Flutter-considered high-speed milling surface roughness prediction method and system

Abstract

The invention provides a flutter-considered high-speed milling surface roughness prediction method and system, and relates to the technical field of milling machining.The method comprises the steps that force signals are continuously collected through a force sensor arranged on a milling workbench, vibration signals are continuously collected through an acceleration sensor arranged on the milling workbench, and the surface roughness of the high-speed milling surface is predicted; extracting the multi-dimensional fine-grained features of the force signal and the vibration signal so as to obtain detailed information representing the surface roughness caused by the inherent process and the flutter effect in the milling process, and effectively fusing the multi-dimensional fine-grained features of the periodic signal and the flutter signal through a fine-grained feature fusion network; according to the method, the vibration signals are collected, the surface roughness is predicted by combining with the vibration signals, and the situation that the prediction of the surface roughness is inaccurate due to the chatter phenomenon in high-speed milling machining is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of milling machining, and specifically to a method and system for predicting the surface roughness of high-speed milling considering chatter. Background Art

[0002] High-speed milling is a key material removal technology in modern manufacturing, and is widely used in fields such as automobiles, aviation, and aerospace. With the continuous improvement of the requirements for high quality and low cost of mechanical products, it has become particularly important to control the machining quality. In actual production, the surface roughness is often used as an evaluation index for measuring the machining quality, which not only affects the operating performance of the product, but also plays an important role in the overall service life of the product. At present, the surface roughness measurement method is usually carried out offline after machining, which brings some limitations, such as low efficiency, high cost, and scratching the surface of the workpiece. Therefore, it is necessary to develop an online surface roughness prediction method, which is of great significance for actual production.

[0003] In recent years, many scholars have begun to study data-driven methods, that is, combining sensing technology and artificial intelligence technology for online surface roughness prediction. For surface roughness prediction, the vibration signal is currently monitored online in milling machining, and the online surface roughness prediction is realized by extracting the features sensitive to the surface roughness and training an artificial neural network. At present, there are also preprocessing the monitored noise, vibration, and texture images based on principal component analysis and grey correlation theory, and using particle swarm optimization and least squares support vector machine to construct a multi-dimensional feature fusion model for surface roughness prediction. At present, there are also extracting the time-frequency features of the force signal by using continuous wavelet transform, and then applying convolutional neural network and transfer learning to complete the surface roughness prediction task in the milling process of γ-TiAl alloy.

[0004] The above research has conducted extensive research on data-driven methods, laying an important foundation for the online surface roughness prediction in the milling machining process, and its basic principle has been quite mature. However, in these studies, there are few reports on the online prediction of surface roughness considering the influence of chatter. In high-speed milling machining, due to the attenuation of the damping effect, the chatter phenomenon often occurs, resulting in chatter marks on the machined surface, which significantly affects the surface roughness and leads to inaccurate surface roughness prediction. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the purpose of the present invention is to propose a method for predicting the surface roughness of high-speed milling considering chatter, including: Step 1: On the milling workbench, perform a plane high-speed milling experiment on the workpiece, continuously collect the force signal through the force sensor configured on the milling workbench, and at the same time, continuously collect the vibration signal through the acceleration sensor configured on the milling workbench; Step 2: Take the continuously collected force signals and continuously collected vibration signals as input samples, and take the corresponding surface roughness of the input samples as output samples, and construct a training set based on the input samples and output samples; Step 3: For the input samples in the training set, perform multi-dimensional fine-grained feature extraction and splicing on the force signals and vibration signals in the input samples to obtain multi-dimensional fine-grained feature components of periodic signals and multi-dimensional fine-grained feature components of chatter signals; Step 4: Input the multi-dimensional fine-grained feature components of periodic signals and multi-dimensional fine-grained feature components of chatter signals into the fine-grained feature fusion network to obtain the predicted value of surface roughness. The fine-grained feature fusion network includes a DREF module, a FASM module, and a Bayesian prediction module; Step 5: Based on the predicted values of multiple surface roughnesses and their corresponding output samples, perform multiple iterative updates on the parameters in the DREF module, FASM module, and Bayesian prediction module in the fine-grained feature fusion network to obtain a trained fine-grained feature fusion network; Step 6: Continuously collect the force signals to be predicted and the vibration signals to be predicted, and then perform Steps 3-4 based on the trained fine-grained feature fusion network to obtain the surface roughness.

