Digital-twin-based adaptive control system and method for numerical control machine tool

Abstract

This invention discloses an adaptive control system and method for CNC machine tools based on digital twins, relating to the field of machine tool control. It includes: a model building module, used to construct a synchronously mapped digital twin model in virtual space based on the physical entity of the CNC machine tool, the three-dimensional model of the workpiece to be processed, and a preset tool processing path, and to discretize the processing path into several processing monitoring time-series points; and a data acquisition module, used to acquire vibration and temperature signals at the tool processing location in real time at each time-series point through vibration and temperature sensors during the processing of the physical machine tool. This invention perceives the processing status in real time and uses a trained risk prediction model to perform in-depth analysis of the fused features, enabling it to predict the drill bit deformation risk coefficient at a specific future time point. The system can anticipate risks before the drill bit actually undergoes significant deformation or damage, greatly avoiding production interruptions and workpiece scrap losses caused by sudden failures.

Description

Technical Field

[0001] This invention relates to the field of machine tool control technology, specifically to an adaptive control system and method for CNC machine tools based on digital twins. Background Technology

[0002] As the manufacturing industry moves towards intelligent manufacturing, CNC machine tools, as core equipment, face higher demands on the stability and quality consistency of their machining processes. In critical processes such as drilling and milling, the cutting tools, as direct actuators, are highly susceptible to minute deformations or damage under high-load conditions due to factors such as vibration, temperature rise, and wear. These microscopic changes are difficult to capture in real time using traditional monitoring methods, yet they can directly lead to workpiece dimensional deviations, decreased surface quality, or even tool breakage and machine downtime, severely impacting production efficiency and product qualification rates. Digital twin technology, as a key enabling technology for the deep integration of the physical and information worlds, provides an innovative solution for constructing real-time interactive virtual machine tool models and, based on this, achieving intelligent prediction and proactive control of the machining process.

[0003] The existing control systems of CNC machine tools have the following main shortcomings: The control methods are mostly simple feedback control based on fixed process parameters, lacking the ability to perceive and predict the intrinsic state of the machining process, such as tool health and cutting conditions, in real time and with precision. The system usually only responds passively after an anomaly occurs, and cannot provide early warning and intervention in the early stages of tool performance degradation or deterioration of working conditions. Existing monitoring methods only monitor spindle power or vibration in a single direction, failing to integrate vibration spectrum, temperature field distribution, and visual characteristics of chip morphology for comprehensive diagnosis, resulting in low accuracy in identifying complex failure modes. The lack of a virtual environment that can accurately map physical entities makes the optimization and adjustment of control strategies heavily reliant on on-site trial and error, making it difficult to achieve truly adaptive optimization control. Summary of the Invention

[0004] (a) Technical problems to be solved In view of the above-mentioned shortcomings of the existing technology, the present invention provides an adaptive control system and method for CNC machine tools based on digital twins, which can effectively solve the problems of the existing technology.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention discloses an adaptive control system and method for CNC machine tools based on digital twins, comprising: The model building module is used to construct a synchronously mapped digital twin model in virtual space based on the physical entity of the CNC machine tool, the 3D model of the workpiece to be processed, and the preset tool processing path, and to discretize the processing path into several processing monitoring time sequence points; The data acquisition module is used to acquire vibration and temperature signals at the tool processing point in real time at various time points during the processing of physical machine tools through vibration and temperature sensors, and to acquire sputtering image data of the processing area simultaneously through a high-speed industrial camera. The feature extraction module is used to perform time-frequency domain analysis on vibration and temperature signals, extract vibration and temperature features, and perform image processing on sputtered image data to extract the directional distribution and range coverage data of the sputtered material. Based on a preset sputtering mode threshold, it determines whether there is a sputtering anomaly. If so, it extracts the abnormal sputtering features. The risk prediction module is used to construct a risk prediction model with vibration characteristics, temperature characteristics, and abnormal sputtering characteristics as input features. The model is trained using historical machining data, integrates and analyzes multi-source features, and predicts and outputs the tool deformation risk coefficient at a specified future time point. The coefficient represents the quantified probability value of the tool undergoing plastic bending or micro-chipping at the specified future time point. After the machining of a single workpiece is completed, the actual measured workpiece hole depth, hole diameter, and surface roughness are compared with the final state predicted by the digital twin model to calculate multi-dimensional errors. Using this error data and the corresponding full-process time-series feature data, the weight parameters of the risk prediction model are fine-tuned through backpropagation, and the adjustment command amplitude in the decision generation module is optimized through reinforcement learning to achieve the self-evolution of the system. The decision generation module is used to receive the tool deformation risk coefficient and generate an adaptive adjustment command for the tool at the future time point according to the preset risk level and action mapping strategy. The adjustment command includes the tool feed rate compensation amount and the spindle speed adjustment amount. The parameter control module is used to calculate the output cooling water flow adjustment coefficient based on the tool deformation risk coefficient and the current temperature characteristics using a preset fuzzy control algorithm. The instruction execution module is used to convert adaptive adjustment instructions and flow adjustment coefficients into control signals that can be recognized by the CNC machine tool controller, and then send them to the actuators of the physical machine tool.

