Intelligent monitoring method for tool wear in laser-assisted micro-milling based on digital twinning
By using digital twin technology and intelligent network models, real-time monitoring and early warning of tool wear during laser-assisted micro-milling were achieved, solving the problem of difficulty in comprehensive display and control in existing technologies, and improving machining accuracy and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN UNIV OF TECH
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-02
AI Technical Summary
Existing tool wear monitoring methods are difficult to achieve comprehensive and intuitive three-dimensional display interaction in laser-assisted micro-milling processes, and lack real-time prediction and control functions, resulting in limited machining accuracy and efficiency.
A digital twin technology is used to establish an intelligent prediction model for tool wear by combining convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM). By collecting triaxial force signals in real time and displaying and predicting them synchronously on the Unity3D platform, real-time monitoring and early warning of tool wear can be achieved.
It enables real-time monitoring and early warning of tool wear during laser-assisted micro-milling, improving machining accuracy and efficiency, reducing the cost of frequent tool changes, and ensuring machining quality and machine tool safety.
Smart Images
Figure CN122125547A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of laser-assisted micro-milling tool wear monitoring, and relates to an intelligent monitoring method for laser-assisted micro-milling tool wear based on digital twins. Background Technology
[0002] Laser-assisted machining (LAM) is a machining concept proposed to reduce cutting forces during the cutting of difficult-to-machine materials. It softens the workpiece by heating, reducing the material's strength and hardness, thereby reducing cutting forces, tool wear, and ultimately improving surface quality and machining efficiency. Laser-assisted micro-milling (LAMM) combines laser and micro-milling technologies, using lasers to soften materials and improve their machinability. It is an effective technique for machining microscale parts from difficult-to-machine materials like Inconel 718. During LAMM, the complex cutting forces and heat generated exacerbate the rapid wear of tools with small cutting diameters. The small diameter and high rotation speed of micro-milling tools make them difficult to monitor, affecting machining accuracy and increasing costs due to frequent tool changes. Therefore, it is necessary to conduct in-depth research on tool wear in laser-assisted micro-milling.
[0003] Most current tool wear monitoring methods only acquire and analyze signals during machine tool processing, focusing solely on data processing and displaying two-dimensional data. They lack interaction with the physical space, and tool wear monitoring in laser-assisted micro-milling is challenging due to the high speed and small diameter of the tools, making it difficult to provide a comprehensive and intuitive analysis of tool wear. Adding a three-dimensional interactive interface can provide more comprehensive information. Digital twin technology, with its multi-dimensional, multi-physical, and multidisciplinary attributes, as well as its high fidelity, virtual-real fusion, and real-time interactive characteristics, is well-suited for tool wear monitoring. Summary of the Invention
[0004] To address the challenge of real-time monitoring in LAMM machining processes, this invention provides a digital twin-based intelligent monitoring method for laser-assisted micro-milling tool wear. The method includes: constructing a 3D model of the laser-assisted micro-milling system using SolidWorks software and performing kinematic analysis on the LAMM machining process; establishing a synchronous motion simulation model of the machining process by driving the 3D model through a motion control program; establishing an intelligent tool wear prediction model based on a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM); collecting triaxial force signals in real-time during machining using a data acquisition card and synchronously transmitting them to the LAMM digital twin monitoring system built using Unity3D via Socket communication; combining the triaxial force data with the intelligent tool wear prediction model to achieve real-time prediction of tool wear during LAMM machining; developing an over-limit alarm function; and combining the correlation between the triaxial force data and the tool diameter reduction rate during LAMM machining to achieve early warning and control of tool wear in the LAMM machining system.
