A numerical control turning and milling machining path optimization and prediction feedback method
By combining digital twin models with multimodal feedback, the problem of insufficient rigidity of CNC lathe systems under dynamic factors was solved, enabling real-time path optimization and safety control, improving machining efficiency and accuracy, and extending tool life.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN VOCATIONAL & TECHNICAL COLLEGE (DALIAN OPEN UNIVERSITY)
- Filing Date
- 2026-01-26
- Publication Date
- 2026-06-05
Smart Images
Figure CN122151713A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of CNC lathe machining technology, and in particular to a method for optimizing and predicting feedback CNC milling and turning machining paths. Background Technology
[0002] CNC lathe machining technology, as a key process in modern mechanical manufacturing, focuses on precisely controlling the machine tool's motion trajectory through digital programs to complete the cutting of various complex geometric shapes. This technology primarily relies on pre-programmed CNC programs (such as ISO standard G-code) to direct the relative movement of the tool and workpiece, achieving the dimensional accuracy and surface quality required by the design drawings. In typical machining scenarios such as rotating parts and irregular curved surfaces, CNC lathes, through the coordinated control of spindle rotation and tool feed, can complete the machining of various features such as external diameters, internal holes, and threads. However, with increasingly complex product structures and ever-increasing machining accuracy requirements, the fixed-program machining mode of traditional CNC systems is showing its limitations in adapting to dynamic factors such as fluctuations in material properties and changes in tool condition.
[0003] Existing CNC systems generally employ an open-loop control architecture, where the machining path relies entirely on preset program instructions, lacking the ability to monitor and dynamically adjust the actual machining state in real time. While this control method may meet the requirements for simple machining with stable cutting conditions, it easily leads to instability in the cutting process when dealing with complex conditions such as material inhomogeneity and intermittent cutting. Specifically, when tool wear intensifies or the workpiece material hardness changes abruptly, a fixed feed rate can cause a sharp increase in cutting force, inducing tool chatter and deterioration of the machined surface roughness. Furthermore, existing systems often employ single-parameter feedback control strategies, such as adjusting the feed rate solely based on spindle current, failing to comprehensively consider the coupled effects of multiple factors such as cutting temperature and tool wear, resulting in limited control accuracy. More importantly, most current CNC systems lack effective machining process prediction models, making it impossible to implement preventative adjustments before abnormal conditions occur. This passive response mode severely restricts the improvement of machining efficiency and quality.
[0004] Therefore, a method for optimizing and predicting feedback CNC milling and turning machining paths is needed. Summary of the Invention
[0005] In view of this, the present invention provides a CNC milling and turning machining path optimization and predictive feedback method. By combining dynamic weight model predictive control and multimodal feedback with safety arbitration, it solves the problems of poor flexibility and low machining efficiency caused by fixed machining paths in the prior art, as well as the difficulty in guaranteeing machining accuracy and severe tool wear caused by the lack of real-time dynamic optimization.
[0006] Therefore, the present invention provides the following technical solution: A method for optimizing and predicting CNC milling and turning machining paths, comprising: A digital twin model is determined based on the confidence level of the 3D model of the parts to be manufactured and the process parameters. Based on the digital twin model, the toolpath for multiple future control cycles is optimized and solved with the goal of minimizing cost to obtain machining instructions; Multi-source data is collected synchronously by multiple sensors and then fused. The multi-source data includes position, force, vibration, and temperature. The actual state is obtained based on the multi-source data through Kalman filtering. The predicted state is obtained based on the fused data through the digital twin model; The deviation between the actual state and the predicted state is calculated, and the processing instructions are corrected by the hybrid controller. The machining is completed using the modified machining instructions.
[0007] Furthermore, it also includes: After processing is completed, record the data collected throughout the process and perform incremental training on the digital twin model.
[0008] Furthermore, it also includes: During processing, if any multi-source data exceeds a preset threshold, a safety warning will be issued and processing will be stopped.
[0009] Furthermore, the cost function includes:
[0010] in, To estimate processing time, To estimate tool wear, To predict trajectory tracking error, To estimate the vibration amplitude; and These are the maximum allowable values for each indicator; and The time-varying weighting coefficients satisfy the following conditions: .
[0011] Furthermore, the time-varying weighting coefficients are adjusted based on the stability of the processing, as expressed by the following formula:
[0012] in, As the base weight, To adjust the factor, For the past The moving average of the absolute value of the cutting force deviation within each control cycle This is the actual cutting force. To predict cutting force, The maximum permissible cutting force; and Constrained within the interval .
