A thin-walled part milling chatter identification method and system based on feature fusion

Abstract

The application provides a thin-walled part milling chatter identification method and system based on feature fusion, wherein the method comprises the following steps: acquiring a first data set, at least comprising machining process data and milling machining equipment data during workpiece machining; and performing data conversion on the first data set to obtain a second data set, processing the second data set to obtain a data matrix with multi-channel spatial features; acquiring image information, which is a workpiece machining surface image taken at the end of machining; processing the image information to obtain image data; establishing a chatter identification model, and taking the data matrix and the image data as inputs of the chatter identification model to obtain an identification result. The chatter identification method provided by the application combines milling signal data and workpiece machining surface image data to jointly train the model, improves the prediction accuracy and robustness of the model, and can solve the problem of insufficient accuracy of the existing chatter detection method.

Description

Technical Field

[0001] This disclosure generally relates to the field of chatter detection technology in thin-walled part milling, and specifically to a method for identifying chatter in thin-walled part milling based on feature fusion. Background Technology

[0002] In aerospace and other engineering fields, thin-walled parts are widely used due to their high strength and light weight. However, thin-walled parts generally have poor rigidity, making them prone to chatter during milling. Chatter not only affects the machining quality and efficiency of the parts but also the normal operation of machine tools and fixtures in the machining system. Therefore, it is necessary to detect chatter in the workpiece.

[0003] Currently, there are two main methods for chatter detection. The first method involves acquiring sensor signal data during the milling process, extracting features from the signal data using time-domain, frequency-domain, and wavelet packet analysis, and then identifying chatter. This method has limitations; it is easily interfered with by factors such as cutting environment noise and ordinary vibrations, leading to insufficient detection accuracy. The second method uses computer vision to process images of the milled workpiece surface, employing convolutional neural networks to process digital images and textures to detect chatter. This method has relatively high accuracy, but training the neural network model solely using images requires a large amount of image data of the machined surfaces of thin-walled parts as the training set, resulting in problems such as insufficient data volume and data imbalance, making model training difficult. Summary of the Invention

[0004] In view of the above-mentioned defects or deficiencies in the prior art, it is desirable to provide a method and system for identifying milling chatter of thin-walled parts based on feature fusion to solve the above problems.

[0005] The first aspect of this application provides a method for identifying milling chatter in thin-walled parts based on feature fusion, comprising the following steps:

[0006] A first data set is obtained, the processing data set including at least: first processing process data, second processing process data, and milling equipment data during the milling of thin-walled workpieces; the first processing process data and the second processing process data are used to characterize the state of the workpiece during processing.

[0007] The first data set is transformed to obtain a second data set, and the second data set is processed to obtain a data matrix with multi-channel spatial features; the second data set includes the coordinates of the workpiece processing position and third processing data, the third processing data being obtained by resampling the first processing data and the second processing data;

[0008] Acquire image information, which is an image of the workpiece's machined surface taken at the end of the machining process; process the image information to obtain image data;

[0009] The data matrix and the image data are input into the flutter recognition model to obtain the recognition result of whether the workpiece is fluttering.

[0010] According to the technical solution provided in the embodiments of this application, the step of performing data transformation on the first data set to obtain a second data set, and processing the second data set to obtain a multi-channel spatial feature data matrix, specifically includes the following steps:

[0011] A first sampling frequency and a second sampling frequency are obtained, and a third sampling frequency is set based on the first sampling frequency and the second sampling frequency; the third sampling frequency is less than or equal to the smallest of the first sampling frequency and the second sampling frequency.

[0012] The first processing data and the second processing data are resampled according to the third sampling frequency to obtain the third processing data.

[0013] Obtain the workpiece coordinate system, and calculate the position coordinates of the third machining process data in the workpiece coordinate system through machine tool kinematics to obtain the second data set;

[0014] Position matching is performed based on the second data set to obtain the data matrix with multi-channel spatial characteristics.

