Precision calibration method and system for pentahedron machining center based on multi-sensor fusion

Abstract

The present invention relates to the field of program control system technology, specifically to a precision calibration method and system for a pentahedron machining center based on multi-sensor fusion, comprising: a sensor system construction and calibration module for on-site calibration, drift correction, and redundant deployment to ensure data accuracy, and utilizing blockchain technology to achieve full traceability of the calibration process; a multi-source data preprocessing and fusion module for synchronizing heterogeneous data in time and space and dynamically fusing multi-source information; an intelligent modeling and state prediction module for real-time, multi-task prediction of key states such as tool wear and thermal deformation; an adaptive compensation and path optimization module for dynamically optimizing tool paths based on prediction results; at the same time, through an online learning mechanism, the calibration model is continuously updated using machining results; a distributed collaborative control module executes data processing and calibration algorithms locally, and collaborates with the numerical control system in making decisions to achieve low-latency, intelligent responses to machining anomalies.

Description

Technical Field

[0001] The present invention relates to the technical field of program control systems, and in particular to a precision calibration method and system for a pentahedron machining center based on multi-sensor fusion. Background Art

[0002] Multi-sensor fusion technology is crucial for the precision calibration of pentahedron machining centers, as these machines' complex structures and multi-axis linkages demand extremely high precision. With the advancement of industrial and intelligent manufacturing, integrating multiple data sources such as laser trackers, high-precision encoders, and temperature sensors can more comprehensively capture geometric and thermal error information, enabling more intelligent and efficient error compensation and significantly improving machining performance.

[0003] However, multi-source data fusion and dynamic error compensation in pentahedron machining centers face significant challenges. Different sensors (for example, high-precision but low-sampling-rate laser trackers, high-speed but noise-susceptible encoders, and temperature sensors with varying response characteristics) vary significantly in data type, sampling frequency, and accuracy, making precise alignment of data in time and space very difficult. Furthermore, the sources of error are complex and multivariate. Geometric, thermal, and mechanical errors are coupled and vary over time, making it difficult for traditional single error models to fully describe these combined effects. Furthermore, calibration methods that rely on offline measurements cannot adapt to the demands of dynamic machining environments, while real-time compensation faces the contradiction between high-frequency data processing and limited computing resources.

[0004] Therefore, a precision calibration method and system for pentahedron machining center based on multi-sensor fusion is proposed. Summary of the Invention

[0005] The present invention aims to solve the problem of reduced machining accuracy of existing pentahedron machining centers due to comprehensive factors during long-term operation. The present invention provides a precision calibration method for pentahedron machining centers based on multi-sensor fusion. The specific steps include:

[0006] Integrate physical quantity measurement sensors in the pentahedron machining center, perform on-site multi-point calibration of the sensors and establish a drift correction model. Use blockchain technology to trace the calibration process and deploy redundant sensor arrays to detect and isolate anomalies.

[0007] An adaptive filtering algorithm is used to preprocess sensor data, establish a global coordinate system, and perform spatiotemporal synchronization and coordinate fusion. Confidence-based fusion and factor graph optimization are used, and the closed-loop feedback mechanism of the CNC system is utilized to generate precision compensation control signals. Calibration is performed using a point cloud error correction model, combined with a digital twin model, to predict machining state parameters in real time, including tool wear, thermal deformation, and material inconsistency errors.

[0008] An online adaptive mechanism is adopted to update the calibration point cloud error correction model and digital twin model in real time based on the precision compensation control signal and processing status parameters, and adjust the tool path; a distributed sensor control architecture is designed, and the calibration compensation algorithm is embedded in the local processing of the sensor node to collaborate with the CNC system.

[0009] Preferably, the physical quantity measurement sensors include laser displacement sensors, temperature sensors, force sensors, vibration sensors, and fiber grating sensors; the fiber grating sensors are used to obtain local deformation data of the workpiece and the tool, and to obtain spatial posture data of the workpiece, the tool, the fixture, and the machine tool through a binocular vision system; and real-time time characteristic calibration is performed by estimating and dynamically compensating for the time delay, jitter, and sampling rate inconsistency time characteristic deviation of the sensor data.

[0010] Preferably, performing on-site multi-point calibration of the sensor and establishing a drift correction model includes: performing on-site multi-point calibration of the physical quantity measurement sensor in a processing environment, detecting and correcting signal drift in real time, and using numerical control system feedback adjustment to ensure sensor accuracy; using historical sensor operation data to predict performance degradation trends and trigger predictive calibration and maintenance; recording sensor calibration data and processes through blockchain technology to ensure data traceability and reliability; deploying redundant sensors, and identifying and isolating abnormal sensors through cross-validation of numerical control system data.

[0011] Preferably, the confidence-based fusion and factor graph optimization includes the following steps: assigning dynamic weights to the fiber Bragg grating sensor and binocular vision system data according to the sensor measurement accuracy and environmental interference, including illumination changes and electromagnetic noise; constructing a factor graph model, using the spatial posture data and local deformation data of the workpiece, tool, fixture and machine tool as nodes, and the sensor measurement values ​​as factors, and optimizing the global accuracy compensation parameters through maximum a posteriori estimation.

[0012] Preferably, the point cloud error correction model includes: registering the point cloud data generated by the binocular vision system, converting the point cloud into a normal distribution model based on the probability density function; and extracting tool wear, thermal deformation and material inconsistency errors through multi-layer convolution and pooling operations.