[0013] Furthermore, the adaptive digital twin model includes: A gated cyclic unit network is used to predict the tool position and speed in future time periods; Predicting cutting force, vibration, and tool temperature based on support vector regression algorithm; The confidence level of the prediction results is evaluated in real time, and the prediction model is incrementally trained and its parameters are updated using the data from the current processing task.
[0014] Advantages and positive effects of the present invention: This invention achieves real-time dynamic optimization of the machining path by introducing an adaptive digital twin model and a dynamic weight model for predictive control. This significantly improves the intelligence and adaptability of the machining process. It can proactively predict key states such as cutting force and vibration, and adjust path parameters online accordingly. This effectively overcomes the rigidity defects of traditional fixed-path machining. This not only greatly improves efficiency while ensuring machining accuracy, but also actively avoids abnormal working conditions such as overcutting and chatter, thereby significantly reducing abnormal tool wear and extending tool life.
[0015] This invention ensures high reliability and robustness of the optimization process by fusing multi-sensor data and setting safety strategies. Even in the event of conflicting or distorted data from some sensors, the system can make safety decisions to ensure the continuity and stability of the processing. At the same time, the system has online self-learning and knowledge archiving functions, enabling the digital twin model to continuously evolve as processing tasks are executed. The accumulated process data constitutes a reusable knowledge base, providing strong support for rapid initialization and offline optimization of subsequent processing, and achieving self-improvement and continuous improvement of performance. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a framework diagram of a CNC milling and turning machining path optimization system and its predictive feedback system.
[0018] Figure 2 The flowchart shows the CNC milling and turning machining path optimization system and its predictive feedback method.
[0019] Figure 3 This is a flowchart of the rolling time-domain optimization sub-process in the CNC milling and turning machining path optimization system and its predictive feedback method.
[0020] Figure 4 This is a flowchart of the multi-source data fusion and prediction sub-process in the CNC milling and turning machining path optimization system and its prediction feedback method.
[0021] Figure 5 This is a flowchart of the feedback correction and safety arbitration sub-process in the CNC milling and turning machining path optimization system and its predictive feedback method.
[0022] Figure 6 This is a flowchart of the online model evolution and knowledge archiving sub-process in the CNC milling and turning machining path optimization system and its predictive feedback method. Detailed Implementation
[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0025] This invention provides a method for optimizing machining paths on a CNC lathe based on predictive control and feedback correction, comprising: S1. Import part model and process parameters, load or initialize digital twin model; if it is a new model or the model confidence is below the threshold, enable the simplified model based on physical formula and start conservative machining mode. S2, Rolling Time Domain Optimization: Based on the current digital twin model, the tool path for several future control cycles is optimized using a multi-objective function; S3. During the processing, position, force, vibration, and temperature data are collected simultaneously, and the data are filtered and fused to obtain the actual state. S4. The digital twin model receives the fused data, performs a one-step forward prediction, and outputs the prediction status and its confidence level. S5. Calculate the deviation between the actual state and the predicted state, and the hybrid controller generates correction instructions; integrate all information to determine whether to perform correction, adjust weights, or trigger safety instructions. S6. Execute the final instructions after arbitration and monitor the machine tool's execution status; S7. After the current optimization cycle ends, the dynamic weight model prediction and control module performs rolling optimization based on the latest status; after all processing tasks are completed, the model self-learning is started, the digital twin model is updated, and the processing data is archived to the central database.
[0026] The multi-objective optimization function in S2 is:
[0027] in, To estimate processing time, To estimate tool wear, To predict trajectory tracking error, To estimate the vibration amplitude; and These are the maximum allowable values for each indicator; and The time-varying weighting coefficients satisfy the following conditions: .
[0028] The adjustment strategy for time-varying weighting coefficients is based on the stability of the processing procedure, and its calculation formula is as follows:
[0029] in, As the base weight, To adjust the factor, For the past The moving average of the absolute value of the cutting force deviation within each control cycle This is the actual cutting force. To predict cutting force, The maximum permissible cutting force; and Constrained within the interval .
[0030] The update formula for predicting the tool motion state using a gated cycle unit in S4 is as follows:
[0031] in, for The input vector at time t, for The hidden state at all times To update the door, To reset the door, In the candidate hidden state, and The parameters to be trained, For the sigmoid function, For Hadamard products.
[0032] The actual state is estimated based on multi-sensor data using the Kalman filter algorithm: State prediction:
[0033]
[0034] Measurement Update:
[0035] Among them, These are state estimates, including: tool position. To estimate the error covariance, Here is the state transition matrix. For the control matrix, To control the input, For process noise covariance, The sensor measurement value, For the observation matrix, To measure the noise covariance, This is the Kalman gain.