[0015] According to the technical solution provided in the embodiments of this application, the step of performing position matching based on the second data set to obtain the data matrix with multi-channel spatial features specifically includes the following steps:

[0016] An envelope box is created based on the second data set, and a sliding window is created based on the milling equipment data.

[0017] The data matrix is ​​obtained by traversing the envelope box using the sliding window and calculating the feature values ​​of all processing data within the sliding window.

[0018] According to the technical solution provided in the embodiments of this application, the method for constructing the flutter recognition model specifically includes the following steps:

[0019] Obtain the neural network model;

[0020] Construct a training set; the training set includes data combinations and real labels, the data combinations being composed of the data matrix and the image data, and used as input to the neural network model;

[0021] The neural network model is trained using the training set to obtain the flutter recognition model.

[0022] According to the technical solution provided in the embodiments of this application, the process by which the flutter recognition model outputs a recognition result based on the data matrix and the image data specifically includes the following steps:

[0023] Feature extraction is performed on the data matrix and the image data to obtain a first feature vector of the data matrix and a second feature vector of the image data;

[0024] The first feature vector and the second feature vector are concatenated, and the concatenated first feature vector and the second feature vector are weighted to obtain a weighted feature vector;

[0025] The recognition result is obtained by analyzing the weighted feature vector.

[0026] According to the technical solution provided in the embodiments of this application, the first feature vector and the second feature vector are weighted through a one-dimensional compression-excitation attention mechanism, specifically including the following steps:

[0027] The first feature vector and the second feature vector are stacked to obtain a fused feature vector;

[0028] The fused feature vector is globally pooled using a one-dimensional Squeeze operation.

[0029] The fused feature vector is weighted by a one-dimensional excitation operation.

[0030] According to the technical solution provided in the embodiments of this application, the process of processing the image information to obtain image data specifically includes the following steps:

[0031] The contrast of the image information is enhanced by histogram equalization.

[0032] The image information is then randomly cropped, vectorized, and normalized sequentially.

[0033] According to the technical solution provided in the embodiments of this application, after inputting the data matrix and the image data into the flutter recognition model to obtain the recognition result of whether the workpiece is fluttering, the specific steps include the following:

[0034] Store the data combination and update the training set;

[0035] The flutter recognition model is retrained using the updated training set.

[0036] A second aspect of this application provides a milling chatter identification system for thin-walled parts based on feature fusion, comprising:

[0037] A first acquisition module is configured to acquire a first data set, the processing data set including at least: first processing process data, second processing process data, and milling equipment data during the milling of thin-walled workpieces; the first processing process data and the second processing process data are used to characterize the state of the workpiece during processing.

[0038] The processing module is configured to perform data transformation on the first data set to obtain a second data set, and process the second data set to obtain a data matrix with multi-channel spatial features; the second data set includes the coordinates of the workpiece processing position and third processing data, the third processing data being obtained by resampling the first processing data and the second processing data;

[0039] The second acquisition module is configured to acquire image information, which is an image of the workpiece machining surface taken at the end of the machining process; and to process the image information to obtain image data.

[0040] The identification module inputs the data matrix and the image data into the flutter identification model to obtain the identification result of whether the workpiece is fluttering.

[0041] Compared with existing technologies, the advantages of this application are as follows: By acquiring a first data set and image information, processing the first data set to obtain a data matrix with multi-channel spatial features, and processing the image information to obtain image data, the data matrix and image data are used together as input to the chatter recognition model to obtain the recognition result of whether the workpiece is experiencing chatter. This effectively avoids the problem of interference from factors such as cutting environment noise and ordinary vibration when using only one type of data for chatter recognition, and also avoids the problem of interference from factors such as insufficient image data. This method converts time-series data into a data matrix and combines it with image data of the workpiece's machining surface to jointly train the model, improving the prediction accuracy and robustness of the chatter recognition model. It can more accurately detect the machining state of the workpiece, thereby improving the machining efficiency and product quality of thin-walled parts. Attached Figure Description

[0042] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0043] Figure 1 A flowchart illustrating the steps of the feature fusion-based chatter identification method for thin-walled parts provided in Embodiment 1 of this application;

[0044] Figure 2 A flowchart illustrating the specific steps of the feature fusion-based chatter recognition method for thin-walled parts provided in Embodiment 1 of this application;

[0045] Figure 3 This is a schematic diagram showing the positional relationship between the rectangular envelope box and the sliding window.