[0005] The technical solution of the present invention is as follows: The intelligent monitoring method for wear of laser-assisted micro-milling tools based on digital twins includes the following steps: Step 1: Use SolidWorks software to create a 3D geometric model of the laser-assisted micro-milling (LAMM) machining system, and use 3DS Max to perform lightweight processing and rendering of the 3D geometric model; import the created 3D geometric model into Unity3D software; Step 2: Analyze the kinematic relationships of the LAMM process based on the three-dimensional geometric model obtained in Step 1, establish the parent-child relationships between the components of each three-dimensional model, and then establish a synchronous motion simulation model of the LAMM process through a program-driven approach. Step 3: Based on the LAMM tool wear pre-experiment, voltage signals during the machining process are acquired by a voltage sensor using a data acquisition card. The 8-channel voltage signals acquired by the data acquisition card are converted using the conversion model provided by Kistler: Fx = (channel 0 + channel 1) × 10N; Fy = (channel 4 + channel 5 + channel 6 + channel 7) × 10N; Fz = (channel 2 + channel 3) × 10N. The data acquisition card is then further developed using C# code to embed the conversion model, enabling the real-time conversion of the acquired voltage signals into three-dimensional cutting force signals. After storing the force data, the data is updated to the dataset in real time and a new CSV file is generated. Here, Fx, Fy, and Fz represent the cutting force in the X, Y, and Z directions, respectively, and N represents Newton, an international unit for measuring the magnitude of force. Step 4: Obtain the cutting force and tool diameter dataset based on the laser-assisted micro-milling tool wear test, and use Matlab software to establish a laser-assisted micro-milling tool wear intelligent prediction model based on cutting force to predict the tool diameter reduction rate; Step 5: Integrate the established synchronous motion simulation model and laser-assisted micro-milling tool wear intelligent prediction model into Unity3D software to form a digital twin model; Step 6: During laser-assisted micro-milling, based on the TCP protocol, the triaxial cutting force data collected in real time by the data acquisition card is synchronously transmitted via Socket communication to the intelligent prediction model for laser-assisted micro-milling tool wear embedded in Unity3D software to achieve real-time monitoring of micro-milling tool wear. At the same time, the real-time triaxial cutting force data collected by the data acquisition card and the predicted micro-milling tool diameter reduction rate data are combined with the digital twin model to dynamically, realistically, and multidimensionally map the actual laser-assisted micro-milling process, realizing the real-time three-dimensional visualization function of laser-assisted micro-milling and the monitoring function of micro-milling tool wear. Step 7: When the reduction rate of the micro-milling tool diameter reaches the set threshold (e.g., 20%), the over-limit alarm function is triggered to provide a reference for laser-assisted micro-milling control.
[0006] The beneficial effects of this invention are: This invention utilizes digital twin technology and leverages convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM) to establish an intelligent prediction model for tool wear. This model enables intelligent monitoring of tool wear using cutting force as the measurement signal. Simultaneously, digital twin technology allows for real-time display of the wear status of micro-milling cutters; real-time simulation of LAMM machining is achieved, tracking the tool trajectory; and visualization and online prediction of tool wear are realized. This invention achieves functions such as synchronous mapping of the machining process, 3D visualization, and early warning and control of tool wear. The monitoring system has been field-verified and can provide a reference for the process control of laser-assisted micro-milling. This monitoring method offers a new approach and concept for comprehensively managing the machining status of high-end CNC machine tools, which is of great significance for improving the automation level of the production process, increasing production efficiency, ensuring product quality, and avoiding damage to machine tools, cutting tools, and workpieces. Attached Figure Description
[0007] Figure 1 This is the overall technical solution of the monitoring model established by the laser-assisted micro-milling tool wear monitoring method based on digital twin according to the present invention; Figure 2 This is a diagram of the CNN-BiLSTM model structure; Figure 3 This is a graph showing the prediction results of the test set of the intelligent prediction model for tool wear; Figure 4 This is a diagram of the Socket communication connection process. Detailed Implementation
[0008] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.