[0036] The present invention also provides a CNC lathe machining path optimization system based on predictive control and feedback correction, comprising: a data acquisition module, an adaptive digital twin model module, a dynamic weight model predictive control module, a multimodal feedback and safety arbitration module, a path execution control module, and a central database and human-computer interaction module.
[0037] 1. Adaptive digital twin model module, including: The tool motion state prediction unit is constructed using a gated cyclic unit network to predict the tool position and speed in future time periods. A multiphysics prediction unit for the cutting process, based on the support vector regression algorithm, is used to predict cutting force, vibration, and tool temperature. The model confidence evaluation and self-learning unit is used to evaluate the confidence of the prediction results in real time, and to perform incremental training and parameter updates on the prediction model using the data from this processing task after the processing task is completed.
[0038] 2. Dynamic weight model prediction and control module: The objective function includes time-varying weight coefficients, which can be dynamically adjusted based on the real-time processing status and the confidence level of the digital twin model.
[0039] 3. The multimodal feedback and security arbitration module includes: A hybrid controller is used to integrate PID control and fuzzy inference to calculate path and parameter correction. The multi-sensor data fusion unit uses the Kalman filter algorithm to perform state estimation on heterogeneous sensor data; The safety arbiter sets the priority of sensor data and commands, and triggers predefined safety policies when data conflicts or exceed limits.
[0040] 4. The central database is used to store historical processing data, model parameters, optimization logs, and fault records, forming a traceable, queryable processing knowledge base that can be used for model initialization and offline analysis.
[0041] Example 1 like Figure 1 As shown, a CNC lathe machining path optimization system based on predictive control and feedback correction includes: The system includes a data acquisition module, an adaptive digital twin model module, a dynamic weight model prediction and control module, a multimodal feedback and security arbitration module, a path execution control module, and a central database and human-computer interaction module.
[0042] 1. Adaptive digital twin model module, including: 1) The tool motion state prediction unit is constructed using a gated cyclic unit network to predict the tool position and speed in future time periods; 2) Multiphysics prediction unit for the cutting process, based on support vector regression algorithm, is used to predict cutting force, vibration and tool temperature; 3) Model confidence evaluation and self-learning unit, used to evaluate the confidence of the prediction results in real time, and to perform incremental training and parameter update of the prediction model using the data from this processing task after the processing task is completed; 2. Dynamic weight model prediction and control module: The objective function of optimization includes time-varying weight coefficients, which can be dynamically adjusted according to the real-time processing status and the confidence level of the digital twin model.
[0043] 3. Multimodal feedback and security arbitration module, including: 1) Hybrid controller, used to integrate PID control and fuzzy inference, to calculate path and parameter correction; 2) A multi-sensor data fusion unit uses a Kalman filter algorithm to estimate the state of heterogeneous sensor data; 3) Safety arbitrator: sets the priority of sensor data and commands, and triggers predefined safety policies when data conflicts or exceed limits. The central database is used to store historical processing data, model parameters, optimization logs and fault records, forming a traceable, queryable processing knowledge base that can be used for model initialization and offline analysis.
[0044] Example 2 like Figure 2 As shown, a method for optimizing the machining path of a CNC lathe based on predictive control and feedback correction includes the following steps: Step a: System initialization and adaptive model loading: Import part model and process parameters, load or initialize digital twin model; if it is a new model or the model confidence is below the threshold, enable the simplified model based on physical formula and start conservative machining mode; Step b: Rolling time-domain optimization: The dynamic weighted model predictive control module optimizes the toolpath for several future control cycles based on the current digital twin model using a multi-objective function; Step c: Real-time acquisition and fusion of multi-source data: During the processing, position, force, vibration and temperature data are acquired simultaneously, and filtered and fused to obtain highly reliable state estimates; Step d: Digital twin synchronous prediction and confidence assessment: The digital twin model receives the fused data, performs a one-step forward prediction, and outputs the prediction result and its confidence level; Step e: Feedback Correction and Safety Arbitration: Calculate the deviation between the actual state and the predicted state, and the hybrid controller generates correction instructions; the safety arbitrator integrates all information and determines whether to execute correction, adjust weights, or trigger safety instructions; Step f: Path execution and status monitoring: The path execution control module executes the final, arbitrated instruction and monitors the machine tool's execution status; Step g: Online model evolution and knowledge archiving: After the current optimization cycle ends, the dynamic weight model prediction and control module performs rolling optimization based on the latest status; after all processing tasks are completed, the model self-learning is started, the digital twin model is updated, and the processing data is archived to the central database.