[0013] Preferably, the calibration and compensation algorithm includes: using a numerical control system to automatically measure the tool geometric parameters, installation position and wear status, dynamically compensate for errors in real time, and optimize the compensation strategy by combining sensor data and digital twin simulation; realizing automatic calibration of a dedicated probe, modeling and compensating for the probe's geometric characteristics, contact force and dynamic errors, and using numerical control system error mapping technology to optimize multi-axis linkage accuracy; performing real-time compensation on multi-component force sensors to offset the influence of non-axial forces and correct thermal drift, and using numerical control system algorithms to decouple axial force components to improve detection accuracy.

[0014] Preferably, the predicted processing state parameters include: real-time prediction of processing accuracy based on multi-sensor data and digital twin models, quantification of error uncertainty using the CNC system, triggering early warning and / or parameter adjustment, and optimizing production efficiency, tool life and surface quality; continuous monitoring of interference between tools, workpieces and machine tools, and automatically triggering hierarchical safety strategies in combination with the physical engine to predict collision risks.

[0015] Preferably, the tool path adjustment includes: integrating multi-sensor data into the CNC system, and dynamically adjusting the tool path using a reinforcement learning algorithm; combining a digital twin model to simulate machining dynamics, predict errors and optimize path parameters in real time, and continuously updating the path planning based on sensor data through a closed-loop feedback mechanism of the CNC controller to achieve adaptive control.

[0016] Preferably, the distributed sensor control architecture is implemented in the following ways: deploying sensor nodes with edge computing capabilities, performing localized data processing and anomaly detection, and collaborating with the CNC system to reduce control delays; detecting and isolating faulty sensors through a cross-validation protocol between sensor nodes; and using a distributed coordination algorithm to enable sensor nodes and CNC controllers to make collaborative decisions, quickly respond to processing anomalies, and achieve distributed control.

[0017] The pentahedron machining center precision calibration system based on multi-sensor fusion includes:

[0018] Sensor system construction and calibration module: Integrate physical quantity measurement sensors in the pentahedron machining center, perform on-site multi-point calibration of the sensors and establish a drift correction model. Use blockchain technology to achieve traceability of the calibration process and deploy redundant sensor arrays for anomaly detection and isolation.

[0019] Multi-source data preprocessing and fusion module: Uses adaptive filtering algorithms to preprocess sensor data, establishes a global coordinate system, and performs spatiotemporal synchronization and coordinate fusion;

[0020] Intelligent modeling and state prediction module: Based on confidence fusion and factor graph optimization, it utilizes the CNC system's closed-loop feedback mechanism to generate precision compensation control signals. It also uses a point cloud error correction model for calibration and, combined with a digital twin model, predicts machining state parameters in real time, including tool wear, thermal deformation, and material inconsistency errors.

[0021] Adaptive compensation and path optimization module: uses an online adaptive mechanism to update the calibration point cloud error correction model and digital twin model in real time based on the precision compensation control signal and processing status parameters, and adjust the tool path;

[0022] And distributed collaborative control module: design a distributed sensor control architecture, embed calibration compensation algorithms in the local processing of sensor nodes, and collaborate with the CNC system.

[0023] Compared with the prior art, the beneficial effects of the present invention are embodied in:

[0024] 1. This application integrates multiple sensors, including laser displacement, temperature, force, vibration, fiber Bragg grating sensors, and binocular vision systems, to monitor various errors during machining in real time. Adaptive filtering and confidence-based factor graph optimization are used to effectively fuse multi-source data, and maximum a posteriori estimation is used to accurately compensate for nonlinear errors such as tool wear, thermal deformation, and material inconsistencies. Point cloud error correction models and digital twin models are introduced to achieve high-precision error prediction and calibration. Online adaptive mechanisms and closed-loop feedback from the CNC system ensure sustained high precision in dynamic machining environments.

[0025] 2. The introduction of blockchain to record sensor calibration data and processes ensures traceability and reliability, preventing data tampering. The redundant sensor array, combined with cross-validation within the CNC system, can identify and isolate abnormal sensors in real time, improving system fault tolerance and data reliability. Furthermore, the system can leverage historical operating data to predict performance degradation trends, triggering predictive calibration and maintenance, effectively extending equipment life, reducing costs, minimizing unplanned downtime, and improving equipment utilization.

[0026] 3. Sensor nodes equipped with edge computing capabilities perform local data processing and collaborate with the CNC system to effectively reduce control latency, improve response speed, and enhance machining efficiency. The system leverages multi-sensor data and reinforcement learning algorithms to dynamically adjust tool paths. Combined with the digital twin model, it predicts errors and optimizes path parameters, improving machining efficiency and product quality. Furthermore, it predicts machining accuracy in real time, quantifies error uncertainty, monitors interference, predicts collision risks, and automatically triggers hierarchical safety strategies, significantly enhancing machining safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flowchart of a pentahedron machining center precision calibration method based on multi-sensor fusion according to the present invention;

[0028] Figure 2 This is a flowchart of the data fusion and compensation stage of the present invention;

[0029] Figure 3 It is a structural schematic diagram of the pentahedron machining center precision calibration system based on multi-sensor fusion of the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] See also Figures 1 to 3 The present invention provides a pentahedron machining center precision calibration method based on multi-sensor fusion, the technical solution is as follows: As an embodiment of the present invention, referring to Figure 1 This is a flowchart of the precision calibration method for a pentahedron machining center based on multi-sensor fusion proposed in this application.

[0032] Integrate physical quantity measurement sensors in the pentahedron machining center, perform on-site multi-point calibration of the sensors and establish a drift correction model. Use blockchain technology to trace the calibration process and deploy redundant sensor arrays to detect and isolate anomalies.

