Composite material machining tool wear monitoring method and system based on mc-sbulstm algorithm

Abstract

The application discloses a kind of composite material processing cutter wear monitoring methods based on MC-SBULSTM algorithm, comprising: S1, the working data of end effector in the machining process in robot machining system is collected;S2, detection input and detection target are formed initial sample data set;S3, initial sample data set is labeled with evaluation index, and sample data set is formed;S4, the prediction network is trained using sample data set, to obtain cutter wear monitoring model for predicting tool wear in machining process;S5, the working data in robot machining system is input to cutter wear monitoring model, to obtain the wear value of cutter in current robot machining system.The application also provides a kind of composite material processing cutter wear monitoring system.The method provided by the application can solve the problem of tool state monitoring of movable airfoil drilling equipment in wing assembly field in industrial production, which needs manual shutdown, and the monitoring effect is too dependent on manual experience and affects drilling.

Description

Technical Field

[0001] This invention belongs to the field of automated processing of composite materials, and particularly relates to a method and system for monitoring tool wear in composite material processing based on the MC-SBULSTM algorithm. Background Technology

[0002] Drilling is one of the most frequent machining operations in large aircraft assembly, accounting for over 40% of cutting operations in the aerospace industry. A large aircraft requires drilling millions of riveting or bolt holes in its main structural materials. Aircraft structural components also have very high requirements for hole quality. Statistics show that fatigue failure is the main cause of failure in aircraft structural components, with approximately 75% to 80% of fatigue failures occurring at the connection points of aircraft structures. Therefore, the quality of connection holes greatly affects the fatigue life of connecting components. The use of automated assembly equipment and the informatization and intelligentization of the entire assembly process provide technical support for improving hole-making efficiency and machining quality in aircraft assembly.

[0003] Real-time production monitoring is a crucial part of the manufacturing process, ensuring production safety and improving efficiency. Traditional machining monitoring systems focus on cutting tools, equipment, and signal processing. However, with the application of data-driven methods such as big data analytics, information fusion, and machine learning, real-time monitoring systems are gradually becoming more information-based and intelligent.

[0004] Patent document CN116008114A describes a method for detecting tool wear during bone material drilling based on sensor signals. This method involves collecting drilling force and vibration signals during bone material drilling, preprocessing the signals, extracting features from the extracted time-series signals and filtering the features, and inputting the filtered feature vectors into a bidirectional long short-term memory network optimized by a multi-head attention mechanism to predict the degree of wear.

[0005] Patent document CN112936425A discloses a device for automatically detecting tool wear during drilling of PPS sheets. The device includes a body, a support spring fixedly mounted below a swing rod, a drill bit movably mounted below the support spring, control rods movably mounted on both sides of the support spring, a control switch fixedly mounted above the control rods, an indicator light fixedly mounted above the control switch, and first sliders movably mounted on both sides of the upper end of the drill bit. A second slider is slidably mounted above the first slider. This device automatically detects tool wear during drilling of PPS sheets. When the drill bit is severely worn, it experiences a reaction force from the sheet material, causing it to move upwards. This causes the first sliders on both sides of the drill bit to move upwards, pushing the second slider upwards. The second slider then pushes the control rods into contact with the control switch, activating the indicator light and reminding the operator to replace the tool in time, thus achieving automatic detection of drill bit wear. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for monitoring tool wear in composite material processing based on the MC-SBULSTM algorithm. This method addresses the problem that tool condition monitoring in wing assembly sites requires manual shutdown, the monitoring effect is overly dependent on human experience, and it affects the drilling process. By extracting as many features as possible from a limited dataset, the method improves monitoring accuracy and robustness, meeting practical application requirements.

[0007] To achieve the first objective of this invention, the following technical solution is provided: a method for monitoring tool wear in composite material machining based on the MC-SBULSTM algorithm, comprising the following steps:

[0008] S1. Based on a pre-built robotic machining system for drilling holes in active wing surfaces, and collect the working data of the end effector in the robotic machining system during the machining process;

[0009] S2. Preprocess the acquired working data, select the spindle power signal, feed motor current signal and processing time from the processed working data as detection inputs, and use the tool wear value during processing as the detection target. Combine the detection inputs and detection targets to form an initial sample dataset.

[0010] S3. Using the drilling diameter as an evaluation index, the integrity of the machined surface, and the presence or absence of burrs at the exit, the initial sample dataset is labeled, and the initial sample dataset and the labels are combined to form a sample dataset.

