Active expansion type end device optimization control method based on deep learning model

Abstract

The invention discloses an active expansion type end device optimization control method based on a deep learning model. The method comprises the following steps: firstly, mounting a high-precision sensor on each sliding block of the active expansion type end head device, and acquiring key data in a test process in real time through a high-speed data acquisition card, including sliding block displacement and speed, shearing force, friction force and the like; then, the collected data is preprocessed; a CNN-LSTM hybrid model is constructed, and the processed data is adopted to carry out training; after training is completed, the model is further verified and optimized; finally, the optimized CNN-LSTM model is applied to a sliding block active control system, the displacement and speed of the sliding block are predicted and adjusted in real time, it is ensured that the sliding block and the sample are highly matched in the large deformation process, and therefore the friction effect is effectively eliminated. High-precision control over the sliding block is achieved through the deep learning model, the accuracy of test data and the adaptive capacity of the system are greatly improved, and an excellent control effect is shown under complex experiment conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning. Background Art

[0002] Geotechnical engineering triaxial test is one of the core methods to study the mechanical properties of soil and rock, and its test results directly affect the reliability and safety of geotechnical engineering design. In triaxial test, the friction effect at the end of the specimen is one of the key issues affecting the accuracy of the test results. Traditional loading methods usually produce large sliding friction at the end of the specimen, resulting in uneven stress and strain distribution of the specimen, affecting the authenticity of the test results.

[0003] In order to solve this problem, CN118883231A proposed an active expansion control end device for geotechnical engineering triaxial compression test, such as Figure 1 , Figure 2 , Figure 3 , Figure 4 As shown, the device structure includes: a circular bearing base, which is used to carry other parts and is provided with corresponding sliding parts; a guide rail system, which includes multiple sets of guide rail blocks and several precision balls, and the guide rail blocks can move along the radial direction on the circular bearing base; a slider system, which includes multiple circles of sliders arranged along the radial direction around the center of the base, the sliders cooperate with the guide rail blocks, and a pressure-resistant arc rubber membrane is provided between the sliders; a slider active control system, including a micro servo motor, a carbon fiber force transmission rod, a sensor, a control unit, etc. The device converts the sliding friction between the sample and the loading head into the rolling friction between the slider and the ball, and combines servo control to realize active control of the slider, so as to better fit the movement law of the sample during large deformation. However, the servo control system has certain limitations in practical applications, especially when facing complex and changeable loading conditions, it is difficult to respond to the dynamic changes of the sample in real time and accurately, which may lead to deviations in the stress and strain state of the sample.

[0004] With the rapid development of deep learning technology, it has demonstrated excellent performance in the fields of pattern recognition, predictive control, and has achieved remarkable results in many engineering practices. Deep learning models can capture the potential laws in complex systems by learning from massive historical data, thereby making more accurate predictions and responses in real-time control processes.

[0005] Therefore, exploring an optimization control method and applying the deep learning model to the active deployment control end device to achieve the loading of the unit mechanical test process while maintaining the unit stress state and uniform deformation conditions is of great significance for studying the mechanical characteristics of geotechnical materials throughout the process. Summary of the invention

[0006] In order to solve the problem of inaccurate test results caused by the end friction effect in geotechnical engineering triaxial compression tests, the present invention provides an optimization control method for an active deployable end device based on a deep learning model, specifically an optimization control method for an active deployable end device based on a hybrid model of a convolutional neural network (CNN) and a long short-term memory network (LSTM). By introducing deep learning technology, especially combining the advantages of CNN and LSTM, the present invention breaks through the limitation of traditional servo control that cannot respond in time under complex, nonlinear and variable loading conditions, realizes real-time high-precision control of the slider in the test, and realizes a more comprehensive elimination of the end friction effect, thereby greatly improving the accuracy and reliability of the test results.

