Digital twin cutter wear life prediction method and system based on physical information neural network

Abstract

The invention relates to a digital twin cutter wear life prediction method and system based on a physical information neural network. The method comprises the steps of s1, cutter data processing; s2, resultant force calculation; s3, the cutting speed is calculated; s4, calculating a sliding distance; s5, load calculation; s6, constructing a neural network; building a physical information neural network based on the main cutting force model and the Archard wear model; s7, establishing a loss function; s8, performing feedback and iteration; s9, calculating the abrasion loss of the cutter; and calculating the cutter abrasion loss through the physical information neural network. According to the method, a physical formula related to the cutter wear stage of the numerical control machine tool is fused into the physical information neural network to serve as a part of a loss function, it can be guaranteed that data is subjected to iterative training on the premise of related physical laws, and the interpretability of the process is enhanced.

Description

Technical Field

[0001] The present invention relates to a method and system for predicting the wear life of a digital twin tool based on a physical information neural network, and belongs to the technical field of intelligent manufacturing tool fault diagnosis and health management. Background Art

[0002] Tool wear is a key factor affecting many aspects of metal cutting, including surface quality, machining efficiency and tool life. During machining, tool wear will reduce the dimensional accuracy and surface integrity of parts. When the tool wear is severe, it may even cause tool breakage, resulting in workpiece scrapping and damage to the machine tool. By predicting the tool wear life, production plans can be arranged more accurately to avoid production interruptions caused by excessive tool wear. This is of great significance for improving production efficiency and reducing downtime. And tool wear life prediction is an important part of realizing intelligent manufacturing. By combining tool wear life prediction with other intelligent manufacturing technologies (such as machine learning, big data analysis, etc.), the production process can be automated and intelligent, production efficiency can be improved, production costs can be reduced, and product quality can be improved. However, most of the existing methods use data-driven prediction methods or methods that combine finite element simulation with data prediction to complete tool wear life prediction, which has certain defects in interpretability and calculation speed. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present invention provides a method for predicting the wear life of a digital twin tool based on a physical information neural network;

[0004] The present invention can integrate the tool wear-related physical formulas into the neural network so that the tool wear data can be predicted under the condition of satisfying the physical laws, thereby ensuring the physical consistency and realizing the dual drive of the data mechanism model in the construction of the digital twin model. First, the physical laws of tool wear are clarified and embedded into the physical information neural network as the loss function. Since the tool wear life prediction is calculated, the feed force model and the Archard wear model are combined in the present invention to predict the tool wear amount using the physical information neural network. The feed force model can provide force information during the cutting process, and the Archard wear model can convert these forces into wear amount. Therefore, during the training process, the loss function includes data errors and physical errors, so that the model follows the physical laws while fitting the data, increases the interpretability, and reduces the calculation time relative to the method combining finite element simulation and data prediction.

[0005] The present invention provides a digital twin tool wear life prediction system based on physical information neural network;

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A digital twin tool wear life prediction method based on physical information neural network, comprising:

[0008] s1: Tool data processing; tool data is the tool wear data set, including the milling force signals of the X-axis, Y-axis, and Z-axis. x ,F y ,F z ) in the feed force model and wear model; (F x ,F y ,F z ) refer to the milling force signals of the X-axis, Y-axis and Z-axis respectively;

[0009] s2: calculation of resultant force;

[0010] s3: cutting speed calculation;

[0011] s4: sliding distance calculation;

[0012] s5: load calculation;

[0013] s6: Neural network construction; build physical information neural network based on main cutting force model and Archard wear model;

[0014] s7: loss function construction;

[0015] The main cutting force model and Archard wear model are combined to calculate the tool wear, so as to determine the physical loss function in the physical information neural network. The calculation formula is as follows:

[0016]

[0017] Among them, F f is the main cutting force, K f is the main cutting force constant, A c is the cutting area, v is the cutting speed, n e is the velocity index, W is the wear amount, k is the wear coefficient, F is the load, s is the sliding distance, and H is the material hardness.

