Numerical control machine tool thermal error compensation method and system based on deep learning

Abstract

The invention relates to the technical field of machine tool control, and discloses a numerical control machine tool thermal error compensation method and system based on deep learning, and the method comprises the steps: carrying out the multi-level causal mining of the multi-source time series data of a numerical control machine tool and the topological knowledge of the machine tool, and obtaining the causal features; performing deep time sequence evolution analysis on the causal features to obtain a causal feature evolution mode; based on the evolution mode, self-adaptive compensation is conducted on the thermal error of the numerical control machine tool, and a real-time control strategy is obtained; based on the numerical control machine tool thermal error compensation historical database, performing directional analysis on the real-time control strategy executable parameter to obtain a thermal error real-time control strategy target decision parameter; based on the target decision parameter, accurately correcting the control parameter of the numerical control machine tool to obtain an optimized control parameter; encoding the optimization control parameter into a numerical control machine tool optimization control instruction, and inputting the numerical control machine tool to obtain a compensated processing state; the control efficiency of thermal error compensation of the numerical control machine tool based on deep learning can be improved.

Description

A Deep Learning-Based Method and System for Thermal Error Compensation in CNC Machine Tools Technical Field

[0001] This invention relates to the field of machine tool control technology, and in particular to a method and system for thermal error compensation of CNC machine tools based on deep learning. Background Technology

[0002] With the increasing demand for high-precision machining, thermal errors generated during the operation of CNC machine tools have become one of the key factors affecting their machining accuracy. Thermal errors originate from the generation and transmission of various heat sources inside the machine tool, causing thermal deformation of the machine tool structural components. Ultimately, this causes the relative position between the tool and the workpiece to deviate from the ideal state, resulting in dimensional deviations. To ensure machining accuracy, effective prediction and compensation of thermal errors is an important issue in the field of precision manufacturing.

[0003] Existing methods cannot use the physical topology and operating mechanism of machine tools as prior knowledge to systematically guide and construct interpretable causal relationships between multi-source data. This results in a superficial understanding of the intrinsic generation and propagation mechanisms of thermal errors and a lack of ability to extract and analyze the in-depth temporal characteristics of the dynamic evolution of thermal errors. Consequently, compensation strategies cannot be proactively and adaptively adjusted based on real-time states and evolution patterns. These limitations restrict further improvements in thermal error compensation effectiveness. Therefore, improving the control efficiency of thermal error compensation in CNC machine tools has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for thermal error compensation of CNC machine tools based on deep learning, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a deep learning-based thermal error compensation method for CNC machine tools, comprising:

[0006] S1. Perform multi-level causal mining on the multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal characteristics of the CNC machine tool;

[0007] S2. Perform deep temporal evolution analysis on the causal features to obtain the evolution pattern of the causal features;

[0008] S3. Based on the evolution mode, adaptive compensation is performed on the thermal error of the CNC machine tool to obtain the real-time control strategy of the CNC machine tool;

[0009] S4. Based on the historical database of thermal error compensation for CNC machine tools, perform targeted analysis on the executable parameters of the real-time control strategy to obtain the target decision parameters of the real-time thermal error control strategy.

[0010] S5. Based on the target decision parameters, the control parameters of the CNC machine tool are precisely corrected to obtain the optimized control parameters of the CNC machine tool;

[0011] S6. Encode the optimized control parameters into optimized control instructions for the CNC machine tool, and input the optimized control instructions into the CNC machine tool to obtain the compensated machining state of the CNC machine tool.

[0012] In a preferred embodiment, the step of performing multi-level causal mining on the multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal features of the CNC machine tool includes:

[0013] Modal decomposition is performed on multi-source time-series data to obtain the modal components of CNC machine tools;

[0014] The knowledge guidance criteria for the CNC machine tool are obtained by regularizing and refining the machine tool topology knowledge.

[0015] Based on the knowledge-guided criteria, causal association mining is performed on the modal components to obtain causal pairs of the multi-source data;

[0016] Hierarchical causal verification is performed on the causal pairs, and spurious associations in the verified causal pairs are removed to construct the causal structure of the CNC machine tool.

[0017] The causal characteristics of the CNC machine tool are obtained by performing significant causal selection on the causal structure.

[0018] In a preferred embodiment, the step of performing deep temporal evolution analysis on the causal features to obtain the evolutionary pattern of the causal features includes:

[0019] The causal features are temporally embedded, and the embedded causal features are weighted by an attention mechanism to obtain the causal enhanced representation of the temporal features;

[0020] Multi-scale time series analysis was performed on the causal reinforcement representation to obtain the evolution trend of the causal reinforcement representation;

[0021] The mutation characterization states in the evolutionary trend are organized and arranged in chronological order to obtain the key state sequence of the causal characteristics;

[0022] By connecting the key state sequences, the evolution trajectory of the causal features is obtained;

[0023] The evolutionary trajectory is abstracted to obtain the evolutionary pattern of the causal features.

[0024] In a preferred embodiment, the step of adaptively compensating for the thermal error of the CNC machine tool based on the evolution pattern to obtain a real-time control strategy for the CNC machine tool includes:

[0025] High-dimensional temporal state analysis is performed on the evolutionary pattern to obtain the key evolutionary features of the evolutionary pattern;

[0026] Based on the aforementioned key evolutionary features, the thermal error compensation trigger node of the CNC machine tool is identified to obtain the identification information of the thermal error compensation trigger node.

[0027] Based on the identification information, extract the relevant strategies associated with the thermal error compensation trigger node from the historical compensation records;

[0028] Based on the thermal state data and operating state parameters of the CNC machine tool, the relevant strategies are adaptively screened to obtain the adaptation strategy for the CNC machine tool.

[0029] The adaptation strategy is converted into online instructions to obtain the real-time control strategy of the CNC machine tool.

[0030] In a preferred embodiment, the calculation formula for the adaptation strategy is as follows:

[0031] ;

[0032] In the formula, This is the optimal adaptation strategy for the CNC machine tool. To perform the maximum value operation, For a single candidate compensation strategy. For deep learning-driven policy adaptability evaluation operators, This is the joint state vector of the CNC machine tool. The non-negative constraint strength coefficient. Candidate compensation strategy Comprehensive constraint quantitative indicators, The non-negative real-time reward coefficient. It is a natural exponential function. Candidate compensation strategy The actual execution delay time, This serves as the baseline threshold for policy execution latency. This is a non-linear, real-time reward item.