[0045] Example 3 A method for optimizing the machining path of a CNC lathe based on predictive control and feedback correction includes the following steps: Step a: Combining Figure 3 As shown, the system initialization and adaptive model loading are as follows: import the part model and process parameters, load or initialize the digital twin model; if it is a new model or the model confidence is lower than the threshold, then enable the simplified model based on physical formulas and start the conservative machining mode. Step b: Rolling Time Domain Optimization: The dynamic weighted model predictive control module optimizes the toolpath for several future control cycles based on the current digital twin model using a multi-objective function. The multi-objective optimization function is:
[0046] in, To estimate processing time, To estimate tool wear, To predict trajectory tracking error, To estimate the vibration amplitude; and These are the maximum allowable values for each indicator; and These are time-varying weighting coefficients. The adjustment strategy for the time-varying weight coefficients is based on the stability of the processing procedure, and its calculation formula is as follows:
[0047] in, As the base weight, To adjust the factor, For the past The moving average of the absolute value of the cutting force deviation within each control cycle This is the actual cutting force. To predict cutting force, The maximum permissible cutting force; and Constrained within the interval .
[0048] Step c: Combining Figure 4 As shown, multi-source data is acquired and fused in real time: during the processing, position, force, vibration and temperature data are acquired simultaneously, and filtered and fused to obtain highly reliable state estimates.
[0049] Step d: Digital twin synchronous prediction and confidence assessment: The digital twin model receives the fused data, and the tool motion state prediction uses a gated loop unit, whose core update formula is:
[0050]
[0051]
[0052]
[0053] in, for The input vector at time t, for The hidden state at all times To update the door, To reset the door, In the candidate hidden state, and The parameters to be trained, For the sigmoid function, Perform a forward prediction using the Hadamard product, and output the prediction result and its confidence level.
[0054] Step e: Combining Figure 5 As shown, feedback correction and safety arbitration: the deviation between the actual state and the predicted state is calculated, and the hybrid controller generates correction instructions; the safety arbitrator integrates all information and determines whether to perform correction, adjust weights, or trigger safety instructions.
[0055] Step f: Path execution and status monitoring: The path execution control module executes the final instructions after arbitration and monitors the machine tool execution status.
[0056] Step g: Combining Figure 6 As shown, the model evolves online and knowledge is archived: after the current optimization cycle ends, the dynamic weight model prediction and control module performs rolling optimization based on the latest state; after all processing tasks are completed, the model self-learning is started, the digital twin model is updated, and the processing data is archived to the central database.
[0057] Example 4 This system optimizes the finishing of alloy blades, including: 1. Initialization and Model Loading: The operator imports the 3D CAD model and initial process parameters of the blade through a human-machine interface; the system retrieves the "adaptive digital twin model" and its parameters from the central database, which has previously processed blades of similar materials. Because it is a mature model, the system's confidence level is high, so this model is directly used for optimization.
[0058] 2. Rolling optimization and prediction: The dynamic weighted model predictive control module begins operation. When machining the complex blade base and back surfaces, the gated loop unit network in the digital twin model predicts the tool's position and speed in real time for the next few steps, while the support vector regression unit predicts the cutting force and temperature. In the initial stage, the system focuses on accuracy and surface quality, therefore the trajectory tracking error weights in the objective function are optimized. and vibration weight The settings are relatively high; when the digital twin predicts that it is about to enter a known tool wear zone (such as the blade root fillet), the system dynamically adjusts the weights. Based on the time-varying weight coefficient formula, the system detects that the moving average of the cutting force prediction deviation begins to increase, and therefore automatically increases the tool wear weight. The weight of the feed rate is used to determine an optimized path that slightly reduces the feed rate while maintaining accuracy, in order to protect the tool.
[0059] 3. Real-time feedback and safety arbitration: The multi-sensor data fusion unit collects data in real time through force sensors, vibration sensors and infrared thermometers installed on the spindle and turret, and uses Kalman filtering algorithm to fuse the data to obtain a more reliable actual cutting state; 1) The hybrid controller (PID + fuzzy inference) calculates minute path correction instructions, including: When the actual cutting force is slightly higher than the predicted value, fuzzy inference will give a small trajectory offset to disperse the cutting force.
[0060] 2) During machining, when the vibration sensor detects abnormal high-frequency chatter, the safety arbitrator is immediately triggered. It determines that the vibration data has the highest priority, exceeding the current optimization command. Therefore, the arbitrator decisively executes the predefined safety strategy: instructing the machine tool to pause the feed and perform a small retraction action. After the vibration disappears, it recalculates the safe path to continue machining, thus avoiding damage to the workpiece surface caused by chatter.
