Cutter wear diagnosis and service life prediction method based on physical guidance online updating

Abstract

The invention discloses a cutter wear diagnosis and life prediction method based on physical guidance online updating, which comprises the following steps: firstly, constructing a training set through sensor signal noise reduction processing, and constructing a dual-module neural network model for pre-training; meanwhile, a staged wear physical model is established, predicted values of the two models are dynamically fused through a particle filter algorithm, and errors are corrected in real time through a state equation. And then a degradation track is constructed based on the dynamically updated wear value to extrapolate the residual life, and closed-loop control from wear diagnosis and life prediction to tool changing decision is realized in combination with a real-time signal. According to the invention, the prediction credibility is improved through multi-source information fusion; dynamic calibration of prediction is realized through particle filtering; and closed-loop optimization from state monitoring to active maintenance is realized. And through a dynamic correction mechanism and a threshold triggering strategy, the false alarm risk is reduced while the precision is ensured, and finally the industrial application value of prolonging the service life of the cutter and reducing non-planned shutdown is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of tool wear monitoring and life prediction, and in particular to a tool wear diagnosis and life prediction method based on physical guidance online updating. Background Art

[0002] With the rapid development of intelligent manufacturing technologies, cutting tools are becoming increasingly important in high-speed milling operations. However, due to the significant differences in tool wear rates at different stages of wear, existing methods struggle to accurately monitor and predict tool wear. Traditional methods rely primarily on empirical and data-driven models to model and predict tool wear, but these approaches have significant limitations in practical applications.

[0003] On the one hand, methods based on physical models often assume that tool wear characteristics are stable under different working conditions, which makes it impossible to accurately predict tool wear status under complex and variable working conditions; on the other hand, although methods based on data-driven models can adapt to a variety of working conditions, due to the lack of physical constraints, they are prone to overfitting, affecting the prediction accuracy.

[0004] In recent years, hybrid approaches combining physical and data-driven models have gained increasing attention. However, existing research has largely neglected the importance of online wear signature updates and lacks effective methods for real-time adjustments to tool wear status. Therefore, an online update method that integrates physical and data-driven models is urgently needed to improve the real-time performance and accuracy of tool wear diagnosis and life prediction. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the existing technology. To achieve the above purpose, a tool wear diagnosis and life prediction method based on physical guidance online update is adopted to solve the problems raised in the above background technology.

[0006] A tool wear diagnosis and life prediction method based on physics-guided online updating includes the following steps:

[0007] Step S1: collecting tool wear signals in real time through sensors, performing noise reduction processing by combining wavelet denoising and bandpass filtering, and constructing a training data set containing tool wear information;

[0008] Step S2: Based on the preprocessed data, a data-driven model for preliminary diagnosis is established, and a feature extraction module and a head decision module are designed to extract time series features from the preprocessed signal and generate wear prediction values;

[0009] Step S3: Using the training data set to train the data-driven model, iteratively optimizing the network weights through the loss function until convergence, to obtain a pre-trained data-driven model;

[0010] Step S4: Based on the tool wear characteristics, a physical model for tool wear diagnosis is constructed, the wear curve is described in stages, and physical prediction values ​​are output to drive the results with data;

[0011] Step S5: The pre-trained data-driven model and the physical model are integrated through a particle filter algorithm, and the wear prediction value is corrected in real time in combination with the state equation;

[0012] Step S6: constructing a degradation trajectory based on the updated wear prediction value, and extrapolating the remaining life as a quantitative basis for tool change decision;

[0013] Step S7: real-time acquisition and pre-processing of new signals, input into the fused model, and dynamically output the prediction results of the current wear value and the remaining life;

[0014] Step S8: Compare the prediction result with the preset threshold value, trigger the shutdown and tool change or continue monitoring, and complete the closed-loop control from diagnosis to maintenance.

[0015] As a further solution of the present invention: the specific steps in step S1 include:

[0016] Step S11: collecting tool wear status related signals in real time through a sensor system installed on the high-speed milling processing equipment;

[0017] Step S12: pre-processing the original sensor signal, using wavelet denoising to eliminate high-frequency noise, and removing invalid frequency band signals through band-pass filtering;

[0018] Step S13: constructing a model training data set based on the tool wear measured data.

