CN-120170547-B - Variable working condition cutter abrasion prediction method based on compound machine tool

CN120170547BCN 120170547 BCN120170547 BCN 120170547BCN-120170547-B

Abstract

The invention relates to the technical field of compound machine tools and discloses a variable working condition cutter abrasion prediction method based on a compound machine tool, which comprises the following steps of s1, collecting dynamic signals and obtaining technological parameters; S2, constructing a cutter abrasion prediction model, and S3, performing multitasking of a meta-learning framework, namely obtaining updated W y and b y through meta-learning. And deploying the cutter wear prediction model in an actual machining system of the compound machine tool, and rapidly outputting a cutter wear prediction value at the current moment according to the dynamic signals and the technological parameters received by implementation. The method has the advantages of no machine tool shutdown and high cutter abrasion prediction accuracy.

Inventors

WANG WENYU
WANG XIAOYU
SU KAI
KONG YAGUANG
TANG GUOHAI
CHEN HONGHUAN
ZHAO HAIJUN
XU FENG
QIU LIUFENG
Yang Canzhong
HUANG HEBIN

Assignees

杭州大天数控机床有限公司
杭州电子科技大学

Dates

Publication Date: 20260508
Application Date: 20250428

Claims (9)

1. The variable working condition cutter abrasion prediction method based on the compound machine tool is characterized by comprising the following steps of: s1, collecting dynamic signals and obtaining technological parameters: Acquiring dynamic signals, namely cutting force (F c,t ), spindle current (I s,t ) and vibration signal (V t ), and acquiring technological parameters, namely cutting speed V c , feeding speed V f , cutting temperature T c and cutter yield strength sigma y ; s2, constructing a cutter wear prediction model: S2-1, constructing a cutter abrasion prediction model input x i based on a long-short-term memory network model LSTM; s2-2, constructing a cutter wear prediction loss function L enhanced , L enhanced ＝L LSTM +λ·L physics ; In the above formula, L physics is a physical model loss function, lambda is a physical loss weight coefficient for adjusting the weight of physical constraint loss, L LSTM is the supervision loss of a long-term memory network model LSTM; s2-3, constructing a cutter abrasion state vector h t , and obtaining a cutter abrasion predicted value by training a long-term and short-term memory network model LSTM: In the above-mentioned method, the step of, In the training process of the short-term memory network model LSTM, according to a cutter abrasion prediction loss function L enhanced established in s2-2, the training aim is to minimize the cutter abrasion prediction loss function L enhanced , and initial W y and b y are obtained through training; And s3, multitasking of the meta-learning framework, namely obtaining updated W y and b y through meta-learning.
2. The variable working condition tool wear prediction method based on the compound machine tool according to claim 1 is characterized in that in step s1, cutting force (F c,t ), spindle current (I s,t ), vibration signal (V t ) and cutting temperature T c are obtained by collecting dynamic data of the compound machine tool in a machining process state through sensors, cutting speed V c and feeding speed V f are obtained through a communication interface of the compound machine tool, and tool yield strength sigma y is obtained by referring to data.
3. The variable-working-condition tool wear prediction method based on the compound machine tool according to claim 1, wherein in the step s2-1, the tool wear prediction model input X i comprises two parts of input dynamic signals and process parameters: In the above-mentioned method, the step of, Representing dynamic signals including cutting force (F c,t ), spindle current (I s,t ), vibration signal (V t ), cutting temperature T c ; Representing technological parameters including cutting speed v c , feeding speed v f and yield strength sigma y , T representing time step of collecting signal, and T representing total number of times of collecting signal.
4. The variable working condition cutter abrasion prediction method based on the compound machine tool according to claim 3, wherein the construction method of the L physics of the step s2-2 is as follows: Constructing a chip force model, namely F c ＝K c ·a p ·f z , and constructing a temperature rise model: In the above formula, F c is cutting force, K c is a material cutting force coefficient, a p is cutting depth, F z is feeding amount per tooth, T c is rising temperature, v c is cutting speed, eta c is thermal efficiency, A c is cutting contact area, and K is material heat conductivity coefficient; Taking the deviation between the model calculation result and the actual acquisition value as one of the loss functions of the cutter abrasion prediction model, and adding the deviation into an optimization target in a regular term form, so that the LSTM model is learned in a direction conforming to a physical rule; constructing a cutting force model loss function: Constructing a temperature rise constraint loss function: In the above-mentioned method, the step of, In order to actually collect the resulting cutting force, Calculating the resulting cutting force for the model; in order to actually collect the temperature rise obtained, Calculating the obtained temperature rise for the model; The physical model loss function L physics is defined as follows: L physics ＝λ 1 L force +λ 2 L temperature ; in the above equation, λ 1 and λ 2 are weight factors used to balance the importance of different physical constraints.