[0007] In order to achieve the above technical objectives, the present invention is implemented through the following technical solutions: An optimization control method for an active deployable terminal device based on a deep learning model, the active deployable terminal device comprising: a circular bearing base, used to carry other component devices and provided with corresponding sliding components; a guide rail system, comprising multiple sets of guide rail blocks and a number of precision balls, the guide rail blocks can move along the radial direction on the circular bearing base; a slider system, comprising multiple circles of sliders arranged along the radial direction around the center of the base, the sliders cooperate with the guide rail blocks, and a pressure-resistant arc-shaped rubber membrane is provided between the sliders; a slider active control system, comprising a micro servo motor, a force transmission rod, a sensor, and a control unit; characterized in that the implementation steps of the control method include: Step 1: Sensor placement and data collection; Step 2: Data preprocessing and dataset establishment; Step 3: Construction of CNN-LSTM hybrid model; Step 4: Training of CNN-LSTM hybrid model; Step 5: Verification and optimization of CNN-LSTM hybrid model; Step 6: Model deployment and real-time control: The verified and optimized CNN-LSTM hybrid model is deployed in the slider active control system of the active deployable terminal device to achieve real-time and high-precision control of the slider movement.

[0008] Advantages and beneficial effects of the present invention: The present invention introduces a CNN-LSTM hybrid model to accurately control the slider movement of the active deployment end device in real time, effectively solving the problem of inaccurate test results caused by the end friction effect in geotechnical engineering triaxial compression tests. The specific effects are as follows: (1) Slider control accuracy is significantly improved The present invention realizes real-time high-precision control of the slider movement through a deep learning model, significantly reducing the impact of friction effects on test data. After adopting the method of the present invention, the standard deviation of the slider displacement control is reduced from the original ±5 μm to ±2 μm, and the error of the speed control is reduced from ±0.2 mm / s to ±0.05 mm / s. This high-precision control ensures the synchronization of the slider and the specimen during the entire large deformation process, significantly reducing the impact of the end friction effect on the test results, and the uniformity of the stress and strain distribution of the specimen during the entire large deformation process is improved by about 15% to 20%.

[0009] (2) Enhanced system adaptability and robustness The CNN-LSTM hybrid model has a strong learning ability and can adapt to different types of geotechnical materials and variable loading conditions. Under different experimental conditions, the CNN-LSTM hybrid model ensures the excellent control effect of the slider active control system under various loading conditions through continuous learning and updating. Practical application shows that the system response time is shortened by about 30% and the control accuracy is improved by about 25%.

[0010] (3) Improvement of experimental efficiency Since the active control system of the slider can accurately control the slider in real time, the experimental error caused by the friction effect is reduced, thereby reducing the need for repeated experiments and improving experimental efficiency. Specifically, the single test time is shortened by about 20%, the test cycle is greatly shortened, and the experimenters can obtain high-precision experimental data more quickly and optimize the entire experimental process.

[0011] (4) Expansion of the scope of application The method of the present invention is not only applicable to traditional geotechnical material tests, but also can adapt to specimens of different types and shapes, as well as different loading conditions, which makes the method have a wide range of application prospects in various complex engineering tests and expands the application scope of geotechnical engineering tests.

[0012] (5) Security and stability are guaranteed The slider active control system has built-in abnormality detection and fault protection mechanisms to ensure the safety of the test process and the stability of the system operation. When the predicted value of the CNN-LSTM hybrid model exceeds the safety range or the system is abnormal, the system can promptly issue an alarm and stop the slider movement to prevent damage to the specimen or equipment failure. Long-term operation tests have shown that the system has no faults within 48 hours of continuous operation, meeting the needs of actual engineering applications.

[0013] (6) Promote the development of geotechnical engineering test technology This invention combines deep learning technology with geotechnical engineering tests and creates a test method based on intelligent control. Through precise slider motion control, it is possible to conduct a more in-depth study of the mechanical behavior of geotechnical materials under complex stress states, provide more reliable test data and scientific basis for the design, analysis and construction of geotechnical engineering, and promote the development of geotechnical engineering test technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is one of the structural schematic diagrams of the active deployable terminal device (structural decomposition schematic diagram).

[0015] Figure 2 This is the second structural diagram of the active deployable terminal device (when the terminal slider is not extended).

[0016] Figure 3 This is the third structural diagram of the active deployable terminal device (when the terminal slider is expanded).

[0017] Figure 4 This is the fourth structural schematic diagram of the active deployable terminal device (schematic diagram of the active control system of the slider).

[0018] Figure 5 It is a schematic diagram of the process of the present invention.

[0019] Figure 6 Schematic diagram of the arrangement of sensors in the method of the present invention.

