A variable working condition tool wear prediction method based on a compound machine tool

Abstract

The application relates to the technical field of compound machine tools and discloses a variable working condition tool wear prediction method based on a compound machine tool, which comprises the following steps: s1: collecting dynamic signals and acquiring process parameters; s2: constructing a tool wear prediction model; s3: multi-task training of a meta-learning framework: through meta-learning, an updated W y And b y The tool wear prediction model is deployed in an actual machining system of the compound machine tool, dynamic signals and process parameters received during implementation are used, and a tool wear prediction value at a current time is quickly output. The application has the beneficial effects that the machine tool does not need to be stopped and the tool wear prediction accuracy is high.

Description

Technical Field

[0001] This invention relates to the field of composite machine tool technology, and in particular to a method for predicting tool wear under varying working conditions based on composite machine tools. Background Technology

[0002] Cutting tools are indispensable core components in CNC machining, and their condition directly affects machining accuracy, workpiece surface quality, and production efficiency. During machining, tool wear gradually accumulates. If the wear condition cannot be monitored and predicted in a timely manner, it may lead to workpiece dimensional deviations, surface defects, or tool breakage, resulting in production interruptions or even equipment damage. This is especially true for composite machine tools, which typically have multiple tool magazines, each storing multiple tools. Multiple tools are used during the machining process, and if the wear of any one of these tools exceeds a preset value, the workpiece machining quality will be substandard. Therefore, conducting research on tool wear detection and prediction is of significant practical importance for improving machining efficiency, ensuring product quality, and reducing production costs.

[0003] Existing tool wear detection methods are mainly divided into two categories: direct methods and indirect methods. Direct methods, represented by image-based methods, acquire images of the tool wear surface using equipment such as microscopes and industrial cameras, and analyze the wear area or amount. This method allows direct observation of the tool wear state, but requires machine downtime for image acquisition, making online detection impossible. Furthermore, it is costly and inefficient, unsuitable for large-scale production applications. Indirect methods, based on machining process parameters such as cutting force, spindle current, and vibration signals, infer the tool wear state through data analysis and physical modeling. However, since the tool wear process is inherently a complex process involving multiple coupled factors and dynamic nonlinearity, physical modeling methods often struggle to achieve high-precision predictions under varying operating conditions, and the models exhibit poor applicability and robustness. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention provides a method for predicting tool wear under varying operating conditions based on composite machine tools. This prediction method does not require machine tool downtime. Based on the combination of deep learning and physical mechanisms, it constructs a tool wear prediction method for dynamic operating conditions through multivariate data fusion and feature extraction, and effectively improves the accuracy of prediction by utilizing meta-learning.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for predicting tool wear under varying working conditions based on composite machine tools includes the following steps:

[0007] s1: Acquiring dynamic signals and process parameters: Acquiring dynamic signals: Cutting force (F) c,t ), spindle current (I) s,t ) and vibration signal (Vt ); Obtain process parameters: cutting speed v c Feed rate: v f Cutting temperature T c Tool yield strength σ y ;

[0008] s2: Constructing a tool wear prediction model: s2-1: Constructing a tool wear prediction model based on a Long Short-Term Memory (LSTM) network model. Input X i ;s2-2: Constructing the tool wear prediction loss function L enhanced L enhanced =L LSTM +λ·L physics In the above formula, L physics L represents the physical model loss function, where λ is the physical loss weight coefficient used to adjust the weights of the physical constraint loss. LSTM The supervision loss for the Long Short-Term Memory (LSTM) network model; s2-3: Constructing the tool wear state vector h t By training a Long Short-Term Memory (LSTM) network model, the predicted tool wear value is obtained: In the above formula, h represents the predicted tool wear value at the current time step t. t W is the tool wear state vector. y Let b be the output layer weight matrix. y The bias vector is used during the training of the LSTM short-term memory network model. This is based on the tool wear prediction loss function L established in s2-2. enhanced The training objective is to minimize the tool wear prediction loss function L. enhanced The initial W is obtained through training. y and b y ;

[0009] s3: Multi-task training of the meta-learning framework: Obtaining the updated W through meta-learning. y and b y .

[0010] Preferably, in step s1, the cutting force (F) c,t ), spindle current (I) s,t ), vibration signal (V) t Cutting temperature T c It is obtained by collecting dynamic data on the machining process of the composite machine tool through sensors; cutting speed v c Feed rate: v f It is obtained through the communication interface built into the composite machine tool; the tool yield strength σ y It was obtained by consulting relevant materials.