[0006] Furthermore, the vibration sensor in the data acquisition module is an accelerometer mounted on the spindle box or turret, used to acquire vibration spectra containing multiple axes; the temperature sensor is an infrared thermal imager, used to acquire the temperature field distribution of the tool working part; the sputtering image data is acquired by an industrial camera with an acquisition frame rate of not less than twice the machining spindle speed, and is equipped with a protective cover to prevent cutting fluid interference and a light source with a preset wavelength.

[0007] Furthermore, the vibration features in the feature extraction module include the root mean square value and peak factor in the time domain, as well as the amplitude at the principal axis rotation frequency obtained by fast Fourier transform and the energy proportion of a specific high-frequency band in the frequency domain; the temperature features are the instantaneous temperature rise rate and average temperature of the drill tip region.

[0008] Furthermore, the feature extraction module's logic for judging sputtering anomalies is as follows: The acquired sputtering image data is preprocessed, including grayscale conversion, noise reduction, and dynamic threshold segmentation, in order to separate the moving sputtering pixel set from the background; Connectivity analysis was performed on the segmented binary image to calculate the total sputtering coverage area at each time point. Based on the spatial coordinates of all sputtering pixels, principal component analysis was used to calculate the mainstream direction angle and its distribution dispersion of the sputtering. The overall coverage area, mainstream direction angle, and distribution dispersion of the sputtered material at the current point are compared with the pre-generated benchmark sputtering model. The benchmark sputtering model is obtained through statistical learning based on historical data of the same tool and the same material under normal processing conditions. It defines the average value of the sputtering area, the expected value of the direction angle, and the threshold range of its allowable fluctuation for each time series point. If the total area of ​​the sputtered material at the current location exceeds ±30% of the area threshold of the corresponding location in the baseline model, the mainstream direction angle deviates from the expected direction by more than 45 degrees, or the dispersion of the direction distribution exceeds the preset dispersion threshold, then the sputtering anomaly at that location is determined to have occurred. When an anomaly is detected, the extraction of the abnormal sputtering features is triggered. These features include the abrupt change in area of ​​the abnormal sputtering, the abrupt change in direction angle, and the image texture feature vector of the abnormal region.

[0009] Furthermore, the construction process of the risk prediction model in the risk prediction module is as follows: The design incorporates a multi-input branch neural network structure. The first branch combines a one-dimensional convolutional layer with a long short-term memory network layer to capture temporal dependencies and frequency domain patterns for vibration feature sequences. The second branch uses a one-dimensional convolutional layer to extract local trends for temperature feature sequences. The third branch uses a two-dimensional convolutional layer to process the spatial information of anomalous sputtering features. The high-dimensional features extracted from each branch are then concatenated in a fusion layer. Collect full-time-series sensor data of historical processing and corresponding actual deformation risk labels of the cutting tools to form a training dataset; use mean square error as the loss function, adopt adaptive moment estimation algorithm to conduct end-to-end supervised training of the model, and use the validation set to stop early to prevent overfitting until the model can establish a mapping relationship from multi-source time-series features to future deformation risks. Vibration feature sequence, temperature feature sequence and image feature vector representing sputtering anomaly are used as independent input channels. After local feature fusion and dimensionality reduction through convolutional layer and pooling layer, global feature synthesis is performed in fully connected layer, and finally a tool deformation risk coefficient between 0 and 1 is output.