[0018] s8: Feedback and iteration, i.e. model evaluation and verification, evaluation and verification of the updated physical information neural network;

[0019] s9: Calculate tool wear; calculate tool wear through physical information neural network.

[0020] Preferably, the resultant force calculation according to the present invention comprises:

[0021] The force on each axis is generalized to obtain the resultant force F c, the calculation formula is as follows:

[0022]

[0023] According to the preferred embodiment of the present invention, the cutting speed calculation comprises:

[0024] For a rotating workpiece, the cutting speed v is calculated as follows:

[0025] v = π·D·n(3);

[0026] Where v is the cutting speed in meters per minute, D is the diameter of the workpiece in meters, and n is the spindle speed in revolutions per minute.

[0027] Preferably, according to the present invention, the sliding distance calculation includes:

[0028] The sliding distance calculation formula is as follows:

[0029]

[0030] Where s is the sliding distance in mm, f is the feed amount per tool pass in mm / rev, N is the total number of revolutions, n is the spindle speed, and t is the machining time in minutes.

[0031] Preferably, the neural network construction according to the present invention refers to: building a physical information neural network based on the main cutting force model and the Archard wear model, integrating physical constraints into the loss function of the physical information neural network, and realizing the prediction of the wear amount; including:

[0032] 1) Data preparation: the data includes sensor data and wear data;

[0033] Sensor data: including force / vibration signals of the three axes (F x ,F y ,F z );

[0034] Wear data: the actual wear value of the tool;

[0035] 2) Physical constraint definition;

[0036] The physical constraint definition part is completed by combining the main cutting force model and the Archard wear model; the main cutting force model is formula (2), and the Archard wear model is formula (1);

[0037] 3) Customized loss function;

[0038] The custom loss function is the differential equation derived from the Archard wear formula according to time, as shown below:

[0039]

[0040] 4) Construction of physical information neural network;

[0041] Build a physical information neural network based on the main cutting force model and Archard wear model;

[0042] 5) Training model;

[0043] The force / vibration signals of the X, Y, and Z axes and their corresponding wear data are input into the physical information neural network for training. The physical information neural network continuously adjusts the parameters by optimizing the custom loss function until the physical information neural network converges; including:

[0044] 6) Wear prediction process;

[0045] After the physical information neural network training is completed, it enters the prediction stage; input the force / vibration signals of the three axes of X, Y, and Z: obtain new force or vibration signals from the sensor, and use the trained physical information neural network to predict the wear amount;

[0046] 7) Compare with actual wear value

[0047] Plot a graph to compare the wear values ​​predicted by the model with the actual wear values.

[0048] Further preferably, comparing with the actual wear value; comprising:

[0049] A. Data acquisition and division:

[0050] Take a part of the prepared data set as test data or verification data;

[0051] B. Obtaining model prediction results:

[0052] The input features of the test data or the verification data are fed into the trained physical information neural network, and the corresponding predicted wear value is obtained through forward propagation;

[0053] C. Comparison curve chart: With the index or time of the data point as the horizontal axis, the actual wear value and the predicted wear value are plotted as two curves for comparison.

[0054] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the processor implements the steps of a digital twin tool wear life prediction method based on a physical information neural network.

[0055] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a digital twin tool wear life prediction method based on a physical information neural network.

[0056] A digital twin tool wear life prediction system based on physical information neural network, comprising:

[0057] The tool data processing module is configured as follows: tool data is a tool wear data set, including milling force signals of the three axes of X-axis, Y-axis and Z-axis, and the force (F x ,F y ,F z ) in the feed force model and wear model; (F x ,F y ,F z ) refer to the milling force signals of the X-axis, Y-axis and Z-axis respectively;

[0058] A calculation module is configured to: calculate resultant force, cutting speed, sliding distance and load;

[0059] The neural network building module is configured to: build a physical information neural network based on the main cutting force model and the Archard wear model;

[0060] The loss function building module is configured to: calculate the tool wear amount by combining the main cutting force model and the Archard wear model, thereby determining the physical loss function in the physical information neural network;

[0061] The tool wear calculation module is configured to calculate the tool wear through a physical information neural network.