[0033] An adaptive filtering algorithm is used to preprocess sensor data, establish a global coordinate system, and perform spatiotemporal synchronization and coordinate fusion. Confidence-based fusion and factor graph optimization are used, and the closed-loop feedback mechanism of the CNC system is utilized to generate precision compensation control signals. Calibration is performed using a point cloud error correction model, combined with a digital twin model, to predict machining state parameters in real time, including tool wear, thermal deformation, and material inconsistency errors.

[0034] An online adaptive mechanism is adopted to update the calibration point cloud error correction model and digital twin model in real time based on the precision compensation control signal and processing status parameters, and adjust the tool path; a distributed sensor control architecture is designed, and the calibration compensation algorithm is embedded in the local processing of the sensor node to collaborate with the CNC system.

[0035] Example 1

[0036] This example uses the precise machining of the cavity of a large automobile bumper injection mold (made of P20 pre-hardened mold steel) on a certain brand's pentahedron machining center as a typical application scenario to detail the specific implementation steps of the present method. This machining task requires the contour accuracy of the key mating surfaces and exterior curved surfaces of the mold cavity to reach ±0.015mm, with a surface roughness Ra <0.4μm, to ensure the assembly precision and appearance quality of the final injection molded product.

[0037] A heterogeneous sensor system for measuring physical quantities is integrated at key locations on the pentahedron machining center. These sensors include: two laser displacement sensors, deployed on the spindle box and column, for non-contact measurement of the real-time displacement of the spindle relative to the worktable, directly monitoring geometric errors caused by thermal effects and structural deformation; 12 temperature sensors, distributed and embedded in key heat sources and heat conduction paths such as the spindle bearings, motor, ball screw, guideway slider, and bed, for real-time monitoring of the machine tool's global temperature distribution; a force sensor, integrated between the machine table and fixture base, for real-time measurement of cutting forces in the X, Y, and Z directions, reflecting tool wear, material hardness changes, and machining load; three vibration sensors, magnetically mounted on the spindle box, toolholder, and workpiece fixture, monitor high-frequency vibrations during machining, identifying and suppressing machining chatter and assisting in diagnosing tool anomalies; and fiber grating sensors: A multi-point FBG sensor array is attached to the non-machining area of ​​the mold base and the toolholder to directly acquire local strain and micro-deformation data of the workpiece and tool caused by cutting and clamping forces.

[0038] The binocular vision system is an industrial camera combined with a structured light projector, installed in a safe position inside the machine tool protective cover. Its field of view covers the entire processing area and is used to obtain the spatial pose data point cloud of the workpiece, tool, fixture and machine tool feature points through shooting and 3D reconstruction.

[0039] Fiber Bragg grating sensors are specifically used to obtain local deformation data of workpieces and tools, which is crucial for fine error compensation, especially deformation caused by thermal deformation and cutting force; while binocular vision systems can obtain spatial posture data of workpieces, tools, fixtures and machine tools, helping to establish an accurate global coordinate system and identify motion errors.

[0040] By integrating multiple sensors to obtain multi-dimensional data, key physical quantities during the machining process, including displacement, temperature, force, vibration, deformation, and spatial position, are more comprehensively and effectively captured, providing a data foundation for subsequent precision calibration. Real-time compensation for sensor data latency, jitter, and sampling rate inconsistencies ensures the spatiotemporal synchronization and consistency of multi-source data, greatly improving the reliability of fused data and avoiding calibration errors caused by time deviations.

[0041] At the standard machine tool ambient temperature of 20±0.5℃, using a laser interferometer as the measurement reference, the laser displacement sensor is calibrated at multiple positions within the working stroke, one point every 200mm along the X-axis, for a total of 10 points, to generate a high-precision error correction lookup table or polynomial model; refer to Figure 2 , which is a process step diagram for the data fusion and compensation stage of this application.

[0042] Furthermore, data from all temperature sensors and laser displacement sensors is collected simultaneously during the machine tool's complete thermal cycle: cold start-up, high-speed idle run, thermal stabilization, and natural cooling. Based on this data, a multivariate nonlinear regression method is used to establish a mapping model between the temperature field {T1, T2, ..., T12} and the geometric drifts {ΔX, ΔY, ΔZ} in each axis. This model is deployed on edge computing nodes to offset measurement drift caused by temperature changes in real time.

[0043] Furthermore, the system continuously analyzes the historical operating data of each sensor, including signal-to-noise ratio, drift, and output stability. When a sensor's performance degradation indicator is detected—for example, if the standard deviation of the drift-corrected residual continues to increase for a week and exceeds a preset threshold of 5%—the system automatically generates a predictive maintenance work order in the maintenance system, prompting technicians to recalibrate or replace the sensor. Each calibration activity, including the operator ID, timestamp, pre- and post-calibration data, the serial number of the reference device used, and the ambient temperature and humidity, is encapsulated into a data block. The hash value of this data block is linked to the hash value of the previous data block, forming an unalterable sensor calibration history chain.

[0044] To identify abnormal sensors, three independent temperature sensors are deployed at key temperature measurement points on the front end of the spindle. A monitoring program running within the CNC system continuously compares these three readings using a median voting method. If a sensor reading deviates from the median by more than 0.5°C for a sustained period, the system identifies it as abnormal, isolates the data, and issues an alert to the operator on the HMI interface, ensuring the overall reliability of the data source.