[0006] Optionally, Step 3 specifically includes: Step 3.1: Use the wavelet packet decomposition method WPD to denoise the vibration signal to obtain the denoised vibration signal, and then extract from the denoised vibration signal through successive variational mode decomposition SVMD to obtain the first periodic signal and the first chatter signal; Step 3.2: Extract from the force signal through successive variational mode decomposition SVMD to obtain the second periodic signal and the second chatter signal; Step 3.3: Perform feature extraction on the first periodic signal, the first chatter signal, the second periodic signal, and the second chatter signal respectively to obtain multi-dimensional fine-grained features of the first periodic signal, multi-dimensional fine-grained features of the first chatter signal, multi-dimensional fine-grained features of the second periodic signal, and multi-dimensional fine-grained features of the second chatter signal. The multi-dimensional fine-grained features include time-domain features and frequency-domain features; Step 3.4: Perform vector splicing on the multi-dimensional fine-grained features of the first periodic signal and the multi-dimensional fine-grained features of the second periodic signal to obtain multi-dimensional fine-grained feature components of periodic signals, and perform vector splicing on the multi-dimensional fine-grained features of the first chatter signal and the multi-dimensional fine-grained features of the second chatter signal to obtain multi-dimensional fine-grained feature components of chatter signals.

[0007] Optionally, Step 4 specifically includes: Step 4.1: According to the DREF module in the fine-grained feature fusion network, extract and fuse the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the flutter signal to obtain the fused features; Step 4.2: Process the fused features through SENet and BiLSTM in the FASM module to obtain the high-level features; Step 4.3: Input the high-level features into the Bayesian prediction module to obtain the predicted value of the surface roughness.

[0008] Optionally, Step 4.1 specifically includes: Step 4.1.1: Perform deep feature extraction on the multi-dimensional fine-grained feature components of the periodic signal to obtain the deep features of the periodic signal. Specifically, perform convolution operation, batch normalization BN, ReLU, max pooling, convolution operation, batch normalization BN, ReLU, and max pooling on the multi-dimensional fine-grained feature components of the periodic signal to obtain the first deep feature. Pass the multi-dimensional fine-grained feature components of the periodic signal through the fully connected layer to obtain the second deep feature. Connect the first deep feature and the second deep feature by addition for residual connection to obtain the deep features of the periodic signal; Step 4.1.2: Process the deep features of the periodic signal through the gating mechanism composed of the sigmoid layer and the tanh layer to obtain the first gating weight matrix , which is specifically implemented through the following formula: ; where μ and ν are learning parameters, is the deep feature of the periodic signal, represents element-wise multiplication, represents the sigmoid layer; Perform convolution on the first gating weight matrix through one-dimensional convolutions with three different receptive fields to obtain the first-scale feature, the second-scale feature, and the third-scale feature. Then, splice the first-scale feature, the second-scale feature, and the third-scale feature to obtain the first multi-scale feature. Input the first multi-scale feature into the FC layer. In the FC layer, the first multi-scale feature passes through the ReLU activation function and the sigmoid activation function to obtain the first adaptive weight; Step 4.1.3: Perform deep feature extraction on the multi-dimensional fine-grained feature components of the flutter signal to obtain the deep features of the flutter signal. Specifically, perform convolution operations, batch normalization BN, ReLU, max pooling, convolution operations, batch normalization BN, ReLU, and max pooling on the multi-dimensional fine-grained feature components of the flutter signal to obtain the third deep feature. Pass the multi-dimensional fine-grained feature components of the flutter signal through a fully connected layer to obtain the fourth deep feature. Residually connect the third deep feature and the fourth deep feature by adding them together to obtain the deep features of the flutter signal; Step 4.1.4: Process the deep features of the flutter signal through a gating mechanism composed of a sigmoid layer and a tanh layer to obtain the second gating weight matrix , which is specifically implemented through the following formula: ; where, is the deep feature of the flutter signal; Perform convolution on the second gating weight matrix through one-dimensional convolutions with three different receptive fields to obtain the fourth-scale feature, the fifth-scale feature, and the sixth-scale feature. Then, concatenate the fourth-scale feature, the fifth-scale feature, and the sixth-scale feature to obtain the second multi-scale feature. Input the second multi-scale feature into the FC layer. In the FC layer, the second multi-scale feature passes through the ReLU activation function and the sigmoid activation function to obtain the second adaptive weight; Step 4.1.5: Perform weighted summation on the deep features of the periodic signal and the deep features of the flutter signal through the first adaptive weight and the second adaptive weight to obtain the fused feature, which is specifically implemented through the following formula: ; where, F represents the fused feature, represents the first adaptive weight, represents the deep feature of the periodic signal, represents the second adaptive weight, represents the deep feature of the flutter signal, represents element-wise addition.

[0009] Optionally, step 4.2 specifically includes: Step 4.2.1: In SENet, pass the fused feature through global pooling, the ReLU activation function, and the sigmoid activation function to obtain the channel attention weight. Perform a dot product operation on the channel attention weight and the fused feature through the Scale operation to obtain the recalibrated fused feature; Step 4.2.2: Process the recalibrated fused feature through BiLSTM to obtain the high-level feature.