[0062] The beneficial effects of the present invention are:

[0063] 1. Interpretability: Integrating the physical formulas related to the tool wear stage of CNC machine tools into the physical information neural network as part of the loss function can ensure that the data is iteratively trained under the premise of relevant physical laws, thereby enhancing the interpretability of the process.

[0064] 2. Improve prediction accuracy: By combining data-driven and mechanism-driven methods, the mechanism-driven method is used to compensate for the errors of data-driven methods, thereby further improving prediction accuracy.

[0065] 3. Improve calculation speed: Compared with traditional finite element simulation for mechanism prediction, the present invention can improve the calculation speed of the mechanism part and perform data mechanism fusion in the overall large network architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1A schematic flow chart of a method for predicting wear life of a digital twin tool based on a physical information neural network according to the present invention;

[0067] Figure 2 This is a diagram of the architecture of the physical information neural network of the present invention;

[0068] Figure 3 This is a schematic diagram of building a physical information neural network based on the main cutting force model and the Archard wear model in the present invention. DETAILED DESCRIPTION

[0069] The present invention will be further defined below in conjunction with the accompanying drawings and embodiments, but is not limited thereto.

[0070] Example 1

[0071] A digital twin tool wear life prediction method based on physical information neural network, comprising:

[0072] s1: Tool data processing; tool data is the tool wear data set, including the milling force signals of the X-axis, Y-axis, and Z-axis. x ,F y ,F z ) in the feed force model and wear model; generally, the feed force is a resultant force that can be obtained by combining forces in all directions. (F x ,F y ,F z ) refer to the milling force signals of the X-axis, Y-axis and Z-axis respectively;

[0073] s2: calculation of resultant force;

[0074] s3: cutting speed calculation;

[0075] s4: sliding distance calculation;

[0076] s5: Load calculation; e.g. Figure 1 As shown in the figure, the load calculation process is as follows: the resultant force Fc and the main cutting force Ff calculated from the vibration force signals of the X, Y, and Z axes are squared and then square rooted to obtain the result.

[0077] s6: Neural network construction; build physical information neural network based on main cutting force model and Archard wear model;

[0078] s7: loss function construction;

[0079] The main cutting force model and Archard wear model are combined to calculate the tool wear, so as to determine the physical loss function in the physical information neural network. The calculation formula is as follows:

[0080]

[0081] Among them, F f is the main cutting force, K f is the main cutting force constant, A c is the cutting area (cutting depth × cutting width), v is the cutting speed, n e is the speed index (empirical value), W is the wear amount, k is the wear coefficient, F is the load, s is the sliding distance, and H is the material hardness.

[0082] s8: Feedback and iteration; that is, model evaluation and verification, evaluation and verification of the updated physical information neural network; the evaluation method is: during the training process, the performance of the physical information neural network is evaluated by the PINN loss indicator. The smaller the PINN loss value, the better the performance; the verification method is: verify the difference between the predicted value and the true value on the verification data set. When the difference is small enough, the physical information neural network meets the requirements. Please explain and supplement; to ensure that the performance and accuracy of the model meet the requirements.

[0083] s9: Calculate tool wear; calculate tool wear through physical information neural network.

[0084] After s9, a digital twin system for tool wear was built based on the physical information neural network, and the CNC machine tool tools were modeled based on the Unreal Engine to complete the visualization part of the digital twin system. The PINN neural network was embedded into the Unreal Engine through the Unreal Engine blueprint function, and socket communication was used to complete the function of data input into the Unreal Engine, realizing the integration of the neural network into the Unreal Engine, completing the data interaction and visualization display parts.