[0045] Furthermore, an adaptive Kalman filter algorithm based on signal variance is applied to the high-frequency dynamic signals collected by the force and vibration sensors. This algorithm dynamically adjusts its internal parameters based on the real-time fluctuations of the signals, effectively filtering out random electrical noise while accurately preserving the transient characteristics of the signals caused by changes in the cutting state, such as entry / exit and minor tool chipping.

[0046] Furthermore, time synchronization involves synchronizing all sensor nodes and the CNC system with the master clock within the local area network (LAN) via the Precision Time Protocol (PTP), ensuring sub-millisecond timestamp accuracy of <1ms for each data source. The inherent image acquisition and data transmission delay of approximately 20-30ms in binocular vision systems is dynamically compensated for by online calibration and model prediction, ensuring precise timeline alignment of its output data with other high-frequency sensor data.

[0047] Spatial synchronization involves placing a standard ball stick or calibration plate on the machine tool's work surface. The machine's built-in touch-trigger probe and binocular vision system jointly measure the 3D coordinates of multiple feature points on the target, accurately solving the homogeneous transformation matrix between the vision coordinate system and the machine's global G54 coordinate system. This matrix is ​​used to synchronize the position and pose data measured by all sensors to the machine tool coordinate system in real time.

[0048] Furthermore, real-time confidence levels are assigned to the measurement data from different sensors as fusion weights. For example, if a large amount of machine tool coolant splashes, causing noise or partial loss in the binocular vision system's point cloud data, the system dynamically reduces its confidence weight based on the point cloud's completeness and signal-to-noise ratio, for example, from 0.9 to 0.5. Similarly, if ground vibration caused by the start-up of a nearby machine tool is detected, the vibration sensor's confidence weight is lowered accordingly.

[0049] On-site multi-point calibration and real-time signal drift correction ensure high accuracy and reliability of sensor data in actual processing environments. Leveraging historical data to predict sensor performance degradation trends triggers predictive calibration and maintenance, transforming reactive maintenance into proactive maintenance, reducing downtime and lowering maintenance costs. Incorporating blockchain technology to record calibration data and processes provides a highly trusted, immutable calibration history, ensuring data traceability and reliability. Deploying redundant sensors promptly identifies and isolates abnormal or faulty sensors, preventing data errors and calibration deviations caused by sensor failure.

[0050] The factor graph model is constructed. Specifically, the nodes define the core state variables of the system, including the three-dimensional position P_tool of the tool cutting point in the machine tool coordinate system, the three-dimensional position P_workpiece of the point to be processed on the mold cavity surface, the spatial displacement vector ΔP_thermal caused by the thermal deformation of the spindle, the effective cutting edge geometric offset ΔP_wear caused by tool wear, and the workpiece force deformation vector ΔP_deformation.

[0051] Each sensor's measurement is constructed as a constraint factor connecting related state nodes. For example, the laser displacement sensor's measurement constitutes a unary factor constraining the ΔP_thermal node; the force sensor's measurement, combined with the tool wear model, forms a factor connecting the P_tool and ΔP_wear nodes; the strain data measured by the fiber Bragg grating sensor forms a factor connecting the P_workpiece and ΔP_deformation nodes; and the point cloud data measured by the binocular vision system forms a global observation factor connecting the P_tool and P_workpiece nodes. Table 1 shows the output of the factor graph optimization at a specific machining moment.

[0052] Table 1: Factor graph optimization output results at a certain moment in processing

[0053]

[0054] Furthermore, using incremental smoothing and a factor graph optimization algorithm, the maximum a posteriori estimation method is used to solve, in real time, the state variable node values ​​that maximize the joint probability of all sensor measurement constraint factors. After integrating all information sources, a globally optimal and most reliable error state estimate is obtained.

[0055] Dynamic weight allocation based on confidence levels can automatically adjust data contribution based on the sensor's real-time measurement accuracy and environmental interference, effectively reducing the impact of noise and abnormal data on the fusion results and improving the robustness of data fusion. The factor graph model uses all relevant spatial position and deformation data as nodes and sensor measurements as factors. It performs global optimization through maximum a posteriori estimation, comprehensively considering all information and more accurately calculating the globally optimal accuracy compensation parameters than local optimization. This optimization method for improving calibration accuracy can more accurately estimate and compensate for complex geometric errors and deformations, significantly improving the overall calibration accuracy of the machining center.

[0056] The point cloud error correction model combines the normal distribution transformation (NDT) module, the convolutional neural network (CNN) module, and the self-attention mechanism (SA) module. First, the model transforms the original point cloud into normal distribution parameters with local geometric features through normal distribution transformation. Subsequently, these normal distribution transformation features are input into the convolutional neural network (CNN). CNN learns and extracts high-level abstract spatial features through multi-layer convolution operations, capturing the correlation between different areas of the point cloud. Finally, the self-attention mechanism is introduced to further enhance the features extracted by CNN.

[0057] The binocular vision system acquires point cloud data of the processing area at a frequency of 5Hz. The system first converts the point cloud into the machine tool coordinate system using the coordinate transformation matrix.

[0058] Furthermore, the tool point cloud acquired in real time is registered with the ideal point cloud generated from the standard CAD model of the tool using a normal distribution transformation. NDT converts the point cloud into a probability density function field. The resulting registration is a transformation matrix and a matching score, which robustly represents complex tool pose deviations caused by factors such as wear and thermal expansion.