[0010] Optionally, step 4.3 specifically includes: In the Bayesian prediction module, the high-level features are converted into one-dimensional vectors through a compression layer. The one-dimensional vectors pass through the ReLU activation function in the fully connected layer to obtain vectors with enhanced non-linearity. The vectors with enhanced non-linearity are processed through approximate Bayesian dropout to obtain the initial surface roughness. Through Monte Carlo (MC) random sampling, the same input sample is repeatedly input into the fine-grained feature fusion network to obtain multiple initial surface roughness values, and the average value of all the initial surface roughness values is calculated to obtain the predicted value of the surface roughness.

[0011] A high-speed milling surface roughness prediction system considering chatter, which is used to implement the high-speed milling surface roughness prediction method considering chatter, includes a data acquisition module, a model training module, a data processing module, and a surface roughness prediction module; The data acquisition module is used to continuously acquire force signals and vibration signals and transmit the force signals and vibration signals to the model training module; The model training module is used to use the continuously acquired force signals and continuously acquired vibration signals as input samples, and the corresponding surface roughness of the input samples as output samples. A training set is constructed based on the input samples and output samples, and the input samples of the training set are transmitted to the data processing module; the model training module is also used to receive the predicted value of the surface roughness, and based on the predicted values of multiple surface roughnesses and their corresponding output samples, the parameters in the fine-grained feature fusion network are iteratively updated multiple times; The data processing module is used to receive the input samples, perform multi-dimensional fine-grained feature extraction and splicing on the force signals and vibration signals in the input samples, and transmit the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the chatter signals to the surface roughness prediction module; The surface roughness prediction module is used to receive the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the chatter signals, input the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the chatter signals into the fine-grained feature fusion network to obtain the predicted value of the surface roughness, and transmit the predicted value of the surface roughness to the model training module.

[0012] The beneficial effects produced by adopting the above technical solutions are as follows: In the present invention, force signals are continuously collected through a force sensor configured on a milling table, and vibration signals are continuously collected through an acceleration sensor configured on the milling table. Furthermore, multi-dimensional fine-grained features of the force signals and vibration signals are extracted, thereby obtaining detailed information characterizing the surface roughness caused by inherent processes and chatter during milling. Then, a fine-grained feature fusion network is used to effectively fuse the multi-dimensional fine-grained features of periodic signals and chatter signals to achieve the prediction of surface roughness. Thus, the present invention collects vibration signals and combines force signals to predict surface roughness, reducing the inaccurate prediction of surface roughness caused by chatter phenomena during high-speed milling. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a schematic flow chart of a method for predicting the surface roughness of high-speed milling considering chatter according to an embodiment of the present invention; Figure 2 is a workpiece machining experimental diagram according to an embodiment of the present invention; Figure 3 is a schematic structural diagram of a fine-grained feature fusion network according to an embodiment of the present invention; Figure 4 is a prediction result diagram of a method for predicting the surface roughness of high-speed milling considering chatter according to an embodiment of the present invention; Figure 5 is a schematic structural diagram of a system for predicting the surface roughness of high-speed milling considering chatter according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The following further describes in detail the specific embodiments of the present invention in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0015] Aiming at the problems existing in the prior art, the present invention provides a method for predicting the surface roughness of high-speed milling considering chatter to improve the accuracy of surface roughness prediction, thereby bringing significant economic benefits to the manufacturing industry by reducing the rejection rate and the number of downtime. Combining Figure 1 , it may include the following steps: Step 1: On the milling table, a plane high-speed milling experiment is carried out on the workpiece. Force signals are continuously collected through a force sensor configured on the milling table, and at the same time, vibration signals are continuously collected through an acceleration sensor configured on the milling table; Combining Figure 2 , in the specific implementation process of the present invention, in a machining center, a plane high-speed milling experiment is carried out through a computer programming control system and a control handle. Data is collected through a force sensor and an acceleration sensor and output to a data acquisition system, and the data acquisition system outputs multi-source signal data, namely force signals and vibration signals.

[0016] Among them, in the specific implementation process of the present invention, a series of high-speed face milling experiments of Al6061 under different machining parameters were carried out on a three-axis vertical machining center CNC-400DBS-S3. A force sensor was installed on the workbench and connected to a charge amplifier to collect x the force signals in the y direction and

[0017] direction. An acceleration sensor was fixed on the surface of the workpiece to collect vibration signals. All the collected signals were sampled by a data acquisition system DH5956 at a frequency of 20 kHz and transmitted to a computer for storage and processing. x The workpiece to be machined was an Al6061 block of 140×140×15 mm, and the cutting tool was an uncoated three-edge flat-bottom milling cutter with a diameter of 4 mm. High-speed milling was carried out along the

[0018] Table 1 Machining Parameters Serial number Spindle speed Cutting depth Milling state 1 15000 0.1 Stable 2 15000 0.2 Stable 3 15000 0.5 Transition 4 15000 0.6 Transition 5 15000 0.9 Severe chatter 6 15600 0.1 Stable 7 15600 0.2 Stable 8 15600 0.5 Transition 9 15600 0.7 Severe chatter 10 15600 1.1 Severe chatter 11 16200 0.2 Stable 12 16200 0.4 Transition 13 16200 0.5 Transition 14 16200 0.7 Severe chatter 15 16200 1.2 Severe chatter Step 2: Take the continuously collected force signals and continuously collected vibration signals as input samples, and take the surface roughness corresponding to the input samples as output samples, and construct a training set based on the input samples and output samples; Step 2.1: Slide on the continuously collected force signals and continuously collected vibration signals through a sliding window to obtain a plurality of force signals and the vibration signals corresponding to the force signals. The force signals and the vibration signals corresponding to the force signals form input samples; In the specific implementation process, the size and step length of the sliding window are both set to 1 s.