[0010] Furthermore, the preset process for the risk level and action mapping strategy in the decision generation module is as follows: When the risk coefficient is below the first threshold, no adjustment command is generated; when the risk coefficient is between the first and second thresholds, a command to reduce the feed rate by 10%-20% is generated; when the risk coefficient is above the second threshold, a compound command to reduce the feed rate by 20%-30% and the spindle speed by 5%-15% is generated, and a warning signal is sent to the designated human-machine interface.

[0011] Furthermore, the fuzzy control algorithm in the parameter control module takes the tool deformation risk coefficient and the current drill tip temperature as input variables and the cooling water flow rate adjustment coefficient as output variables; it defines fuzzy sets of low, medium, and high risk, as well as low temperature, normal temperature, and high temperature, and presets a fuzzy rule base. Through the flow rate adjustment coefficient output by the defuzzification operation, it realizes the adaptive adjustment of cooling intensity.

[0012] Furthermore, the instruction execution module is interconnected with an update feedback module via a wireless network. The update feedback module is used to highlight the stress concentration area and thermal load distribution of the tool in the virtual model using color gradient or deformation amplification based on the data acquired by the data acquisition module and the execution results of the instruction execution module, and based on the measured vibration and temperature data; dynamically simulate the chip formation and removal process on the virtual workpiece surface based on the sputtering image analysis results; and update the tool's motion trajectory and process parameters in the virtual model in real time according to the issued control commands.

[0013] Furthermore, the model building module and the data acquisition module are interconnected via a wireless network, the feature extraction module is interconnected with the data acquisition module and the risk prediction module via a wireless network, the risk prediction module is interconnected with the decision generation module via a wireless network, and the decision generation module is interconnected with the parameter control module and the instruction execution module via a wireless network.

[0014] The second invention discloses an adaptive control method for CNC machine tools based on digital twins, comprising the following steps: Step 1: Construct a digital twin model that is synchronously mapped to the physical CNC machine tool and the workpiece to be processed, and discretize the preset tool processing path into several processing monitoring time sequence points; Step 2: During the physical machining process, the vibration signal, temperature signal and sputtering image data of the machining area are collected in real time at each time point; Step 3: Process the vibration and temperature signals to extract vibration and temperature features; process the sputtered image data to extract sputtering direction and range data, and determine whether there are sputtering anomalies. If so, extract the abnormal sputtering features. Step 4: Input the vibration characteristics, temperature characteristics and abnormal sputtering characteristics into the pre-trained convolutional neural network prediction model. The model outputs the predicted value of the tool deformation risk coefficient at a specified time point in the future. Step 5: Based on the tool deformation risk coefficient, generate adaptive adjustment instructions for the tool at future time points, and dynamically calculate the cooling water flow adjustment coefficient based on the risk coefficient and current temperature characteristics. Step 6: Send the adaptive adjustment command and flow adjustment coefficient to the physical CNC machine tool for execution, and synchronously drive the digital twin model to update its status. At the same time, based on the deviation between the actual machining results and the model prediction, iteratively optimize the prediction model and control strategy.

[0015] (III) Beneficial Effects Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: 1. By constructing a high-fidelity digital twin model and integrating multi-source sensor data such as vibration, temperature, and vision, the processing status can be perceived in real time. By using a trained risk prediction model to perform in-depth analysis of the integrated features, the risk coefficient of drill bit deformation at a specific point in the future can be predicted in advance. The system can anticipate risks before the drill bit actually undergoes significant deformation or damage, greatly avoiding production interruptions and workpiece scrap losses caused by sudden failures.

[0016] 2. By adopting a preset risk level and action mapping strategy and fuzzy control algorithm based on the predicted risk coefficient, the system intelligently generates adaptive adjustment commands including feed rate, spindle speed compensation, and cooling water flow coefficient. Based on the quantitative evaluation of the current and future states, the control parameters can be dynamically matched with the actual machining conditions and tool status, maximizing machining efficiency while ensuring machining quality. Furthermore, by optimizing cooling and load, the tool life can be effectively extended. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a schematic diagram of the framework of the adaptive control system for CNC machine tools in this invention.

[0019] The labels in the diagram represent: 1. Model building module; 2. Data acquisition module; 3. Feature extraction module; 4. Risk prediction module; 5. Decision generation module; 6. Parameter control module; 7. Instruction execution module; 8. Update feedback module. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] The present invention will be further described below with reference to embodiments.