[0011] S4. Construct a prediction network based on a multi-scale convolutional neural network and a stacked bidirectional long short-term memory network, and train the prediction network using a sample dataset to obtain a tool wear monitoring model for predicting tool wear during machining.

[0012] S5. Input the working data of the robot machining system into the tool wear monitoring model to obtain the current tool wear value in the robot machining system, providing a reference for whether to replace the tool in the future.

[0013] This invention collects detection inputs with time characteristics and predicts the tool wear in the robotic machining system through a pre-built tool wear monitoring model, providing a reference for whether to replace the tool in the future.

[0014] Specifically, the robotic processing system is a robotic hole-making system for making holes in composite material panels and internal skeletons of 14 movable wing surfaces such as flaps, outer flaps, and ailerons. It consists of a high-precision industrial robot, an end effector, a moving device, and an integrated control system. The industrial robot delivers the end effector to a designated position in space, and the end effector completes the hole-making task of the movable wing surfaces in a certain posture.

[0015] Specifically, the end effector is the execution component for drilling holes on movable wing surfaces, and mainly includes: actuator base, pressure foot unit, drilling unit, vision measurement unit, hole diameter and depth measurement unit, station switching unit, and robot quick change device.

[0016] Specifically, the preprocessing includes using a five-point cubic smoothing method and Pearson correlation analysis to reduce noise and dimensionality of the working data.

[0017] Specifically, the preprocessing includes data preprocessing, feature filtering, and correlation analysis, wherein data preprocessing includes data truncation, error removal, and noise removal.

[0018] Specifically, the prediction network uses a multi-scale convolutional neural network to extract multi-scale local features of the detection input, and uses a stacked bidirectional unidirectional long short-term memory network to obtain time-series features related to tool wear changes. Finally, the obtained multi-scale local features and time-series features are output as predicted tool wear values ​​through a fully connected network.

[0019] Specifically, the tool wear value is obtained through the following evaluation method:

[0020] The qualified hole quality standards should include a hole diameter tolerance of less than H11, a surface roughness of less than Ra3.2μm, and a burr height of less than 200μm. After measuring the three indicators of hole making on the test piece, it was found that burrs first restrict tool life, and then surface roughness is the second restrictive factor.

[0021] Since the maximum number of holes limited by the two indicators differs greatly, and there is a deburring process after drilling, we seek a wear threshold between the two to minimize the amount of deburring work while ensuring that the hole diameter and surface roughness are qualified.

[0022] In order to maximize tool life, we need to find the critical number of holes where the burr growth rate changes drastically.

[0023] The average value of the tool wear corresponding to the number of holes where the burr height change rate increases sharply is taken as the wear threshold of this tool in the drilling test piece of the robotic drilling equipment.

[0024] Specifically, the multi-scale convolutional neural network extracts multi-scale features at three spatial scales: 3, 5, and 7, and outputs the corresponding multi-scale local features by vertically concatenating and combining them.

[0025] Specifically, the fully connected network connects multi-scale local features and time-series features through the ReLU inactivation function.

[0026] Specifically, during training, the Adam optimizer is selected as the optimization algorithm, and the learning rate is adaptively adjusted.

[0027] Specifically, during training, the parameters of the prediction network are updated using a loss function and gradient calculation.

[0028] Specifically, the expressions for the loss function and gradient calculation are as follows:

[0029]

[0030] Where n is the number of samples, y i For actual data, y i ′ represents the predicted data, and y represents the loss value;

[0031] After the loss is calculated, the gradient is further calculated to reduce the loss value.

[0032] To achieve the second objective of this invention, the following technical solution is provided: a composite material machining tool wear monitoring system, implemented using the tool wear monitoring model provided above, comprising:

[0033] The data acquisition module is used to acquire the detection inputs set in the robot processing system;

[0034] The data analysis module analyzes the tool wear in the robot machining system based on the detection input and outputs the tool wear value.

[0035] The visualization module is used to display changes in the detected input values ​​and changes in tool wear values.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] A tool wear monitoring dataset for drilling carbon fiber laminates was established through hole-making experiments. Pearson correlation analysis showed that the monitoring signals in the data were strongly correlated with the tool wear values.

[0038] Based on process standards, a comprehensive evaluation method for hole quality considering drilling diameter, surface integrity and exit burrs is proposed, and the tool wear threshold for field-used tools is determined using this method.