[0019] Step 2.2: Continuously measure the surface roughness Ra of the workpiece by a roughness meter, determine the surface roughness corresponding to the input samples, and take the surface roughness corresponding to the input samples as output samples; In the specific implementation process, the surface roughness of the workpiece is measured three times, and the average value of the three measurements is calculated as the final surface roughness, and then the surface roughness corresponding to the input samples is determined.

[0020] Step 2.3: The input samples and the output samples form training samples, and a plurality of training samples form a sample set. The sample set is divided into a training set and a test set according to a preset ratio. Specifically, 70% of the sample set is used as the training set, and 30% is used as the test set. The training set is used for model training, and the test set is used for model testing.

[0021] Step 3: For the input samples in the training set, perform multi-dimensional fine-grained feature extraction and splicing on the force signals and vibration signals in the input samples to obtain the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the chatter signal; Step 3.1: Use the wavelet packet decomposition method WPD to denoise the vibration signal to obtain the denoised vibration signal, and then extract from the denoised vibration signal through successive variational mode decomposition SVMD to obtain the first periodic signal and the first chatter signal; Since the force signal has strong stability, while the vibration signal is easily interfered by environmental noise. Therefore, first use the wavelet packet decomposition method (WPD) to denoise the vibration signal. WPD is an improvement of WT and can decompose the input signal into multiple frequency bands, thus effectively filtering out most of the noise.

[0022] Step 3.2: Extract from the force signal through successive variational mode decomposition SVMD to obtain the second periodic signal and the second chatter signal; Step 3.3: Perform feature extraction on the first periodic signal, the first chatter signal, the second periodic signal, and the second chatter signal respectively to obtain the multi-dimensional fine-grained features of the first periodic signal, the multi-dimensional fine-grained features of the first chatter signal, the multi-dimensional fine-grained features of the second periodic signal, and the multi-dimensional fine-grained features of the second chatter signal. The multi-dimensional fine-grained features include time-domain features and frequency-domain features; Among them, the time-domain features include standard deviation, peak value, root mean square amplitude, absolute average amplitude, root mean square, peak-to-peak value, kurtosis, and kurtosis factor, and the frequency-domain feature is the average frequency.

[0023] Among them, the multi-dimensional fine-grained features of the periodic signal mainly reflect the periodic fluctuations of the surface quality caused by the inherent milling process, while the multi-dimensional fine-grained features of the chatter signal reveal the non-periodic changes of the surface quality caused by the chatter effect.

[0024] Step 3.4: Normalize and vector splice the multi-dimensional fine-grained features of the first periodic signal and the multi-dimensional fine-grained features of the second periodic signal to obtain the multi-dimensional fine-grained feature components of the periodic signal, and normalize and vector splice the multi-dimensional fine-grained features of the first chatter signal and the multi-dimensional fine-grained features of the second chatter signal to obtain the multi-dimensional fine-grained feature components of the chatter signal.

[0025] Among them, in order to balance the influence of each feature on the model, normalization is performed to eliminate the dimensional differences between them.

[0026] Step 4: Combine Figure 3, input the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the flutter signal into the fine-grained feature fusion network to obtain the predicted value of the surface roughness. The fine-grained feature fusion network includes a DREF module, a FASM module, and a Bayesian prediction module; Step 4.1: According to the DREF module in the fine-grained feature fusion network, extract and fuse the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the flutter signal to obtain the fused feature; Among them, the DREF module is designed to deeply explore the coupling relationship of the multi-dimensional fine-grained features of the periodic signal component and the flutter signal component in the surface roughness prediction, and is composed of two parts: deep feature extraction and feature fusion.

[0027] Step 4.1.1: Perform deep feature extraction on the multi-dimensional fine-grained feature components of the periodic signal to obtain the deep features of the periodic signal. Specifically, perform convolution operation, batch normalization BN, ReLU, max pooling, convolution operation, batch normalization BN, ReLU, and max pooling on the multi-dimensional fine-grained feature components of the periodic signal to ensure the stability and efficiency of feature extraction, and obtain the first deep feature. Pass the multi-dimensional fine-grained feature components of the periodic signal through the fully connected layer to obtain the second deep feature. Connect the first deep feature and the second deep feature by addition to perform residual connection to obtain the deep features of the periodic signal, thereby preventing gradient disappearance.