[0046] Figure 4 This is a schematic diagram of the flutter recognition model.

[0047] Figure 5 A schematic diagram of the structure of the flutter recognition system for thin-walled parts based on feature fusion provided in Embodiment 2 of this application. Detailed Implementation

[0048] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0049] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0050] Example 1

[0051] Please refer to Figure 1 and Figure 2 This embodiment provides a method for identifying milling chatter in thin-walled parts based on feature fusion, including the following steps:

[0052] S1. Obtain a first data set, the processing data set including at least: first processing process data, second processing process data and milling equipment data during the milling of thin-walled workpieces; the first processing process data and the second processing process data are used to characterize the state of the workpiece during processing;

[0053] S2. Perform data transformation on the first data set to obtain a second data set, and process the second data set to obtain a data matrix with multi-channel spatial features; the second data set includes the coordinates of the workpiece processing position and third processing data, the third processing data being obtained by resampling the first processing data and the second processing data;

[0054] S3. Acquire image information, wherein the image information is an image of the workpiece machining surface taken at the end of machining; process the image information to obtain image data;

[0055] S4. Input the data matrix and the image data into the flutter recognition model to obtain the recognition result of whether the workpiece is fluttering.

[0056] Specifically, in step S1, the first machining process data is machining process position data, which includes at least: the feed axis position and machining trajectory collected by the machine tool CNC system. The second machining process data refers to other process data excluding machining process position data, and includes at least: cutting force signals and vibration signals collected by detection sensors; the milling equipment data includes at least: the type of CNC machine tool, the type of cutting tool, and the cutting edge radius of the cutting tool. The first machining process data and the second machining process data can be collectively referred to as machining process data.

[0057] The specific type and model of the sensor can be selected according to the type of data to be collected. For example, a cutting force sensor can collect cutting force signals, and a vibration sensor can collect vibration signals.

[0058] Since different thin-walled parts have different processing techniques and precision requirements, different sensor combinations can be installed on the corresponding machine tools. For example, a three-axis force gauge and a three-axis vibration sensor can be installed on the tool holder to measure the cutting force and vibration signals in the X, Y, and Z axes, respectively.

[0059] In addition, it also includes machine tool machining data; the machine tool machining data includes at least the coordinate axis positions of the machine tool in the machine tool coordinate system, the coordinate axis positions in the workpiece coordinate system, and the coordinate axis positions in the tool coordinate system; here, the coordinate axis positions of the machine tool coordinate system are the coordinate axis positions marked on each machine tool body, and correspondingly, the coordinate axis directions of the workpiece coordinate system and the coordinate axis directions of the tool coordinate system are consistent with the coordinate axis directions of the machine tool coordinate system.

[0060] Step S2 specifically includes:

[0061] S21. Obtain a first sampling frequency and a second sampling frequency, and set a third sampling frequency based on the first sampling frequency and the second sampling frequency; the third sampling frequency is less than or equal to the smallest of the first sampling frequency and the second sampling frequency.

[0062] S22. Resample the first processing data and the second processing data according to the third sampling frequency to obtain the third processing data;

[0063] S23. Obtain the workpiece coordinate system, and calculate the position coordinates of the third machining process data in the workpiece coordinate system through machine tool kinematics to obtain the second data set;

[0064] S24. Perform position matching based on the second data set to obtain the data matrix with multi-channel spatial features.