[0059] The deviation distribution map after NDT registration, the 2D thermal map generated by interpolating the temperature sensor array data, and the frequency spectrum generated by the short-time Fourier transform of the force sensor time series signal are fed as multi-channel inputs into a deep convolutional neural network (CNN). Through multiple layers of convolution and pooling operations, the CNN automatically extracts the deep nonlinear characteristics of this heterogeneous data, which are related to tool wear, thermal deformation, and uneven material hardness, which manifest as abnormal fluctuations in cutting forces.

[0060] A self-attention module is applied on top of the deep feature map extracted by CNN. This module can calculate the intrinsic correlation between features corresponding to different error sources at different spatial positions or channels in the feature map, and assign attention weights accordingly.

[0061] When finishing large mold surfaces, thermal deformation is the primary source of error. The SA module assigns higher weight to feature channels associated with the thermal map (attention score > 0.8). When machining complex curved surfaces or sharp corners, deformation caused by tool wear and cutting forces is more critical, so the SA module dynamically shifts attention to the corresponding features.

[0062] By converting binocular point cloud data into a normal distribution model and utilizing NDT methods, efficient and robust point cloud registration and feature extraction are achieved, providing high-quality data for subsequent error analysis. The powerful feature extraction capabilities of multi-layer convolutional neural networks enable effective identification and learning of complex nonlinear error characteristics such as tool wear, thermal deformation, and material inconsistencies from sensor data, which is difficult to achieve using traditional methods. Through weighted feature fusion, with a particular focus on tool wear as a key error factor, the model can more accurately identify and compensate for machining errors caused by tool wear, extending tool life and improving machining quality.

[0063] Furthermore, a high-fidelity virtual model driven by real-time data of the physical machining center is constructed, which not only includes the geometric, thermodynamic, and dynamic models of the machine tool, but also integrates process parameters and material properties.

[0064] The output of the point cloud error correction model, such as the predicted tool wear, the thermal deformation field of the entire machine, and real-time sensor data including force, vibration, and temperature, are used as input to drive the digital twin model to perform "virtual machining" simulation.

[0065] The digital twin model simulates the cutting process in real time under the influence of the current combined error, predicting the deviation between the machined surface profile of the next toolpath and the theoretical CAD profile. The system quantifies the uncertainty of this predicted error. If the upper bound of the 95% confidence interval of the predicted error exceeds 80% of the tolerance band, or 0.012mm, an alert or automatic adjustment is triggered.

[0066] Combined with a highly efficient physical collision detection engine, the digital twin model continuously monitors the dynamic distances between the tool, toolholder, fixture, and workpiece. If a collision risk is predicted within the next three seconds and the safety distance falls below a set threshold of 2mm, the system automatically triggers safety strategies, including reducing the feed rate or pausing the program.

[0067] Furthermore, the system integrates the global error estimation results of factor graph optimization and the prediction deviation of the digital twin model to generate a real-time, three-dimensional accuracy compensation vector C=(Cx, Cy, Cz).

[0068] This compensation vector is transmitted in real time to the CNC system via a high-speed Ethernet interface. The system's built-in real-time error compensation dynamically superimposes this vector on the servo axis position commands, thereby correcting the tool tip's three-dimensional position in real time at the interpolation level. This deeply integrated compensation approach essentially expands the CNC system's "sense-act" capabilities, enabling traditional open-loop CNC machine tools to intelligently adapt to complex working conditions.

[0069] Furthermore, the tool compensation is based on the preset strategy of the CNC system. For example, the internal probe is automatically called to measure the actual length and radius of the tool every 2 hours of processing or when the cumulative cutting load reaches a threshold. The tool compensation library is updated in real time based on the wear amount inferred from the force sensor data to achieve dynamic and accurate tool compensation.

[0070] Probe calibration is an automated calibration macro built into the system. It uses a standard ball to perform multi-directional touch-trigger measurements on the touch-trigger probe. It establishes and compensates for the probe's geometric characteristics, contact forces and dynamic errors to ensure accurate on-machine measurement.

[0071] Furthermore, the compensation algorithm running in the CNC system performs real-time decoupling calculations on the original signal of the three-component dynamometer, eliminating the inter-axis crosstalk caused by eccentric sensor installation or changes in the force application point, and compensates for its thermal drift based on the temperature sensor reading, thereby obtaining more accurate axial and radial cutting force components.

[0072] Comprehensive tool error compensation automatically measures tool parameters and wear status, and combines sensors and digital twins for real-time dynamic compensation, effectively eliminating tool-induced machining errors and extending tool life. Combined with automated calibration of dedicated probes, it not only accounts for geometric, contact force, and dynamic errors, but also optimizes multi-axis linkage accuracy through CNC system error mapping, improving machining precision. Real-time compensation of multi-component force sensors offsets non-axial forces, corrects thermal drift, and decouples axial force components, providing more accurate input for cutting force control and wear monitoring.

[0073] The previously mentioned factor graph optimization module obtains the fused estimated key error state vectors in real time. These primarily include the spindle thermal deformation vector ΔP_thermal caused by machine tool heat generation and the tool wear geometric offset ΔP_wear caused by tool cutting. ΔP_thermal and ΔP_wear are used as dynamic parameters and input into the digital twin model in real time. The digital twin model reads the next one or more NC program codes to be executed in the CNC system. In a virtual environment, the error vectors are superimposed in real time on the ideal tool path of the NC code, generating a "true predicted trajectory" that reflects the errors in the real world.

[0074] Furthermore, the "virtual surface model" created after simulating the "real predicted trajectory" is compared in three dimensions with the pre-loaded theoretical CAD model of the workpiece, and the predicted shape deviation is calculated point by point. The covariance matrix output from the factor graph optimization process is used to assess the uncertainty of each predicted shape deviation.