[0011] Preferably, in step s2-1, the tool wear prediction model is input X. i It includes two parts: input dynamic signals and process parameters. In the above formula, Represents dynamic signals, including cutting force (F) c,t ), spindle current (I) s,t ), vibration signal (V) t Cutting temperature T c ; This indicates process parameters, including cutting speed v. c Feed rate v f Yield strength σ y t represents the time step of a certain signal acquisition, and T represents the total number of signal acquisitions.

[0012] Preferably, L in step s2-2 physics The construction method is as follows:

[0013] Constructing the chip force model: F c =K c ·a p ·f z Construct a temperature rise model:

[0014] In the above formula, F c It is the cutting force, K c a is the material cutting force coefficient. p f is the depth of cut. z T is the feed per tooth. c It's the temperature rise, v c It is the cutting speed, η c For thermal efficiency, A c Where k is the cutting contact area and k is the thermal conductivity of the material.

[0015] The deviation between the model calculation results and the actual collected values ​​is used as one of the loss functions of the tool wear prediction model and added to the optimization objective in the form of a regularization term, so that the LSTM model learns in a direction that conforms to physical laws.

[0016] Constructing the loss function for the cutting force model:

[0017] Construct a temperature rise-constrained loss function:

[0018] In the above formula, The actual cutting force obtained. The cutting force calculated from the model; This refers to the actual temperature rise measured. The temperature rise calculated by the model;

[0019] Define the physical model loss function L physics as follows:

[0020] L physics =λ1L force +λ2L temperature ;

[0021] In the above formula, λ1 and λ2 are weighting factors used to balance the importance of different physical constraints.

[0022] Preferably, in steps s2-3, the tool wear state vector h t The construction method is as follows: In the LSTM model, dynamic signals and process parameters are used as independent long short-term memory network channels to extract time series representations, specifically defined as follows:

[0023]

[0024] In the above formula, This method uses dynamic signals as channels in a long short-term memory network to extract time-series representations. The process parameters are used as channels in a Long Short-Term Memory network to extract time series representations; the outputs of the two channels are fused in the hidden layer to form the tool wear state vector h. t The representation of:

[0025] Preferably, in steps s2-3, the entire training process optimizes parameters using backpropagation and gradient descent algorithms. The tool wear prediction loss function is differentiated with respect to all trainable parameters in the long short-term memory network model, the gradient of each parameter is calculated, and each parameter is updated along the negative gradient direction.

[0026] In the above formula, η is the learning rate, and θ lstm The set of all learnable parameters in the LSTM model, including W. y and b y .

[0027] As a preferred option, the multi-task training method of the meta-learning framework in step s3 is as follows:

[0028] s3-1: Divide the samples into multiple tasks T i Within each task, the data is divided into training subsets. and verification subset

[0029]

[0030] In the above formula, This indicates that the i-th task is in the training set of the LTSM model. This represents the training and validation sets for the i-th task in the LTSM model;

[0031] s3-2: Randomly obtain the shared initialization parameters θ for meta-learning, and update them through the inner loop of meta-learning to obtain the personalized model parameters θ for this task. i Based on personalized model parameters θ i The updated globally shared parameter θ is obtained through meta-learning outer loop updates. new ; where θ, θ i ′、θ new All are sets of all learnable parameters in the LSTM model, including parameter W. y and b y .

[0032] As a preferred embodiment, in step s3-2, the inner loop is updated as follows:

[0033]

[0034] In the above formula, α is the inner loop learning rate, used for gradient updates within the task. For task T i Supervised loss of the model on the training data, For task T i The corresponding physical model loss is given above, where λ is the physical loss weighting coefficient.

[0035] Based on θ i The outer loop is updated as follows:

[0036]

[0037] In the above formula, β is the outer loop learning rate, used for global updates; For task T i The validation loss is used to guide the update of the globally shared parameter θ; The loss of the physical model is used to ensure the generalization ability of θ to physical laws.

[0038] The updated globally shared parameter θ is obtained through the outer loop. new To obtain the updated W y and b y This allows us to obtain the predicted tool wear value after meta-learning.