[0020] In the figure: 1—circular bearing base plate, 2—permeable stone, 3—slider stop rod; Rectangular groove: 4-long rectangular groove, 5-medium rectangular groove, 6-short rectangular groove; Guide rail: 7-long guide rail, 8-middle guide rail, 9-short guide rail; Guide rail blocks: 10—first guide rail block, 11—second guide rail block, 12—third guide rail block, 13—fourth guide rail block, 14—fifth guide rail block, 15—sixth guide rail block, 16—seventh guide rail block; 17—arc groove, 18—rectangular hole, 19—mortise and tenon groove; Circle sliders: 21 - first slider, 22 - second slider, 23 - third slider, 24 - fourth slider, 25 - fifth slider, 26 - sixth slider, 27 - seventh slider; 28—tenon, 29—sector-shaped body, 30—post; Micro servo motors: 31 - a first micro servo motor, 32 - a second micro servo motor, 33 - a third micro servo motor, 34 - a fourth micro servo motor, 35 - a fifth micro servo motor, 36 - a sixth micro servo motor, 37 - a seventh micro servo motor; 38—carbon fiber force transmission rod, 39—hinge support, 40—limiting ring, 41—pressure-resistant arc-shaped rubber membrane, 42—drain hole, 43—high-precision micro sensor; 431—displacement sensor, 432—velocity sensor, 433—pressure sensor, 434—shear force sensor, 435—friction force sensor. DETAILED DESCRIPTION

[0021] CN118883231A proposes an active expansion control end device for geotechnical engineering triaxial compression test, which includes a circular bearing base, a guide rail system, a slider system, and a slider active control system, wherein the slider active control system includes a micro servo motor, a carbon fiber force transmission rod, a sensor, a control unit, etc. The device converts the sliding friction between the sample and the loading head into the rolling friction between the slider and the ball, and realizes active control of the slider in combination with servo control.

[0022] The present invention introduces a deep learning model into the active control system of the slider of the above-mentioned active deployment end device. On the basis of servo control, the deep learning model is combined to perform high-precision real-time control of the slider movement, which can better cope with the complex changes of the sample during the whole process of large deformation, further optimize the stress-strain state of the sample, and ensure the scientificity and reliability of the triaxial test results of geotechnical engineering. The present invention will provide a more reliable scientific basis for basic research and engineering design in the field of geotechnical engineering.

[0023] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.

[0024] The present invention proposes an optimization control method for an active deployable end device based on a deep learning model, which realizes real-time and high-precision control of the slider by combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) to ensure the stress-strain uniformity of the sample throughout the large deformation process.

[0025] like Figure 5 As shown, the specific implementation steps of the method of the present invention include: Step 1: Sensor placement and data collection On each slider of the active deployable end device (the specific structure of the device is shown in CN118883231A), a variety of high-precision sensors are arranged, including displacement sensors, velocity sensors, pressure sensors, shear force sensors and friction force sensors, to collect key multi-dimensional data during the test in real time.

[0026] Specifically, Figure 6 As shown, the arrangement of each sensor is: The displacement sensor 431 is installed on the curved side of the slider, facing the movement direction of the slider, and measures the displacement change of the slider in real time with an accuracy of ±0.1 μm.

[0027] The speed sensor 432 (photoelectric encoder) is installed on the corresponding guide rail block of the slider to obtain the real-time speed information of the slider. The encoder resolution is 1000 P / R, and combined with the counter, high-precision speed measurement can be achieved.

[0028] The pressure sensor 433 is embedded in the end surface where the slider contacts the sample, and measures the normal pressure applied by the slider to the sample, with a measuring range of 0-10 MPa and an accuracy of ±0.5% FS.

[0029] The shear force sensor 434 and the friction force sensor 435 are embedded in the end surface where the slider contacts the sample, and are used to measure the shear force and friction force between the slider and the sample, respectively. The sensor range is 0-5 kN, and the accuracy is ±0.2% FS.

[0030] Each slider is installed with a displacement sensor, a velocity sensor, a pressure sensor, a shear force sensor and a friction force sensor. As an embodiment, there are 126 sliders in total, and correspondingly, 126 displacement sensors, 126 velocity sensors, 126 pressure sensors, 126 shear force sensors and 126 friction force sensors are installed.

