Method for detecting milling chatter region of thin-walled part

Abstract

The application discloses a kind of thin-walled part milling chatter region detection methods, it is related to thin-walled part milling chatter detection technical field, the detection method includes: obtaining the machining process data and milling machining equipment data when thin-walled part milling and carries out data processing, obtains multichannel space feature data matrix;According to the data matrix is labeled to obtain data set according to chatter occurrence position information;Establishes chatter region detection model and utilizes data set to train, obtains thin-walled part milling chatter region detection model.This method converts time series data into data matrix by data processing, solves the problem that time series data loses a large amount of location relationship information, establishes chatter region detection model and uses data matrix to train, the chatter region detection model obtained can accurately detect the position of thin-walled part in milling process Chatter occurs, so as to more effectively detect and prevent the chatter phenomenon in the milling process of thin-walled part.

Description

Technical Field

[0001] This invention generally relates to the field of milling chatter detection technology, and specifically to a method for detecting milling chatter areas in thin-walled parts. Background Technology

[0002] Milling, due to its high precision and production efficiency, has been widely used in various manufacturing fields. However, chatter frequently occurs during milling. Chatter refers to the severe vibration caused by the relative movement of the tool and workpiece during machining, and it is a common problem in milling. Given the widespread use of thin-walled parts and the severe impact of chatter, chatter detection in thin-walled parts is a significant and urgent problem to be solved.

[0003] To more effectively detect and prevent chatter, we need to accurately pinpoint the specific location of chatter occurrences, rather than simply focusing on whether chatter occurs during a specific period or across the entire workpiece during machining. This introduces the concept of chatter region detection. This method allows us to precisely determine the exact location of chatter on the workpiece surface, helping engineers take more targeted measures and better understand and resolve chatter problems. Existing chatter detection methods typically analyze time-series signal data. However, this data often loses significant spatial location information. Directly detecting chatter regions using time-series signal data only provides the time of chatter occurrence, not its specific location. Furthermore, discontinuities in some abnormal signals within the time-series data can lead to the omission of certain chatter regions, failing to meet the requirements of chatter region detection. Therefore, there is an urgent need for a chatter region detection method for thin-walled part milling that can address these issues. 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 for detecting chatter areas in the milling of thin-walled parts, which can accurately locate chatter areas.

[0005] In a first aspect, the present invention provides a method for detecting chatter regions during milling of thin-walled parts, the detection method comprising:

[0006] Acquire first machining process data, second machining process data, and milling equipment data during the milling of thin-walled parts; the first machining process data and the second machining process data are respectively used to characterize different machining information of the thin-walled parts during the milling process.

[0007] Data processing is performed based on the first processing process data, the second processing process data, and the milling equipment data to obtain a multi-channel spatial feature data matrix;

[0008] A chatter region detection model is established, and the multi-channel spatial feature data matrix is ​​input into the chatter region detection model. The chatter region detection model analyzes and obtains the regions where chatter occurs during the milling process of the thin-walled part.

[0009] According to the technical solution provided by the embodiments of the present invention, data processing is performed based on the first processing process data, the second processing process data, and the milling equipment data to obtain a multi-channel spatial feature data matrix, specifically including the following steps:

[0010] Obtain the first and second sampling frequencies, and set the third sampling frequency;

[0011] According to the third sampling frequency, the data of the first processing process and the data of the second processing process are resampled to obtain the data of the third processing process.

[0012] Obtain the workpiece coordinate system of the thin-walled part, calculate the machining position coordinates of the third machining process data in the workpiece coordinate system, and obtain a data set containing the machining position coordinates and the third machining process data;

[0013] The multi-channel spatial feature data matrix is ​​calculated based on the data set and the milling equipment data.

[0014] According to the technical solution provided by the embodiments of the present invention, the multi-channel spatial feature data matrix is ​​calculated based on the data set and the milling equipment data, specifically including the following steps:

[0015] An envelope box is established based on the processing position coordinates of the dataset, and a sliding window is created based on the milling equipment data;

[0016] The envelope box is traversed using the sliding window, and the feature values ​​of all the third processing data within each sliding window are calculated to obtain the multi-channel spatial feature data matrix.

[0017] According to the technical solution provided in the embodiments of the present invention, a flutter region detection model is established, specifically including the following steps:

[0018] The chatter region of the milled thin-walled part is marked, the machining surface information of the thin-walled part is obtained, and the relative position information of the chatter region with respect to the machining surface of the thin-walled part is calculated.