[0033] In a preferred embodiment, the step of performing targeted analysis on the executable parameters of the real-time control strategy based on the historical database of thermal error compensation for CNC machine tools to obtain the target decision parameters of the real-time thermal error control strategy includes:

[0034] The feature vector of the time-series aligned historical compensation record is obtained by performing feature dimension mapping on the time-series aligned historical compensation record of the CNC machine tool thermal error compensation historical database.

[0035] Based on the feature vector, the executable parameters of the real-time control strategy are adaptively matched and evaluated to obtain the performance fit characterization of the real-time control strategy.

[0036] The performance matching characteristics are sorted by parameter priority to obtain the position arrangement list of the real-time control strategy;

[0037] The positional arrangement list is optimized by performing parameter optimization to obtain the core strategy parameters of the positional arrangement list;

[0038] The core parameters of the strategy are fused to obtain the target decision parameters of the real-time thermal error control strategy.

[0039] In a preferred embodiment, the step of precisely correcting the control parameters of the CNC machine tool based on the target decision parameters to obtain optimized control parameters for the CNC machine tool includes:

[0040] Multi-level attention allocation is performed on the target decision parameters to obtain the attention weight distribution of the target decision parameters;

[0041] Based on the attention weight distribution, the control parameters of the CNC machine tool are adjusted by weighting to obtain preliminary correction parameters for the control parameters;

[0042] The consistency of the time series composed of the preliminary correction parameters is verified to obtain the verification result of the preliminary correction parameters.

[0043] Based on the verification results, the preliminary correction parameters are optimized by error feedback to obtain the optimized control parameters of the CNC machine tool.

[0044] In a preferred embodiment, the step of weighting and adjusting the control parameters of the CNC machine tool based on the attention weight distribution to obtain preliminary correction parameters for the control parameters includes:

[0045] Read the control parameters of the CNC machine tool;

[0046] The attention weight distribution is weighted and fused with the control parameters to obtain the weighted result of the current control parameters;

[0047] Based on the weighted result, the current control parameters are overridden to obtain preliminary correction parameters for the control parameters;

[0048] In a preferred embodiment, encoding the optimized control parameters into optimized control instructions for the CNC machine tool and inputting the optimized control instructions into the CNC machine tool to obtain the compensated machining state of the CNC machine tool includes:

[0049] The optimized control parameters are structured and encapsulated to obtain a data block of the optimized control parameters;

[0050] The data block is encoded into optimized control instructions for the CNC machine tool, and the optimized control instructions are arranged to obtain the optimized control instruction sequence for the CNC machine tool.

[0051] The optimized control command sequence is imported into the control terminal of the CNC machine tool to obtain the real-time control strategy of the CNC machine tool;

[0052] According to the real-time control strategy, the CNC machine tool is controlled to perform production processing, and the processing status feedback of the CNC machine tool is obtained;

[0053] By integrating the machining status feedback of the CNC machine tool, the compensated machining status of the CNC machine tool is obtained.

[0054] To address the above problems, this invention also provides a deep learning-based thermal error compensation system for CNC machine tools, the system comprising:

[0055] The multi-source deep causal mining module is used to perform multi-level causal mining on multi-source time-series data and machine tool topology knowledge of CNC machine tools to obtain the causal features of the CNC machine tools.

[0056] The causal feature deep evolution module is used to perform deep temporal evolution analysis on the causal features to obtain the evolution mode of the causal features;

[0057] A thermal error compensation strategy module is used to adaptively compensate for the thermal error of the CNC machine tool based on the evolution mode, so as to obtain the real-time control strategy of the CNC machine tool.

[0058] The parameter depth orientation analysis module is used to perform orientation analysis on the executable parameters of the real-time control strategy based on the historical database of thermal error compensation of CNC machine tools, so as to obtain the target decision parameters of the real-time thermal error control strategy.

[0059] The control parameter precision correction module is used to precisely correct the control parameters of the CNC machine tool based on the target decision parameters, so as to obtain the optimized control parameters of the CNC machine tool.

[0060] The control instruction encoding and execution module is used to encode the optimized control parameters into optimized control instructions for the CNC machine tool, and input the optimized control instructions into the CNC machine tool to obtain the compensated machining state of the CNC machine tool.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. This invention extracts the topology and operating principle of machine tools into clear knowledge-guiding criteria, systematically guiding the causal mining and hierarchical verification of multi-source time-series data. This process ensures that the discovered causal relationships are not only based on statistical data patterns, but also strictly conform to physical common sense and engineering logic, thereby constructing a causal feature system that combines data credibility and physical interpretability. This fundamentally solves the shortcomings of traditional methods that rely on statistical correlation and cannot identify false relationships, providing a solid and reliable basis for a thorough understanding of the generation and propagation mechanism of thermal errors.

[0063] 2. This invention performs multi-scale time-series analysis and evolutionary pattern abstraction on causal characteristics, thereby deeply capturing the dynamic laws governing the evolution of thermal errors with operating conditions and time. Based on this evolutionary pattern, the system can assess the current state in real time and predict future trends, thus generating a forward-looking adaptive compensation strategy. This transforms thermal error compensation from traditional passive response and hysteresis correction to proactive intervention and precise adjustment based on state prediction, significantly improving the timeliness, accuracy, and overall control efficiency of compensation. Attached Figure Description

[0064] Figure 1 is a flowchart illustrating a deep learning-based thermal error compensation method for CNC machine tools according to an embodiment of the present invention.

[0065] Figure 2 is a functional block diagram of a deep learning-based CNC machine tool thermal error compensation system provided in an embodiment of the present invention;

[0066] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0067] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0068] This application provides a deep learning-based method for thermal error compensation in CNC machine tools. The execution entity of this deep learning-based CNC machine tool thermal error compensation method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the deep learning-based CNC machine tool thermal error compensation method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0069] Referring to Figure 1, a flowchart illustrating a deep learning-based thermal error compensation method for CNC machine tools according to an embodiment of the present invention is shown. In this embodiment, the deep learning-based thermal error compensation method for CNC machine tools includes:

[0070] S1. Perform multi-level causal mining on the multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal characteristics of the CNC machine tool.