[0009] The present invention provides a digital twin model of tool wear that features strong real-time performance, high visualization, and good intuitiveness, such as... Figure 1 As shown, the LAMM tool wear system uses synchronous motion simulation to dynamically drive the 3D model, visually simulating the machining process. This achieves synchronous mapping between the virtual model and the physical entity, and enhances the realism and real-time performance of micro-milling tool wear. Specifically: A 1:1 3D model of the laser-assisted micro-milling system was constructed using SolidWorks software. The established 3D geometric model was then imported into 3ds Max software to set materials and colors, and underwent lightweight processing. The processed 3D model was then imported into Unity3D software. In addition to the 3D model, kinematic analysis of the machining process was performed. The movement of the micro-milling machine tool and its laser emitter consists of four motion axes. Motion analysis was conducted on each axis of the laser-assisted micro-milling system. The X-axis is the machine tool's transverse feed axis, controlled by a servo motor, moving the stage along the bed's crossbeam guide rail to achieve the X-axis feed function of the micro-milling cutter relative to the workpiece. The Y-axis is the machine tool's feed axis, controlled by a servo motor, moving the stage along the bed's Y-axis guide rail to achieve the Y-axis feed function of the micro-milling cutter relative to the workpiece. The Z-axis movement is achieved by the movement of the machine tool's headstock on the vertical guide rail, requiring stable and accurate motion. The spindle is a crucial component of the machine tool, its function being to drive the tool to rotate at high speed and transmit torque. The spindle speed of a milling machine determines the cutting speed and the magnitude of the cutting force, which in turn affects the machining efficiency and machining quality. The process of establishing the synchronous motion simulation model: Combining the established 3D model with kinematic analysis, the kinematic relationships between models are established. The motion relationships between laser-assisted micro-milling models are described through parent-child construction relationships. A tree structure of the motion model of the laser-assisted micro-milling machining system is established based on Unity3D software. After the parent-child construction relationships between model components are established, the control of its child objects is obtained level by level through scripts on the parent object. In Unity3D software, the Transform component is used to control the movement and rotation of each part model. Based on kinematic analysis, C# programs to control the motion of each model are written and dragged onto the corresponding models in the virtual scene. Box Collider and Rigid Body components are added to simulate the collision and gravity effects of the real world. Synchronous mapping of the laser-assisted micro-milling machining process is achieved with the help of the engine. At the same time, the model motion is controlled by adjusting the cutting process parameters on the Inspector component. Different instructions are used as needed to control the changes in motion and keep it synchronized with the moving objects in the real physical scene.
[0010] In this embodiment, Incone 1 718, a nickel-based superalloy used in key components in the aerospace field, was selected as the machining material. The workpiece was prepared for continuous groove milling, with air cooling employed during the process. Before conducting the LAMM experiment, several laser parameters were determined, including laser power, laser incident angle, and the distance between the laser spot and the cutting zone. The laser power was 35W, the laser incident angle was 31°, and the laser spot size was 400μm.
[0011] To obtain information on tool wear changes, machining needs to be stopped, the tool measured, and machining resumed after obtaining the wear measurement value. This process is repeated to obtain the tool wear state at different times. However, this causes the accumulated cutting heat during the cutting process to dissipate, and cutting heat is one of the main factors causing tool wear. The wear state produced by cutting after the tool has stopped differs significantly from that produced by continuous cutting. Therefore, 40 sets of laser-assisted micro-milling experiments were conducted, with cutting times ranging from 6s to 240s, and the cutting time increment between adjacent sets being 6s. The MX230 infinitely coated (mainly composed of TiAlN) double-flute carbide milling tool manufactured by Nissin was selected, with a tool diameter of 400µm. The cutting parameters recommended by Nissin were: spindle speed 60000r / min, feed per tooth 0.5µm, depth of cut 20µm, feed rate 1mm / s, and tool overhang 15mm. After each cutting experiment, the tool diameter was measured using a VHX-600E ultra-depth-of-field microscope, and the diameter reduction rate was calculated to characterize the tool's wear state. The tool diameter reduction rate is calculated as shown in formula (1). The cutting force at different times was recorded to establish the correlation between the two, providing a data basis for predicting the tool wear state in the later stage.
[0012] (1) in, The initial diameter of the tool. This is the diameter of the tool after wear.
[0013] The data obtained during the experiment are summarized in the table below.