[0085] Example 2

[0086] The difference between the digital twin tool wear life prediction method based on physical information neural network described in Example 1 is that:

[0087] Calculation of resultant forces; including:

[0088] The force on each axis is generalized to obtain the resultant force F c , the calculation formula is as follows:

[0089]

[0090] Cutting speed calculation; including:

[0091] Cutting speed is the speed at which the tool moves relative to the workpiece, usually expressed in meters per minute (m / min). During machining, the calculation of cutting speed depends on the rotation speed of the workpiece (spindle speed) and the diameter of the workpiece. For rotating workpieces (such as turning), the calculation formula for cutting speed v is as follows:

[0092] v = π·D·n(3);

[0093] Where v is the cutting speed in meters per minute (m / min), D is the diameter of the workpiece in meters (m), and n is the spindle speed in revolutions per minute (RPM).

[0094] Slide distance calculation; including:

[0095] Sliding distance s calculation, during the tool processing, the sliding distance is directly related to the feed rate of each tool. The sliding distance is calculated by the tool feed rate, that is, the total moving distance of the tool relative to the workpiece during the entire processing process. The sliding distance calculation formula is as follows:

[0096]

[0097] Where s is the sliding distance in mm, f is the feed amount per tool pass in mm / rev, N is the total number of revolutions, n is the spindle speed, and t is the machining time in minutes.

[0098] Neural network construction: refers to: building a physical information neural network based on the main cutting force model and Archard wear model, integrating physical constraints into the loss function of the physical information neural network, and realizing the prediction of wear amount; including:

[0099] In the present invention, since tool wear life prediction is performed, the initial condition is 0, and the boundary condition is the maximum wear value when machining is impossible.

[0100] The first part is the data training part, which is to train the neural network through the input data. The second part is the PINN-specific part, which includes the PINN loss, which is composed of the main cutting force model and the Archard model. Its function is to complete the data mechanism fusion part by fitting the results with the PINN loss continuously on the basis of the data training in the first part.

[0101] 1) Data preparation: the data includes sensor data and wear data;

[0102] Sensor data: Receives sensor data such as force signals and vibration signals from machine operation, including force / vibration signals of the three axes (F x ,F y ,F z );

[0103] Wear data: the actual wear value of the tool; obtained through measurement or experiment as a supervision signal.

[0104] 2) Physical constraint definition;

[0105] The physical constraint definition part is completed by combining the main cutting force model and the Archard wear model; the main cutting force model is formula (2), and the Archard wear model is formula (1);

[0106] 3) Customized loss function;

[0107] PINN combines the loss functions of data error and physical error to optimize the model during training: Data loss: For example, the mean square error (MSE) or mean absolute error (MAE) is used to measure the gap between the wear value predicted by the model and the actual measured value. Physical loss: The error is calculated according to the physical formula and incorporated into the optimization target through a custom loss function.

[0108] The custom loss function is the differential equation derived from the Archard wear formula according to time, as shown below:

[0109]

[0110] 4) Construction of physical information neural network;

[0111] Build a physical information neural network based on the main cutting force model and Archard wear model; build a neural network model, such as LSTM network to process time series data, and combine historical force signals to predict wear;

[0112] like Figure 2 and Figure 3 As shown in the figure, the RNN network composed of LSTM is used to predict the wear value. The mechanism module in the network guides the data to be trained under physical constraints according to the physical formula. In this process, physical knowledge is integrated. The data loss is the loss obtained by training the data in the data-driven model. The final loss is composed of the two.

[0113] 5) Training model;

[0114] The force / vibration signals of the X, Y, and Z axes and their corresponding wear data are input into the physical information neural network for training. The physical information neural network continuously adjusts parameters by optimizing the custom loss function (including data error and physical constraints) until the physical information neural network converges; including:

[0115] Parameter initialization: Randomly initialize the neural network weights and biases (such as Xavier initialization).