[0071] In this embodiment of the invention, the step of performing multi-level causal mining on the multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal features of the CNC machine tool includes:

[0072] Modal decomposition is performed on multi-source time-series data to obtain the modal components of CNC machine tools;

[0073] The knowledge guidance criteria for the CNC machine tool are obtained by regularizing and refining the machine tool topology knowledge.

[0074] Based on the knowledge-guided criteria, causal association mining is performed on the modal components to obtain causal pairs of the multi-source data;

[0075] Hierarchical causal verification is performed on the causal pairs, and spurious associations in the verified causal pairs are removed to construct the causal structure of the CNC machine tool.

[0076] The causal characteristics of the CNC machine tool are obtained by performing significant causal selection on the causal structure.

[0077] The multi-source time-series data collected during the operation of CNC machine tools are processed. These data include various types of continuous time-series data such as temperature data, vibration data, speed data, and load data of various machine tool components. By decomposing these complex mixed data according to their inherent frequency characteristics and fluctuation patterns, the signals from different sources and of different properties that were originally intertwined are separated into independent basic data units with clear physical meanings. These basic data units are the modal components of the CNC machine tool.

[0078] Machine tool topology knowledge encompasses the connection methods, transmission paths, structural layouts, functional attributes, and working principles of various machine tool components. By systematically organizing and summarizing this scattered topology information, clarifying the interaction logic, energy transfer paths, and signal feedback mechanisms between components, and transforming this organized information into unified and clear judgment criteria and behavioral guidelines, these criteria and principles constitute the knowledge guidance guidelines for CNC machine tools.

[0079] Based on the extracted knowledge-guided criteria, the mutual influence relationships between each modal component are analyzed one by one. Combining the interaction logic of components in machine tool topology, the variation law of physical quantities corresponding to different modal components is observed. It is determined whether the change of one modal component will directly or indirectly cause the regular change of another modal component. Two modal components that have been verified by the knowledge criteria and have a clear mutual influence relationship are paired to form causal pairs of multi-source data.

[0080] According to the structural hierarchy and operational logic of the machine tool, each causal pair is verified at different levels. Under different operating conditions and load conditions of the machine tool, the consistency and correlation stability of the changes of the two modal components in the causal pair are continuously observed. For those correlations that only appear under specific accidental conditions, do not conform to the machine tool topology knowledge and operating rules, and cannot be repeatedly verified under multiple operating conditions, they are judged as false correlations and are eliminated. The real causal pairs verified at each level are organized in an orderly manner according to the structural framework and operational logic of the machine tool to construct an overall framework that can comprehensively reflect the real causal relationship between the multi-source data of the CNC machine tool, that is, the causal structure of the CNC machine tool.

[0081] A comprehensive analysis of the constructed causal structure is conducted to evaluate the degree of influence of each causal pair on thermal errors during machine tool operation, the frequency of occurrence under different operating conditions, and the stability of its effect. The focus is on screening out those causal relationships that play a key role in the formation of machine tool thermal errors, can exist stably under various operating conditions, and have significant effects. These causal relationships with core influence are taken as the core content, and finally the causal characteristics of CNC machine tools are formed.

[0082] The beneficial effects include simplifying the complexity of multi-source time-series data, providing a regular and effective data foundation, ensuring that deep learning accurately processes machine tool operation data, providing clear guidance for causal mining, avoiding blind spots, enabling deep learning to organically combine with machine tool structural and functional knowledge, improving the rationality of mining results, establishing a bridge for multi-source data association, providing basic features for causal meaning, helping deep learning understand the intrinsic connections of data, ensuring the authenticity and reliability of causal relationships, reducing interference from invalid information, improving the training efficiency and effectiveness of deep learning models, focusing on core influencing factors, reducing the computational complexity of deep learning, and improving the efficiency and accuracy of thermal error compensation analysis and decision-making.

[0083] S2. The causal features are subjected to deep temporal evolution analysis to obtain the evolution pattern of the causal features.

[0084] In this embodiment of the invention, the causal features are temporally embedded, and the embedded causal features are weighted by an attention mechanism to obtain the causal enhanced representation represented by the temporal features;

[0085] Multi-scale time series analysis was performed on the causal reinforcement representation to obtain the evolution trend of the causal reinforcement representation;

[0086] The mutation characterization states in the evolutionary trend are organized and arranged in chronological order to obtain the key state sequence of the causal characteristics;

[0087] By connecting the key state sequences, the evolution trajectory of the causal features is obtained;

[0088] The evolutionary trajectory is abstracted to obtain the evolutionary pattern of the causal features.

[0089] Three different time spans—short, medium, and long—were selected as the analytical scales. The short time scale focused on minute-level changes in causal enhancement characteristics, the medium time scale focused on hour-level changes, and the long time scale focused on day-level trends. By observing, comparing, and analyzing causal enhancement characteristics at these three different time scales, we comprehensively captured the overall direction and pattern of change over time, and thus obtained the evolutionary trend of causal enhancement characteristics.

[0090] First, the evolutionary trend is monitored time by time. When a significant change is found in the causal enhancement representation that is significantly different from the state at previous consecutive time points, the representation state at that time point is determined as a mutation representation state. Then, these mutation representation states are arranged in strict chronological order according to their actual occurrence time to form an ordered sequence set, which is the key state sequence of causal features.

[0091] Each mutation in the key state sequence is represented as an independent node. Based on the temporal logical relationship between these nodes, all nodes are connected sequentially by a continuous path to form a continuous path that can fully reflect the change process of causal features from the first key state to the last key state. This path is the evolutionary trajectory of causal features.

[0092] By conducting a comprehensive and detailed observation of the evolutionary trajectory, identifying common characteristics such as state transition patterns and repetitive key states, and then summarizing, refining, and simplifying these representative common characteristics, and removing irrelevant details from the trajectory, a general pattern expression that can accurately reflect the changing patterns of causal characteristics is ultimately formed. This pattern is the evolutionary pattern of causal characteristics.