[0039] Preferably, the following steps are also included:

[0040] S4: Rapid fine-tuning and deployment of new working conditions and tasks:

[0041] s4-1: When the model encounters new and unseen tasks in real-world applications, it undergoes rapid fine-tuning through gradient updates without altering the model structure. The fine-tuning process utilizes the globally shared parameters θ learned in step s3. new And based on the small sample data of the new working condition task, the following updates are performed:

[0042]

[0043] In the above formula, θ′ new The model parameters are for rapid adaptation to new working conditions and tasks. The supervised loss for the training set of the new task. The physical model loss term for the new task is used to finally obtain the fine-tuned model parameters θ′. new That is, the parameters θ of the Long Short-Term Memory network model actually deployed for prediction. lstm That is, to obtain the parameter W under the new working condition task. y and b y ;

[0044] s4-2: Tool Wear Prediction Model After multi-task element learning and rapid fine-tuning under new working conditions, it is deployed in the actual machining system of the composite machine tool, receiving dynamic signals and process parameters in real time, based on the updated model parameters θ. lstm Quickly output the predicted tool wear value at the current moment.

[0045] Therefore, the present invention has the following beneficial effects: (1) By introducing dynamic signals and process parameters and combining them with an LSTM model, a tool wear prediction loss function is constructed, and the tool wear prediction value is obtained by training the LSTM model. And update parameter W through meta-learning y and b y (2) Multi-task training using a meta-learning framework, with backpropagation and gradient descent algorithms used for parameter optimization during meta-learning, and combined with inner and outer loop updates to continuously optimize model parameters so that they can extract common features between different work conditions and further improve the accuracy of prediction; (3) For new work conditions, without changing the model structure, rapid fine-tuning is achieved through gradient updates to obtain θ′. new This enables rapid adaptation to new working conditions. Attached Figure Description

[0046] Figure 1 This is a block diagram of a long short-term memory network model.

[0047] Figure 2 This is a flowchart of meta-learning and meta-testing.

[0048] Figure 3 This is a comparison chart of the predicted tool wear value and the actual tool wear value. Detailed Implementation

[0049] To make the technical problems to be solved, the technical solutions, and the beneficial technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and several exemplary embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0050] It should be understood that the LSTM model involved in this embodiment is short for Long Short-Term Memory Network Model. The terms involved in the embodiment, Long Short-Term Memory Network Model, Long Short-Term Memory Network Model LSTM, and LSTM Model all refer to the Long Short-Term Memory Network Model.

[0051] Example 1: A method for predicting tool wear under varying working conditions based on a composite machine tool, comprising the following steps:

[0052] s1: Acquiring dynamic signals and process parameters: Acquiring dynamic signals: Cutting force (F) c,t ), spindle current (I) s,t ) and vibration signal (V t ); Obtain process parameters: cutting speed v c Feed rate: v f Cutting temperature T c Tool yield strength σ y ;

[0053] Cutting force (F) c,t ), spindle current (I) s,t ), vibration signal (V) t Cutting temperature T c It is obtained by collecting dynamic data on the machining process of the composite machine tool through sensors; cutting speed v c Feed rate: v f It is obtained through the communication interface built into the composite machine tool; the tool yield strength σ y It was obtained through consulting materials;

[0054] s2: Construct a tool wear prediction model, such as Figure 1 As shown:

[0055] s2-1: Constructing a tool wear prediction model based on the Long Short-Term Memory (LSTM) network model. Input X i Tool wear prediction model input X i It includes two parts: input dynamic signals and process parameters.

[0056]

[0057] In the above formula, Represents dynamic signals, including cutting force (F) c,t ), spindle current (I) s,t ), vibration signal (V) t Cutting temperature T c ; This indicates process parameters, including cutting speed v. c Feed rate v f Yield strength σ y ; t represents the time step of a certain signal acquisition, and T represents the total number of signal acquisitions;

[0058] s2-2: Constructing the tool wear prediction loss function L enhanced L enhanced =L LSTM +λ·L physics In the above formula, L physics L represents the physical model loss function, where λ is the physical loss weight coefficient used to adjust the weights of the physical constraint loss. LSTM The supervision loss for the Long Short-Term Memory (LSTM) network model;

[0059] L in step s2-2 physics The construction method is as follows:

[0060] Constructing the chip force model: F c =K c ·a p ·f z Construct a temperature rise model:

[0061] In the above formula, F c It is the cutting force, K c a is the material cutting force coefficient. p f is the depth of cut. z T is the feed per tooth. c It's the temperature rise, v c It is the cutting speed, η c For thermal efficiency, A c Where k is the cutting contact area and k is the thermal conductivity of the material.