[0028] Step 4.1.2: Process the deep features of the periodic signal through the gating mechanism composed of the sigmoid layer and the tanh layer to obtain the first gating weight matrix , which is specifically implemented through the following formula: ; Among them, μ and ν are learning parameters, is the deep feature of the periodic signal, represents element-wise multiplication, represents the sigmoid layer; Perform convolution on the first gating weight matrix through one-dimensional convolution with three different receptive fields to obtain the first scale feature, the second scale feature, and the third scale feature. Then, splice the first scale feature, the second scale feature, and the third scale feature to obtain the first multi-scale feature to enhance the non-linear expression ability. Input the first multi-scale feature into the FC layer. In the FC layer, the first multi-scale feature passes through the ReLU activation function and the sigmoid activation function to obtain the first adaptive weight; Step 4.1.3: Perform deep feature extraction on the multi-dimensional fine-grained feature components of the flutter signal to obtain the deep features of the flutter signal. Specifically, perform convolution operations, batch normalization (BN), ReLU, max pooling, convolution operations, batch normalization (BN), ReLU, and max pooling on the multi-dimensional fine-grained feature components of the flutter signal to obtain the third deep feature. Pass the multi-dimensional fine-grained feature components of the flutter signal through a fully connected layer to obtain the fourth deep feature. Connect the third deep feature and the fourth deep feature through residual connection by addition to obtain the deep features of the flutter signal; Step 4.1.4: Process the deep features of the flutter signal through a gating mechanism composed of a sigmoid layer and a tanh layer to obtain the second gating weight matrix , which is specifically implemented through the following formula: ; where are the deep features of the flutter signal; Perform convolution on the second gating weight matrix through one-dimensional convolutions with three different receptive fields to obtain the fourth-scale feature, the fifth-scale feature, and the sixth-scale feature. Then, splice the fourth-scale feature, the fifth-scale feature, and the sixth-scale feature to obtain the second multi-scale feature. Input the second multi-scale feature into the FC layer. In the FC layer, the second multi-scale feature passes through the ReLU activation function and the sigmoid activation function to obtain the second adaptive weight; Step 4.1.5: Perform weighted summation on the deep features of the periodic signal and the deep features of the flutter signal through the first adaptive weight and the second adaptive weight to obtain the fusion feature, which is specifically implemented through the following formula: ; where F represents the fusion feature, represents the first adaptive weight, represents the deep features of the periodic signal, represents the second adaptive weight, represents the deep features of the flutter signal, represents addition.

[0029] Step 4.2: Process the fusion feature through the SENet and BiLSTM in the FASM module to obtain high-level features; Step 4.2.1: In the SENet, pass the fusion feature through global pooling, the ReLU activation function, and the sigmoid activation function to obtain the channel attention weight. Perform a dot product operation on the channel attention weight and the fusion feature through the Scale operation to obtain the recalibrated fusion feature; Among them, SENet can also be understood as consisting of Squeeze, Excitation, and Scale operations, which suppress the influence of irrelevant or redundant information by focusing on different parts of the fused features. In the Squeeze operation, the global information of the input features is embedded into the network through global pooling. Next, in the Excitation operation, two FC layers are established, and the two FC layers respectively include the ReLU activation function and the sigmoid activation function. By establishing the two FC layers, the dependence relationship between channels is learned and channel attention weights are generated. The Scale operation is to perform a dot product operation on the obtained channel attention weights and the fused features, and output the recalibrated fused features.

[0030] Step 4.2.2: Process the recalibrated fused features through BiLSTM to obtain high-level features.

[0031] Among them, BiLSTM is the final component of the FASM module and consists of two layers containing 50 neurons each. Each BiLSTM layer contains a forward LSTM and a backward LSTM, which store the long-term temporal information of the fused features. LSTM is a special recurrent neural network that solves the problems of gradient vanishing and explosion faced by traditional RNNs when dealing with long sequences by introducing gating units. The gating units consist of a forget gate, an input gate, an output gate, and a memory unit. The forget gate is used to clear the information that is no longer useful in the historical state, the input gate is used to determine the new information that can be added to the memory unit, and the output gate is used to control the output of the information in the memory unit at the current time step.

[0032] Step 4.3: Input the high-level features into the Bayesian prediction module to obtain the predicted value of the surface roughness.

[0033] In the Bayesian prediction module, the high-level features are converted into a one-dimensional vector through a compression layer, and the one-dimensional vector passes through the ReLU activation function in the fully connected layer to enhance the nonlinear expression ability of the model. The number of neurons is set to 30 to obtain the vector with enhanced nonlinearity; the vector with enhanced nonlinearity is processed through approximate Bayesian dropout to obtain the initial surface roughness. Through Monte Carlo (MC) random sampling, the same input sample is repeatedly input into the fine-grained feature fusion network to obtain multiple initial surface roughness values, and the average value of all the initial surface roughness values is calculated to obtain the predicted value of the surface roughness.