[0093] The beneficial effects are as follows: Leveraging deep learning's ability to process time-series data, and through weighted processing via temporal embedding and attention mechanisms, core temporal information related to thermal errors is highlighted, laying the foundation for subsequent evolutionary pattern analysis. Utilizing deep learning's advantage in capturing multi-dimensional information, multi-scale time-series analysis comprehensively covers changes across different time spans, avoiding information omissions and making evolutionary trends more accurate. Based on deep learning's sensitive capture of abnormal changes in time-series data, key nodes in the evolution of causal features are accurately located, ensuring that key state sequences fully reflect important changes and providing precise support for constructing evolutionary trajectories. With deep learning's modeling capabilities for sequence data, the evolutionary trajectory formed by path connections clearly presents the correlation between key states, providing an intuitive and complete analytical basis for pattern abstraction. By leveraging deep learning's ability to summarize data patterns, pattern abstraction transforms complex trajectories into general patterns, facilitating subsequent thermal error compensation decisions and improving the scientific rigor and efficiency of strategy formulation.

[0094] S3. Based on the evolution mode, adaptive compensation is performed on the thermal error of the CNC machine tool to obtain the real-time control strategy of the CNC machine tool.

[0095] In this embodiment of the invention, the step of adaptively compensating for the thermal error of the CNC machine tool based on the evolution mode to obtain a real-time control strategy for the CNC machine tool includes:

[0096] High-dimensional temporal state analysis is performed on the evolutionary pattern to obtain the key evolutionary features of the evolutionary pattern;

[0097] Based on the aforementioned key evolutionary features, the thermal error compensation trigger node of the CNC machine tool is identified to obtain the identification information of the thermal error compensation trigger node.

[0098] Based on the identification information, extract the relevant strategies associated with the thermal error compensation trigger node from the historical compensation records;

[0099] Based on the thermal state data and operating state parameters of the CNC machine tool, the relevant strategies are adaptively screened to obtain the adaptation strategy for the CNC machine tool.

[0100] The adaptation strategy is converted into online instructions to obtain the real-time control strategy of the CNC machine tool.

[0101] The specific calculation formula for the adaptation strategy is as follows:

[0102] ;

[0103] In the formula, This is the optimal adaptation strategy for the CNC machine tool. To perform the maximum value operation, For a single candidate compensation strategy. For deep learning-driven policy adaptability evaluation operators, This is the joint state vector of the CNC machine tool. The non-negative constraint strength coefficient. Candidate compensation strategy Comprehensive constraint quantitative indicators, The non-negative real-time reward coefficient. It is a natural exponential function. Candidate compensation strategy The actual execution delay time, This serves as a baseline threshold for policy execution latency. This is a non-linear, real-time reward item.

[0104] The evolutionary model is comprehensively analyzed for time-related information, and is broken down into layers according to time and state dimensions. State data corresponding to each time node is collected one by one. By comparing the differences in state data at different time nodes, state information that can directly affect the change of thermal error is selected. This selected information is systematically integrated and summarized to form key evolutionary features that can accurately reflect the core change law of the evolutionary model.

[0105] The key evolutionary features obtained are compared one by one with the preset thermal error compensation trigger conditions. These preset trigger conditions are based on the summary of typical scenarios in the historical operation of CNC machine tools where thermal errors occur and compensation is required. When one or more of the key evolutionary features meet the preset trigger conditions, the corresponding node can be identified as a thermal error compensation trigger node. Then, a unique identifier containing core information such as its occurrence time and the corresponding thermal error-related state is assigned to the trigger node to ensure that the node can be accurately distinguished and identified.

[0106] Based on the identification information of the thermal error compensation trigger node, a precise search is performed in the database that stores complete records of past thermal error compensation. During the search, the core content of the identification information is used as the matching basis to find all thermal error compensation records in the historical records that are consistent with or highly related to the identification information of the current trigger node. The corresponding compensation strategies are completely extracted from these matched records and integrated to form a set of relevant strategies.

[0107] Comprehensive data on the current thermal state of the CNC machine tool is collected, including real-time temperature and temperature change rate of each core component. At the same time, the current operating parameters of the machine tool are acquired, covering processing speed, load, and runtime. These real-time collected data are compared in a comprehensive and detailed manner with the historical application data of each strategy in the relevant strategy set. The applicability of each relevant strategy under the current operating conditions is analyzed. Strategies that differ significantly from the current operating conditions and cannot meet the current thermal error compensation requirements are eliminated. Strategies that can accurately match the current operating conditions and effectively solve the current thermal error problem are retained and identified as suitable strategies.

[0108] The determined adaptation strategy is broken down in detail, clarifying each specific execution step and operation requirement. These execution steps and operation requirements are then transformed into standardized operation instructions that the CNC machine tool control system can directly recognize and execute. According to the control logic and normal operation process of the CNC machine tool, these standardized instructions are arranged in a reasonable order to ensure smooth connection and no conflicts between instructions. Subsequently, the integrity of the arranged instructions is verified to confirm that all operations required to implement the adaptation strategy have been transformed into corresponding instructions without omission, and that the instruction format fully meets the receiving requirements of the control system. Finally, a real-time control strategy that can be directly issued to the CNC machine tool is formed.

[0109] In the calculation formula of the adaptation strategy, This is the optimal adaptation strategy for CNC machine tools, used to directly guide the adaptive thermal error compensation operation, match the best compensation requirements under the current working conditions, and ensure that the compensation effect is accurate and meets the standards. The operation that maximizes the value is used to select the best-performing strategy from all candidate compensation strategies, thus avoiding the selection of ineffective strategies. Each candidate compensation strategy is derived from a set of strategies associated with the thermal error compensation trigger node in historical compensation records. These strategies have all been validated in practical applications and possess a certain degree of reliability. This is a deep learning-driven policy fit evaluation operator used to quantitatively evaluate the degree of fit between candidate policies and the current machine tool state. A higher value indicates a stronger fit. It is a joint state vector for CNC machine tools, which integrates core information such as the current thermal state data and operating state parameters of the machine tool. It is the basic data support for adaptability assessment and ensures that the assessment results are consistent with the actual working conditions. It is a non-negative constraint strength coefficient, which is determined based on factors such as the hardware performance limitations of the machine tool, processing technology requirements, and energy consumption control standards. It is used to adjust the influence of the comprehensive constraint quantification index on strategy selection and avoid selecting strategies that exceed the machine tool's carrying capacity. Candidate compensation strategy The comprehensive constraint quantification index is calculated by comprehensively considering factors such as energy consumption costs, equipment wear and tear, and process compatibility levels required during strategy execution. It reflects the constraint costs of strategy execution; a higher value indicates more restrictive conditions for strategy execution. This is a non-negative real-time reward coefficient, set according to the machine tool's response speed requirements for thermal error compensation. The more urgent the compensation requirement, the larger the coefficient value, used to strengthen the weight of the impact of strategy execution delay on the selection result. This is a natural exponential function used to perform non-linear processing on real-time reward items, making the reward effect more closely match actual needs. Candidate compensation strategy The actual execution delay time is obtained by statistically analyzing the entire process from instruction issuance to execution completion, directly reflecting the execution speed of the strategy. The baseline threshold for strategy execution delay is determined by comprehensively considering the optimal compensation response time in the machine tool's historical operation and the maximum allowable delay time corresponding to the machining accuracy requirements. This serves as the core standard for measuring whether the strategy execution delay meets the target. It is a non-linear, real-time reward item, when Less than When this value is close to 1, the reward effect is significant. Less than As τs increases, this value gradually decreases, and the reward effect weakens. This guides the system to prioritize strategies with shorter execution delays. The overall formula calculates the suitability evaluation value of each candidate compensation strategy, deducts the constraint cost of its execution, and adds the real-time reward. Finally, it selects the strategy with the highest comprehensive score as the optimal suitability strategy, which ensures the suitability of the strategy with the current working conditions and takes into account the feasibility and real-time performance of the strategy execution.