[0014] As a key design feature of this invention, the intelligent prediction model for laser-assisted micro-milling tool wear is based on a CNN-BiLSTM network to predict tool wear. The CNN-BiLSTM network structure is as follows: Figure 2 As shown, a CNN network is first used to extract features from the three-dimensional cutting force and tool diameter data. Then, the feature sequences extracted by the CNN are input into a BiLSTM network, which is used to model the sequence data and learn the temporal features of tool wear state. The CNN-BiLSTM model combines convolutional and recurrent structures, enabling it to capture multi-scale features in the data at different levels.
[0015] The structure of the intelligent prediction model for laser-assisted micro-milling tool wear based on a CNN-BiLSTM network includes a data input layer, convolutional layers, pooling layers, unfolded layers, LSTM layers, dropout layers, fully connected layers, and an output layer. The input layer receives data; the convolutional and pooling layers are responsible for feature extraction and dimensionality reduction; the unfolded layer inputs the extracted feature sequences into the BiLSTM layer; the dropout layer prevents overfitting and monitors the tool wear state; the fully connected layer is used for feature fusion and output; the model is trained using an adaptive Adam optimization algorithm; and finally, the output layer provides the predicted wear state.
[0016] 28 sets of data from 40 sets of data obtained in the experiment were randomly selected as the training set, and 12 sets were selected as the test set. The time lag order was set to 1:5. During training, the Adam optimizer was used, with a maximum training epoch of 400, a batch size of 128, an initial learning rate of 5e-3, and piecewise learning rate decay with a decay factor of 0.9 and a decay period of 800. A training progress chart was plotted. After approximately 100 training iterations, the root mean square error tended to converge, and the model's predicted results were compared with the actual results. The comparison results are shown below. Figure 3 As shown, the root mean square error (RMSE) in the prediction results is 0.0083, the mean absolute error (MAE) is 0.0071, and the mean absolute percentage error (MAPE) is 5.94%. The error results are small, indicating that the established LAMM tool wear prediction model is effective and has obtained good training results.
[0017] The Advantech USB5817 data acquisition card's data acquisition system underwent secondary development to realize the data communication function of the monitoring system. A Kistler 9256C1 six-component force gauge was connected to a Kistler 5080A charge amplifier. The acquired 8-channel voltage signals were transmitted in real time to the Advantech USB5817 data acquisition card. The read voltage data was converted into triaxial force data using the conversion model provided by Kistler: Fx = (channel 0 + channel 1) × 10N, Fy = (channel 4 + channel 5 + channel 6 + channel 7) × 10N, Fz = (channel 2 + channel 3) × 10N. C# programming was used to convert the voltage data into triaxial force data, displaying the cutting force data in real time and saving it to a CSV file. The acquired triaxial force data was then transmitted in real time to a digital twin monitoring system developed in Unity3D via Socket communication technology. Figure 4 As shown. The specific implementation method of the communication function is as follows: First, the data acquisition card creates a network service using a Socket handle, then uses the bind() function to bind the service to the host's IP address and assign a port number, and uses the listen() function to establish real-time listening to the monitoring system. Meanwhile, the monitoring system also starts a network service using a Socket handle and uses the connect() function to send a connection request to the Advantech data acquisition system. Once the Advantech data acquisition system successfully listens for the connection request, it uses the accept() function to accept the connection request and establish a network connection. At this point, the tool wear monitoring system and the Advantech data acquisition system can exchange data. During the laser-assisted micro-milling process, the cutting three-way cutting force data collected by the Advantech data acquisition system is saved in string type, and then the encode("utf-8") method is called to convert the string type to bytes type. After that, the Socket object can be used to call the send(msg) function to send the tool wear data in bytes type to the data buffer; Len(buff) represents the remaining available byte capacity of the Socket send buffer, and Len(msg) represents the total byte length of the bytes type cutting force data to be sent. When there is enough cache in the send buffer, that is, Len(buff)>Len(msg), msg can be sent out at one time; when there is not enough cache in the send buffer, that is, Len(buff)<Len(msg), only a part of the data of msg is sent out in one call, and the send(msg) interface needs to be called repeatedly many times to send out all the data; the tool wear digital twin monitoring system, as the receiving end, calls the recv(msg) function to receive the buffer data, where Len(buff) represents the byte length of the data to be read that has been stored in the Socket receive buffer, and Len(msg) represents the target data byte length expected to be read by a single recv call. That is, when Len(buff)>Len(msg), more than one complete data is received when the recv(msg) interface returns; when the data in the receive buffer is insufficient, that is, Len(buff)<Len(msg), incomplete data is received when the recv(msg) returns, and the recv(msg) interface needs to be called repeatedly many times to receive all the complete data; through the above transmission method, the cutting force data collected in real time by the Advantech USB5817 acquisition card is synchronously transmitted to the tool wear monitoring system to realize the real-time display of the three-way force during laser-assisted micro-milling.