[0019] The relative positions of the flutter regions are matched with the multi-channel spatial feature data matrix obtained through the data processing to obtain a set of data matrices with the flutter regions marked by annotation boxes;

[0020] The flutter region detection model is obtained by training the data matrix set.

[0021] According to the technical solution provided in the embodiments of the present invention, the data matrix set of flutter regions is marked with annotation boxes, and the positions of the annotation boxes are obtained by the following formula:

[0022]

[0023]

[0024]

[0025]

[0026] Among them, g x ,g y g represents the coordinates of the top-left corner of the label box on the data matrix. w ,g h The width and height of the annotation box are represented by m; the number of rows in the data matrix is ​​represented by n; the number of columns in the data matrix is ​​represented by w0; and the height of the machined surface is represented by h0.

[0027] According to the technical solution provided in the embodiments of the present invention, the flutter region detection model further includes calculating the confidence level using the following formula:

[0028] s c =p o ×iou(F,T)+α×b w ×b h (5)

[0029] Among them, s c p represents the confidence score. o The probability of a flutter region existing within the predicted bounding box generated by the model is represented by F and T, respectively. iou(F,T) represents the intersection-union ratio (IU / U) of the predicted and ground truth bounding boxes, and α represents the area weighting coefficient (updated during neural network training). w b represents the width of the prediction box. h This indicates the height of the prediction box.

[0030] According to the technical solution provided in the embodiments of the present invention, the training of the flutter region detection model includes the following steps:

[0031] Initialize the area threshold of the prediction box;

[0032] If the area of ​​the predicted box obtained in the current training is less than the area threshold, the predicted box is discarded directly, and the process continues until the final target predicted box is obtained.

[0033] During the training process, the area threshold is constantly being updated.

[0034] According to the technical solution provided in the embodiments of the present invention, the process of filtering until the final target prediction box is obtained specifically includes the following steps:

[0035] Calculate the confidence score for each predicted bounding box during training;

[0036] The confidence scores corresponding to each prediction box are sorted from high to low, and a rejection IOU threshold is set; the rejection IOU threshold is used to filter out prediction boxes with an overlap rate exceeding the threshold and a low confidence score;

[0037] The formula for calculating IOU is:

[0038] I = |x 12 -x 21 |×|y 12 -y 21 | (6)

[0039] U = |x 12 -x 11 |×|y 12 -y 11 |+|x 22 -x 21 |×|y 22 -y 21 |-I (7)

[0040]

[0041] Where I represents the area of ​​the intersection of the predicted boxes; U represents the area of ​​the union of the two predicted boxes; iou represents the intersection-union ratio of the two predicted boxes; (x 11 ,y 11 ) and (x 12 ,y 12 (x) represents the diagonal vertex of a prediction box; 21 ,y 21 ) and (x 22 ,y 22 () represents the diagonal vertex of another prediction box.

[0042] In summary, this technical solution specifically discloses a method for detecting chatter regions during milling of thin-walled parts. The detection method includes: acquiring machining process data and milling equipment data during the milling of thin-walled parts and processing the data to obtain a multi-channel spatial feature data matrix; labeling the data matrix according to the chatter occurrence location information to obtain a dataset; establishing a chatter region detection model and training it using the dataset to obtain a trained chatter region detection model for milling of thin-walled parts.

[0043] Existing chatter detection methods typically analyze time-series signal data. However, this data often loses significant spatial location information. Directly detecting chatter regions using time-series data only provides the time of occurrence, not the specific location. Furthermore, discontinuities in some anomalous signals within the time-series data can lead to the omission of certain chatter regions, failing to meet the requirements of chatter region detection. This new method transforms time-series data into a data matrix through data processing, resolving the issue of lost positional information. A chatter region detection model is established and trained using the data matrix. This model can accurately detect the location of chatter during the milling process of thin-walled parts, thus more effectively detecting and preventing chatter phenomena during the milling of thin-walled parts. Attached Figure Description

[0044] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0045] Figure 1 This is a flowchart illustrating a method for detecting chatter areas during milling of thin-walled parts.

[0046] Figure 2 This is a flowchart illustrating step S200 in a method for detecting chatter areas during milling of thin-walled parts.

[0047] Figure 3 This is a schematic diagram of the overall process for a method to detect chatter areas during milling of thin-walled parts.

[0048] Figure 4 This is a schematic diagram of the prediction box filtering algorithm in a method for detecting chatter areas during milling of thin-walled parts.