[0065] Specifically, in step S21, since the sampling frequencies of the detection sensor and the machine tool CNC system are different, "first" and "second" are used to distinguish them. The first sampling frequency is the sampling frequency of the machining position information, that is, the sampling frequency of the machine tool feed axis. The second sampling frequency is other sampling frequencies (possibly multiple) besides the first sampling frequency. The third sampling frequency is a resampling frequency set based on the first sampling frequency and the second sampling frequency. The third sampling frequency is less than or equal to the smallest of the first sampling frequency and the second sampling frequency. Under the above limitations, the third sampling frequency is set as large as possible to ensure that more data is acquired and thus more information is stored.

[0066] In step S22, the first processing data and the third processing data are resampled according to the third sampling frequency, and the resampled results are combined to form the third processing data.

[0067] In this process, the maximum sampling frequency of the feed axis of a certain machine tool CNC system is used as the first sampling frequency, for example, 10kHz. The second sampling frequency of the detection sensor is, for example, 20kHz. Then, the third sampling frequency is set to 10kHz. The average value of every two second machining process data is calculated according to time. The first machining process data and the second machining process data are integrated to obtain the third machining process data.

[0068] In step S23, the coordinates of the third machining process data in the workpiece coordinate system are calculated based on the machine tool's forward kinematics, and the third machining process data is transferred to the workpiece surface based on the coordinates.

[0069] Taking a five-axis CNC milling machine tool with the AC axis as an example, but not limited to five-axis CNC milling machine tools with the AC axis as the rotary axis, data conversion is performed through forward kinematics. Based on the feed rates (X, Y, Z) of the translational axes (X-axis, Y-axis, Z-axis) and the feed rates (α, β) of the rotary axes (A-axis, C-axis) in the third machining process data, the machining position coordinates (x, y, z) corresponding to the third machining process data are obtained.

[0070] Please refer to Figure 3 Step S24 specifically includes the following steps:

[0071] An envelope box is created based on the second data set, and a sliding window is created based on the milling equipment data.

[0072] The data matrix is ​​obtained by traversing the envelope box using the sliding window and calculating the feature values ​​of all processing data within the sliding window.

[0073] Specifically, firstly, a rectangular envelope is established based on the position coordinates of several data points. Then, a circular sliding window with a specific radius (the specific radius needs to be based on the size of the machining tool) is slid in the rectangular envelope with a specific step size S until the entire rectangular envelope is browsed. The average value of the vibration signal of all the data points in the sliding window after each movement, or the maximum value of the cutting force signal, is calculated. The vibration signal and cutting force signal in different directions are used as different channels to obtain the data matrix with multi-channel spatial characteristics.

[0074] Step S3 specifically includes the following steps:

[0075] The contrast of the image information is enhanced by histogram equalization.

[0076] The image information is then randomly cropped, vectorized, and normalized sequentially.

[0077] Specifically, due to unavoidable random interference during imaging and transmission, the raw images acquired by the imaging system will encounter factors detrimental to image processing during conversion and transmission. For example, during photography, optical system distortion and measurement system vibration can blur the image; noise is introduced during image digitization (scanning, sampling, quantization); and noise contamination during transmission can degrade image quality. Therefore, a series of preprocessing operations are necessary before subsequent image processing and analysis.

[0078] First, histogram equalization is used to enhance the contrast of the workpiece machining surface image: First, a histogram of the image information is calculated, that is, the distribution of the number of pixels at different gray levels. This step is achieved by statistically analyzing the r value for each gray level. k The number of pixels n in the image k This is done first; then, the cumulative distribution function is calculated to measure the pixel distribution at each gray level in the image. The cumulative distribution function is obtained by summing the number of pixels in the histogram. Then, each pixel value of the original image is re-converted to a new gray level. The conversion process is as follows: First, the normalized version of the cumulative distribution function C′(r) is calculated using the following formula. k ):

[0079]

[0080] Wherein, C(r) k ) represents the unnormalized cumulative distribution function, indicating that the pixel value is less than or equal to r.k The cumulative number of pixels, C min It is the minimum value of the cumulative distribution function (usually 0), and M×N is the total number of pixels in the image.

[0081] Finally, each pixel value r in the original image is converted to a new grayscale level.