[0062] The deviation between the model calculation results and the actual collected values ​​is used as one of the loss functions of the tool wear prediction model and added to the optimization objective in the form of a regularization term, so that the LSTM model learns in a direction that conforms to physical laws.

[0063] Constructing the loss function for the cutting force model:

[0064] Construct a temperature rise-constrained loss function:

[0065] In the above formula, The actual cutting force obtained. The cutting force calculated from the model; This refers to the actual temperature rise measured. The temperature rise calculated by the model;

[0066] Define the physical model loss function L physics as follows:

[0067] L physics =λ1L force +λ2L temperature ;

[0068] In the above formula, λ1 and λ2 are weighting factors used to balance the importance of different physical constraints;

[0069] s2-3: Construct the tool wear state vector h t By training a Long Short-Term Memory (LSTM) network model, the predicted tool wear value is obtained:

[0070] In the above formula, h represents the predicted tool wear value at the current time step t (i.e., time t). t W is the tool wear state vector. y Let b be the output layer weight matrix. y The bias vector is used during the training of the LSTM short-term memory network model. This is based on the tool wear prediction loss function L established in s2-2. enhanced The training objective is to minimize the tool wear prediction loss function L. enhanced The initial W is obtained through training. y and b y ;

[0071] Tool wear state vector h t The construction method is as follows: In the LSTM model, dynamic signals and process parameters are used as independent long short-term memory network channels to extract time series representations, specifically defined as follows:

[0072]

[0073] In the above formula, This method uses dynamic signals as channels in a long short-term memory network to extract time-series representations. The process parameters are used as channels in a Long Short-Term Memory network to extract time series representations; the outputs of the two channels are fused in the hidden layer to form the tool wear state vector h. t The representation of:

[0074] During the training process, the established tool wear prediction loss function L is used. enhanced The model not only minimizes the predicted tool wear value Supervision loss L between the actual tool wear label y and the tool wear value LSTM It also combines the establishment of a physical model loss function L physics The intermediate features output by the model are physically consistent. The entire training process optimizes parameters using backpropagation and gradient descent algorithms. The training objective is to minimize the tool wear prediction loss function L. enhanced ;

[0075] Specifically, in each forward propagation, the Long Short-Term Memory (LSTM) network models the dynamic signal and process parameters in time series, outputting a fused tool state feature vector, and obtaining the wear prediction value at the current time step t through a fully connected layer. Subsequently, the system calculates the error between the tool wear prediction value and the actual tool wear label, and constructs a physical consistency loss based on the deviation between the physical model prediction value and the measured data. The backpropagation process relies on the chain rule, taking the derivative of the tool wear prediction loss function with respect to all trainable parameters in the LSM network model to calculate the gradient of each parameter. To ensure that the learning direction conforms to the optimization objective, a gradient descent algorithm is introduced, updating each parameter along the negative gradient direction.

[0076]

[0077] In the above formula, η is the learning rate, and θ lstm The set of all learnable parameters in the LSTM model, including W. y and b y ;

[0078] s3: Multi-task training of the meta-learning framework, obtaining the updated W through meta-learning. y and b y ,like Figure 2 As shown:

[0079] s3-1: Divide the samples into multiple tasks T i Within each task, the data is divided into training subsets. and verification subset

[0080]

[0081] In the above formula, This indicates that the i-th task is in the training set of the LTSM model. This represents the training and validation sets for the i-th task in the LTSM model;

[0082] s3-2: Randomly obtain the shared initialization parameters θ for meta-learning, and update them through the inner loop of meta-learning to obtain the personalized model parameters θ for this task. i The inner loop is updated as follows:

[0083]

[0084] In the above formula, α is the inner loop learning rate, used for gradient updates within the task. For task T i Supervised loss of the model on the training data, For task T i The corresponding physical model loss is given above, where λ is the physical loss weighting coefficient.

[0085] Based on personalized model parameters θ′ i The updated globally shared parameter θ is obtained through meta-learning outer loop updates. new Based on θ i The outer loop is updated as follows:

[0086]

[0087] In the above formula, β is the outer loop learning rate, used for global updates; For task T i The validation loss is used to guide the update of the globally shared parameter θ; The loss of the physical model is used to ensure the generalization ability of θ to physical laws.

[0088] Where θ, θ i ′、θ new All are sets of all learnable parameters in the LSTM model, including parameter W. y and b y The updated globally shared parameter θ is obtained through the outer loop. new To obtain the updated W y and b y Thus, the predicted tool wear value is obtained after meta-learning. The tool wear prediction function updated by meta-learning: Deployed in the actual machining system of a composite machine tool, it receives dynamic signals and process parameters in real time, based on the updated model parameters θ. lstm Quickly output the predicted tool wear value at the current moment.