[0022] Example 1 This embodiment presents an adaptive control system for CNC machine tools based on digital twins, such as... Figure 1 As shown, it includes: Model building module 1 is used to construct a synchronously mapped digital twin model in virtual space based on the physical entity of the CNC machine tool, the three-dimensional model of the workpiece to be processed, and the preset tool processing path, and to discretize the processing path into several processing monitoring time sequence points.

[0023] Data acquisition module 2 is used to acquire vibration and temperature signals at the tool machining point in real time at various time points during the machining process of the physical machine tool through vibration and temperature sensors, and to simultaneously acquire sputtering image data of the machining area through a high-speed industrial camera; the vibration sensor is an accelerometer mounted on the spindle box or tool turret, used to acquire vibration spectrum containing multiple axes; the temperature sensor is an infrared thermal imager, used to acquire the temperature field distribution of the tool working part; the sputtering image data is acquired through an industrial camera, with an acquisition frame rate of not less than twice the machining spindle speed, and is equipped with a protective cover to prevent cutting fluid interference and a light source with a preset wavelength.

[0024] Feature extraction module 3 is used to perform time-frequency domain analysis on vibration and temperature signals, extract vibration and temperature features, and simultaneously perform image processing on sputtering image data to extract the directional distribution and range coverage data of the sputtering. Based on a preset sputtering mode threshold, it determines whether there are sputtering anomalies. If so, it extracts the abnormal sputtering features. Vibration features include the root mean square value and peak factor in the time domain, and the amplitude at the principal axis rotation frequency obtained by fast Fourier transform and the energy proportion of specific high-frequency bands in the frequency domain. Temperature features are the instantaneous temperature rise rate and average temperature of the drill tip area. The logic for judging sputtering anomalies is as follows: The acquired sputtering image data is preprocessed, including grayscale conversion, noise reduction, and dynamic threshold segmentation, in order to separate the moving sputtering pixel set from the background; Connectivity analysis was performed on the segmented binary image to calculate the total sputtering coverage area at each time point. Based on the spatial coordinates of all sputtering pixels, principal component analysis was used to calculate the mainstream direction angle and its distribution dispersion of the sputtering. The overall coverage area, mainstream direction angle, and distribution dispersion of the sputtered material at the current point are compared with the pre-generated benchmark sputtering model. The benchmark sputtering model is obtained through statistical learning based on historical data of the same tool and the same material under normal processing conditions. It defines the average value of the sputtering area, the expected value of the direction angle, and the threshold range of its allowable fluctuation for each time point. If the total area of ​​the sputtered material at the current location exceeds ±30% of the area threshold of the corresponding location in the baseline model, the mainstream direction angle deviates from the expected direction by more than 45 degrees, or the dispersion of the direction distribution exceeds the preset dispersion threshold, then the sputtering anomaly at that location is determined to have occurred. When an anomaly is detected, the extraction of abnormal sputtering features is triggered. These features include the abrupt change in area of ​​the abnormal sputtering, the abrupt change in direction angle, and the image texture feature vector of the abnormal region.

[0025] Risk prediction module 4 is used to construct a risk prediction model with vibration characteristics, temperature characteristics, and abnormal sputtering characteristics as input features. The model is trained through historical machining data, integrates and analyzes multi-source features, and predicts and outputs the tool deformation risk coefficient at a specified future time point. The coefficient represents the quantified probability value of the tool undergoing plastic bending or micro-chipping at the specified future time point. After the machining of a single workpiece is completed, the actual measured workpiece hole depth, hole diameter, and surface roughness are compared with the final state predicted by the digital twin model to calculate multi-dimensional errors. Using this error data and the corresponding full-process time-series feature data, the weight parameters of the risk prediction model are fine-tuned through backpropagation, and the adjustment command amplitude in the decision generation module 5 is optimized through reinforcement learning to achieve the self-evolution of the system. The process of constructing a risk prediction model is as follows: The design incorporates a multi-input branch neural network structure. The first branch combines a one-dimensional convolutional layer with a long short-term memory network layer to capture temporal dependencies and frequency domain patterns for vibration feature sequences. The second branch uses a one-dimensional convolutional layer to extract local trends for temperature feature sequences. The third branch uses a two-dimensional convolutional layer to process the spatial information of anomalous sputtering features. The high-dimensional features extracted from each branch are then concatenated in a fusion layer. Collect full-time-series sensor data of historical processing and corresponding actual deformation risk labels of the cutting tools to form a training dataset; use mean square error as the loss function, adopt adaptive moment estimation algorithm to conduct end-to-end supervised training of the model, and use the validation set to stop early to prevent overfitting until the model can establish a mapping relationship from multi-source time-series features to future deformation risks. Vibration feature sequence, temperature feature sequence and image feature vector representing sputtering anomaly are used as independent input channels. After local feature fusion and dimensionality reduction through convolutional layer and pooling layer, global feature synthesis is performed in fully connected layer, and finally a tool deformation risk coefficient between 0 and 1 is output.