[0034] Step 5: Based on the predicted values of multiple surface roughnesses and their corresponding output samples, iteratively update the parameters in the DREF module, FASM module, and Bayesian prediction module in the fine-grained feature fusion network multiple times to obtain the trained fine-grained feature fusion network; Specifically, calculate the mean square error between the predicted value of the surface roughness and the output sample to obtain the loss function value. Based on the loss function value, use the Adam optimizer to update the parameters in the DREF module, FASM module, and Bayesian prediction module in the fine-grained feature fusion network, and obtain the next training sample in the training set. Return to step 3 until the number of parameter updates reaches the preset number of iterations to obtain the trained fine-grained feature fusion network; During the training process, the model training is executed based on the TensorFlow 1.14.0 framework with Keras 2.2.4. The learning rate and batch size are set to 0.001 and 64 respectively, and the preset number of iterations is 30.

[0035] Combined with Figure 4 , the predicted value obtained by the present invention is highly consistent with the true value measured by the roughness meter, fully reflecting its potential and value in industrial applications. To verify the performance advantages of the proposed method in surface roughness prediction, five classic deep learning methods are selected for comparison, including CapsNet, DCNN, BiLSTM, BiGRU, and ResNet. The evaluation metrics are R 2 , RMSE, MAPE, and MAE. The comparison results are shown in Table 2. Obviously, the proposed method obtains the optimal results in surface roughness prediction. R 2 is increased by 6.36% - 24.52%, RMSE is reduced by 50.7% - 70.19%, MAPE is reduced by 51.97% - 80.5%, and MAE is reduced by 55.08% - 70.84%.

[0036] Table 2 Comparison results with typical deep learning methods Method <![CDATA[R 2 > RMSE MAPE(%) MAE The present invention 0.9812 0.0493 6.5347 0.0345 CapsNet 0.8734 0.1279 19.0277 0.1003 DCNN 0.8364 0.1454 33.5193 0.1089 BiLSTM 0.9225 0.1 13.8407 0.0768 BiGRU 0.9081 0.1090 13.6052 0.0824 ResNet 0.7880 0.1654 29.0784 0.1183 Step 6: Continuously collect the force signal to be predicted and the vibration signal to be predicted, and then execute steps 3 - 4 based on the trained fine-grained feature fusion network to obtain the surface roughness. Specifically, perform multi-dimensional fine-grained feature extraction and splicing on the force signal and the vibration signal corresponding to the force signal to obtain the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the chatter signal. Input the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the chatter signal into the trained fine-grained feature fusion network to obtain the surface roughness.

[0037] A high-speed milling surface roughness prediction system considering chatter is used to implement a high-speed milling surface roughness prediction method considering chatter. Combined with Figure 5 , it includes a data acquisition module, a model training module, a data processing module, and a surface roughness prediction module; The data acquisition module is used to continuously acquire force signals and vibration signals, and transmit the force signals and vibration signals to the model training module; The model training module is used to use the continuously acquired force signals and continuously acquired vibration signals as input samples, use the corresponding surface roughness of the input samples as output samples, construct a training set based on the input samples and output samples, and transmit the input samples of the training set to the data processing module; the model training module is also used to receive the predicted values of the surface roughness, and based on the predicted values of multiple surface roughnesses and their corresponding output samples, perform multiple iterative updates on the parameters in the fine-grained feature fusion network; The data processing module is used to receive the input samples, perform multi-dimensional fine-grained feature extraction and splicing on the force signals and vibration signals in the input samples, and transmit the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the chatter signals to the surface roughness prediction module; The surface roughness prediction module is used to receive the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the chatter signals, input the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the chatter signals into the fine-grained feature fusion network, obtain the predicted value of the surface roughness, and transmit the predicted value of the surface roughness to the model training module.

[0038] It should be noted that the interaction process between the data processing module and the surface roughness prediction module and the model training module only exists during the training process of the fine-grained feature fusion network. During the process of specific prediction after the training of the fine-grained feature fusion network is completed, in combination with step 6, specifically, step 6 continuously acquires the force signal to be predicted and the vibration signal to be predicted through the data acquisition module. Furthermore, the data acquisition module transmits the force signal to be predicted and the vibration signal to be predicted to the data processing module. The data processing module receives the force signal to be predicted and the vibration signal to be predicted, performs multi-dimensional fine-grained feature extraction and splicing on the force signal and the vibration signal, obtains the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the chatter signal, and transmits the multi-dimensional fine-grained feature components of the periodic signal and the multi-dimensional fine-grained feature components of the chatter signal to the surface roughness prediction module. After being processed by the surface roughness prediction module, the surface roughness is obtained.