[0019] As a further solution of the present invention: the specific steps in step S2 include:

[0020] Step S21: constructing a data-driven model for tool wear diagnosis, wherein the model includes a feature extraction module and a head decision module;

[0021] Step S22: using the designed feature extraction module to extract the time series features related to wear from the preprocessed sensor signal;

[0022] Step S23: Generate a staged tool wear prediction value based on the time series characteristics through the constructed head decision module.

[0023] As a further solution of the present invention: the specific steps in step S3 include:

[0024] Step S31: input the training data set into the data-driven model for training;

[0025] Step S32: Calculate the network weight gradient using the loss function;

[0026] Step S33: Iterate and optimize until the model converges, and save the pre-trained optimal weights and model.

[0027] As a further solution of the present invention: the specific steps in step S4 include:

[0028] Construct a physical model for tool wear diagnosis. Based on the tool flank wear model with adjustable model coefficients, use logarithmic polynomials to model the early and late wear curves of the tool to describe the characteristics of different wear stages and calculate the physical predicted wear value of stage i. The formula is:

[0029]

[0030] Where A, B, C, and D represent the parameters of the model, In(·) is the logarithmic function, and t i is a moment in stage i.

[0031] As a further solution of the present invention: the specific steps in step S5 include:

[0032] Step S51: construct an online update model for tool wear diagnosis, whose state transition equation and observation equation are:

[0033]

[0034] Where, represents the physical predicted wear value of the i-1th stage, h represents the time step, K1, K2, K3, K4 represent the coefficient values ​​of the fourth-order Runge-Kutta algorithm, u i-1 and v i They represent the state transition noise of the i-1th stage and the observation noise of the i-th stage respectively;

[0035] Step S52: Use particle filtering to update the predicted tool wear state in real time, achieve real-time correction, and quickly obtain the updated tool wear value prediction value W. i .

[0036] As a further solution of the present invention: the derivation steps of the state transfer equation are:

[0037] The formula of tool wear variation over time is converted into the form of ordinary differential equation, which is:

[0038]

[0039] Where, represents the differential of the physical predicted wear value output by the physical module, and F(·) represents the ordinary differential equation;

[0040] The numerical solution of the ordinary differential equation is obtained by the fourth-order Runge-Kutta algorithm, which is expressed as:

[0041]

[0042] Where, It represents the physical predicted wear value of the i+1th stage. The expressions of K1, K2, K3, and K4 are:

[0043]

[0044] As a further solution of the present invention: the specific steps in step S6 include:

[0045] Step S61: Based on the updated tool wear value prediction value W i , get the tool life of the current i-th stage The formula is:

[0046]

[0047] Where, v E and v L represent the average wear growth rates in the early and late stages, respectively;

[0048] Step S62: according to the tool life of the current stage Predict the remaining tool life at stage i The formula is:

[0049]

[0050] Where, T Max Indicates the total tool life.

[0051] Compared with the prior art, the present invention has the following technical effects by adopting the above technical solution:

[0052] Deep fusion of physics-guided and data-driven approaches: Combining the physical characteristics of tool wear with deep learning models, and using physical laws to constrain deep learning networks, the models can not only learn complex nonlinear features but also maintain high prediction accuracy and stability under variable working conditions.

[0053] Real-time online update and high-precision prediction: The particle filter algorithm is used to realize online update of tool wear status. When the wear status changes, it can quickly respond and adjust the prediction model, significantly improving the real-time and dynamic response capabilities of the model and adapting to changes in different processing environments.

[0054] Multi-stage wear modeling and dynamic adjustment: Logarithmic polynomial curves are used to model the characteristics of the tool in the early, stable, and rapid wear periods. A two-layer Bi-LSTM is used to perform time series analysis on the wear characteristics. This allows for rapid adjustment of the prediction model when the wear state changes suddenly, improving adaptability and prediction accuracy under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings:

[0056] Figure 1 This is a schematic diagram of the steps of the prediction method disclosed in the embodiment of this application.