[0049] Figure 5 This is a structural diagram of a convolutional neural network model used in a method for detecting chatter areas during milling of thin-walled parts. Detailed Implementation

[0050] The present invention 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 for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] Example 1

[0053] Please refer to Figure 1 and Figure 3 The flowchart and overall flowchart of a method for detecting chatter areas in milling of thin-walled parts provided in this embodiment are shown. The detection method includes:

[0054] S100: Acquire the first machining process data, the second machining process data, and the milling equipment data during the milling of thin-walled parts;

[0055] The first machining process data refers to machining process position data, mainly feed axis position data and machining trajectory data during machining; the second machining process data refers to other process data excluding machining process position data; for example, the first machining process data includes at least: feed axis position and machining trajectory collected by the machine tool CNC system; the second machining process data includes at least: cutting force signal and vibration signal collected by detection sensors; the milling equipment data includes at least: the type of CNC machine tool, the type of tool, and the cutting edge radius of the tool.

[0056] During the workpiece milling experiment, data such as cutting force, vibration signals, spindle current, feed rate, and rotational speed can be collected through sensors and the machine tool CNC system. Simultaneously, the machine tool CNC system can also collect milling equipment data such as tool type, size, cutting edge radius, and the coordinate axis positions of the machine tool, workpiece, and tool.

[0057] The detection sensors can be force sensors, vibration sensors, current sensors, acoustic sensors, etc., depending on the type of signal data acquisition. When dealing with thin-walled parts, different thin-walled parts have different processing techniques and precision requirements. Therefore, different sensor combinations should be installed on the corresponding machine tools. For example, a three-dimensional force gauge and a three-dimensional vibration sensor can be installed on the tool holder to measure the vibration signals and cutting force signals in the X, Y, and Z directions of the tool. The specific type and model of the data acquisition sensors can be selected according to the actual situation. Finally, the signal data collected by each sensor needs to be summarized.

[0058] Combination Figure 2 It can be seen that, S200: Data processing is performed based on the first processing process data, the second processing process data and the milling equipment data to obtain a multi-channel spatial feature data matrix; the multi-channel spatial feature data matrix is ​​composed of processing information of at least one of the processing signals, and different processing signals serve as different channels of the data matrix; the processing signal is, for example, a vibration signal, a cutting force signal, etc.

[0059] Furthermore, step S200 can be further elaborated as follows:

[0060] S201: Obtain the first sampling frequency and the second sampling frequency, and set the third sampling frequency;

[0061] S202: Resample the first processing data and the second processing data according to the third sampling frequency to obtain the third processing data;

[0062] The first sampling frequency is the sampling frequency of the machining position information, i.e. 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 the resampling frequency set based on the first and second sampling frequencies. The third sampling frequency is less than or equal to the first and second sampling frequencies. The first machining process data and the second machining process data are resampled according to the third sampling frequency.

[0063] Since the acquisition frequencies of the internal CNC system and the external detection sensors are different, it is first necessary to resample the data at different acquisition frequencies based on the acquisition frequency of the feed axis in the internal CNC system of the machine tool, so that the machining process data at different acquisition frequencies correspond one-to-one. Then, based on the acquisition frequency of the feed axis, the signal data acquired by the detection sensor with the higher acquisition frequency is resampled.

[0064] For example, in a preferred embodiment, the maximum sampling frequency of the CNC system feed axis can only reach 1kHz (limited by the CNC system hardware), while the sampling frequency of vibration and cutting force signals can reach up to 20kHz. Further, assuming that the sampling frequencies of vibration and cutting force signals are both 20kHz, and the CNC system feed axis data sampling frequency is 1kHz, the third sampling frequency is set to 1kHz, and the first machining process data and the second machining process data are resampled and aligned according to time.

[0065] S203: Obtain the workpiece coordinate system of the thin-walled part, calculate the machining position coordinates of the third machining process data in the workpiece coordinate system, and obtain a data set containing the machining position coordinates and the third machining process data;

[0066] Taking a five-axis CNC milling machine tool with the AC rotary axis as an example, but not limited to five-axis CNC milling machine tools with the AC rotary 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 machining process data, the machining position coordinates (x,y,z) of each data point in the workpiece coordinate system are obtained.

[0067] S204: Calculate the multi-channel spatial feature data matrix based on the data set and the milling equipment data;

[0068] Because the sampling data points are not evenly distributed on adjacent machining paths during workpiece milling, it is necessary to perform position matching operations on the machining process data based on the machining position coordinates.