[0039] In summary, the present invention extracts the multi-dimensional fine-grained features of the force signal and the vibration signal based on SVMD and WPD, thereby effectively characterizing the detailed information of the surface roughness caused by the inherent process and chatter effect during the milling process. The present invention constructs a novel fine-grained feature fusion network to effectively fuse the multi-dimensional fine-grained features of the periodic signal and the chatter signal, and realizes intelligent surface roughness prediction.

[0040] The advantages of the present invention are that it can improve the accuracy of surface roughness prediction in the milling process, providing an effective solution to the problem of surface roughness prediction for chatter that may occur in real scenarios. This will effectively ensure the stability of surface quality during the machining process, thereby reducing the scrap rate and the number of downtimes.

[0041] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method for predicting the surface roughness of high-speed milling considering chatter, characterized in that, Including: Step 1: Conduct a plane high-speed milling experiment on the workpiece on the milling table. Continuously collect force signals through the force sensor configured on the milling table, and at the same time, continuously collect vibration signals through the acceleration sensor configured on the milling table. Step 2: Use the continuously collected force signals and vibration signals as input samples, and use the corresponding surface roughness of the input samples as output samples to construct a training set based on the input samples and output samples. Step 3: For the input samples in the training set, perform multi-dimensional fine-grained feature extraction and splicing on the force signals and vibration signals in the input samples to obtain multi-dimensional fine-grained feature components of periodic signals and multi-dimensional fine-grained feature components of chatter signals. Step 4: Input the multi-dimensional fine-grained feature components of periodic signals and multi-dimensional fine-grained feature components of chatter signals into the fine-grained feature fusion network to obtain the predicted value of surface roughness. The fine-grained feature fusion network includes a DREF module, a FASM module, and a Bayesian prediction module. Step 5: Based on the predicted values of multiple surface roughnesses and their corresponding output samples, perform multiple iterative updates on the parameters in the DREF module, FASM module, and Bayesian prediction module in the fine-grained feature fusion network to obtain a trained fine-grained feature fusion network. Step 6: Continuously collect the force signal to be predicted and the vibration signal to be predicted, and then execute Steps 3-4 based on the trained fine-grained feature fusion network to obtain the surface roughness.

2. A method for predicting the surface roughness of high-speed milling considering chatter, as described in claim 1, wherein Step 3 specifically includes: Step 3.1: Use the wavelet packet decomposition method WPD to denoise the vibration signal to obtain the denoised vibration signal, and then extract the first periodic signal and the first chatter signal from the denoised vibration signal through successive variational mode decomposition SVMD. Step 3.2: Extract the second periodic signal and the second chatter signal from the force signal through successive variational mode decomposition SVMD. Step 3.3: Perform feature extraction on the first periodic signal, the first chatter signal, the second periodic signal, and the second chatter signal respectively to obtain the multi-dimensional fine-grained features of the first periodic signal, the multi-dimensional fine-grained features of the first chatter signal, the multi-dimensional fine-grained features of the second periodic signal, and the multi-dimensional fine-grained features of the second chatter signal. The multi-dimensional fine-grained features include time-domain features and frequency-domain features. Step 3.4: Perform vector splicing on the multi-dimensional fine-grained features of the first periodic signal and the multi-dimensional fine-grained features of the second periodic signal to obtain the multi-dimensional fine-grained feature components of periodic signals, and perform vector splicing on the multi-dimensional fine-grained features of the first chatter signal and the multi-dimensional fine-grained features of the second chatter signal to obtain the multi-dimensional fine-grained feature components of chatter signals.

3. A method for predicting the surface roughness of high-speed milling considering chatter, as claimed in claim 1, wherein Step 4 specifically includes: Step 4.1: According to the DREF module in the fine-grained feature fusion network, perform feature extraction and fusion on the multi-dimensional fine-grained feature components of periodic signals and multi-dimensional fine-grained feature components of chatter signals to obtain fused features. Step 4.2: Process the fused features through the SENet and BiLSTM in the FASM module to obtain high-level features. Step 4.3: Input the high-level features into the Bayesian prediction module to obtain the predicted value of the surface roughness.