[0082] Through the conversion process, histogram equalization technology is used to process the milling images of thin-walled parts, which improves the contrast and quality of the images by redistributing the pixel values.

[0083] After histogram equalization improves image quality, the image is then randomly cropped, vectorized, and normalized. Data augmentation enhances the diversity and applicability of the image, while providing consistent input data to the neural network to improve model performance.

[0084] The method for constructing the flutter recognition model specifically includes the following steps:

[0085] Obtain the neural network model;

[0086] Construct a training set; the training set includes data combinations and real labels, the data combinations being composed of the data matrix and the image data, and used as input to the neural network model;

[0087] The neural network model is trained using the training set to obtain the flutter recognition model.

[0088] Specifically, firstly, a neural network model is built. After the model is built, it is trained using a prepared training set. By selecting appropriate loss functions, optimization algorithms, and hyperparameters, the model is iteratively optimized on the training set, and the model parameters are continuously adjusted to improve the model's accurate ability to identify workpiece flutter states, thus completing the model training and optimization. The structure of the flutter recognition model is as follows: Figure 4 As shown.

[0089] In step S4, the process of outputting the flutter recognition result based on the data matrix and the image data specifically includes the following steps:

[0090] The data matrix and the image data are input into a neural network model, and feature extraction is performed on the data combination to obtain a first feature vector of the data matrix and a second feature vector of the image data.

[0091] The first feature vector and the second feature vector are concatenated, and the concatenated first feature vector and the second feature vector are weighted to obtain a weighted feature vector;

[0092] The recognition result is obtained by analyzing the weighted feature vector.

[0093] Furthermore, the first feature vector and the second feature vector are weighted using a one-dimensional compression-excitation attention mechanism, specifically including the following steps:

[0094] The first feature vector and the second feature vector are stacked to obtain a fused feature vector;

[0095] The fused feature vector is globally pooled using a one-dimensional Squeeze operation.

[0096] The fused feature vector is weighted by a one-dimensional excitation operation.

[0097] Specifically, the design focus of the flutter recognition model is on how to effectively fuse the first feature vector extracted from the data matrix with the second feature vector extracted from the image data, that is, to build a model that uses two types of data to determine whether flutter has occurred during workpiece processing.

[0098] First, the data matrix and the image data of fixed size are input into the neural network model. They are first processed by convolutional layers for feature extraction. After passing through a series of convolutional layers, pooling layers and fully connected layers, two feature vectors are obtained and then concatenated.

[0099] Then, a one-dimensional squeeze-and-excitation attention mechanism is used to weight the concatenated feature vectors in order to highlight the flutter features.

[0100] Specifically, the weighted operation using a one-dimensional compression-incentive attention mechanism includes the following steps:

[0101] The first feature vector and the second feature vector are stacked to obtain a fused feature vector.

[0102] The one-dimensional Squeeze operation performs global average pooling on the input fused feature vector to obtain an average value (scalar) that represents the global information of the entire vector. This operation helps to capture the importance of the overall features.

[0103]

[0104] Where z represents the average value after compression, x i It is the i-th element in the feature vector, and N is the length of the feature vector.

[0105] One-dimensional Excitation operation: The average value obtained from the compression operation is input into a small fully connected layer. The fully connected layer learns the weights of each element, and a non-linear activation function is introduced to perform a non-linear transformation to obtain the weight factor of each element. Each element is multiplied by the obtained weights, and each element of the original fused feature vector is weighted, strengthening the information of important elements and weakening the information of unimportant elements.

[0106] By acquiring global information through a one-dimensional Squeeze operation and generating weights through a one-dimensional Excitation operation, the original fused feature vector is then weighted. This one-dimensional compression-excitation attention mechanism helps the model capture important features more effectively, thereby improving model performance.