[0031] During the data acquisition process, a high-speed data acquisition card was used to synchronously collect data from the above sensors at a sampling rate of 2000 Hz, which included multi-dimensional information during the movement of the slider, including displacement, velocity, pressure, shear force, friction, etc., to ensure that comprehensive and accurate information on the test process was obtained to improve the generalization ability of the model.

[0032] Furthermore, the data acquisition card is equipped with an anti-interference design to ensure the stability and consistency of data in complex experimental environments.

[0033] Step 2: Data preprocessing and dataset creation Step 2.1 Preprocess the collected raw sensor data to improve data quality and the effect of subsequent model training. Data preprocessing includes the following steps: First, data cleaning: For the noise and outliers that may exist in the sensor data, signal processing methods such as median filtering and Kalman filtering are used to clean the data. The median filter effectively removes spike noise through the sliding window technology to maintain the smoothness of the data, while the Kalman filter is used to smooth the time series data, reduce the impact of random noise, and ensure the continuity and accuracy of the data.

[0034] Next, normalization processing: Use the Min-Max normalization method to scale the feature data of different dimensions and magnitudes to the range of [0,1], eliminate the scale differences between the data, and prevent a certain feature from having too much influence on the loss function during model training. Normalization processing not only speeds up the convergence of the model, but also improves the stability of training.

[0035] Then, feature extraction: using feature selection algorithms such as principal component analysis (PCA) and mutual information method, key features that have a significant impact on the movement of the slider are extracted. For example, PCA projects high-dimensional data into a low-dimensional space by reducing the dimension while retaining the main information of the data; the mutual information law selects features with high mutual information values, such as displacement change rate, velocity fluctuation, and shear force peak, by quantifying the correlation between each feature and the target variable. Feature extraction not only reduces the dimension of the data and the complexity of the model, but also improves the training efficiency and prediction accuracy of the model.

[0036] In addition, in order to improve the generalization ability of the model and prevent overfitting, data enhancement technology is used. By performing operations such as translating, scaling, and adding noise to the original data, more training samples are generated, the diversity of the data set is enriched, and the robustness of the model to input changes is enhanced.

[0037] Step 2.2: Create a data set based on the preprocessed data and divide the data set into training set, validation set, and test set in proportion. The specific proportions can be divided as follows: Training set: accounts for 70% of the total data and is used for model training.

[0038] Validation set: 15% of the total data, used for model hyperparameter tuning and early stopping strategy implementation.

[0039] Test set: 15% of the total data, used for the final evaluation of model performance.

[0040] Organize the data into a format suitable for model input. For example, structure the time series data into a three-dimensional array, where the dimensions include the number of samples, time steps, and number of features. Make sure the data format meets the input requirements of the CNN-LSTM hybrid model.

[0041] Step 3: Construction of CNN-LSTM hybrid model The CNN-LSTM hybrid model includes a CNN part and a LSTM part, wherein the CNN part is used to extract the spatial features of the slider motion, and the LSTM part is used to capture the advantages of time series features.

[0042] The CNN part includes multiple layers of one-dimensional convolutional layers, pooling layers and batch normalization layers, where: The first convolution layer uses 64 one-dimensional convolution kernels, each of which is 3×3 in size, to extract low-level features of the slider motion. This layer captures local features and performs feature mapping through convolution operations. The output shape is (L-3+1, 64), where L is the input length and the activation function uses ReLU (linear rectification function) to introduce nonlinearity.

[0043] The first max pooling layer follows the first convolutional layer, with a pooling window size of 2, which aims to reduce feature dimensions, reduce computational complexity, and prevent overfitting. After pooling, the output shape is ((L-3+1) / 2, 64).

[0044] The second convolution layer uses 128 one-dimensional convolution kernels, each of which is 3×3 in size, to further extract deep features. This layer enhances the model's ability to understand the spatial motion of the slider. The output shape is (((L-3+1) / 2-3+1), 128), and the activation function also uses ReLU.

[0045] The second maximum pooling layer follows the second convolutional layer and continues to reduce the feature dimension. The output shape becomes (((L-3+1) / 2-3+1) / 2, 128).

[0046] The batch normalization layer is added after each convolutional layer to stabilize the network training process, prevent internal covariate shift, and improve the training speed and stability of the model.