[0116] Forward Pass: The pre-processed X, Y, Z vibration / force signals are input into PINN to obtain the predicted wear amount.

[0117] Calculate the loss function:

[0118] The data error LData is calculated based on the predicted and actual wear amount. The physical residual Lphysics is calculated based on the network output and the built-in physical equation. The weighted sum of the two is used to obtain the total loss Loss.

[0119] Backpropagation: Use automatic differentiation techniques to calculate the gradient of the total loss with respect to the network parameters.

[0120] Parameter Update: Use an optimization algorithm (such as Adam or LBFGS) to update the network parameters based on the gradient information so that the total loss continues to decrease.

[0121] Epochs: Repeat the above steps for several rounds (Epochs) until the network parameters converge, that is, the total loss or validation set error no longer decreases significantly, or the preset training stop condition is reached.

[0122] Convergence and Verification:

[0123] In the early stages of training, the network parameters are relatively disordered, and can neither fit the data well nor strictly satisfy the physical equations. As the training iterations proceed, under the joint constraints of Ldata and Lphysics, the network parameters are gradually adjusted so that the prediction results fit the measured data and satisfy the physical laws at the same time.

[0124] When the loss function reaches stability, the gradient changes tend to be stable, or the performance of the validation set reaches a satisfactory level, the network can be considered to have converged. At this time, the obtained PINN model not only fits the training data well, but can also be extrapolated to unseen data to a certain extent, and maintains the internal consistency of physical laws.

[0125] 6) Wear prediction process;

[0126] After the physical information neural network training is completed, it enters the prediction stage; input the force / vibration signals of the three axes of X, Y, and Z: obtain new force or vibration signals from the sensor, and use the trained physical information neural network to predict the wear amount;

[0127] 7) Compare with actual wear value

[0128] Plot a graph to compare the wear values ​​predicted by the model with the actual wear values.

[0129] Comparison with actual wear values; including:

[0130] A. Data acquisition and division:

[0131] Take a part of the prepared data set as test data or validation data; this data is not used to update the model parameters during the training process, so it can be used to objectively evaluate the predictive performance of the model.

[0132] B. Obtaining model prediction results:

[0133] Feed the input features of the test data or validation data into the trained physical information neural network (PINN), and obtain the corresponding predicted wear value through forward propagation; this will generate an array or list of "predicted wear values". At the same time, there is also the corresponding "actual wear value" (real measurement value) in the test data set.

[0134] C. Line / Comparison Plot: With the index or time of the data point as the horizontal axis, the actual wear value and the predicted wear value are plotted as two curves for comparison. If the two curves are close, it means that the prediction is more accurate.

[0135] The PINN-based data mechanism dual-driven digital twin tool wear life prediction method has broad prospects in practical industrial applications.

[0136] Here are a few specific application examples:

[0137] 1. Aviation manufacturing industry:

[0138] Turbine blade processing: Turbine blade processing requires high precision and stability. Using PINN to predict tool wear, processing parameters can be adjusted in real time to ensure processing quality and extend tool life.

[0139] Composite material processing: The processing of composite materials has high requirements on tool wear. Through digital twin technology, tool wear can be monitored and predicted in real time, the processing technology can be optimized, and downtime can be reduced.

[0140] 2. Automobile Manufacturing Industry:

[0141] Engine parts processing: For high-precision parts such as crankshafts and camshafts, it is crucial to control tool wear. Through the PINN model, tool wear can be predicted in real time during the production process, and tools can be replaced in time to ensure product quality.

[0142] Car body manufacturing: In the car body manufacturing process, using PINN to predict tool wear can optimize the cutting and stamping processes, improve production efficiency and product consistency.

[0143] 3. Mould manufacturing:

[0144] Injection mold processing: The processing of injection molds has very high requirements for tools. Through PINN technology, the status of the tool can be monitored in real time, and its wear life can be predicted, so as to optimize the processing parameters and extend the service life of the tool.