[0107] In step S4, during the actual workpiece milling process, data is collected and processed using the same method as the method used to obtain the training set to ensure data consistency. The data matrix and the image data are input into the chatter recognition model to obtain the recognition result of whether the workpiece is experiencing chatter. The chatter recognition result includes at least the chatter state, transition state, and steady state. By fusing the features of the data matrix and the image data, the interference of the environment during milling is reduced, and richer and more diverse feature information is obtained. At the same time, the introduced one-dimensional compression-excitation attention mechanism enables the model to adaptively adjust the weights of different features, making the model pay more attention to key features and improving the robustness and prediction accuracy of the model.

[0108] Further, after inputting the data matrix and the image data into the flutter recognition model to obtain the recognition result of whether the workpiece is fluttering, the specific steps include the following:

[0109] Store the data combination and update the training set;

[0110] The flutter recognition model is retrained using the updated training set.

[0111] Specifically, after each chatter recognition using the data matrix and the image data, the data combination consisting of the data matrix and the image data is stored, and the stored data combination is added to the training set to update the training set. The chatter recognition model is periodically retrained using the updated training set. This training and update strategy helps to continuously optimize the model to adapt to the changes and challenges in the actual milling process.

[0112] Example 2

[0113] Please refer to Figure 5This embodiment provides a method for identifying milling chatter in thin-walled parts based on feature fusion, including:

[0114] A first acquisition module is configured to acquire a first data set, the processing data set including at least: first processing process data, second processing process data, and milling equipment data during the milling of thin-walled workpieces; the first processing process data and the second processing process data are used to characterize the state of the workpiece during processing.

[0115] The processing module is configured to perform data transformation on the first data set to obtain a second data set, and process the second data set to obtain a data matrix with multi-channel spatial features; the second data set includes the coordinates of the workpiece processing position and third processing data, the third processing data being obtained by resampling the first processing data and the second processing data;

[0116] The second acquisition module is configured to acquire image information, which is an image of the workpiece machining surface taken at the end of the machining process; and to process the image information to obtain image data.

[0117] The identification module inputs the data matrix and the image data into the flutter identification model to obtain the identification result of whether the workpiece is fluttering.

[0118] Specifically, the system provided in this embodiment is used to implement the feature fusion-based chatter recognition method for thin-walled part milling as described in the above embodiments.

[0119] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for identifying milling chatter in thin-walled parts based on feature fusion, characterized in that, Includes the following steps: A first data set is obtained, the processing data set including at least: first processing process data, second processing process data, and milling equipment data during the milling of thin-walled workpieces; the first processing process data and the second processing process data are used to characterize the state of the workpiece during processing. The first data set is transformed to obtain a second data set, and the second data set is processed to obtain a data matrix with multi-channel spatial features; the second data set includes the coordinates of the workpiece processing position and third processing data, the third processing data being obtained by resampling the first processing data and the second processing data; Acquire image information, which is an image of the workpiece's machined surface taken at the end of the machining process; process the image information to obtain image data; The data matrix and the image data are input into the flutter recognition model to obtain the recognition result of whether the workpiece is fluttering, including: Feature extraction is performed on the data matrix and the image data to obtain a first feature vector of the data matrix and a second feature vector of the image data; The first feature vector and the second feature vector are concatenated, and the concatenated first feature vector and the second feature vector are weighted to obtain a weighted feature vector; The recognition result is obtained by analyzing the weighted feature vector; The weighting of the first feature vector and the second feature vector using a one-dimensional compression-excitation attention mechanism specifically includes the following steps: The first feature vector and the second feature vector are stacked to obtain a fused feature vector; The fused feature vector is globally pooled using a one-dimensional Squeeze operation. The fused feature vector is weighted by a one-dimensional excitation operation; The first machining process data is machining process position data, which includes at least: the feed axis position and machining trajectory collected by the machine tool CNC system; The second machining process data refers to other process data excluding machining process position data. The second machining process data includes at least: cutting force signals and vibration signals collected by detection sensors; the milling equipment data includes at least: the type of CNC machine tool, the type of cutting tool, and the cutting edge radius of the cutting tool.