5. The variable working condition tool wear prediction method based on the compound machine tool according to claim 4, wherein in the step s2-3, the construction method of the tool wear state vector h t is as follows, in an LSTM model, dynamic signals and process parameters are respectively used as independent long-short-term memory network channel extraction time series representations, and the method is specifically defined as follows: In the above-mentioned method, the step of, To extract a time series representation with a dynamic signal as a long and short memory network channel, Extracting a time sequence representation by taking the process parameters as long-term and short-term memory network channels; The output of the two channels is subjected to feature fusion on the hidden layer to form a representation of a cutter abrasion state vector h t :
6. The variable-working-condition tool wear prediction method based on the compound machine tool according to claim 1 or 5, wherein in the step s2-3, the whole training process performs parameter optimization through back propagation and gradient descent algorithm, the tool wear prediction loss function derives against all trainable parameters in the long-short-period memory network model, the gradient of each parameter is calculated, and each parameter is updated along the negative direction of the gradient: In the above formula, η is the learning rate, θ lstm is the set of all the learnable parameters in the LSTM model, including W y and b y .
7. The variable working condition cutter wear prediction method based on the compound machine tool according to claim 1, wherein the multitasking training method of the meta-learning framework in step s3 is as follows: s3-1 dividing the sample into a plurality of tasks T i , within each task, the data is divided into training subsets And verification subset D i val : In the above-mentioned method, the step of, Representing the training set of the ith task in the LTSM model, Representing an ith task training set verification set in the LTSM model; s3-2, randomly obtaining a sharing initialization parameter theta of meta-learning, obtaining a personalized model parameter theta ' i under the task through meta-learning inner circulation updating, and obtaining an updated global sharing parameter theta new through meta-learning outer circulation updating based on the personalized model parameter theta ' i , wherein theta and theta ' i 、θ new are all the sets of all the learnable parameters in the LSTM model and comprise parameters W y and b y .
8. The variable-working-condition tool wear prediction method based on the compound machine tool according to claim 7, wherein in the step s3-2, the internal circulation is updated as follows: In the above formula, alpha is the internal circulation learning rate, which is used for gradient update in the task, Model supervision loss for training data on task T i , The loss of the corresponding physical model on the task T i is represented by lambda, which is a physical loss weight coefficient; The outer loop update based on θ' i is as follows: in the above formula, beta is the outer circulation learning rate and is used for global updating; the verification loss of the task T i is used for guiding the update of the global sharing parameter theta; the generalization capability of theta on the physical rule is ensured for the loss of a physical model; The updated global sharing parameter theta new is obtained through external circulation, the updated W y and b y are obtained, and then the cutter abrasion predicted value after meta-learning is obtained
9. The variable working condition cutter wear prediction method based on the compound machine tool according to claim 1, further comprising the steps of: s4, quick fine adjustment and deployment of new working condition tasks: s4-1, when the model is deployed in an actual application scene and meets a new unseen working condition task, on the premise of not changing the model structure, completing quick fine adjustment through gradient update, wherein the fine adjustment process uses the global sharing parameter theta new learned in the step s3, and performs the following update by combining small sample data of the new working condition task: In the above description, θ' new is the model parameter after the new working condition task is quickly adapted, For the supervision loss of the new task training set, The method is characterized in that the method is used for obtaining a physical model loss item of a new task, and finally obtaining a finely-adjusted model parameter theta' new which is a long-period memory network model parameter theta lstm which is actually deployed for prediction, namely obtaining parameters W y and b y under the new working condition task; s4-2 tool wear prediction model After multi-task element learning and quick fine adjustment of new working condition tasks, the tool wear prediction value is deployed in an actual machining system of a compound machine tool, receives dynamic signals and process parameters in real time, and quickly outputs the tool wear prediction value at the current moment based on updated model parameters theta lstm