[0110] S4. Based on the historical database of thermal error compensation for CNC machine tools, perform targeted analysis on the executable parameters of the real-time control strategy to obtain the target decision parameters of the real-time thermal error control strategy.

[0111] In this embodiment of the invention, the step of performing targeted analysis on the executable parameters of the real-time control strategy based on the historical database of thermal error compensation for CNC machine tools to obtain the target decision parameters of the real-time thermal error control strategy includes:

[0112] The feature vector of the time-series aligned historical compensation record is obtained by performing feature dimension mapping on the time-series aligned historical compensation record of the CNC machine tool thermal error compensation historical database.

[0113] Based on the feature vector, the executable parameters of the real-time control strategy are adaptively matched and evaluated to obtain the performance fit characterization of the real-time control strategy.

[0114] The performance matching characteristics are sorted by parameter priority to obtain the position arrangement list of the real-time control strategy;

[0115] The positional arrangement list is optimized by performing parameter optimization to obtain the core strategy parameters of the positional arrangement list;

[0116] The core parameters of the strategy are fused to obtain the target decision parameters of the real-time thermal error control strategy.

[0117] The historical compensation records in the CNC machine tool thermal error compensation historical database, which are organized in chronological order, are processed. These historical records contain information such as thermal error data of CNC machine tools under different working conditions, executed compensation parameters, and machining status after compensation. Various types of information in each historical compensation record, such as temperature changes during machine tool operation, speed data, type and value of compensation parameters, and error improvement after compensation, are transformed into feature information of a unified dimension. This results in each historical compensation record corresponding to a vector containing these feature information, thus obtaining the feature vector of the time-aligned historical compensation records.

[0118] The executable parameters of the real-time control strategy are the various control parameters to be executed that are currently planned to compensate for the thermal error of the CNC machine tool, such as the planned adjustment of the machine tool speed, temperature control range, and machining interval time. According to the feature dimension mapping method determined in the first step, these executable parameters are transformed into corresponding feature vectors. Then, these vectors are compared one by one with the feature vectors of historical compensation records to analyze the fit between the current executable parameters and the parameters in each historical record in terms of working condition adaptability and expected compensation effect. For example, whether the temperature rise rate under the current working condition is similar to the temperature rise rate of a certain historical record, and whether the adjustment direction of the current executable parameters is consistent with the adjustment direction of the effective compensation parameters in the historical record. Through such comparative analysis, an effectiveness fit characterization that can intuitively reflect the adaptability of the current executable parameters is formed.

[0119] The performance fit characterization encompasses information such as the degree of fit between each executable parameter and historical effective compensation records, the expected thermal error reduction effect, and the operational difficulty during actual execution. These information are ranked according to their importance to the final thermal error compensation effect. For example, if the performance fit characterization of an executable parameter shows that it is highly compatible with the current operating conditions and can significantly reduce thermal error, then that parameter has a higher priority than other parameters. If the performance fit characterization of a parameter shows that its compatibility is poor and the expected compensation effect is weak, then its priority is lower. All executable parameters are arranged from high to low according to this rule, forming an ordered ranking list.

[0120] For the high-priority parameters ranked at the top of the sequence list, and considering the real-time operating status of the CNC machine tool, including the current machining temperature, operating speed, type of workpiece being machined, and machining progress, we analyze the actual operability of each high-priority parameter and its key role in the overall thermal error compensation effect. We eliminate parameters that, although high in priority, cannot be executed under the current operating conditions or contribute very little to the overall compensation effect after execution. We then select parameters that can directly determine the core effect of thermal error compensation, are highly adapted to the current operating conditions, and can be effectively executed. These parameters are the core parameters of the strategy.

[0121] The core parameters of the strategy involve different key aspects of thermal error compensation for CNC machine tools, which may include multiple types such as speed adjustment, temperature control threshold, and machining rhythm adjustment. The interaction between these core parameters is comprehensively considered. For example, adjusting a certain speed parameter may cause a change in the rate of temperature change of the machine tool, thereby affecting the execution effect of the temperature control parameter. These core parameters are integrated and analyzed to coordinate the adaptation relationship between the parameters, clarify the specific value and execution order of each core parameter, and finally form a set of unified and coordinated target decision parameters that can maximize the effect of thermal error compensation.

[0122] The beneficial effects include providing a standardized data foundation for deep learning, facilitating the efficient use of historical data, improving the standardization and efficiency of data processing, capturing feature adaptation patterns through deep learning to make performance evaluation more accurate and objective, analyzing multi-dimensional features with deep learning to accurately identify key parameters, eliminate invalid interference, and improve the targeting and efficiency of parameter processing, mining correlation patterns between operating conditions and historical data with deep learning to accurately select core parameters and ensure that parameters are adapted to current operating conditions, and coordinating the correlation of core parameters through deep learning to form coordinated and effective target decision parameters, thereby improving the scientific nature of thermal error compensation strategies.

[0123] S5. Based on the target decision parameters, the control parameters of the CNC machine tool are precisely corrected to obtain the optimized control parameters of the CNC machine tool.

