axis to collect angle deviation data during multi-axis linkage motion. A real-time data acquisition system uses EtherCAT or CAN bus communication protocols to connect to the above sensors and transmit the collected data to the machine tool controller; The data fusion module is embedded in the machine tool controller and uses the Kalman filter algorithm to reduce noise, correct, and classify and store the collected error data. The error analysis module calculates the changing trends of the geometric error, thermal error, dynamic error and coupling error of the five-axis machine tool based on sensor data, and inputs the error results into the motion compensation module of the controller.

3. The multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool according to claim 1, characterized in that: The implementation method of step (2) is: 1) Use Kalman filter algorithm to reduce noise and preprocess the collected data; 2) Construct a full-dimensional nonlinear error model for the five-axis linkage, use sparse tensor decomposition technology to reduce the dimensionality of the high-dimensional error tensor, and extract the main error coupling relationships; 3) By constructing a differentiable error propagation function and using the inverse error propagation algorithm to trace the source of the machining error, adaptive weight allocation is performed according to the contribution rate of the error source, and critical errors are compensated first; 4) Generate multiple sets of error compensation strategy samples using a variational autoencoder and optimize the initial compensation value through deep learning of historical processing data; 5) Use genetic algorithms to globally optimize the compensation strategy, use the machining accuracy after error compensation as the fitness function, and dynamically adjust the compensation amount; 6) The optimized compensation amount is input into the machine tool controller in real time through the edge computing module to adjust the tool posture, feed speed and five-axis linkage motion parameters to achieve dynamic compensation of machining errors.

4. The multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool according to claim 1, characterized in that: The implementation method of step (3) is: 1) Based on the collected data, the causal relationship between different error sources is identified through the Granger causality test method, and the initial topology of the Bayesian network is generated using a greedy algorithm; 2) Using variational inference algorithms to optimize the joint probability distribution model of the Bayesian network to construct the causal relationship and joint probability distribution of each error source; 3) Using real-time sensor data to update the node states in the Bayesian network, the comprehensive error of the five-axis CNC machine tool under specific machining conditions is calculated based on Bayesian reasoning; 4) Based on the Bayesian inference results, the tool position, rotation angle and five-axis linkage motion parameters are dynamically adjusted to achieve real-time compensation of machining errors; 5) The compensation results are updated to the Bayesian network model through a feedback loop embedded in the machine tool controller, and the causal relationship model and error compensation strategy are optimized by combining multi-task learning.

5. The multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool according to claim 1, characterized in that: The implementation method of step (4) is: 1) The disturbance signal is decomposed into several intrinsic mode functions using empirical mode decomposition technology, and the characteristic frequency and amplitude change trend of the disturbance signal are extracted in combination with wavelet transform; 2) Construct a trajectory disturbance pattern recognition model based on a long short-term memory network, input the extracted disturbance features into the model, and dynamically update the model parameters through an online learning mechanism to adapt to changes in real-time processing conditions; 3) Based on the disturbance pattern recognition results, the reinforcement learning algorithm is used to optimize and adjust the five-axis linkage motion parameters. The tool posture and motion path are optimized for compensable disturbances, and the active vibration suppression function is used to reduce the impact of difficult-to-compensate disturbances. 4) Through the closed-loop feedback mechanism, the compensated processing trajectory data is input into the disturbance pattern recognition model in real time to further improve the model's adaptability to complex disturbance conditions.

6. The multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool according to claim 1, characterized in that: The implementation method of step (5) is: 1) Cluster analysis of the error data in the workspace of the five-axis machine tool is performed based on the density clustering algorithm, and the workspace is divided into multiple partitions with consistent error characteristics; 2) Use the Voronoi diagram optimization algorithm to adjust the partition boundaries to make the error transition area between partitions smooth; 3) Using the local polynomial regression method to construct an error distribution model in each partition, and dynamically updating the partition error model in combination with real-time sensor data; 4) According to the current position of the tool and the machining path, the error compensation parameters of the corresponding partition are called, and the error compensation transition at the partition boundary is smoothed by the Bezier curve interpolation algorithm; 5) For cross-partition processing paths, a dynamic path planning algorithm is used to optimize the processing path, and the deep reinforcement learning algorithm is combined to dynamically adjust the error compensation strategy between partitions.

7. The multi-dimensional real-time coupling error compensation method for a five-axis CNC machine tool according to claim 1, characterized in that: The implementation method of step (6) is: 1) Embed a dynamic compensation module in the machine tool CNC system. This module implements real-time control of error compensation through a three-layer structure consisting of a perception layer, a decision layer, and an execution layer. 2) Using the fuzzy logic control algorithm, fuzzy rules are generated according to the sources and magnitudes of geometric errors, thermal errors, dynamic errors, and coupling errors, and the final error compensation is calculated through a dynamic weighted fusion method; 3) Dynamically optimize the machining path through reinforcement learning algorithms to avoid the machining path passing through areas with significant errors. Combined with the adaptive adjustment function of motion parameters, the feed speed and tool posture are adjusted in real time. 4) Integrate active vibration suppression function and use predictive control algorithm to adjust the feed rhythm and reduce dynamic error superposition; 5) Through the closed-loop feedback mechanism, the actual error value after processing is fed back to the dynamic compensation module, the fuzzy rules and compensation strategy are updated, and the compensation accuracy is optimized through the real-time evaluation module.

Images