[0057] Figure 2 This is a network diagram of a data-driven model according to an embodiment disclosed in this application.

[0058] Figure 3 This is a schematic diagram of the online update model architecture based on physical guidance according to the embodiment disclosed in this application.

[0059] Figure 4 This is a comparison chart of tool wear status prediction results for the embodiments disclosed in this application.

[0060] Figure 5 This is a comparison chart of the tool remaining life prediction results of the embodiment disclosed in this application.

[0061] Figure 6 This is a schematic diagram of tool wear status monitoring under different parameter working conditions of the embodiment disclosed in this application.

[0062] Figure 7 Schematic diagram of the experimental verification platform for the embodiments disclosed in this application. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] Please refer to Figure 1 In an embodiment of the present invention, a tool wear diagnosis and life prediction method based on physical guidance online update includes the following steps:

[0065] Step S1: Use sensors to collect tool wear signals in real time, combine wavelet denoising and bandpass filtering to perform noise reduction processing, and construct a training data set containing tool wear information. The specific steps include:

[0066] Step S11: collecting tool wear status related signals in real time through a sensor system installed on the high-speed milling processing equipment;

[0067] Step S12: pre-processing the original sensor signal, using wavelet denoising to eliminate high-frequency noise, and removing invalid frequency band signals through band-pass filtering;

[0068] Step S13: constructing a model training data set based on the tool wear measured data.

[0069] In this embodiment, the collected original sensor signals are preprocessed, and wavelet denoising and bandpass filtering methods are used to remove noise and invalid signals to obtain pure sensor signals. The model training data set D is constructed in combination with the tool wear information. Exp ;

[0070] Step S2: Based on the preprocessed data, a data-driven model for preliminary diagnosis is established, and a feature extraction module and a head decision module are designed to extract time series features from the preprocessed signal and generate wear prediction values. The specific steps include:

[0071] like Figure 2 As shown in the figure, it is a schematic diagram of the data-driven model network, which uses a deep neural network (DRSN) and a bidirectional long short-term memory network (Bi-LSTM) for wear feature extraction and regression prediction.

[0072] Step S21: constructing a data-driven model for tool wear diagnosis, wherein the model includes a feature extraction module and a head decision module;

[0073] Step S22: Process the clean sensor signal using the designed feature extraction module, and extract the wear-related timing features from the pre-processed sensor signal;

[0074] Step S23: Generate the wear prediction value of the tool in each stage based on the time series characteristics through the constructed head decision module to obtain the wear prediction value W of the i-th stage data. i Data .

[0075] Step S3: Use the training data set to train the data-driven model, iteratively optimize the network weights through the loss function until convergence, and obtain a pre-trained data-driven model. The specific steps include:

[0076] Step S31: input the training data set into the data-driven model for training;

[0077] Step S32: Calculate the network weight gradient using the loss function;

[0078] Step S33: Iterate and optimize until the model converges, and save the pre-trained optimal weights and model.

[0079] In this embodiment, the training data set D Exp Input the data-driven module and calculate the loss function to pre-train the initial weights of the network until the model converges, thereby obtaining the pre-trained network weights and their corresponding optimal pre-trained data model;

[0080] Step S4: Based on the tool wear characteristics, a physical model for tool wear diagnosis is constructed, the wear curve is described in stages, and physical prediction values ​​are output to drive the results with data. The specific steps include:

[0081] like Figure 3 As shown in the figure, it is a schematic diagram of the architecture of the online update model based on physics guidance, which includes the process of integrating the physical model and the data-driven model.

[0082] Construct a physical model for tool wear diagnosis. Based on the tool flank wear model with adjustable model coefficients, use logarithmic polynomials to model the early and late wear curves of the tool to describe the characteristics of different wear stages and calculate the physical predicted wear value of stage i. The formula is:

[0083]

[0084] Where A, B, C, and D represent the parameters of the model, In(·) is the logarithmic function, and t i is a moment in stage i.