2. The method for identifying milling chatter in thin-walled parts based on feature fusion according to claim 1, characterized in that, The step of transforming the first data set to obtain a second data set, and then processing the second data set to obtain a multi-channel spatial feature data matrix, specifically includes the following steps: A first sampling frequency and a second sampling frequency are obtained, and a third sampling frequency is set based on the first sampling frequency and the second sampling frequency; the third sampling frequency is less than or equal to the smallest of the first sampling frequency and the second sampling frequency. The first processing data and the second processing data are resampled according to the third sampling frequency to obtain the third processing data. Obtain the workpiece coordinate system, and calculate the position coordinates of the third machining process data in the workpiece coordinate system through machine tool kinematics to obtain the second data set; Position matching is performed based on the second data set to obtain the data matrix with multi-channel spatial characteristics.

3. The method for identifying milling chatter in thin-walled parts based on feature fusion according to claim 2, characterized in that, The step of performing position matching based on the second data set to obtain the data matrix with multi-channel spatial features specifically includes the following steps: An envelope box is created based on the second data set, and a sliding window is created based on the milling equipment data. The data matrix is ​​obtained by traversing the envelope box using the sliding window and calculating the feature values ​​of all processing data within the sliding window.

4. The method for identifying milling chatter in thin-walled parts based on feature fusion according to claim 1, characterized in that, The method for constructing the flutter recognition model specifically includes the following steps: Obtain the neural network model; Construct a training set; the training set includes data combinations and real labels, the data combinations being composed of the data matrix and the image data, and used as input to the neural network model; The neural network model is trained using the training set to obtain the flutter recognition model.

5. The method for identifying milling chatter in thin-walled parts based on feature fusion according to claim 1, characterized in that, The process of processing the image information to obtain image data specifically includes the following steps: The contrast of the image information is enhanced by histogram equalization. The image information is then randomly cropped, vectorized, and normalized sequentially.

6. The method for identifying milling chatter in thin-walled parts based on feature fusion according to claim 5, characterized in that, After inputting the data matrix and the image data into the flutter recognition model to obtain the recognition result of whether the workpiece is fluttering, the specific steps include the following: Store the data combination and update the training set; The flutter recognition model is retrained using the updated training set.

7. A chatter recognition system for thin-walled parts milling based on feature fusion, characterized in that, include: A first acquisition module is configured to acquire a first data set, the processing data set including at least: first processing process data, second processing process data, and milling equipment data during the milling of thin-walled workpieces; the first processing process data and the second processing process data are used to characterize the state of the workpiece during processing. The processing module is configured to perform data transformation on the first data set to obtain a second data set, and process the second data set to obtain a data matrix with multi-channel spatial features; the second data set includes the coordinates of the workpiece processing position and third processing data, the third processing data being obtained by resampling the first processing data and the second processing data; The processing module is further configured to: extract features from the data matrix and the image data to obtain a first feature vector of the data matrix and a second feature vector of the image data; concatenate the first feature vector and the second feature vector; weight the concatenated first feature vector and the second feature vector to obtain a weighted feature vector; and analyze the weighted feature vector to obtain the recognition result. The processing module is further configured to: weight the first feature vector and the second feature vector using a one-dimensional compression-excitation attention mechanism, specifically including the following steps: The first feature vector and the second feature vector are stacked to obtain a fused feature vector; the fused feature vector is globally pooled by a one-dimensional Squeeze operation; the fused feature vector is weighted by a one-dimensional Excitation operation; the first machining process data is machining process position data, which includes at least: the feed axis position and machining trajectory collected by the machine tool CNC system; The second machining process data refers to other process data excluding machining process position data. The second machining process data includes at least: cutting force signals and vibration signals collected by detection sensors; the milling equipment data includes at least: the type of CNC machine tool, the type of cutting tool, and the cutting edge radius of the cutting tool. The second acquisition module is configured to acquire image information, which is an image of the workpiece machining surface taken at the end of the machining process; and to process the image information to obtain image data. The identification module inputs the data matrix and the image data into the flutter identification model to obtain the identification result of whether the workpiece is fluttering.

Images