[0075] A confidence interval with a specific confidence level, such as 95%, is generated for the prediction error to indicate the reliability of the prediction result. When the upper or lower bounds of the confidence interval for the prediction error exceed the preset tolerance band, the system triggers an information alert, prompting the operator to pay attention and continue processing. When the mean of the prediction error exceeds a certain percentage of the preset tolerance band, such as 80%, the system triggers an action alert and automatically calls the reinforcement learning module to request adjustment of subsequent processing parameters.

[0076] This closed-loop prediction and control ensures machining accuracy remains within the permitted range, directly improving the final surface quality of the mold. Furthermore, by predicting tool wear, predictive tool management is implemented, maximizing tool life and improving overall production efficiency. Within the digital twin model's physics engine, a precise 3D model is constructed, encompassing the machine's main components, spindle, worktable, tooling, toolholders, and fixtures.

[0077] The core of the process is to build a real-time work-in-progress model of the workpiece. This model subtracts the processed material from the blank model in real time based on the progress of the virtual machining simulation, dynamically updating the workpiece's immediate geometry. The physics engine reads and pre-calculates the motion trajectory for the next few seconds from the CNC system's command buffer.

[0078] High-speed geometric interference detection is continuously performed at a frequency of 1000Hz on all model entities in the dynamic simulation scene, especially between the tool holder, spindle head and the real-time workpiece model and fixture.

[0079] When the physics engine predicts a potential collision risk, the system automatically implements the following classification strategies based on the imminence of the risk:

[0080] The first-level safety strategy is information warning, the second-level safety strategy is automatic feed hold, and the third-level safety strategy is emergency stop.

[0081] Based on multi-sensor data and digital twin models, it predicts machining accuracy in real time and, by quantifying error uncertainty, promptly triggers warnings or adjusts machining parameters, significantly optimizing production efficiency, extending tool life, and improving surface quality. It continuously monitors potential interference between tools, workpieces, and machine tools, accurately predicts collision risks using a physics engine, and automatically triggers hierarchical safety strategies, such as deceleration and pauses, effectively avoiding costly collisions and protecting both equipment and workpieces.

[0082] Furthermore, the actual mold cavity error data detected by a high-precision three-dimensional coordinate measuring machine after machining is completed is fed back to the system as a true label for online fine-tuning of the point cloud error correction model weights and correction of the relevant physical parameters of the digital twin model. Through this process of continuous learning and self-evolution, the model becomes increasingly adaptable to specific machine tools and machining tasks. The system integrates a reinforcement learning agent whose state is the current sensor data, the predicted state of the digital twin model, and the remaining machining path information; the action is to fine-tune the key parameters of the next NC program, including the feed rate F, spindle speed S, and cutting step Ap / Ae; the reward is a comprehensive score that comprehensively predicts the improvement in machining accuracy, machining efficiency, and tool health status, which can dynamically optimize the compensation strategy to achieve the optimal balance between machining quality, production efficiency, and cost control.

[0083] Furthermore, when the system predicts that chatter marks may be generated during the finish machining of the side wall of a deep cavity in the mold due to excessive tool overhang, the RL agent will automatically decide to change the original large cutting depth and low feed strategy to a small cutting depth and high feed multi-level milling strategy, sacrificing a small amount of theoretical time in exchange for higher surface quality and processing stability, ultimately reducing the workload of subsequent manual polishing.

[0084] Deeply integrating multi-sensor data with the CNC system and introducing reinforcement learning algorithms enables dynamic, adaptive adjustment of tool paths, enabling real-time optimization based on actual machining conditions, improving machining efficiency and precision. Simulating machining dynamics with a digital twin model enables early prediction of potential errors and real-time optimization of path parameters, avoiding the lag of waiting to correct machining errors until problems arise. This allows the machining process to self-correct and optimize based on actual conditions, significantly improving machining stability and final product quality.

[0085] The distributed sensor control architecture involves connecting each group or key sensor, such as the temperature and vibration sensors mounted on the main shaft, to an edge computing node equipped with edge computing capabilities: an embedded computer powered by an ARM Cortex-A series processor. This node locally performs data cleaning, filtering, feature extraction, and the aforementioned drift correction algorithms. This architecture handles the bulk of computing tasks at the source of the data, significantly reducing the computational load on the main controller and network communication latency, reducing the real-time response latency of the entire system from tens of milliseconds in centralized processing to less than 10ms.

[0086] Distributed collaboration uses a distributed coordination protocol to exchange information and make collaborative decisions between sensor nodes and between them and the CNC controller. For example, if a spindle vibration node detects a spectrum anomaly (suspected chatter) and a force sensor node simultaneously detects periodic and dramatic fluctuations in cutting force, the two nodes confirm the occurrence of the event through a cross-validation protocol and collaboratively request the CNC system to adjust the spindle speed in real time to avoid the chatter zone. This distributed collaboration mechanism responds much faster than the traditional model, which aggregates all raw data to a central processor for analysis and decision-making.

[0087] Deploying sensor nodes with edge computing capabilities can perform data preprocessing and anomaly detection locally. The cross-validation protocol between sensor nodes can promptly detect and isolate faulty sensors. The use of a distributed coordination algorithm enables sensor nodes and CNC controllers to make collaborative decisions and quickly respond to processing anomalies, thereby enhancing the agility of responding to emergencies.