[0124] In this embodiment of the invention, the step of precisely correcting the control parameters of the CNC machine tool based on the target decision parameters to obtain the optimized control parameters of the CNC machine tool includes:

[0125] Multi-level attention allocation is performed on the target decision parameters to obtain the attention weight distribution of the target decision parameters;

[0126] Based on the attention weight distribution, the control parameters of the CNC machine tool are adjusted by weighting to obtain preliminary correction parameters for the control parameters;

[0127] The consistency of the time series composed of the preliminary correction parameters is verified to obtain the verification result of the preliminary correction parameters.

[0128] Based on the verification results, the preliminary correction parameters are optimized by error feedback to obtain the optimized control parameters of the CNC machine tool.

[0129] The step of weighting and adjusting the control parameters of the CNC machine tool based on the attention weight distribution to obtain preliminary correction parameters for the control parameters includes:

[0130] Read the control parameters of the CNC machine tool;

[0131] The attention weight distribution is weighted and fused with the control parameters to obtain the weighted result of the current control parameters;

[0132] Based on the weighted result, the current control parameters are overridden to obtain preliminary correction parameters for the control parameters.

[0133] When allocating attention to target decision parameters in multiple levels, different levels are first divided according to the dimensions such as the thermal error compensation links and the scope of influence associated with the parameters. Each level focuses on the target decision parameters within the corresponding range. Then, the correlation between each parameter and the thermal error compensation effect in the historical compensation data analyzed by deep learning, as well as the actual performance of each parameter under the current machine tool operating state, are combined to determine the importance of each target decision parameter in the corresponding level, thereby forming a complete attention weight distribution. This ensures that the weight allocation can accurately reflect the priority of each parameter's influence on the correction of control parameters.

[0134] Through the control terminal data interface of the CNC machine tool, all control parameters currently in use during the machine tool's operation can be retrieved in real time, including the machine tool's speed setting, feed rate, temperature regulation related settings, and all other operation control data related to thermal error compensation, ensuring that the read control parameters can completely and accurately reflect the current control status of the machine tool.

[0135] The previously obtained attention weight distribution is matched one by one with the read control parameters. Each control parameter is matched with its corresponding weight value. Then, the parameters are combined according to the relationship between their values ​​and corresponding weights. The higher the weight value of a parameter, the greater its influence on the final result during the combination process. Through this corresponding combination method, the weighted processing result of each control parameter is obtained.

[0136] Based on the weighted fusion results of each parameter, the control parameters currently in use by the CNC machine tool are replaced one by one. Each control parameter is replaced by its corresponding weighted result. The replacement process strictly follows the principle of one-to-one correspondence, without omitting any relevant control parameter, and finally forms the preliminary correction parameters.

[0137] The preliminary correction parameters are arranged in chronological order to form a complete time sequence. Then, the changes in the values ​​of the preliminary correction parameters at adjacent time points in the sequence are checked to determine whether the magnitude of the changes conforms to the objective laws of machine tool operation. At the same time, it is checked whether there are any conflicts between the values ​​of different parameters or any mismatch with the current thermal state and operating requirements of the machine tool, so as to fully verify the consistency and rationality of the parameter sequence.

[0138] Based on the results of the consistency check, if the preliminary correction parameters are found to have numerical abrupt changes, contradictions, or do not meet the actual operating requirements of the machine tool, the problematic parameters are fine-tuned in a targeted manner, referring to the historical compensation and optimization experience accumulated by deep learning and the current real-time thermal error data of the machine tool. This ensures that each parameter can be accurately adapted to the current operating status and thermal error of the machine tool, and finally obtains optimized control parameters that meet the thermal error compensation requirements.

[0139] The beneficial effects include: accurately identifying key parameters for thermal error compensation, meeting the needs of deep learning feature extraction and priority classification, providing scientific guidance for control parameter correction, acquiring real-time machine tool control status data to avoid correction deviations, laying a reliable data foundation for subsequent steps, fully considering the importance of each parameter to avoid indiscriminate adjustments, improving the scientific nature and pertinence of control parameter correction, rapidly updating control parameters, efficiently forming preliminary correction results, providing clear adjustment targets for subsequent verification and optimization, identifying unreasonable parameters to avoid affecting the thermal error compensation effect, ensuring the reliability and rationality of correction parameters, addressing parameter issues in a targeted manner, and combining deep learning experience with real-time data fine-tuning to improve the accuracy and adaptability of control parameters and ensure the thermal error compensation effect.

[0140] S6. Encode the optimized control parameters into optimized control instructions for the CNC machine tool, and input the optimized control instructions into the CNC machine tool to obtain the compensated machining state of the CNC machine tool.

[0141] In this embodiment of the invention, the step of encoding the optimized control parameters into optimized control instructions for the CNC machine tool, and inputting the optimized control instructions into the CNC machine tool to obtain the compensated machining state of the CNC machine tool includes:

[0142] The optimized control parameters are structured and encapsulated to obtain a data block of the optimized control parameters;

[0143] The data block is encoded into optimized control instructions for the CNC machine tool, and the optimized control instructions are arranged to obtain the optimized control instruction sequence for the CNC machine tool.

[0144] The optimized control command sequence is imported into the control terminal of the CNC machine tool to obtain the real-time control strategy of the CNC machine tool;

[0145] According to the real-time control strategy, the CNC machine tool is controlled to perform production processing, and the processing status feedback of the CNC machine tool is obtained;

[0146] By integrating the machining status feedback of the CNC machine tool, the compensated machining status of the CNC machine tool is obtained.

[0147] When encapsulating optimized control parameters in a structured manner, the scattered optimized control parameters are classified and organized according to the functional logic of each control module of the CNC machine tool. For example, parameters related to spindle operation, feed system, and cooling system are classified separately. At the same time, each parameter is labeled with a clear functional identifier and associated component information, ultimately forming a data set with a fixed hierarchical structure and a unified format, namely the data block of optimized control parameters. This ensures that the data block contains all the key parameters required to achieve thermal error compensation, and that the logical relationships between the parameters are clearly identifiable.