[0047] The features extracted by the CNN part are then input into the LSTM part.

[0048] The LSTM part includes two layers of LSTM units, where: The first layer of LSTM units contains 256 hidden units, which can effectively process the time information in the input sequence data. This layer is set to return sequences (return_sequences=1), so that the output of each time step is passed to the next layer, thereby retaining the complete time series information. This is very important for the subsequent LSTM layers because it can learn the short-term and long-term dependencies of the slider movement.

[0049] The second layer of LSTM units contains 128 hidden units, which further deepens the understanding of time series data. This layer does not return a sequence (return_sequences=0), but only outputs the state of the last time step to meet the input requirements of the fully connected layer.

[0050] There is a fully connected layer after the second layer of LSTM units, and the output of the fully connected layer is two neurons, which correspond to the optimal displacement and speed control instructions of the slider respectively. In the above two neurons, the activation function adopts a linear function to realize the direct output of continuous control parameters.

[0051] The two-layer LSTM structure enables the CNN-LSTM hybrid model to more effectively capture complex motion patterns and changing trends, thereby improving the accuracy of predictions.

[0052] The LSTM unit uses a forget gate, input gate, and output gate mechanism to dynamically adjust the flow of information at each time step to ensure that important information is retained and unimportant information is forgotten. This mechanism is particularly effective in capturing long-term dependencies, allowing the model to perform well in dealing with nonlinear and time-varying characteristics during slider movement.

[0053] Step 4: Training of CNN-LSTM hybrid model The CNN-LSTM hybrid model has sensor data as input and optimal displacement and speed control instructions for the slider as output. The trained CNN-LSTM hybrid model can effectively extract the spatial and temporal characteristics of the slider motion to achieve accurate prediction and control of the slider displacement and speed.

[0054] The data content of the training set is used to train the CNN-LSTM hybrid model, in which the adaptive moment estimation (AdamW) optimization algorithm is used, combined with the weight decay strategy to prevent overfitting and improve the generalization ability of the CNN-LSTM hybrid model. The loss function uses a weighted combination of mean square error (MSE) and mean absolute error (MAE), and adjusts the ratio of MSE and MAE to take into account both error sensitivity and robustness.

[0055] During the training process, the cosine annealing learning rate scheduling strategy is used to gradually reduce the learning rate during the training process to promote model convergence. At the same time, an early stopping strategy is set to stop training when the loss of the validation set does not decrease within 15 consecutive epochs to prevent the model from overfitting. In addition, K-fold cross-validation (for example, K=5) is used to evaluate the performance of the model under different data set partitions to ensure the stability and generalization ability of the model.

[0056] Step 5: CNN-LSTM hybrid model verification and optimization After the training of the CNN-LSTM hybrid model is completed, the CNN-LSTM hybrid model is further rigorously verified and optimized to ensure its stability and accuracy in practical applications.

[0057] First, the CNN-LSTM hybrid model is evaluated using the test set, and indicators such as MSE, MAE, and determination coefficient (R²) of the CNN-LSTM hybrid model on the test set are calculated.

[0058] Then, residual analysis is performed. If it is found that the prediction error of the CNN-LSTM hybrid model is normally distributed and the mean is close to zero, it means that the CNN-LSTM hybrid model has a high degree of credibility in predicting the slider movement. The experimental results show that the CNN-LSTM hybrid model exhibits high prediction accuracy and good generalization ability on the test set.

[0059] If the performance on the test set is not satisfactory, you can optimize it in the following ways: adjust the hyperparameters of the CNN-LSTM hybrid model, such as increasing or decreasing the number of LSTM units, adjusting the size and number of convolution kernels, changing the learning rate, etc.; increase the amount of training data to provide more samples for the CNN-LSTM hybrid model to learn; or introduce regularization methods, such as the Dropout layer, to further prevent overfitting. Through repeated iterative optimization, a CNN-LSTM hybrid model with excellent performance is finally obtained.

[0060] Step 6: Model deployment and real-time control The verified and optimized CNN-LSTM hybrid model will be deployed in the active control system of the slider of the active deployable terminal device to achieve real-time and high-precision control of the slider movement.