[0085] Step S5: The pre-trained data-driven model and the physical model are integrated through the particle filter algorithm, and the wear prediction value is corrected in real time in combination with the state equation. The specific steps include:

[0086] Step S51: construct an online update model for tool wear diagnosis, whose state transition equation and observation equation are:

[0087]

[0088] Where, represents the physical predicted wear value of the i-1th stage, h represents the time step, K1, K2, K3, K4 represent the coefficient values ​​of the fourth-order Runge-Kutta algorithm, u i-1 and v i They represent the state transition noise of the i-1th stage and the observation noise of the i-th stage respectively;

[0089] Step S52: Use particle filtering to update the predicted tool wear state in real time, achieve real-time correction, and quickly obtain the updated tool wear value prediction value W. i .

[0090] The derivation steps of the state transfer equation are:

[0091] Step a1: Convert the tool wear variation over time formula (1) into the ordinary differential equation form (4), which is:

[0092]

[0093] In formula (4), represents the differential of the physical predicted wear value output by the physical module, and F(·) represents the ordinary differential equation;

[0094] Step a2: Obtain a numerical solution to the ordinary differential equation using the fourth-order Runge-Kutta algorithm. The expression is:

[0095]

[0096] In formula (5), It represents the physical predicted wear value of the i+1th stage. The expressions of K1, K2, K3, and K4 are:

[0097]

[0098] like Figure 4 As shown, the figure is a comparison diagram of the tool wear state prediction results, which compares the prediction results of the model of the present invention, actual measurement values, traditional data-driven models and physical models.

[0099] Step S6: construct a degradation trajectory based on the updated wear prediction value, and extrapolate the remaining life as a quantitative basis for tool change decision.

[0100] In this embodiment, according to the updated tool wear value prediction value W i Further predict the remaining life of the current tool

[0101] Predict the remaining life of the current tool. The specific steps include:

[0102] like Figure 5 As shown, the figure is a comparison diagram of the tool remaining life prediction results, which compares the prediction of the model of the present invention and the actual measurement value.

[0103] Step S61: Based on the updated tool wear value prediction value W i , use formula (3) to obtain the tool life of the current i-th stage The formula is:

[0104]

[0105] Where, v E and v L represent the average wear growth rates in the early and late stages, respectively;

[0106] Step S62: according to the tool life of the current stage Further use formula (7) to predict the remaining tool life of the i-th stage The formula is:

[0107]

[0108] Where, T Max Indicates the total tool life.

[0109] Step S7: real-time acquisition and pre-processing of new signals, input into the fused model, and dynamically output the prediction results of the current wear value and the remaining life;

[0110] In this embodiment, the sensor signals generated during the machining process are collected in real time, and the collected real-time cutting force signals are pre-processed and input into the optimal tool wear diagnosis and life prediction model for processing, and the current tool wear prediction value, tool life and remaining life prediction value are output;

[0111] Step S8: Compare the prediction result with the preset threshold value, trigger the shutdown and tool change or continue monitoring, and complete the closed-loop control from diagnosis to maintenance.

[0112] like Figure 6 As shown, the figure is a schematic diagram of tool wear status monitoring under different parameter working conditions, which reflects the adaptability and real-time advantages of the present invention under variable working conditions.

[0113] In this embodiment, if the current tool remaining life prediction value reaches a predetermined tool remaining life critical threshold, the machine is stopped and the tool is replaced; otherwise, the process returns to step S7 to continue monitoring.

[0114] like Figure 7 As shown in the figure, it is a schematic diagram of the experimental verification platform, showing the layout of the sensor monitoring system and data acquisition equipment.

[0115] Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents, and all should be included within the scope of protection of the present invention.