[0069] Step S204 specifically includes the following steps one and two.

[0070] Step 1: Establish an envelope box based on the processing position coordinates of the dataset, and create a sliding window based on the milling equipment data;

[0071] Step 2: Use the sliding window to traverse the envelope box, calculate the feature values ​​of all the third processing data in each sliding window, and obtain the multi-channel spatial feature data matrix.

[0072] Specifically, the process involves: first, establishing a rectangular envelope based on the position coordinates of the data points; then, using a circular sliding window of a specific radius (the radius of which is the same as the radius of the machining tool), sliding sequentially within the envelope at a specific step size s until the entire envelope is viewed; calculating the average value of the vibration signal of all data points within each sliding window, or the maximum value of the cutting force signal, etc., where vibration signals and cutting force signals in different directions serve as different channels, thus obtaining a multi-channel spatial feature data matrix that reconstructs the spatial positional relationships.

[0073] S300: Establish a chatter region detection model and input the multi-channel spatial feature data matrix into the chatter region detection model. The chatter region detection model analyzes and obtains the regions where chatter occurs during the milling process of the thin-walled part.

[0074] Because the data matrix obtained after data processing has an unbalanced aspect ratio and the special characteristic of separable chatter regions, it is necessary to construct a suitable network model and train the model. At the same time, it is necessary to verify the positioning accuracy and reliability of the chatter region detection model for milling thin-walled parts. Therefore, the specific steps for establishing the chatter region detection model in step S300 are as follows:

[0075] (1) Dataset labeling:

[0076] Step 1: Mark the chatter area on the milled thin-walled part, obtain the machining surface information of the thin-walled part, and calculate the relative position information of the chatter area with respect to the machining surface of the thin-walled part.

[0077] For thin-walled parts that have been milled, firstly, based on the machining condition of the thin-walled workpiece, mark the chatter area on the machined surface of the workpiece, and measure the diagonal coordinates (x, y) of the marking frame. max ,y max ) and (x min ,ymin The relative position of the chattering region on the workpiece machining surface is obtained by taking the length s0 and height h0 of the workpiece machining surface. This relative position information is also obtained based on the workpiece coordinate system.

[0078] Step 2: Match the relative positions of the flutter regions with the multi-channel spatial feature data matrix obtained through the data processing to obtain a set of data matrices with the flutter regions marked by annotation boxes;

[0079] Here, the processing method for the current machining process data (part milling machining used to establish the chatter area detection model) is the same as the aforementioned processing method. Then, the chatter area is marked by the following formulas (1) to (4) to form a data matrix set. The data matrix set is used to train the chatter area detection model.

[0080]

[0081]

[0082]

[0083]

[0084] Among them, g x ,g y g represents the coordinates of the top-left corner of the label box on the data matrix. w ,g h The width and height of the annotation box are represented by m; the number of rows in the data matrix is ​​represented by n; the number of columns in the data matrix is ​​represented by w0; and the height of the machined surface is represented by h0.

[0085] (2) Feature extraction:

[0086] In a preferred embodiment, the flutter region detection model employs a deep convolutional neural network as its backbone, comprising 18 convolutional layers and 2 pooling layers. The model first extracts features through multiple convolutional layers. These layers capture features at different scales by using convolutional kernels of varying sizes and numbers. Simultaneously, a fine-grained feature fusion (pass-through) operation is introduced to fuse feature maps of different resolutions, thereby capturing information at different scales at different levels and improving the model's detection capability. The specific network model structure is as follows: Figure 5 As shown.

[0087] (3) Prediction box generation

[0088] To better adapt to the unbalanced aspect ratio of the data matrix, the prior anchors are pre-defined: K-means clustering is used to generate n prior anchor widths and heights suitable for flutter region detection. After the data matrix is ​​processed by a convolutional neural network, it is divided into 13×13 squares, each square generating a prediction box. The model predicts the offset t of the center point coordinates of each prediction box. x and t y and the width and height offset t of the prediction box w and t h Then, the center point coordinates and width and height of the final prediction box are calculated by the following formulas (9) to (12).