[0039] The proposed tool wear monitoring algorithm MC-SBULSTM, which combines multi-scale convolutional neural networks and stacked long short-term memory neural networks, can extract as many features as possible from a limited dataset, improving monitoring accuracy and robustness and meeting practical application requirements. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the technical process for the composite material machining tool wear monitoring method provided in this embodiment;

[0041] Figure 2This is a structural diagram of the robot processing system provided in this embodiment;

[0042] Figure 3 This is a graph showing the correlation analysis results between the spindle signal and the feed motor signal provided in this embodiment;

[0043] Figure 4 This is a graph showing the correlation analysis results of spindle power signal, feed motor current signal, machining time and tool wear provided in this embodiment;

[0044] Figure 5 This is a diagram of the MC-SBULSTM network structure provided in this embodiment. Detailed Implementation

[0045] To make the technical solution and beneficial effects of the present invention clearer, the present invention will be described in detail below with reference to the specific implementation method of tool monitoring in the part machining process and the accompanying drawings. This embodiment is based on the technical solution of the present invention and provides detailed implementation methods and specific operating procedures, but the scope of protection of the present invention is not limited to the following embodiment.

[0046] like Figure 1 As shown, this embodiment provides a method for monitoring tool wear in composite material machining. The specific steps are as follows:

[0047] S1. Based on a pre-built robotic machining system for drilling holes in active wing surfaces, and collect the working data of the end effector in the robotic machining system during the machining process;

[0048] Furthermore, this robotic processing system, such as Figure 2 As shown, it includes a track laid on the base surface, a moving device moving along the track direction, and a high-precision industrial robot mounted on the moving device. The high-precision industrial robot is equipped with an end effector, which is the execution component for hole making on the movable wing surface. It mainly includes modules such as: actuator base, pressure foot unit, hole making unit, vision measurement unit, hole diameter and depth measurement unit, station switching unit, and robot quick change device. It has functions such as automatic drilling, countersinking, reference hole alignment and error compensation, automatic normal alignment, automatic hole diameter and depth detection, automatic pressure foot clamping, pressure foot protection, automatic tool status monitoring and protection, automatic tool changing and tool testing, automatic lubrication and cooling, and automatic dust collection.

[0049] The specific process of S1 is as follows:

[0050] S11: The robotic hole-making system for movable wing surfaces is a robotic hole-making system for making holes in composite material panels and internal skeletons of 14 movable wing surfaces such as flaps, outer flaps, and ailerons. It consists of a high-precision industrial robot, an end effector, a moving device, and an integrated control system.

[0051] S12: SINUMERIK Edge is an industrial-grade smart gateway designed by Siemens AG specifically for the machine tool industry. It integrates functions such as data acquisition, edge computing, data transmission, remote operation and maintenance, and security defense. It can achieve a data acquisition frequency of up to 1000Hz for machine tool data.

[0052] S13: The composite material specimen is a T300 carbon fiber reinforced composite laminate with a size of 250mm×250mm×4mm. Its upper surface is covered with epoxy preimpregnated expanded copper foil, and its lower surface is covered with a prepreg layer made of epoxy preimpregnated glass fiber fabric.

[0053] S14: The drilling tool is a special drilling tool for carbon fiber reinforced composite materials. The coating is diamond coating, the substrate is cemented carbide, the tool diameter is 4.2mm, the cutting length is 13mm, and the countersinking angle is 130°.

[0054] S2. Preprocess the acquired working data, select the spindle power signal, feed motor current signal and processing time from the processed working data as detection inputs, and use the tool wear value during processing as the detection target. Combine the detection inputs and detection targets to form an initial sample dataset.

[0055] The specific process of S2 is as follows:

[0056] S21: SINUMERIK Edge serves as a data acquisition gateway. To the south, a network cable and a switch are used to connect the XP1 interface to the SINUMERIK robot, and to the north, a router and a switch are used to connect to the edge server.

[0057] After establishing a local area network, log in to the Edge interface using the edge server, select the variables to be collected. In this example, select the power Z1 pow, current Z1 cur, torque Z1 tor, and load Z1 load of the feed motor, the power SP pow, current SP cur, torque SP tor, and load SP load of the electric spindle, the processing time time, and set the collection frequency to 1000Hz.