Description

Variable working condition cutter abrasion prediction method based on compound machine tool Technical Field The invention relates to the technical field of compound machine tools, in particular to a variable working condition cutter abrasion prediction method based on a compound machine tool. Background The cutter is an indispensable core component in numerical control machining, and the state of the cutter directly influences machining precision, workpiece surface quality and production efficiency. During the machining process, tool wear gradually builds up, and if the wear state cannot be monitored and predicted in time, the workpiece may be out of tolerance, surface defects or tool breakage, thereby causing production interruption and even equipment damage. In particular to a compound machine tool, a plurality of tool magazines are usually arranged on the compound machine tool, a plurality of tools are stored in each tool magazine, a plurality of tools are used in the processing process of the compound machine tool, and the processing quality of a workpiece cannot reach the standard when any tool is worn out beyond a preset value. The development of tool wear detection and prediction research has important practical significance for improving the processing efficiency, ensuring the product quality and reducing the production cost. The existing tool wear detection method is mainly divided into a direct method and an indirect method. The direct method is represented by an image method, and an image of the tool wear surface is obtained by a microscope, an industrial camera, or the like, and the wear area or the wear amount is analyzed. The method can directly observe the abrasion state of the cutter, but needs to stop for image acquisition, cannot realize online detection, has higher cost and lower detection efficiency, and is not suitable for large-scale production and application. The indirect rule is based on machining process parameters such as cutting force, spindle current, vibration signals and the like, and the cutter abrasion state is estimated through data analysis and physical modeling. However, since the tool wear process is essentially a complex process of multi-factor coupling and dynamic nonlinearity, the physical modeling method often has difficulty in obtaining a high-precision prediction result under a variable working condition, and the applicability and the robustness of the model are poor. Disclosure of Invention The invention aims to solve the problems in the prior art and provides a variable working condition cutter abrasion prediction method based on a compound machine tool, which does not need machine tool shutdown, on the basis of combining deep learning with a physical mechanism, a cutter abrasion prediction method of a dynamic working condition is constructed through multi-element data fusion and feature extraction, and the accuracy of prediction is effectively improved through meta learning. In order to achieve the above purpose, the present invention adopts the following technical scheme: a variable working condition cutter abrasion prediction method based on a compound machine tool comprises the following steps: S1, acquiring dynamic signals, namely cutting force (F c,t), spindle current (I s,t) and vibration signals (V t), and acquiring process parameters, namely cutting speed V c, feeding speed V f, cutting temperature T c and cutter yield strength sigma y; S2, constructing a cutter abrasion prediction model, wherein S2-1, constructing cutter abrasion prediction model input X i based on a long-short-term memory network model LSTM, S2-2, constructing a cutter abrasion prediction loss function L enhanced,Lenhanced＝LLSTM+λ·Lphysics, wherein L physics is a physical model loss function, lambda is a physical loss weight coefficient and is used for adjusting the weight of physical constraint loss, L LSTM is the supervision loss of the long-term memory network model LSTM, and S2-3, constructing a cutter abrasion state vector h t, and obtaining a cutter abrasion predicted value by training the long-short-term memory network model LSTM: In the above-mentioned method, the step of, In the training process of the short-term memory network model LSTM, according to a cutter abrasion prediction loss function L enhanced established in s2-2, the training aim is to minimize the cutter abrasion prediction loss function L enhanced, and initial W y and b y are obtained through training; And s3, multitasking of the meta-learning framework, namely obtaining updated W y and b y through meta-learning. Preferably, in step s1, the cutting force (F c,t), the spindle current (I s,t), the vibration signal (V t) and the cutting temperature T c are obtained by collecting dynamic data of the machining process state of the compound machine tool through sensors, the cutting speed V c and the feeding speed V f are obtained through a communication interface of the compound machine tool, and the tool yield stren