[0145] Stamping die processing: During the stamping die processing, tool wear directly affects product quality. Using digital twin technology and PINN model, real-time monitoring and prediction of tool status can be achieved to ensure efficient and stable die processing.

[0146] 4. Energy Industry:

[0147] Wind turbine blade processing: The processing of wind turbine blades involves a large number of composite materials, and tool wear is a key issue. Through PINN technology, tool wear can be predicted in real time, the processing process can be optimized, and production efficiency can be improved.

[0148] Oil drilling equipment manufacturing: When manufacturing oil drilling equipment, tool wear directly affects the service life of the equipment. Using digital twins and PINN models, tool wear can be predicted, the manufacturing process can be optimized, and the high quality of the equipment can be ensured.

[0149] 5. Electronics manufacturing industry:

[0150] Semiconductor manufacturing: In the semiconductor manufacturing process, the accuracy and life of the tool are crucial. Through the PINN model, the wear of the tool can be predicted in real time, and maintenance and replacement can be carried out in time to ensure production continuity and product consistency.

[0151] Circuit board processing: The processing of circuit boards requires high-precision tools. Through digital twin technology, the tool status can be monitored in real time and its wear life can be predicted, thereby optimizing the processing technology and improving product quality.

[0152] These examples demonstrate the application potential of the PINN-based data mechanism dual-driven digital twin tool wear life prediction method in various industries. By combining physical information with data-driven models, accurate prediction and optimization of tool status can be achieved, improving production efficiency and product quality.

[0153] Example 3

[0154] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a digital twin tool wear life prediction method based on a physical information neural network described in Example 1 or 2 are implemented.

[0155] Example 4

[0156] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a digital twin tool wear life prediction method based on a physical information neural network as described in Example 1 or 2.

[0157] Example 5

[0158] A digital twin tool wear life prediction system based on physical information neural network, comprising:

[0159] The tool data processing module is configured as follows: tool data is a tool wear data set, including milling force signals of the three axes of X-axis, Y-axis and Z-axis, and the force (F x ,F y ,F z ) in the feed force model and wear model; (F x ,F y ,F z ) refer to the milling force signals of the X-axis, Y-axis and Z-axis respectively;

[0160] A calculation module is configured to: calculate resultant force, cutting speed, sliding distance and load;

[0161] The neural network building module is configured to: build a physical information neural network based on the main cutting force model and the Archard wear model;

[0162] The loss function building module is configured to: calculate the tool wear amount by combining the main cutting force model and the Archard wear model, thereby determining the physical loss function in the physical information neural network;

[0163] The tool wear calculation module is configured to calculate the tool wear through a physical information neural network.

Claims

1. A digital twin tool wear life prediction method based on physical information neural network, characterized in that: include: s1: tool data processing; Tool data refers to tool wear data set, including milling force signals of X-axis, Y-axis and Z-axis. x ,F y ,F z ) in the feed force model and wear model; (F x ,F y ,F z ) refer to the milling force signals of the X-axis, Y-axis and Z-axis respectively; s2: resultant force calculation; s3: cutting speed calculation; s4: sliding distance calculation; s5: load calculation; s6: Neural network construction; build physical information neural network based on main cutting force model and Archard wear model; s7: loss function construction; The main cutting force model and Archard wear model are combined to calculate the tool wear, so as to determine the physical loss function in the physical information neural network. The calculation formula is as follows: Among them, F f is the main cutting force, K f is the main cutting force constant, A c is the cutting area, v is the cutting speed, n e is the velocity index, W is the wear amount, k is the wear coefficient, F is the load, s is the sliding distance, and H is the material hardness; s8: Feedback and iteration, i.e. model evaluation and verification, evaluation and verification of the updated physical information neural network; s9: Calculate tool wear; calculate tool wear through physical information neural network.