4. A method for predicting the surface roughness of high-speed milling considering chatter, according to claim 3, characterized in that Step 4.1 specifically includes: Step 4.1.1: Perform deep feature extraction on the multi-dimensional fine-grained feature components of the periodic signal to obtain the deep features of the periodic signal. Specifically, perform convolution operations, batch normalization (BN), ReLU, max pooling, convolution operations, batch normalization (BN), ReLU, and max pooling on the multi-dimensional fine-grained feature components of the periodic signal to obtain the first deep feature. Pass the multi-dimensional fine-grained feature components of the periodic signal through a fully connected layer to obtain the second deep feature. Connect the first deep feature and the second deep feature in a residual manner by addition to obtain the deep features of the periodic signal; Step 4.1.2: Process the depth features of the periodic signal through a gating mechanism composed of a sigmoid layer and a tanh layer to obtain the first gating weight matrix , which is specifically implemented by the following formula: ； Among them, μ and ν are learning parameters, is the depth feature of the periodic signal, represents element-wise multiplication, represents the sigmoid layer; Perform convolution on the first gating weight matrix through one-dimensional convolution with three different receptive fields to obtain the first-scale feature, the second-scale feature, and the third-scale feature. Then, concatenate the first-scale feature, the second-scale feature, and the third-scale feature to obtain the first multi-scale feature. Input the first multi-scale feature into the FC layer. In the FC layer, the first multi-scale feature passes through the ReLU activation function and the sigmoid activation function to obtain the first adaptive weight; Step 4.1.3: Perform deep feature extraction on the multi-dimensional fine-grained feature components of the flutter signal to obtain the deep features of the flutter signal. Specifically, perform convolution operations, batch normalization (BN), ReLU, max pooling, convolution operations, batch normalization (BN), ReLU, and max pooling on the multi-dimensional fine-grained feature components of the flutter signal to obtain the third deep feature. Pass the multi-dimensional fine-grained feature components of the flutter signal through a fully connected layer to obtain the fourth deep feature. Connect the third deep feature and the fourth deep feature in a residual manner by addition to obtain the deep features of the flutter signal; Step 4.1.4: Process the depth features of the flutter signal through a gating mechanism composed of a sigmoid layer and a tanh layer to obtain a second gating weight matrix , which is specifically implemented through the following formula: ； Among them, is the depth feature of the flutter signal; Perform convolution on the second gating weight matrix through one-dimensional convolution with three different receptive fields to obtain the fourth-scale feature, the fifth-scale feature, and the sixth-fifth scale feature. Then, splice the fourth-scale feature, the fifth-scale feature, and the sixth-fifth scale feature to obtain the second multi-scale feature. Input the second multi-scale feature into the FC layer. In the FC layer, the second multi-scale feature passes through the ReLU activation function and the sigmoid activation function to obtain the second adaptive weight; Step 4.1.5: Weightedly sum the deep features of the periodic signal and the deep features of the flutter signal through the first adaptive weight and the second adaptive weight to obtain the fused features, which is specifically implemented by the following formula: ； Among them, F represents the fused feature, represents the first adaptive weight, represents the depth feature of the periodic signal, represents the second adaptive weight, represents the depth feature of the flutter signal, represents element-wise addition.

5. A method for predicting the surface roughness of high-speed milling considering chatter, according to claim 3, characterized in that Step 4.2 specifically includes: Step 4.2.1: In SENet, pass the fused features through global pooling, the ReLU activation function, and the sigmoid activation function to obtain the channel attention weights. Perform a dot product operation on the channel attention weights and the fused features through the Scale operation to obtain the recalibrated fused features; Step 4.2.2: Process the recalibrated fused features through BiLSTM to obtain the high-level features.

6. A method for predicting the surface roughness of high-speed milling considering chatter, as described in claim 3, wherein Step 4.3 specifically includes: In the Bayesian prediction module, convert the high-level features into a one-dimensional vector through a compression layer. Pass the one-dimensional vector through the ReLU activation function in the fully connected layer to obtain the vector with enhanced non-linearity. Process the vector with enhanced non-linearity through approximate Bayesian dropout to obtain the initial surface roughness. Through Monte Carlo (MC) random sampling, repeatedly input the same input sample into the fine-grained feature fusion network to obtain multiple initial surface roughness values, and calculate the average value of all the initial surface roughness values to obtain the predicted value of the surface roughness.

7. A high-speed milling surface roughness prediction system considering flutter, characterized in that, A method for predicting the surface roughness of high-speed milling considering flutter, which is used to implement the method described in claim 1, includes a data acquisition module, a model training module, a data processing module, and a surface roughness prediction module; The data acquisition module is used to continuously collect force signals and vibration signals and transmit the force signals and vibration signals to the model training module; The model training module is used to take the continuously collected force signals and continuously collected vibration signals as input samples, take the surface roughness corresponding to the input samples as output samples, construct a training set based on the input samples and output samples, and transmit the input samples of the training set to the data processing module; the model training module is also used to receive the predicted values of the surface roughness, and based on the predicted values of multiple surface roughnesses and their corresponding output samples, perform multiple iterative updates on the parameters in the fine-grained feature fusion network; The data processing module is used to receive the input samples, perform multi-dimensional fine-grained feature extraction and splicing on the force signals and vibration signals in the input samples, and transmit the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the flutter signals to the surface roughness prediction module; The surface roughness prediction module is used to receive the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the flutter signals, input the multi-dimensional fine-grained feature components of the periodic signals and the multi-dimensional fine-grained feature components of the flutter signals into the fine-grained feature fusion network, obtain the predicted values of the surface roughness, and transmit the predicted values of the surface roughness to the model training module.

Images