Claims

1. A tool wear diagnosis and life prediction method based on physical guidance online update, characterized in that: The following steps are involved: Step S1: collecting tool wear signals in real time through sensors, performing noise reduction processing by combining wavelet denoising and bandpass filtering, and constructing a training data set containing tool wear information; Step S2: Based on the preprocessed data, a data-driven model for preliminary diagnosis is established, and a feature extraction module and a head decision module are designed to extract time series features from the preprocessed signal and generate wear prediction values; Step S3: Using the training data set to train the data-driven model, iteratively optimizing the network weights through the loss function until convergence, to obtain a pre-trained data-driven model; Step S4: Based on the tool wear characteristics, a physical model for tool wear diagnosis is constructed, the wear curve is described in stages, and physical prediction values ​​are output to drive the results with data; Step S5: The pre-trained data-driven model and the physical model are integrated through a particle filter algorithm, and the wear prediction value is corrected in real time in combination with the state equation; Step S6: constructing a degradation trajectory based on the updated wear prediction value, and extrapolating the remaining life to serve as a quantitative basis for tool change decision; Step S7: real-time acquisition and pre-processing of new signals, input into the fused model, and dynamically output the prediction results of the current wear value and the remaining life; Step S8: Compare the prediction result with the preset threshold value, trigger the shutdown and tool change or continue monitoring, and complete the closed-loop control from diagnosis to maintenance.

2. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 1, characterized in that: The specific steps in step S1 include: Step S11: collecting tool wear status related signals in real time through a sensor system installed on the high-speed milling processing equipment; Step S12: pre-processing the original sensor signal, using wavelet denoising to eliminate high-frequency noise, and removing invalid frequency band signals through band-pass filtering; Step S13: constructing a model training data set based on the tool wear measured data.

3. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 1, characterized in that: The specific steps in step S2 include: Step S21: constructing a data-driven model for tool wear diagnosis, wherein the model includes a feature extraction module and a head decision module; Step S22: using the designed feature extraction module to extract the time series features related to wear from the preprocessed sensor signal; Step S23: Generate a staged tool wear prediction value based on the time series characteristics through the constructed head decision module.

4. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 1, characterized in that: The specific steps in step S3 include: Step S31: input the training data set into the data-driven model for training; Step S32: Calculate the network weight gradient using the loss function; Step S33: Iterate and optimize until the model converges, and save the pre-trained optimal weights and model.

5. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 1, characterized in that: The specific steps in step S4 include: Construct a physical model for tool wear diagnosis. Based on the tool flank wear model with adjustable model coefficients, use logarithmic polynomials to model the early and late wear curves of the tool to describe the characteristics of different wear stages and calculate the physical predicted wear value of stage i. The formula is: Where A, B, C, and D represent the parameters of the model, In(·) is the logarithmic function, and t i is a moment in stage i.

6. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 1, characterized in that: The specific steps in step S5 include: Step S51: construct an online update model for tool wear diagnosis, whose state transition equation and observation equation are: Where, represents the physical predicted wear value of the i-1th stage, h represents the time step, K1, K2, K3, K4 represent the coefficient values ​​of the fourth-order Runge-Kutta algorithm, u i-1 and v i They represent the state transition noise of the i-1th stage and the observation noise of the i-th stage respectively; Step S52: Use particle filtering to update the predicted tool wear state in real time, achieve real-time correction, and quickly obtain the updated tool wear value prediction value W. i .

7. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 6, characterized in that: The derivation steps of the state transfer equation are: The formula of tool wear variation over time is converted into the form of ordinary differential equation, which is: Where, represents the differential of the physical predicted wear value output by the physical module, and F(·) represents the ordinary differential equation; The numerical solution of the ordinary differential equation is obtained by the fourth-order Runge-Kutta algorithm, which is expressed as: Where, It represents the physical predicted wear value of the i+1th stage. The expressions of K1, K2, K3, and K4 are:

8. The tool wear diagnosis and life prediction method based on physical guidance online update according to claim 1, characterized in that: The specific steps in step S6 include: Step S61: Based on the updated tool wear value prediction value W i , get the tool life of the current i-th stage The formula is: Where, v E and v L represent the average wear growth rates in the early and late stages, respectively; Step S62: according to the tool life of the current stage Predict the remaining tool life at stage i The formula is: Where, T Max Indicates the total tool life.

Images