[0026] Decision generation module 5 receives the tool deformation risk coefficient and, based on the preset risk level and motion mapping strategy, generates adaptive adjustment instructions for the tool at future time points. These instructions include tool feed rate compensation and spindle speed adjustment. The preset process for the risk level and motion mapping strategy is as follows: When the risk coefficient is below the first threshold, no adjustment command is generated; when the risk coefficient is between the first and second thresholds, a command to reduce the feed rate by 10%-20% is generated; when the risk coefficient is above the second threshold, a compound command to reduce the feed rate by 20%-30% and the spindle speed by 5%-15% is generated, and a warning signal is sent to the designated human-machine interface.

[0027] The parameter control module 6 is used to calculate and output the cooling water flow adjustment coefficient based on the tool deformation risk coefficient and the current temperature characteristics using a preset fuzzy control algorithm. The fuzzy control algorithm takes the tool deformation risk coefficient and the current drill tip temperature as input variables and the cooling water flow adjustment coefficient as output variables. It defines fuzzy sets for low, medium, and high risk, as well as low temperature, normal temperature, and high temperature, and presets a fuzzy rule base. Through defuzzification calculation, the flow adjustment coefficient output achieves adaptive adjustment of cooling intensity.

[0028] The instruction execution module 7 is used to convert the adaptive adjustment instruction and flow adjustment coefficient into control signals that can be recognized by the CNC machine tool controller, and send them to the actuator of the physical machine tool.

[0029] The instruction execution module 7 is connected to the update feedback module 8 via a wireless network. The update feedback module 8 is used to highlight the stress concentration area and thermal load distribution of the tool in the virtual model by color gradient or deformation magnification based on the data obtained by the data acquisition module 2 and the execution results of the instruction execution module 7, and based on the measured vibration and temperature data; dynamically simulate the chip formation and discharge process on the virtual workpiece surface based on the sputtering image analysis results; and update the tool motion trajectory and process parameters in the virtual model in real time according to the issued control commands.

[0030] Model building module 1 and data acquisition module 2 are interconnected via a wireless network. Feature extraction module 3, data acquisition module 2, and risk prediction module 4 are interconnected via a wireless network. Risk prediction module 4 and decision generation module 5 are interconnected via a wireless network. Decision generation module 5, parameter control module 6, and instruction execution module 7 are interconnected via a wireless network.

[0031] Compared with existing technologies, this embodiment integrates sputtering images with vibration and temperature features by constructing a word twin model and uses convolutional neural networks for temporal correlation analysis, thereby enabling the prediction of drill bit deformation risks. The system can not only generate adaptive machining parameter adjustment instructions, but also link the cooling system for nonlinear precise control, improving the intelligence level, process stability and tool life of the machining process, and fundamentally avoiding quality accidents and downtime losses caused by sudden tool failure.

[0032] Example 2 At other levels, this embodiment also provides another optimization mechanism based on Embodiment 1, specifically a digital twin-based adaptive control method for CNC machine tools, including the following steps: Step 1: Construct a digital twin model that is synchronously mapped to the physical CNC machine tool and the workpiece to be processed, and discretize the preset tool processing path into several processing monitoring time sequence points; Step 2: During the physical machining process, the vibration signal, temperature signal and sputtering image data of the machining area are collected in real time at each time point; Step 3: Process the vibration and temperature signals to extract vibration and temperature features; process the sputtered image data to extract sputtering direction and range data, and determine whether there are sputtering anomalies. If so, extract the abnormal sputtering features. Step 4: Input the vibration characteristics, temperature characteristics and abnormal sputtering characteristics into the pre-trained convolutional neural network prediction model. The model outputs the predicted value of the tool deformation risk coefficient at a specified time point in the future. Step 5: Based on the tool deformation risk coefficient, generate adaptive adjustment instructions for the tool at future time points, and dynamically calculate the cooling water flow adjustment coefficient based on the risk coefficient and current temperature characteristics. Step 6: Send the adaptive adjustment command and flow adjustment coefficient to the physical CNC machine tool for execution, and synchronously drive the digital twin model to update its status. At the same time, based on the deviation between the actual machining results and the model prediction, iteratively optimize the prediction model and control strategy.