[0089] The label feature data of the flutter region detection model includes: the coordinates of the center point and the width and height of the predicted bounding box of the flutter region;

[0090] The coordinates of the center point and the width and height values ​​are calculated using the following formula:

[0091] b x =σ(t) x )+c x (9)

[0092] b y =σ(t) y )+c x (10)

[0093]

[0094]

[0095] Where, p w and p h c represents the width and height of the prior bounding box. x and c y t represents the coordinates of the top-left corner of the square. x ,t y ,t w ,t h These represent the offsets of the center point coordinates and the width and height of the predicted bounding box, respectively, and σ(x) represents the sigmoid function. x ,c y b was obtained after calculation. x and b y b represents the coordinates of the center point of the actual predicted bounding box; w and b h These represent the width and height of the actual predicted bounding box, respectively.

[0096] (4) Prediction box filtering

[0097] Since the flutter region is separable, dividing a flutter region into two still results in a flutter region. Therefore, to avoid discarding larger boxes that are closer to the true label box in the subsequent prediction box selection step due to the higher confidence of smaller boxes, a method for calculating the confidence of prediction boxes in the model is defined, and the specific formula (6) is as follows:

[0098] s c =p o ×iou(F,T)+α×b w ×b h (6)

[0099] Among them, s c p represents the confidence score. o Let F and T represent the predicted bounding box generated by the model and the ground truth bounding box, respectively. Let iou(F,T) represent the intersection-union ratio (IU / IU) of the predicted and ground truth bounding boxes, and α be the area weighting coefficient to be trained. w b represents the width of the prediction box. h This indicates the height of the prediction box.

[0100] Meanwhile, the flutter region detection model obtained through training also includes the following steps for filtering prediction boxes:

[0101] Initialize the area threshold S of the prediction box min If the area of ​​the predicted bounding box obtained in the current training is less than the area threshold S min If the predicted box is not found, the predicted box is discarded directly, and the filtering continues until the final target predicted box is obtained.

[0102] During training, the area threshold is continuously updated until the model's loss reaches an acceptable range, ultimately yielding a suitable area threshold S. min .

[0103] Combination Figure 4 As can be seen, since the flutter region detection task only requires one class, the filtering process until the final target prediction box is obtained includes the following steps:

[0104] Step 1: Calculate the confidence score of each prediction box during training;

[0105] Step 2: Sort the confidence scores of each prediction box from high to low, and set a rejection IOU threshold; the rejection IOU threshold is used to filter out prediction boxes with an overlap rate exceeding the threshold and a low confidence score.

[0106] The formula for calculating IOU is:

[0107] I = |x 12 -x21 |×|y 12 -y 21 | (7)

[0108] U = |x 12 -x 11 |×|y 12 -y 11 |+|x 22 -x 21 |×|y 22 -y 21 |-I (8)

[0109]

[0110] Where I represents the area of ​​the intersection of the predicted boxes; U represents the area of ​​the union of the two predicted boxes; iou represents the intersection-union ratio of the two predicted boxes; (x 11 ,y 11 ) and (x 12 ,y 12 (x) represents the diagonal vertex of a prediction box; 21 ,y 21 ) and (x 22 ,y 22 () represents the diagonal vertex of another prediction box.

[0111] In actual thin-walled part milling operations, data is collected and processed using the same method. Specifically, using the same machining equipment, data acquisition method, and data processing method, the obtained data is processed and then input into the trained thin-walled part milling chatter region detection model to obtain the chatter occurrence area during the milling process. To improve the model's accuracy in detecting and locating chatter regions, machining process data collected during each thin-walled workpiece milling operation is added to the dataset, and the model is retrained at specific intervals to update the model's parameters.

[0112] In summary, this invention employs a machine tool CNC system and various external detection sensors to collect machining process data during milling. It then performs data processing operations such as signal resampling, data conversion, and position matching to transform one-dimensional time-series signal data into a multi-channel signal matrix. This eliminates the need for fixed-window segmentation of the time-series signal data, enabling the extraction of signal data from non-fixed spatial regions. This flexible data extraction method can accurately cover more discontinuous chatter signals in the time series, better extract chatter signal features, reduce manual intervention, avoid information loss issues caused by fixed windows, and improve the accuracy and stability of subsequent chatter region detection.

[0113] Secondly, since some of the obtained signal matrix data has an unbalanced aspect ratio, the prior boxes (anchors) of the traditional object detection model need to be optimized to prevent them from failing to correctly fit the fluttering regions. Furthermore, since the fluttering regions are separable (i.e., one fluttering region can be divided into two parts), to prevent larger boxes that are closer to the true labeled boxes from being discarded during subsequent non-maximum suppression (NMS), the calculation method for the confidence score and the predicted box selection mechanism in the model need to be improved.