[0148] When encoding data blocks into optimized control instructions for CNC machine tools, the machine language encoding standard commonly used in the CNC machine tool industry is adopted. Based on the type, function, and value of each parameter in the data block, it is converted one by one into the corresponding binary machine instruction, ensuring that each instruction can be accurately recognized and parsed by the control unit of the CNC machine tool. During the instruction arrangement process, the machining process and execution logic of the CNC machine tool are strictly followed. The encoded individual instructions are arranged sequentially according to the order of "parameter initialization instruction—core control instruction—auxiliary execution instruction—status monitoring instruction," forming a continuous and orderly sequence of optimized control instructions, avoiding conflicts in instruction execution order or overlapping functions.

[0149] When importing optimized control command sequences into the CNC machine tool control terminal, a data transmission channel is established through the standard communication interface provided by the CNC machine tool. During transmission, a data verification mechanism is employed to verify the integrity and accuracy of each command sequence segment, preventing data loss or tampering. After successfully receiving the command sequence, the control terminal stores it in a dedicated execution buffer and performs a secondary verification. Once confirmed to be error-free, the import operation is completed, preparing for the subsequent generation and execution of real-time control strategies.

[0150] When controlling CNC machine tool production according to real-time control strategies, the control terminal retrieves an optimized control command sequence from the execution buffer and sends control signals to each execution component one by one according to the command arrangement order. This drives components such as the spindle, feed axis, and cooling system to work collaboratively according to the optimized parameters. During machining, various sensors on the CNC machine tool continuously collect real-time data, including the actual temperature of each component, the tool feed position, and the dimensional accuracy of the machined workpiece. This data is fed back to the control terminal in real time, forming complete machining status feedback information.

[0151] When integrating CNC machine tool machining status feedback, the control terminal categorizes and filters all collected feedback data, eliminating invalid data caused by momentary sensor interference and retaining valid data that accurately reflects the machining status. The valid data is then summarized and statistically analyzed from multiple dimensions, including the effect of thermal error compensation, machine tool operational stability, and machining accuracy compliance, to clarify the actual machining performance of the machine tool after compensation. Finally, a comprehensive and detailed post-compensation machining status report is generated, fully presenting the machining results after thermal error compensation.

[0152] The beneficial effects include: accurately identifying key parameters, meeting the needs of deep learning feature extraction, providing scientific guidance for control parameter correction, acquiring real control data in real time to avoid correction deviations, laying a reliable data foundation, taking into account the importance of parameters to avoid indiscriminate adjustments, improving the scientific nature and pertinence of corrections, quickly forming preliminary correction results, providing clear adjustment targets for subsequent verification and optimization, identifying unreasonable parameters, ensuring the reliability and rationality of correction parameters, avoiding affecting the compensation effect, addressing parameter issues in a targeted manner, and improving parameter accuracy and adaptability by combining deep learning experience with real-time data to ensure the compensation effect.

[0153] Figure 2 shows a functional block diagram of a deep learning-based CNC machine tool thermal error compensation system provided in an embodiment of the present invention.

[0154] The deep learning-based CNC machine tool thermal error compensation system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the deep learning-based CNC machine tool thermal error compensation system 100 may include a multi-source deep causal mining module 101, a causal feature deep evolution module 102, a thermal error compensation strategy module 103, a parameter deep orientation analysis module 104, a control parameter precise correction module 105, and a control instruction encoding and execution module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0155] In this embodiment, the functions of each module / unit are as follows:

[0156] The multi-source deep causal mining module 101 is used to perform multi-level causal mining on the multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal features of the CNC machine tool.

[0157] The causal feature deep evolution module 102 is used to perform deep temporal evolution analysis on the causal features to obtain the evolution mode of the causal features;

[0158] The thermal error compensation strategy module 103 is used to adaptively compensate for the thermal error of the CNC machine tool based on the evolution mode, so as to obtain the real-time control strategy of the CNC machine tool.

[0159] The parameter depth orientation analysis module 104 is used to perform orientation analysis on the executable parameters of the real-time control strategy based on the CNC machine tool thermal error compensation historical database, so as to obtain the target decision parameters of the thermal error real-time control strategy.

[0160] The control parameter precision correction module 105 is used to precisely correct the control parameters of the CNC machine tool based on the target decision parameters, so as to obtain the optimized control parameters of the CNC machine tool.

[0161] The control instruction encoding and execution module 106 is used to encode the optimized control parameters into optimized control instructions for the CNC machine tool, and input the optimized control instructions into the CNC machine tool to obtain the compensated machining state of the CNC machine tool.

[0162] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0163] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0164] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0165] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0166] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning-based method for thermal error compensation in CNC machine tools, characterized in that, The method includes: S1, performing multi-level causal mining on multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal characteristics of the CNC machine tool; S2, performing deep time-series evolution analysis on the causal characteristics to obtain the evolution mode of the causal characteristics; S3, based on the evolution mode, performing adaptive compensation for the thermal error of the CNC machine tool to obtain the real-time control strategy of the CNC machine tool; S4, based on the CNC machine tool thermal error compensation historical database, performing directional analysis on the executable parameters of the real-time control strategy to obtain the target decision parameters of the thermal error real-time control strategy; S5, based on the target decision parameters, performing precise correction on the control parameters of the CNC machine tool to obtain the optimized control parameters of the CNC machine tool; S6, encoding the optimized control parameters into optimized control instructions for the CNC machine tool, and inputting the optimized control instructions into the CNC machine tool to obtain the post-compensation machining state of the CNC machine tool.

2. The deep learning-based CNC machine tool thermal error compensation method as described in claim 1, characterized in that, The method of performing multi-level causal mining on multi-source time-series data and machine tool topology knowledge of CNC machine tools to obtain causal features of the CNC machine tools includes: performing modal decomposition on the multi-source time-series data of the CNC machine tools to obtain modal components of the CNC machine tools; performing regularization and refinement on the machine tool topology knowledge to obtain knowledge guidance criteria for the CNC machine tools; performing causal association mining on the modal components based on the knowledge guidance criteria to obtain causal pairs of the multi-source time-series data; performing hierarchical causal verification on the causal pairs and removing false associations in the verified causal pairs to construct the causal structure of the CNC machine tools; and performing significant causal selection on the causal structure to obtain the causal features of the CNC machine tools.