[0018] The trained tool wear prediction model and function program in MATLAB are packaged and encapsulated into a DLL file, and this DLL file and the MATLAB-related DLL library files are imported into the Unity3D software together. In Unity3D, code for calling the DLL file is written in C# language to drive the tool wear monitoring model. After the tool wear monitoring system receives the real-time transmitted three-way force signal data, the three-way force data during the processing is displayed in real time in the tool wear monitoring system using GUI, and at the same time, the established prediction model is called to quickly predict the tool diameter and the tool diameter reduction rate at the current processing moment, and the real-time predicted tool wear amount is dynamically presented in the digital twin tool wear monitoring system in the form of a line chart through the XCharts component for visual analysis.
[0019] In a laser-assisted micro-milling tool wear monitoring system, a wear threshold alarm function was developed. When the tool wear prediction model predicts that the tool diameter reduction rate exceeds 20%, a warning window pops up to remind the operator to replace the tool in time and stop the machine to wait for the next operation. This effectively reduces accidents and losses caused by tool wear during the machining process, thereby improving machining efficiency.
[0020] The LAMM tool wear monitoring system based on digital twins is built with the Unity3D engine. Its functions such as synchronous mapping of the machining process, 3D visualization of the machining system, and early warning and control of tool wear status can help operators issue timely control and emergency commands to the physical entity during the LAMM process, adjust cutting parameters, laser incident angle, and laser power combination to keep the tool wear status within a reasonable range during the machining process, and improve machining quality.
[0021] This invention can verify the intelligent tool wear monitoring system using a micro-milling machine tool and a laser emitting device. The constructed intelligent monitoring system for laser-assisted micro-milling tool wear, based on digital twins, continuously transmits triaxial force data from the Advantech acquisition system to the monitoring system via Socket communication during LAMM machining, predicting the tool diameter reduction rate at a prediction frequency of 1Hz. This enables motion simulation, 3D visualization, and early warning and control of tool wear in the laser-assisted micro-milling machining system.
Claims
1. A method for intelligent monitoring of wear of laser-assisted micro-milling tools based on digital twins, characterized in that, The specific steps are as follows: Step 1: Use SolidWorks software to create a 3D geometric model of the laser-assisted micro-milling (LAMM) machining system, and use 3DS Max to perform lightweight processing and rendering of the 3D geometric model; import the created 3D geometric model into Unity3D software; Step 2: Analyze the kinematic relationships of the LAMM process based on the three-dimensional geometric model obtained in Step 1, establish the parent-child relationships between the components of each three-dimensional model, and then establish a synchronous motion simulation model of the LAMM process through a program-driven approach. Step 3: Based on the LAMM tool wear pre-experiment, the voltage signals during the machining process are acquired by the voltage sensor through the data acquisition card. The 8-channel voltage signals acquired by the data acquisition card are converted into triaxial cutting force signals in real time through the conversion model provided by Kistler's official website. The data acquisition card is then further developed using C# code to embed the conversion model, so as to realize the real-time conversion of the acquired voltage signals into triaxial cutting force signals, store the force data, update the dataset in real time, and generate a new CSV file. Step 4: Obtain the cutting force and tool diameter dataset based on the laser-assisted micro-milling tool wear test, and use Matlab software to establish a laser-assisted micro-milling tool wear intelligent prediction model based on cutting force to predict the tool diameter reduction rate; Step 5: Integrate the established synchronous motion simulation model and laser-assisted micro-milling tool wear intelligent prediction model into Unity3D software to form a digital twin model; Step 6: During laser-assisted micro-milling, based on the TCP protocol, the triaxial cutting force data collected in real time by the data acquisition card is synchronously transmitted via Socket communication to the intelligent prediction model for laser-assisted micro-milling tool wear embedded in Unity3D software to achieve real-time monitoring of micro-milling tool wear. At the same time, the real-time triaxial cutting force data collected by the data acquisition card and the predicted micro-milling tool diameter reduction rate data are combined with the digital twin model to dynamically, realistically, and multidimensionally map the actual laser-assisted micro-milling process, realizing the real-time three-dimensional visualization function of laser-assisted micro-milling and the monitoring function of micro-milling tool wear. Step 7: When the reduction rate of the micro-milling tool diameter reaches the set threshold, the over-limit alarm function is triggered to provide a reference for the control of laser-assisted micro-milling.