[0033] In summary, this invention achieves precise prediction of future drill bit deformation risk by finely sensing and extracting features of multi-dimensional states such as vibration, temperature and sputtering during the processing, and using sputtering anomalies as a key signal for predicting drill bit deformation risk. By introducing a risk prediction model, the system can perform deep fusion analysis on massive time-series feature data, thereby achieving accurate prediction of drill bit deformation risk and breaking through the limitations of traditional threshold-based alarm-based lagging control. The system can autonomously generate and execute composite adjustment commands that include drill bit motion parameters and cooling flow based on the predicted risk coefficient. The risk prediction model is continuously optimized based on actual feedback. This system fundamentally improves the intelligence level of CNC machine tools under complex working conditions.

[0034] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive control system for CNC machine tools based on digital twins, characterized in that, include: The model building module is used to construct a synchronously mapped digital twin model in virtual space based on the physical entity of the CNC machine tool, the 3D model of the workpiece to be processed, and the preset tool processing path, and to discretize the processing path into several processing monitoring time sequence points; The data acquisition module is used to acquire vibration and temperature signals at the tool processing location in real time at various time points during the machining process of the physical machine tool, and simultaneously acquire sputtering image data of the processing area. The feature extraction module is used to perform time-frequency domain analysis on vibration and temperature signals, extract vibration and temperature features, and perform image processing on sputtered image data to extract the directional distribution and range coverage data of the sputtered material. Based on a preset sputtering mode threshold, it determines whether there is a sputtering anomaly. If so, it extracts the abnormal sputtering features. The risk prediction module is used to build a risk prediction model with vibration characteristics, temperature characteristics and abnormal sputtering characteristics as input features. The model predicts and outputs the tool deformation risk coefficient at a specified time point in the future. The decision generation module is used to receive the tool deformation risk coefficient and generate adaptive adjustment instructions for the tool at the future time sequence points according to the preset risk level and action mapping strategy. The parameter control module is used to calculate the output cooling water flow adjustment coefficient based on the tool deformation risk coefficient and the current temperature characteristics using a preset fuzzy control algorithm. The instruction execution module is used to convert adaptive adjustment instructions and flow adjustment coefficients into control signals that can be recognized by the CNC machine tool controller, and then send them to the actuators of the physical machine tool.

2. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The vibration sensor in the data acquisition module is an accelerometer mounted on the spindle box or turret, used to acquire vibration spectra containing multiple axes; the temperature sensor is an infrared thermal imager, used to acquire the temperature field distribution of the tool's working part; the sputtered image data is acquired through an industrial camera.

3. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The vibration features in the feature extraction module include the root mean square value and peak factor in the time domain, and the amplitude at the principal axis rotation frequency obtained by fast Fourier transform and the energy proportion of a specific high-frequency band in the frequency domain; the temperature features are the instantaneous temperature rise rate and average temperature of the drill tip region.

4. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The feature extraction module's logic for judging sputtering anomalies is as follows: Preprocess the acquired sputtering image data; Connectivity analysis was performed on the segmented binary image to calculate the total sputtering coverage area at each time point. Based on the spatial coordinates of all sputtering pixels, the mainstream direction angle and its distribution dispersion of the sputtering were calculated. The total coverage area, mainstream direction angle, and distribution dispersion of the sputtered material at the current location are compared with the pre-generated benchmark sputtering model. If the total area of ​​the sputtered material at the current location exceeds ±30% of the area threshold of the corresponding location in the baseline model, the mainstream direction angle deviates from the expected direction by more than 45 degrees, or the dispersion of the direction distribution exceeds the preset dispersion threshold, then the sputtering anomaly at that location is determined to have occurred. When an anomaly is detected, the extraction of the abnormal sputtering features is triggered. These features include the abrupt change in area of ​​the abnormal sputtering, the abrupt change in direction angle, and the image texture feature vector of the abnormal region.

5. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The construction process of the risk prediction model in the risk prediction module is as follows: The design incorporates a multi-input branch neural network structure. The first branch combines a one-dimensional convolutional layer with a long short-term memory network layer for vibration feature sequences. The second branch uses a one-dimensional convolutional layer to extract local trends for temperature feature sequences. The third branch uses a two-dimensional convolutional layer to process the spatial information of abnormal sputtering features. The high-dimensional features extracted from each branch are spliced ​​together in the fusion layer; The training dataset is constructed by collecting full-time sensor data of the historical machining process and the corresponding actual deformation risk labels of the cutting tools; the model is trained end-to-end using the mean square error as the loss function and the adaptive moment estimation algorithm is used. Vibration feature sequence, temperature feature sequence and image feature vector representing sputtering anomaly are used as independent input channels. After local feature fusion and dimensionality reduction through convolutional layer and pooling layer, global feature synthesis is performed in fully connected layer to output tool deformation risk coefficient.

6. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The preset process for the risk level and action mapping strategy in the decision generation module is as follows: When the risk coefficient is below the first threshold, no adjustment command is generated; when the risk coefficient is between the first and second thresholds, a command to reduce the feed rate by 10%-20% is generated; when the risk coefficient is above the second threshold, a compound command to reduce the feed rate by 20%-30% and the spindle speed by 5%-15% is generated, and a warning signal is sent to the designated human-machine interface.

7. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The fuzzy control algorithm in the parameter control module takes the tool deformation risk coefficient and the current drill tip temperature as input variables and the cooling water flow rate adjustment coefficient as output variables; it defines fuzzy sets of low, medium, and high risk, as well as low temperature, normal temperature, and high temperature, and presets a fuzzy rule base, and outputs the flow rate adjustment coefficient through defuzzification operation.

8. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The instruction execution module is interconnected with an update feedback module via a wireless network. The update feedback module is used to highlight the stress concentration area and thermal load distribution of the tool in the virtual model by using color gradient or deformation magnification based on the data acquired by the data acquisition module and the execution results of the instruction execution module, and based on the measured vibration and temperature data; dynamically simulate the chip formation and removal process on the virtual workpiece surface based on the sputtering image analysis results; and update the tool's motion trajectory and process parameters in the virtual model in real time according to the issued control commands.

9. The adaptive control system for CNC machine tools based on digital twins according to claim 1, characterized in that, The model building module and the data acquisition module are interconnected via a wireless network. The feature extraction module, the data acquisition module, and the risk prediction module are interconnected via a wireless network. The risk prediction module and the decision generation module are interconnected via a wireless network. The decision generation module, the parameter control module, and the instruction execution module are interconnected via a wireless network.

10. A digital twin-based adaptive control method for CNC machine tools, wherein the method is an implementation method of a digital twin-based adaptive control system for CNC machine tools according to any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Construct a digital twin model that is synchronously mapped to the physical CNC machine tool and the workpiece to be processed, and discretize the preset tool processing path into several processing monitoring time sequence points; Step 2: During the physical machining process, the vibration signal, temperature signal and sputtering image data of the machining area are collected in real time at each time point; Step 3: Process the vibration and temperature signals to extract vibration and temperature features; The sputtering image data is processed to extract the sputtering direction and range data, and the presence of sputtering anomalies is determined accordingly. If anomalies are found, the abnormal sputtering features are extracted. Step 4: Input the vibration characteristics, temperature characteristics and abnormal sputtering characteristics into the pre-trained convolutional neural network prediction model. The model outputs the predicted value of the tool deformation risk coefficient at a specified time point in the future. Step 5: Based on the tool deformation risk coefficient, generate adaptive adjustment instructions for the tool at future time points, and dynamically calculate the cooling water flow adjustment coefficient based on the risk coefficient and current temperature characteristics. Step 6: Send the adaptive adjustment command and flow adjustment coefficient to the physical CNC machine tool for execution, and synchronously drive the digital twin model to update its status. At the same time, based on the deviation between the actual machining results and the model prediction, iteratively optimize the prediction model and control strategy.

Images