[0061] Step 6.1 Model deployment: Convert the trained CNN-LSTM model into a format suitable for real-time control (such as TensorFlow Lite) and embed it into the control unit of the slider active control system of the active deployable terminal device, including: Model conversion: Convert the trained CNN-LSTM hybrid model into a format suitable for embedded systems, and optimize it to reduce the model size and increase the inference speed. For example, use TensorFlow Lite to convert the trained CNN-LSTM model into a lightweight model suitable for real-time inference requirements of embedded devices.

[0062] Embedded platform selection: Choose a suitable embedded platform (such as Raspberry Pi, NVIDIA Jetson, or a dedicated industrial controller) as the hardware basis for model deployment. Install necessary software libraries and dependencies, such as TensorFlow Lite runtime, on the selected embedded platform to ensure that the converted model can run efficiently on the device.

[0063] Software integration: Write embedded software to implement real-time reasoning of the model and generation of control instructions.

[0064] Step 6.2 Real-time control, the specific implementation steps are as follows: First, the control unit of the slider active control system receives various sensor data in real time, and after data synchronization (ensuring that the collection and transmission of sensor data is consistent with the time step of model reasoning to avoid data delay or loss) and data preprocessing, the consistency of the data input into the CNN-LSTM hybrid model is ensured; Secondly, the preprocessed data is input into the CNN-LSTM hybrid model for real-time reasoning to generate the optimal displacement and speed control instructions of the slider; Then, the control unit sends the optimal displacement and speed control instructions output by the CNN-LSTM hybrid model to the servo motor. The servo motor adjusts its output torque to drive the slider to move accordingly, thereby adjusting the movement path and speed of the slider to ensure that the movement of the slider is highly synchronized with the deformation process of the sample. Finally, the information of the slider movement is collected in real time by the sensor and transmitted to the control unit of the slider active control system to form a closed-loop control. The CNN-LSTM hybrid model continuously adjusts the control instructions to ensure the timeliness and accuracy of the response.

[0065] Furthermore, in order to ensure the safety and stability of the active control system of the slider of the active deployable terminal device, an abnormality detection and fault protection mechanism is also set up. Specific measures include: Anomaly detection: Real-time monitoring of the CNN-LSTM hybrid model prediction values ​​and actual control instructions to detect whether there are any anomalies or situations that exceed the preset range.

[0066] Failure protection: When an abnormal situation is detected, the active control system of the slider of the active deployment end device automatically sounds an alarm and stops all slider movement to prevent specimen damage or equipment failure.

[0067] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. An optimization control method for an active deployable terminal device based on a deep learning model, the active deployable terminal device comprising: A circular bearing base is used to bear other components and is provided with corresponding sliding parts; The guide rail system includes multiple sets of guide rail blocks and a number of precision balls, and the guide rail blocks can move along the radial direction on the circular bearing base; the slider system includes multiple circles of sliders arranged along the radial direction around the center of the base, the sliders cooperate with the guide rail blocks, and a pressure-resistant arc rubber membrane is arranged between the sliders; the slider active control system includes a micro servo motor, a force transmission rod, a sensor, and a control unit; characterized in that the implementation steps of the control method include: Step 1: Sensor placement and data collection; Step 2: Data preprocessing and dataset establishment; Step 3: Construction of CNN-LSTM hybrid model; Step 4: Training of CNN-LSTM hybrid model; Step 5: Verification and optimization of CNN-LSTM hybrid model; Step 6: Model deployment and real-time control: The verified and optimized CNN-LSTM hybrid model is deployed in the slider active control system of the active deployable terminal device to achieve real-time and high-precision control of the slider movement.

2. The method for optimizing control of an active deployable terminal device based on a deep learning model as claimed in claim 1, characterized in that: Step 1: On each slider of the active deployment end device, a variety of sensors are arranged, including displacement sensors, velocity sensors, pressure sensors, shear force sensors and friction force sensors. The various types of sensors are arranged as follows: The displacement sensor is installed on the side of the slider, facing the movement direction of the slider, and measures the displacement change of the slider in real time; The speed sensor is installed on the guide rail block corresponding to the slider to obtain the real-time speed information of the slider; The pressure sensor is embedded in the end surface of the slider that contacts the sample to measure the normal pressure applied by the slider to the sample; The shear force sensor and the friction force sensor are embedded in the end surface where the slider contacts the sample, and are used to measure the shear force and friction force between the slider and the sample respectively; During the test, the required data is synchronously collected through a high-speed data acquisition card when the slider moves.