[0088] As an embodiment of the present invention, refer to Figure 3 The schematic diagram of the multi-sensor fusion-based precision calibration system for pentahedron machining centers is shown in the figure below. This system, through a modular design, integrates the key technologies from the aforementioned methods to form a complete and efficient precision calibration solution. This system, through collaborative operation, forms a complete closed loop from data acquisition, processing, intelligent analysis, to final control feedback, significantly improving the machining accuracy, efficiency, stability, and intelligence level of pentahedron machining centers.

[0089] In the case of machining a large automotive mold cavity, compared with traditional offline compensation and manual experience-based adjustments, the accuracy of the final mold cavity's key surface contours was significantly improved, significantly increasing the first-pass yield. Online adaptive path optimization avoided speed reductions or rework due to issues like chatter, improving overall machining efficiency. Furthermore, the improved surface quality reduced subsequent manual grinding and polishing. Tool life was extended due to effective monitoring and optimized use. Blockchain-based calibration records and full-process data monitoring establish a complete and trusted digital manufacturing history for each expensive mold, providing strong data support for mold delivery acceptance, subsequent maintenance, and repair.

[0090] Example 2:

[0091] This embodiment must simultaneously address four highly coupled challenges: macroscopic geometric accuracy, dynamic process stability, management of extreme tool conditions, and microscopic surface integrity. First, an acoustic emission (AE) sensor is added to capture high-frequency signals from tool chipping and material microcracks. The force and vibration sensors are upgraded to a sampling rate of >20kHz to accurately capture chatter characteristics. In addition to routine calibration, the AE sensor undergoes a "broken lead" test to calibrate its sensitivity.

[0092] Furthermore, continuous wavelet transform is used to perform real-time time-frequency analysis on high-frequency dynamic signals such as force, vibration, and AE to extract key features that can characterize the initiation and development of flutter.

[0093] In the factor graph model, new state variable nodes for the workpiece modal amplitude (A_modal) and surface residual stress (σ_residual) are added. Time-frequency analysis characteristics and AE signal energy are used as new factors to constrain these new nodes. Through maximum a posteriori estimation optimization, a global optimal estimate of the geometric, dynamic, and physical states is achieved.

[0094] Furthermore, the NDT-CNN-SA model's input is expanded to include multimodal data including images, thermal maps, and time-spectrograms. The network is trained for multi-task prediction, with outputs including not only geometric errors such as tool wear and thermal deformation, but also the probability of chatter occurring within the next few seconds and the level of surface residual stress in the current machining area.

[0095] The digital twin integrates the workpiece's finite element model and performs real-time coupled simulation, including modal analysis, predicting chatter risk based on current cutting parameters, dynamically generating machining stability lobe diagrams, and predicting residual stress on the machined surface through thermal-mechanical-solid coupled simulation. A three-dimensional compensation vector, C, is generated, and closed-loop feedback from the CNC system is used to correct tool paths in real time to compensate for geometric errors.

[0096] Furthermore, when a high risk of chatter or unfavorable residual stress is predicted, the system, through a reinforcement learning algorithm, goes beyond simply correcting the position and instead autonomously optimizes and adjusts the spindle speed S and feed rate F, proactively avoiding chatter zones and controlling cutting heat to achieve dynamic stability during the machining process. After machining is complete, the actual geometric dimensions and residual stresses are measured using equipment such as a coordinate measuring machine (CMM) and an X-ray diffractometer. This real-world data serves as labels for online fine-tuning of the NDT-CNN-SA model and the digital twin model, enabling continuous learning and evolution of the system.

[0097] Through multi-sensor data fusion and real-time dynamic compensation technology, compared to traditional static compensation and manual experience-based adjustment methods, comprehensive improvements in machining accuracy, surface quality, process stability, and overall efficiency are achieved. Table 2 compares the machining performance indicators of the inventive method with those of non-intelligent, non-real-time compensation methods. Specifically, blade profile accuracy is improved to a stable ±0.015 mm, surface roughness is optimized to an Ra of less than 0.4 microns, the chatter rate is controlled below 5%, the first-pass machining pass rate is 95%, tool life is extended by 40%, and overall machining efficiency is increased by 30%.

[0098] These improvements stem from the collaborative optimization and intelligent control capabilities of the entire system. Through real-time data acquisition, multi-source information fusion, and dynamic compensation mechanisms, they achieve comprehensive improvements in machining accuracy, process stability, and production efficiency. The system integrates functional modules such as error modeling, process control, and predictive maintenance to form a closed-loop intelligent machining system.

[0099] Table 2: Comparison of performance indicators between the method of the present invention and non-intelligent, non-real-time compensation processing

[0100]

[0101] Note: Positive stress (+) helps the crack "open", negative stress (-) helps the crack "close".

[0102] In summary, this embodiment successfully expands the scope of precision calibration from static geometric compensation to active control of dynamic process stability and surface integrity through comprehensive deepening of sensor systems, data fusion, intelligent modeling, and control strategies, thus solving the complex parts processing problems commonly found in high-end manufacturing.