2. According to the method for predicting the wear life of a digital twin tool based on a physical information neural network according to claim 1, it is characterized in that: Calculation of resultant forces; including: The force on each axis is generalized to obtain the resultant force F c , the calculation formula is as follows:

3. The method for predicting the wear life of a digital twin tool based on a physical information neural network according to claim 1 is characterized in that: Cutting speed calculation; including: For a rotating workpiece, the cutting speed v is calculated as follows: v = π·D·n (3); Where v is the cutting speed in meters per minute, D is the diameter of the workpiece in meters, and n is the spindle speed in revolutions per minute.

4. The method for predicting the wear life of a digital twin tool based on a physical information neural network according to claim 1 is characterized in that: Slide distance calculation; including: The sliding distance calculation formula is as follows: Where s is the sliding distance in mm, f is the feed amount per tool pass in mm / rev, N is the total number of revolutions, n is the spindle speed, and t is the machining time in minutes.

5. The method for predicting the wear life of a digital twin tool based on a physical information neural network according to claim 2 is characterized in that: Neural network construction: refers to: building a physical information neural network based on the main cutting force model and Archard wear model, integrating physical constraints into the loss function of the physical information neural network, and realizing the prediction of wear amount; including: 1) Data preparation: the data includes sensor data and wear data; Sensor data: including force / vibration signals of the three axes (F x ,F y ,F z ); Wear data: the actual wear value of the tool; 2) Physical constraint definition; The physical constraint definition part is completed by combining the main cutting force model and the Archard wear model; the main cutting force model is formula (2), and the Archard wear model is formula (1); 3) Customized loss function; The custom loss function is the differential equation derived from the Archard wear formula according to time, as shown below: 4) Construction of physical information neural network; Build a physical information neural network based on the main cutting force model and Archard wear model; 5) Training model; The force / vibration signals of the X, Y, and Z axes and their corresponding wear data are input into the physical information neural network for training. The physical information neural network continuously adjusts the parameters by optimizing the custom loss function until the physical information neural network converges; including: 6) Wear prediction process; After the physical information neural network training is completed, it enters the prediction stage; input the force / vibration signals of the three axes of X, Y, and Z: obtain new force or vibration signals from the sensor, and use the trained physical information neural network to predict the wear amount; 7) Compare with actual wear value Plot a graph to compare the wear values ​​predicted by the model with the actual wear values.

6. The method for predicting the wear life of a digital twin tool based on a physical information neural network according to claim 5 is characterized in that: Comparison with actual wear values; including: A. Data acquisition and division: Take a part of the prepared data set as test data or verification data; B. Obtaining model prediction results: The input features of the test data or the verification data are fed into the trained physical information neural network, and the corresponding predicted wear value is obtained through forward propagation; C. Comparison curve chart: With the index or time of the data point as the horizontal axis, the actual wear value and the predicted wear value are plotted as two curves for comparison.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the processor implements the steps of a digital twin tool wear life prediction method based on physical information neural network.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for predicting the wear life of a digital twin tool based on a physical information neural network are implemented.

9. A digital twin tool wear life prediction system based on physical information neural network, characterized in that: include: The tool data processing module is configured as follows: tool data is a tool wear data set, including milling force signals of the three axes of X-axis, Y-axis and Z-axis, and the force (F x ,F y ,F z ) in the feed force model and wear model; (F x ,F y ,F z ) refer to the milling force signals of the X-axis, Y-axis and Z-axis respectively; A calculation module is configured to: calculate resultant force, cutting speed, sliding distance and load; The neural network building module is configured to: build a physical information neural network based on the main cutting force model and the Archard wear model; The loss function building module is configured to: calculate the tool wear amount by combining the main cutting force model and the Archard wear model, thereby determining the physical loss function in the physical information neural network; The tool wear calculation module is configured to calculate the tool wear through a physical information neural network.

Images