[0061] 4. Model Evolution and Archiving: After the blade is processed, the system uses the data collected throughout the process to incrementally train the gated recurrent unit and support vector regression model in the digital twin model, making the model more accurate in predicting the processing characteristics of this specific material. All parameters optimized, sensor data logs, and model update records are archived to the central database, providing richer knowledge reserves for processing the same type of blade in the future.
[0062] Example 5 This system enables the production of drive shaft components, including: 1. Initialization and Model Loading: When the production line switches to a new driveshaft model, the system imports the new model. Since there is no complete machining history for this model, the loaded generic model has low confidence. According to the strategy, the system enables a simplified model based on physical formulas and initiates a conservative machining mode, which uses lower feed and depth of cut parameters to start the first piece machining to ensure safety.
[0063] 2. Rolling optimization and self-learning: 1) During the first piece machining process, the system quickly collects the actual cutting data of the part at various geometric features through multi-sensor data fusion.
[0064] 2) The dynamic weight model prediction and control module performs small-scale optimizations based on real-time data and conservative parameters, including: In the optical axis section with a small machining allowance, the system predicts that the cutting force is far below the threshold and dynamically increases the machining time weight. This optimization resulted in a faster feed rate and improved efficiency.
[0065] 3) As several parts are processed consecutively, the digital twin model rapidly learns online using the collected data, and its confidence level steadily improves. The system gradually loosens optimization restrictions and begins to more actively adjust paths and parameters.
[0066] 3. Feedback correction and tool wear compensation: After producing hundreds of parts in a row, the multiphysics prediction unit of the digital twin model shows that, under the same machining path, the predicted cutting force shows a slow but continuous upward trend, which is a typical characteristic of tool wear. 1) Based on this trend, the hybrid controller no longer just makes minor path corrections, but begins to generate adjustment instructions for tool compensation (such as tool offset value) to actively counteract dimensional deviations caused by tool wear, ensuring that the diameter of each machined part is near the center of the tolerance zone.
[0067] 2) The safety arbiter continuously monitors the estimated tool wear. When the system predicts that the wear amount is close to the life cycle threshold, the arbiter will not stop immediately (to avoid disrupting the production cycle). Instead, after completing the machining of the current part, it will issue a warning to the operator to replace the tool and record it in the optimization log.
[0068] 4. Knowledge Consolidation and Efficiency Improvement: After a batch of production tasks is completed, the system has learned and consolidated a set of optimal processing strategies for that model of drive shaft, which is stored in the central database. When the next batch of this model is produced, the system can directly call the high-confidence model and use efficient and safe parameters for processing from the beginning, achieving a dual improvement in production efficiency and quality stability.
[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for optimizing and predicting feedback CNC milling and turning machining paths, characterized in that, include: A digital twin model is determined based on the confidence level of the 3D model of the parts to be manufactured and the process parameters. Based on the digital twin model, the toolpath for multiple future control cycles is optimized and solved with the goal of minimizing cost to obtain machining instructions; Multi-source data is collected synchronously by multiple sensors, and the data is then fused. The multi-source data includes: position, force, vibration, and temperature; the actual state is obtained based on the multi-source data through Kalman filtering; The predicted state is obtained based on the fused data through the digital twin model; The deviation between the actual state and the predicted state is calculated, and the processing instructions are corrected by the hybrid controller. The machining is completed using the modified machining instructions.
2. The method according to claim 1, characterized in that, Also includes: After processing is completed, record the data collected throughout the process and perform incremental training on the digital twin model.
3. The method according to claim 1, characterized in that, Also includes: During processing, if any multi-source data exceeds a preset threshold, a safety warning will be issued and processing will be stopped.
4. The method according to claim 1, characterized in that, The cost function includes: in, To estimate processing time, To estimate tool wear, To predict trajectory tracking error, To estimate the vibration amplitude; and These are the maximum allowable values for each indicator; and The time-varying weighting coefficients satisfy the following conditions: .
5. The method according to claim 4, characterized in that, The time-varying weighting coefficients are adjusted based on the stability of the processing, as expressed by the formula: in, As the base weight, To adjust the factor, For the past The moving average of the absolute value of the cutting force deviation within each control cycle This is the actual cutting force. To predict cutting force, The maximum permissible cutting force; and Constrained within the interval .
6. The method according to claim 1, characterized in that, The adaptive digital twin model includes: A gated cyclic unit network is used to predict the tool position and speed in future time periods; Predicting cutting force, vibration, and tool temperature based on support vector regression algorithm; The confidence level of the prediction results is evaluated in real time, and the prediction model is incrementally trained and its parameters are updated using the data from the current processing task.