[0103] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A pentahedron machining center precision calibration method based on multi-sensor fusion, characterized in that: The specific steps include: Integrate physical quantity measurement sensors in the pentahedron machining center, perform on-site multi-point calibration of the sensors and establish a drift correction model. Use blockchain technology to trace the calibration process and deploy redundant sensor arrays to detect and isolate anomalies. An adaptive filtering algorithm is used to preprocess sensor data, establish a global coordinate system, and perform spatiotemporal synchronization and coordinate fusion. Confidence-based fusion and factor graph optimization are used, and the closed-loop feedback mechanism of the CNC system is utilized to generate precision compensation control signals. Calibration is performed using a point cloud error correction model, combined with a digital twin model, to predict machining state parameters in real time, including tool wear, thermal deformation, and material inconsistency errors. An online adaptive mechanism is adopted to update the calibration point cloud error correction model and digital twin model in real time based on the precision compensation control signal and processing status parameters, and adjust the tool path; a distributed sensor control architecture is designed, and the calibration compensation algorithm is embedded in the local processing of the sensor node to collaborate with the CNC system.

2. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1 is characterized in that: The physical quantity measurement sensors include laser displacement sensors, temperature sensors, force sensors, vibration sensors, and fiber Bragg grating sensors; the fiber Bragg grating sensors are used to obtain local deformation data of workpieces and tools, and to obtain spatial posture data of workpieces, tools, fixtures, and machine tools through a binocular vision system; real-time time characteristic calibration is performed by estimating and dynamically compensating for the time delay, jitter, and sampling rate inconsistency time characteristic deviations of the sensor data.

3. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: On-site multi-point calibration of sensors and establishment of a drift correction model include: on-site multi-point calibration of the physical quantity measurement sensor in a processing environment, real-time detection and correction of signal drift, and use of CNC system feedback adjustment to ensure sensor accuracy; using historical sensor operation data to predict performance degradation trends and trigger predictive calibration and maintenance; recording sensor calibration data and processes through blockchain technology to ensure data traceability and reliability; deploying redundant sensors, and identifying and isolating abnormal sensors through cross-validation of CNC system data.

4. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: The confidence-based fusion and factor graph optimization includes the following steps: Dynamic weights are assigned to the fiber Bragg grating sensor and binocular vision system data based on sensor measurement accuracy and environmental interference, including illumination changes and electromagnetic noise. A factor graph model is constructed, with the spatial pose data and local deformation data of the workpiece, tool, fixture, and machine tool as nodes and the sensor measurement values ​​as factors, and the global accuracy compensation parameters are optimized through maximum a posteriori estimation.

5. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: The point cloud error correction model includes: The point cloud data generated by the binocular vision system is registered and converted into a normal distribution model based on the probability density function. Through multi-layer convolution and pooling operations under the convolutional neural network, tool wear, thermal deformation and material inconsistency errors are extracted.

6. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: The calibration and compensation algorithm includes: using a numerical control system to automatically measure tool geometric parameters, installation position and wear status, dynamically compensating errors in real time, and optimizing compensation strategies by combining sensor data and digital twin simulation; realizing automated calibration of dedicated probes, modeling and compensating for probe geometric characteristics, contact forces and dynamic errors, and using numerical control system error mapping technology to optimize multi-axis linkage accuracy; performing real-time compensation on multi-component force sensors to offset the effects of non-axial forces and correct thermal drift, and using numerical control system algorithms to decouple axial force components to improve detection accuracy.

7. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: The predicted machining state parameters include: real-time prediction of machining accuracy based on multi-sensor data and digital twin models, quantification of error uncertainty using the CNC system, triggering early warnings and / or parameter adjustments, and optimizing production efficiency, tool life, and surface quality; continuous monitoring of interference between tools, workpieces, and machine tools, and automatically triggering hierarchical safety strategies based on prediction of collision risks using a physical engine.

8. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: The tool path adjustment includes: integrating multi-sensor data into the CNC system and dynamically adjusting the tool path using a reinforcement learning algorithm; simulating machining dynamics with a digital twin model, predicting errors and optimizing path parameters in real time, and continuously updating the path planning based on sensor data through a closed-loop feedback mechanism of the CNC controller to achieve adaptive control.

9. The precision calibration method for a pentahedron machining center based on multi-sensor fusion according to claim 1, characterized in that: The distributed sensor control architecture is achieved through the following methods: deploying sensor nodes with edge computing capabilities to perform localized data processing and anomaly detection, collaborating with the CNC system to reduce control latency; detecting and isolating faulty sensors through a cross-validation protocol between sensor nodes; and using a distributed coordination algorithm to enable sensor nodes and CNC controllers to make collaborative decisions, quickly respond to machining anomalies, and achieve distributed control.

10. The pentahedron machining center precision calibration system based on multi-sensor fusion is characterized by: include: Sensor system construction and calibration module: Integrates physical quantity measurement sensors into the pentahedron machining center, performs on-site multi-point calibration of the sensors and establishes a drift correction model. Calibration process traceability is achieved through blockchain technology, and redundant sensor arrays are deployed for anomaly detection and isolation. Multi-source data preprocessing and fusion module: Uses adaptive filtering algorithms to preprocess sensor data, establishes a global coordinate system, and performs spatiotemporal synchronization and coordinate fusion. Intelligent modeling and state prediction module: Based on confidence fusion and factor graph optimization, it utilizes the CNC system's closed-loop feedback mechanism to generate precision compensation control signals. It also uses a point cloud error correction model for calibration and, combined with a digital twin model, predicts machining state parameters in real time, including tool wear, thermal deformation, and material inconsistency errors. Adaptive compensation and path optimization module: This module uses an online adaptive mechanism to update the calibration point cloud error correction model and digital twin model in real time based on the precision compensation control signal and machining status parameters, and adjusts the tool path. Also, it uses a distributed collaborative control module: This module designs a distributed sensor control architecture, embeds the calibration compensation algorithm in the local processing of sensor nodes, and collaborates with the CNC system.

Images