[0114] Finally, to ensure the accuracy and precision of the neural network detection after training, the machining process data of the actual thin-walled workpiece milling process is obtained and stored. At set intervals, the trained neural network model is retrained based on the expanded dataset.

[0115] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes 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-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A method for detecting chatter areas during milling of thin-walled parts, characterized in that, The detection method includes: Acquire first machining process data, second machining process data, and milling equipment data during the milling of thin-walled parts; the first machining process data and the second machining process data are respectively used to characterize different machining information of the thin-walled parts during the milling process. Data processing is performed based on the first processing process data, the second processing process data, and the milling equipment data to obtain a multi-channel spatial feature data matrix; A chatter region detection model is established, and the multi-channel spatial feature data matrix is ​​input into the chatter region detection model. The chatter region detection model analyzes and obtains the regions where chatter occurs in the thin-walled part during milling. The process of processing data based on the first processing data, the second processing data, and the milling equipment data to obtain a multi-channel spatial feature data matrix includes the following steps: Obtain the first and second sampling frequencies, and set the third sampling frequency; According to the third sampling frequency, the data of the first processing process and the data of the second processing process are resampled to obtain the data of the third processing process. Obtain the workpiece coordinate system of the thin-walled part, calculate the machining position coordinates of the third machining process data in the workpiece coordinate system, and obtain a data set containing the machining position coordinates and the third machining process data; The multi-channel spatial feature data matrix is ​​calculated based on the data set and the milling equipment data. The establishment of a flutter region detection model includes the following steps: The chatter region of the milled thin-walled part is marked, the machining surface information of the thin-walled part is obtained, and the relative position information of the chatter region with respect to the machining surface of the thin-walled part is calculated. The relative positions of the flutter regions are matched with the multi-channel spatial feature data matrix obtained through the data processing to obtain a set of data matrices with the flutter regions marked by annotation boxes; The flutter region detection model is obtained by training the data matrix set.

2. The method for detecting chatter areas during milling of thin-walled parts according to claim 1, characterized in that, The multi-channel spatial feature data matrix is ​​calculated based on the dataset and the milling equipment data, specifically including the following steps: An envelope box is established based on the processing position coordinates of the dataset, and a sliding window is created based on the milling equipment data; The envelope box is traversed using the sliding window, and the feature values ​​of all the third processing data within each sliding window are calculated to obtain the multi-channel spatial feature data matrix.

3. The method for detecting chatter areas during milling of thin-walled parts according to claim 1, characterized in that, The data matrix set of flutter regions is marked with annotation boxes. The positions of the annotation boxes are obtained using the following formula: in, This represents the coordinates of the top-left corner of the label box on the data matrix. Indicates the width and height of the annotation box; Indicates the number of rows in the data matrix; Indicates the number of columns in the data matrix; Indicates the width of the machined surface; Indicates the height of the machined surface.

4. The method for detecting chatter areas during milling of thin-walled parts according to claim 3, characterized in that, The flutter region detection model also includes calculating the confidence level using the following formula: in, This represents the confidence score. This indicates the probability that a flutter region exists within the predicted bounding box generated by the model. F , T These represent the predicted bounding boxes and the ground truth label boxes generated by the model, respectively. This represents the intersection-union ratio (IoU) between the predicted bounding box and the ground truth label box. The area weights are the coefficients to be trained. This indicates the width of the prediction box. This indicates the height of the prediction box.

5. The method for detecting chatter areas during milling of thin-walled parts according to claim 4, characterized in that, The training of the flutter region detection model includes the following steps: Initialize the area threshold of the prediction box; If the area of ​​the predicted box obtained in the current training is less than the area threshold, the predicted box is discarded directly, and the process continues until the final target predicted box is obtained. During the training process, the area threshold is constantly being updated.

6. The method for detecting chatter areas during milling of thin-walled parts according to claim 5, characterized in that, The process of filtering until the final target prediction box is obtained includes the following steps: Calculate the confidence score for each predicted bounding box during training; Sort the confidence scores corresponding to each prediction box from high to low, and set elimination criteria. Threshold; the elimination The threshold is used to filter out predicted bounding boxes with an overlap rate exceeding the threshold and a low confidence score; The The calculation formula is: in, This represents the area of ​​the intersection of the predicted bounding boxes; This represents the area of ​​the union of two prediction boxes; This indicates the intersection-union ratio of two predicted boxes; and This represents the diagonal vertex of a prediction box; and This represents the diagonal vertex of another prediction box.

Images