3. The deep learning-based CNC machine tool thermal error compensation method as described in claim 1, characterized in that, The step of performing deep temporal evolution analysis on the causal features to obtain the evolutionary pattern of the causal features includes: temporally embedding the causal features and weighting the embedded causal features using an attention mechanism to obtain a causal enhancement representation represented by the temporal features; performing multi-scale temporal analysis on the causal enhancement representation to obtain the evolutionary trend of the causal enhancement representation; organizing and arranging the mutation representation states in the evolutionary trend according to time order to obtain the key state sequence of the causal features; connecting the key state sequence to obtain the evolutionary trajectory of the causal features; and abstracting the evolutionary trajectory to obtain the evolutionary pattern of the causal features.

4. The deep learning-based CNC machine tool thermal error compensation method as described in claim 1, characterized in that, The method of adaptively compensating for the thermal error of the CNC machine tool based on the evolutionary pattern to obtain the real-time control strategy of the CNC machine tool includes: performing high-dimensional temporal state analysis on the evolutionary pattern to obtain the key evolutionary features of the evolutionary pattern; based on the key evolutionary features, performing key identification on the thermal error compensation trigger node of the CNC machine tool to obtain the identification information of the thermal error compensation trigger node; extracting relevant strategies associated with the thermal error compensation trigger node from historical compensation records according to the identification information; adaptively screening the relevant strategies based on the thermal state data and operating state parameters of the CNC machine tool to obtain the adaptation strategy of the CNC machine tool; and online commanding the adaptation strategy to obtain the real-time control strategy of the CNC machine tool.

5. The deep learning-based CNC machine tool thermal error compensation method as described in claim 4, characterized in that, The specific calculation formula for the adaptation strategy is as follows: In the formula, This is the optimal adaptation strategy for the CNC machine tool. To perform the maximum value operation, For a single candidate compensation strategy. For deep learning-driven policy adaptability evaluation operators, This is the joint state vector of the CNC machine tool. The non-negative constraint strength coefficient. Candidate compensation strategy Comprehensive constraint quantitative indicators, The non-negative real-time reward coefficient. It is a natural exponential function. Candidate compensation strategy The actual execution delay time, This serves as a baseline threshold for policy execution latency. This is a non-linear, real-time reward item.

6. The CNC machine tool thermal error compensation method based on deep learning as described in claim 1, characterized in that, The method involves using a historical database of thermal error compensation for CNC machine tools to perform targeted analysis on the executable parameters of the real-time control strategy, thereby obtaining the target decision parameters of the real-time thermal error control strategy. This includes: mapping the time-series aligned historical compensation records in the historical database of thermal error compensation for CNC machine tools to obtain feature vectors for the time-series aligned historical compensation records; performing adaptive matching evaluation on the executable parameters of the real-time control strategy based on the feature vectors to obtain the performance fit characterization of the real-time control strategy; prioritizing the parameters of the performance fit characterization to obtain the positional arrangement list of the real-time control strategy; optimizing the parameters of the positional arrangement list to obtain the core strategy parameters of the positional arrangement list; and performing decision fusion on the core strategy parameters to obtain the target decision parameters of the real-time thermal error control strategy.

7. The deep learning-based CNC machine tool thermal error compensation method as described in claim 1, characterized in that, The step of precisely correcting the control parameters of the CNC machine tool based on the target decision parameters to obtain optimized control parameters for the CNC machine tool includes: performing multi-level attention allocation on the target decision parameters to obtain an attention weight distribution of the target decision parameters; performing weighted adjustment on the control parameters of the CNC machine tool based on the attention weight distribution to obtain preliminary corrected parameters of the control parameters; performing consistency verification on the time sequence formed by the preliminary corrected parameters to obtain a verification result of the preliminary corrected parameters; and performing error feedback optimization on the preliminary corrected parameters based on the verification result to obtain optimized control parameters for the CNC machine tool.

8. The deep learning-based CNC machine tool thermal error compensation method as described in claim 7, characterized in that, The step of adjusting the control parameters of the CNC machine tool based on the attention weight distribution to obtain preliminary correction parameters for the control parameters includes: reading the control parameters of the CNC machine tool; weighting and fusing the attention weight distribution with the control parameters to obtain a weighted result of the current control parameters; and applying parameter overlay to the current control parameters according to the weighted result to obtain preliminary correction parameters for the control parameters.

9. The deep learning-based CNC machine tool thermal error compensation method as described in claim 1, characterized in that, The step of encoding the optimized control parameters into optimized control instructions for the CNC machine tool and inputting the optimized control instructions into the CNC machine tool to obtain the post-compensation machining state of the CNC machine tool includes: structuring the optimized control parameters to obtain data blocks of the optimized control parameters; encoding the data blocks into optimized control instructions for the CNC machine tool and arranging the optimized control instructions to obtain an optimized control instruction sequence for the CNC machine tool; importing the optimized control instruction sequence into the control terminal of the CNC machine tool to obtain a real-time control strategy for the CNC machine tool; controlling the CNC machine tool to perform production processing according to the real-time control strategy to obtain processing status feedback of the CNC machine tool; and integrating the processing status feedback of the CNC machine tool to obtain the post-compensation machining state of the CNC machine tool.

10. A CNC machine tool thermal error compensation system based on deep learning, characterized in that, To implement the deep learning-based CNC machine tool thermal error compensation method of claim 1, the system comprises: a multi-source deep causal mining module, used to perform multi-level causal mining on multi-source time-series data and machine tool topology knowledge of the CNC machine tool to obtain the causal features of the CNC machine tool; a causal feature deep evolution module, used to perform deep time-series evolution analysis on the causal features to obtain the evolution mode of the causal features; a thermal error compensation strategy module, used to adaptively compensate the thermal error of the CNC machine tool based on the evolution mode to obtain the real-time control strategy of the CNC machine tool; and a parameter deep directional solution. The system comprises the following modules: an analysis module for performing targeted analysis on the executable parameters of the real-time control strategy based on the historical database of thermal error compensation for CNC machine tools, to obtain the target decision parameters of the real-time thermal error control strategy; a control parameter precision correction module for precisely correcting the control parameters of the CNC machine tool based on the target decision parameters, to obtain the optimized control parameters of the CNC machine tool; and a control instruction encoding and execution module for encoding the optimized control parameters into optimized control instructions for the CNC machine tool, and inputting the optimized control instructions into the CNC machine tool to obtain the post-compensation machining state of the CNC machine tool.

Images