[0058] Then set the acquisition conditions. In order for Edge to automatically acquire data after the hole-making process begins, it is necessary to set the acquisition trigger conditions. Set the tool position -205.5mm as the acquisition start point. The acquisition operation will be activated whenever the tool feeds to this position. The acquisition will end when the tool position is greater than -205.5mm. This will enable the acquisition of hole-making process data.

[0059] S22: In this example, the KEYENCE VHX-7000 ultra-depth-of-field microscope is used to observe the tool. The flank wear VB has become an internationally recognized standard for measuring tool wear. It is defined in the standard ISO 8688:1 as follows: "Flavor wear is measured in a direction parallel to the wear surface and perpendicular to the initial cutting edge, and is the maximum value of the wear surface from the initial cutting edge to the initial flank face."

[0060] Due to uneven wear of the main cutting edge and chipping on some surfaces, this example measures three different locations on each of the two main cutting edges and takes the average value to reduce the randomness of the measurement.

[0061] Furthermore, the collected data was denoised and dimensionality reduced using the five-point cubic smoothing method and Pearson correlation analysis, and the spindle power signal, feed motor current signal, and machining time were selected as the monitoring input signals.

[0062] The process is as follows:

[0063] S23: Data interception. The distribution of spindle power and feed axis current presents five main stages: idling, drill bit entry, drill bit exit, countersinking, and rapid retraction. However, the data that truly affects hole quality should be between drill bit entry and exit. Therefore, data interception is required to remove the entry and retraction stages and the countersinking stage.

[0064] S24: Error point removal. During signal acquisition, if the equipment or acquisition system is subjected to sudden interference or transient anomaly, the data will undergo a brief change. For example, if a data point deviates significantly from the range of neighboring points, this point is an error point. When processing the data, points that exceed three times the standard deviation of the neighboring data are removed as error points and replaced with the average of the four neighboring points.

[0065] S25: Noise removal. Different processing processes exist in the workshop, and vibrations are inevitable during the processing. At the same time, the transmission chain of the equipment itself may also be a source of vibration.

[0066] This example uses a five-point cubic smoothing method to filter and average the data, making it smoother and more stable. Its basic principle is to use n equidistant sample points x0 to x... n-1 Data y0~y were collected. n-1 For each data point, two adjacent points are taken on each side, and a cubic polynomial is used to approximate the data.

[0067] S26: Feature screening. The sources of signal acquisition are divided into two categories, namely the feed motor signal and the spindle signal, and the types are divided into four types, namely power, current, torque, and load. In order to improve the operation speed, while ensuring the accuracy of the algorithm, it is necessary to reduce the feature dimension of the input data. In this example, the Pearson correlation analysis method is used to analyze the correlation of the input signals, and the features with low correlation are screened out to reduce the feature dimension;

[0068] Pearson correlation analysis is a commonly used statistical method for measuring the strength and direction of the linear relationship between two variables. It is based on the concept of covariance. By calculating the covariance of two variables divided by the product of their respective standard deviations, a correlation coefficient ranging from -1 to 1 is obtained. The calculation formula is:

[0069]

[0070] According to the absolute value of the correlation coefficient, the strength of the correlation can be judged. Generally speaking, a correlation coefficient greater than 0.7 or less than -0.7 is considered a strong correlation, between 0.4 and 0.7 or between -0.4 and -0.7 is considered a medium correlation, and between 0 and 0.4 or between -0 and -0.4 is considered a weak correlation;

[0071] As Figure 3 shown, the correlations between eight features are calculated. It can be concluded from this that the four types of signals of the spindle are strongly correlated, and the four types of signals of the feed motor are also strongly correlated. In this example, the spindle power signal and the feed motor current signal are selected as the inputs of the algorithm model. At the same time, the processing time of the tool is also important for wear prediction. Generally speaking, the longer the processing time, the more serious the tool wear. Finally, the input signals of the tool wear monitoring algorithm are determined to be the spindle power signal, the feed motor current signal, and the processing time. The Pearson analysis of these three types of signals and tool wear is as Figure 4 shown.