[0089] Example 2: Building upon the meta-learning in Example 1, when the model encounters a new, unseen task in a real-world application scenario, it is updated as follows:

[0090] S4: Rapid fine-tuning and deployment of new working conditions and tasks:

[0091] s4-1: When the model encounters new and unseen tasks in real-world applications, it undergoes rapid fine-tuning through gradient updates without altering the model structure. The fine-tuning process utilizes the globally shared parameters θ learned in step s3. new And based on the small sample data of the new working condition task, the following updates are performed:

[0092]

[0093] In the above formula, θ′ new The model parameters are for rapid adaptation to new working conditions and tasks. The supervised loss for the training set of the new task. The physical model loss for the new task is used to finally obtain the fine-tuned model parameters θ′. new That is, the parameters θ of the Long Short-Term Memory network model actually deployed for prediction. lstm That is, to obtain the parameter W under the new working condition task. y and b y ;

[0094] s4-2: Tool Wear Prediction Model After multi-task element learning and rapid fine-tuning under new working conditions, it is deployed in the actual machining system of the composite machine tool, receiving dynamic signals and process parameters in real time, based on the updated model parameters θ. lstm Quickly output the predicted tool wear value at the current moment. This enables the monitoring and early warning of tool status, and features high precision, strong robustness, and good adaptability to working conditions.

[0095] like Figure 3 The figure shows the predicted tool wear values ​​under four different working conditions. A comparison chart of the physical model prediction (i.e., the physical model prediction value) and the actual tool wear label y (the experimental value of the wear width).

[0096] Although specific embodiments of the invention have been described in detail herein, they are given for illustrative purposes only and should not be construed as limiting the scope of the invention. Various substitutions, alterations, and modifications can be conceived without departing from the spirit and scope of the invention.

Claims

1. A method for predicting tool wear under varying working conditions based on composite machine tools, characterized in that, Includes the following steps: s1: Acquiring dynamic signals and process parameters: Acquire dynamic signals: cutting force (F) c,t ), spindle current (I) s,t ) and vibration signal (V t ); Obtain process parameters: cutting speed v c Feed rate: v f Cutting temperature T c Tool yield strength σ y ; s2: Constructing a tool wear prediction model: s2-1: Constructing a tool wear prediction model based on the Long Short-Term Memory (LSTM) network model. Input x i ; s2-2: Constructing the tool wear prediction loss function L enhanced , L enhanced =L LSTM +λ·L physics ; In the above formula, L physics L represents the physical model loss function, where λ is the physical loss weight coefficient used to adjust the weights of the physical constraint loss. LSTM The supervision loss for the Long Short-Term Memory (LSTM) network model; s2-3: Construct the tool wear state vector h t By training a Long Short-Term Memory (LSTM) network model, the predicted tool wear value is obtained: In the above formula, h represents the predicted tool wear value at the current time step t. t W is the tool wear state vector. y Let b be the output layer weight matrix. y The bias vector is used during the training of the LSTM short-term memory network model. This is based on the tool wear prediction loss function L established in s2-2. enhanced The training objective is to minimize the tool wear prediction loss function L. enhanced The initial W is obtained through training. y and b y ; s3: Multi-task training of the meta-learning framework: Obtaining the updated W through meta-learning. y and b y .

2. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 1, characterized in that the steps are as follows: In s1, the cutting force (F) c,t ), spindle current (I) s,t ), vibration signal (V) t Cutting temperature T c It is obtained by collecting dynamic data on the machining process of the composite machine tool through sensors; cutting speed v c Feed rate: v f It is obtained through the communication interface built into the composite machine tool; the tool yield strength σ y It was obtained by consulting relevant materials.

3. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 1, characterized in that, Step s2-1, Input X into the tool wear prediction model i It includes two parts: input dynamic signals and process parameters. In the above formula, Represents dynamic signals, including cutting force (F) c,t ), spindle current (I) s,t ), vibration signal (V) t Cutting temperature T c ; This indicates process parameters, including cutting speed v. c Feed rate v f Yield strength σ y t represents the time step of a certain signal acquisition, and T represents the total number of signal acquisitions.

4. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 3, characterized in that, L in step s2-2 physics The construction method is as follows: Constructing the chip force model: F c =K c ·a p ·f z Construct a temperature rise model: In the above formula, F c It is the cutting force, K c a is the material cutting force coefficient. p f is the depth of cut. z T is the feed per tooth. c It's the temperature rise, v c It is the cutting speed, η c For thermal efficiency, A c Where k is the cutting contact area and k is the thermal conductivity of the material. The deviation between the model calculation results and the actual collected values ​​is used as one of the loss functions of the tool wear prediction model and added to the optimization objective in the form of a regularization term, so that the LSTM model learns in a direction that conforms to physical laws. Constructing the loss function for the cutting force model: Construct a temperature rise-constrained loss function: In the above formula, The actual cutting force obtained. The cutting force calculated from the model; This refers to the actual temperature rise measured. The temperature rise calculated by the model; Define the physical model loss function L physics as follows: L physics =λ1L force +λ2L temperature ; In the above formula, λ1 and λ2 are weighting factors used to balance the importance of different physical constraints.

5. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 4, characterized in that, In steps s2-3, the tool wear state vector h t The construction method is as follows: In the LSTM model, dynamic signals and process parameters are used as independent long short-term memory network channels to extract time series representations, specifically defined as follows: In the above formula, This method uses dynamic signals as channels in a long short-term memory network to extract time-series representations. This involves using process parameters as channels in a long short-term memory network to extract time series representations. The outputs of the two channels are fused in the hidden layer to form the tool wear state vector h. t The representation of:

6. A method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 1 or 5, characterized in that, In steps s2-3, the entire training process optimizes parameters using backpropagation and gradient descent algorithms. The tool wear prediction loss function is differentiated with respect to all trainable parameters in the Long Short-Term Memory network model, and the gradient of each parameter is calculated. Each parameter is then updated along the negative gradient direction. In the above formula, η is the learning rate, and θ lstm The set of all learnable parameters in the LSTM model, including W. y and b y .

7. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 1, characterized in that, The multi-task training method of the meta-learning framework in step s3 is as follows: s3-1: Divide the samples into multiple tasks T i Within each task, the data is divided into training subsets. and verification subset D i val : In the above formula, This indicates that the i-th task is in the training set of the LTSM model. This represents the training and validation sets for the i-th task in the LTSM model; s3-2: Randomly obtain the shared initialization parameters θ for meta-learning, and update them through the inner loop of meta-learning to obtain the personalized model parameters θ′ for this task. i Based on personalized model parameters θ′ i The updated globally shared parameter θ is obtained through meta-learning outer loop updates. new ; where θ and θ′ i θ new All are sets of all learnable parameters in the LSTM model, including parameter W. y and b y .

8. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 7, characterized in that, In step s3-2, the inner loop is updated as follows: In the above formula, α is the inner loop learning rate, used for gradient updates within the task. For task T i Supervised loss of the model on the training data, For task T i The corresponding physical model loss is given above, where λ is the physical loss weighting coefficient. Based on θ′ i The outer loop is updated as follows: In the above formula, β is the outer loop learning rate, used for global updates; For task T i The validation loss is used to guide the update of the globally shared parameter θ; The loss of the physical model is used to ensure the generalization ability of θ to physical laws. The updated globally shared parameter θ is obtained through the outer loop. new To obtain the updated W y and b y This allows us to obtain the predicted tool wear value after meta-learning.

9. The method for predicting tool wear under varying working conditions based on a composite machine tool according to claim 1, characterized in that, It also includes the following steps: S4: Rapid fine-tuning and deployment of new working conditions and tasks: s4-1: When the model encounters new and unseen tasks in real-world applications, it undergoes rapid fine-tuning through gradient updates without altering the model structure. The fine-tuning process utilizes the globally shared parameters θ learned in step s3. new And based on the small sample data of the new working condition task, the following updates are performed: In the above formula, θ′ new The model parameters are for rapid adaptation to new working conditions and tasks. The supervised loss for the training set of the new task. The physical model loss term for the new task is used to finally obtain the fine-tuned model parameters θ′. new That is, the parameters θ of the Long Short-Term Memory network model actually deployed for prediction. lstm That is, to obtain the parameter W under the new working condition task. y and b y ; s4-2: Tool Wear Prediction Model After multi-task element learning and rapid fine-tuning under new working conditions, it is deployed in the actual machining system of the composite machine tool, receiving dynamic signals and process parameters in real time, based on the updated model parameters θ. lstm Quickly output the predicted tool wear value at the current moment.

Images