2. The intelligent monitoring method for laser-assisted micro-milling tool wear based on digital twin as described in claim 1, characterized in that, Step 2 involves establishing the synchronous motion simulation model: combining the established 3D model with kinematic analysis, establishing kinematic relationships between models. The motion relationships between laser-assisted micro-milling models are described through parent-child construction relationships. A tree structure of the motion model of the laser-assisted micro-milling machining system is established based on Unity3D software. After establishing the parent-child construction relationships between model components, the control of its child objects is obtained level by level through scripts on the parent object. In Unity3D software, the Transform component is used to control the movement and rotation of each part model. Based on kinematic analysis, C# programs to control the motion of each model are written and dragged onto the corresponding model in the virtual scene. Box Collider and Rigid Body components are added to simulate the collision and gravity effects of the real world. Synchronous mapping of the laser-assisted micro-milling machining process is achieved with the help of the engine. At the same time, the model motion is controlled by adjusting the cutting process parameters on the Inspector component. Different instructions are used as needed to control the changes in motion, keeping it synchronized with objects in motion in the real physical scene.
3. The intelligent monitoring method for laser-assisted micro-milling tool wear based on digital twin as described in claim 1, characterized in that, The conversion model in step 3 is: Fx = (channel 0 + channel 1) × 10N; Fy = (channel 4 + channel 5 + channel 6 + channel 7) × 10N; Fz = (channel 2 + channel 3) × 10N, where Fx, Fy and Fz represent the cutting force in the X direction, the cutting force in the Y direction and the cutting force in the Z direction, respectively, and N represents Newton, which is an international unit for measuring the magnitude of force.
4. The intelligent monitoring method for laser-assisted micro-milling tool wear based on digital twin as described in claim 1, characterized in that, In step 4, the intelligent prediction model for laser-assisted micro-milling tool wear is based on a CNN-BiLSTM network to predict tool wear. First, the CNN network is used to extract features from the three-dimensional chip force and tool diameter data. Then, the feature sequence extracted by the CNN is input into the BiLSTM network, and the BiLSTM network is used to model the sequence data to learn the temporal features of the tool wear state.
5. The intelligent monitoring method for laser-assisted micro-milling tool wear based on digital twin according to claim 1, characterized in that, The structure of the intelligent prediction model for laser-assisted micro-milling tool wear based on CNN-BiLSTM network includes a model data input layer, convolutional layer, pooling layer, unfolded layer, LSTM layer, dropout layer, fully connected layer, and output layer. The input layer receives data, the convolutional and pooling layers are responsible for feature extraction and dimensionality reduction, the unfolded layer inputs the extracted feature sequence into the BiLSTM layer, the dropout layer is used to prevent overfitting and realize the monitoring of tool wear state, the fully connected layer is used for feature fusion and output, the model is trained using the adaptive learning rate Adam optimization algorithm, and finally the output layer gives the prediction result of wear state.