[0072] S3. Using whether the drilling diameter meets the dimensional tolerance, the integrity of the machined surface, and whether there are burrs at the exit as evaluation indicators, label the initial sample dataset, and form a sample dataset by combining the initial sample dataset and the labels;

[0073] Furthermore, the specific steps of S3 are as follows:

[0074] S31. Select the drilling diameter, surface integrity, and burr height at the exit as the quality evaluation criteria for drilling composite material holes. For riveting holes with 3 < d < 6mm, the hole diameter accuracy is taken as H11, and the tolerance is 0 - 0.075mm. At present, the maximum roughness of carbon fiber composite materials acceptable in the aviation field is 3.2μm, and the burr height is below 200μm;

[0075] S32: This example determines the tool wear threshold by studying the critical value of hole quality not meeting process standards due to tool wear. The hole diameter, roughness and burr height were measured by using an inside micrometer, a surface roughness meter and an ultra-depth-of-field microscope, respectively.

[0076] Analysis of the measurement data showed that the surface roughness of the three tools exceeded the process standard when drilling 840, 857, and 807 holes, respectively, and the burr height exceeded the process standard when drilling 542, 429, and 452 holes, respectively. The hole diameters were all within the allowable tolerance range. Based on this, it can be concluded that burrs are the primary factor restricting tool life. Considering the deburring process after drilling, this example seeks the tool wear value corresponding to the critical number of holes when the burr growth rate changes drastically as the tool wear threshold. Finally, the average wear value of the three tools corresponding to 620 holes, 85.7 μm, was selected as the tool wear threshold.

[0077] S4. Construct a prediction network based on a multi-scale convolutional neural network and a stacked bidirectional long short-term memory network, and train the prediction network using a sample dataset to obtain a tool wear monitoring model for predicting tool wear during machining.

[0078] Furthermore, the specific process of S4 is as follows:

[0079] S41: Multi-scale Convolutional Neural Network (MSCNN) consists of three parallel CNN layers. Each CNN layer can extract a feature map at one scale. The spatial feature lengths extracted at different scales are different, which preserves the global and local features of the network and enhances the feature extraction capability of the network. It has stronger discriminability and robustness than CNN.

[0080] S42: Stacked Bidirectional Unidirectional LSTM Network (SBULSTM). The core idea is to enhance the network's ability to extract deep temporal features of sample data by stacking multiple layers of BiLSTM and LSTM. In this example, a stacked bidirectional unidirectional LSTM network is constructed by using two layers of BiLSTM and one layer of LSTM.

[0081] S43: By combining MSCNN and SBULSTM, spatial and temporal features can be effectively extracted. The feature vector generated after the original signal passes through the MSCNN and SBULSTM networks is finally input into a fully connected layer to map and generate tool wear monitoring results, such as... Figure 5 The diagram shows the MC-SBULSTM network structure provided in this embodiment.

[0082] The specific working process of the MC-SBULSTM network is as follows:

[0083] The multi-channel input data first passes through a convolutional module to reduce the dimensionality and fuse the data from each channel. This module includes a one-dimensional convolutional layer, a ReLU inactive function, and a max pooling layer. The convolutional layer has a kernel size of 3, an output dimension of 64, a stride of 2, and padding of 1. The max pooling layer has a window size of 3 and a stride of 2.

[0084] The feature maps after dimensionality reduction and fusion are input into MSCNN, where spatial features of different scales are extracted using convolutional kernels of sizes 3, 5, and 7. MSCNN employs parallel computation, with three convolutional modules in each computation path. Each convolutional module consists of a convolutional layer, a ReLU inactive function, and a max pooling layer. The output dimensions of the convolutional layers are 64, 128, and 128, respectively. Finally, the output features from the three paths are vertically concatenated and combined, ready to be fed into the subsequent BiLSTM layer.

[0085] The input vector of SBULSTM is 128, so the input_size of the first BiLSTM layer is 128. The size of the hidden layer depends on several factors, including the size of the dataset, the number of features, and the availability of memory and computing resources. It usually needs to be tested experimentally. The hidden layer size of this BiLSTM layer is set to 256. The input_size of the second BiLSTM layer is 256, and the hidden layer size is set to 512. The input_size of the top LSTM layer is 512, and the hidden layer size is also set to 512. Therefore, the output of the SBULSTM network is the hidden state of the last time step of the ULSTM layer, and its dimension is 512. After the top LSTM layer, two fully connected layers are used to complete the feature mapping.

[0086] The parameters of the two fully connected layers are (512, 64) and (64, 1), respectively. A ReLU non-activation function is added in the middle to increase the non-linearity of the output. The final output wear value will be lost by the loss function along with the label value, and the gradient will be calculated to update the optimizer and enter the next iteration until the loss value reaches the set requirement or the number of iterations is used up.

[0087] To verify the performance of the tool wear monitoring model provided in this embodiment, a comparison experiment was conducted with the prediction accuracy of the existing mainstream algorithms and its model. The existing mainstream algorithms include shallow machine learning algorithms, single CNN and LSTM network algorithms, stacked bidirectional unidirectional LSTM and CNN and LSTM combined algorithms. The prediction accuracy was measured by the root mean square error (RMSE) and the mean absolute error (MAE). The comparison results are shown in Table 1.

[0088] Table 1

[0089]

[0090] S5. Input the working data of the robot machining system into the tool wear monitoring model to obtain the current tool wear value in the robot machining system, providing a reference for whether to replace the tool in the future.

[0091] This embodiment also provides a tool wear monitoring system for composite material machining, implemented using the tool wear monitoring model provided in the above embodiments, including:

[0092] The data acquisition module is used to acquire the detection inputs set in the robot processing system;

[0093] The data analysis module analyzes the tool wear in the robot machining system based on the detection input and outputs the tool wear value.

[0094] The visualization module is used to display changes in the detected input values ​​and changes in tool wear values.

[0095] Furthermore, the terms "upper," "lower," "inner," "outer," "front," and "rear" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0096] Of course, the above description is only a specific embodiment of the present invention and is not intended to limit the scope of the present invention. All equivalent changes or modifications made to the structure, features and principles described in the claims of the present invention should be included in the scope of the claims of the present invention.

[0097] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring tool wear in composite material machining based on the MC-SBULSTM algorithm, characterized in that, Includes the following steps: S1. Based on a pre-built robotic machining system for drilling holes in active wing surfaces, and collect the working data of the end effector in the robotic machining system during the machining process; S2. Preprocess the acquired working data, select the spindle power signal, feed motor current signal and processing time in the processed working data as detection inputs, and take the tool wear value during processing as the detection target. Combine the detection input and detection target to form an initial sample dataset. The preprocessing includes using the five-point cubic smoothing method and Pearson correlation analysis method to reduce noise and dimension of the working data. S3. Using the drilling diameter as an evaluation index, the integrity of the machined surface, and the presence or absence of burrs at the exit, the initial sample dataset is labeled, and the initial sample dataset and the labels are combined to form a sample dataset. S4. A prediction network is constructed based on a multi-scale convolutional neural network and a stacked bidirectional long short-term memory network, and the prediction network is trained using a sample dataset to obtain a tool wear monitoring model for predicting tool wear during machining. The prediction network extracts multi-scale local features of the detection input using a multi-scale convolutional neural network, and obtains time-series features related to tool wear changes through a stacked bidirectional unidirectional long short-term memory network. Finally, the obtained multi-scale local features and time-series features are output through a fully connected network to output the predicted tool wear value. The multi-scale convolutional neural network extracts multi-scale features at three spatial scales: 3, 5, and 7, and outputs the corresponding multi-scale local features by vertical concatenation. The fully connected network connects the multi-scale local features and time-series features through the ReLU inactive function. S5. Input the working data of the robot machining system into the tool wear monitoring model to obtain the current tool wear value in the robot machining system, providing a reference for whether to replace the tool in the future.

2. The method for monitoring tool wear in composite material machining based on the MC-SBULSTM algorithm according to claim 1, characterized in that, During training, the Adam optimizer is selected as the optimization algorithm, and the learning rate is adaptively adjusted.

3. The method for monitoring tool wear in composite material machining based on the MC-SBULSTM algorithm according to claim 1, characterized in that, During training, the parameters of the prediction network are updated using a loss function and gradient calculation.

4. The method for monitoring tool wear in composite material machining based on the MC-SBULSTM algorithm according to claim 3, characterized in that, The expressions for the loss function and gradient calculation are as follows: Where n is the number of samples, For actual data, For predictive data, The loss value is calculated first; after the loss is calculated, the gradient is further calculated to reduce the loss value.

5. A tool wear monitoring system for composite material machining, characterized in that, This is achieved through the tool wear monitoring model as described in claim 1, including: The data acquisition module is used to acquire the detection inputs set in the robot processing system; The data analysis module analyzes the tool wear in the robot machining system based on the detection input and outputs the tool wear value. The visualization module is used to display changes in the detected input values ​​and changes in tool wear values.

Images