3. The method for optimizing control of an active deployable terminal device based on a deep learning model as claimed in claim 1, characterized in that: Step 2: Preprocess the collected raw sensor data, including data cleaning, normalization, and feature extraction: A dataset is established based on the preprocessed data, and the dataset is divided into a training set, a validation set, and a test set.

4. The method for optimizing and controlling an active deployable terminal device based on a deep learning model as claimed in claim 1, characterized in that: Step 3: The CNN-LSTM hybrid model includes a CNN part and an LSTM part, wherein the CNN part is used to extract the spatial features of the slider motion, and the LSTM part is used to capture the advantages of the time series features; The CNN part includes multiple layers of one-dimensional convolutional layers, pooling layers and batch normalization layers, where: The first convolution layer uses 64 one-dimensional convolution kernels, each of which is 3×3 in size, to extract low-level features of the slider motion. This layer captures local features and performs feature mapping through convolution operations. The output shape is (L-3+1, 64), where L is the input length. The activation function uses ReLU to introduce nonlinearity. The first maximum pooling layer follows the first convolutional layer, with a pooling window size of 2, which aims to reduce feature dimensions, reduce computational complexity, and prevent overfitting. After pooling, the output shape is ((L-3+1) / 2, 64). The second convolution layer uses 128 one-dimensional convolution kernels, each of which is 3×3 in size, to further extract deep features. This layer enhances the model's ability to understand the spatial motion of the slider. The output shape is (((L-3+1) / 2-3+1), 128), and the activation function uses ReLU. The second maximum pooling layer follows the second convolutional layer, further reducing the feature dimension, and the output shape becomes (((L-3+1) / 2-3+1) / 2, 128); Batch normalization layers are added after each convolutional layer to stabilize the network training process, prevent internal covariate shift, and improve the training speed and stability of the model; The features extracted by the CNN part are then input into the LSTM part; The LSTM part includes two layers of LSTM units, where: The first layer of LSTM units contains 256 hidden units, which can effectively process the time information in the input sequence data; this layer is set to return sequences (return_sequences=1), so that the output of each time step is passed to the next layer, thereby retaining the complete time series information; The second layer of LSTM units contains 128 hidden units, which further deepens the understanding of time series data; this layer does not return sequences (return_sequences=0), but only outputs the state of the last time step to meet the input requirements of the fully connected layer; There is a fully connected layer after the second layer of LSTM units, and the output of the fully connected layer is two neurons, which correspond to the optimal displacement and speed control instructions of the slider respectively; in the above two neurons, the activation function adopts a linear function to realize the direct output of continuous control parameters.

5. The method for optimizing control of an active deployable terminal device based on a deep learning model as claimed in claim 1, characterized in that: Step 4: The data content of the training set is used to train the CNN-LSTM hybrid model, in which the adaptive moment estimation (AdamW) optimization algorithm is used in combination with the weight decay strategy to prevent overfitting and improve the generalization ability of the CNN-LSTM hybrid model; the loss function adopts a weighted combination of the mean square error (MSE) and the mean absolute error (MAE); During the training process, the cosine annealing learning rate scheduling strategy is adopted to gradually reduce the learning rate during the training process to promote model convergence. At the same time, an early stopping strategy is set to stop training when the loss of the validation set does not decrease within 15 consecutive epochs to prevent model overfitting. In addition, K-fold cross-validation is used to evaluate the performance of the model under different data set divisions to ensure the stability and generalization ability of the model.

6. The method for optimizing and controlling an active deployable terminal device based on a deep learning model as claimed in claim 1, characterized in that: Step 5: First, the CNN-LSTM hybrid model is evaluated using the test set, and the MSE, MAE, and coefficient of determination (R²) indicators of the CNN-LSTM hybrid model on the test set are calculated; Then, residual analysis is performed. If it is found that the prediction error of the CNN-LSTM hybrid model is normally distributed and the mean is close to zero, it indicates that the CNN-LSTM hybrid model has a high degree of credibility in predicting the slider movement; On the contrary, optimization is performed through the following methods, including: adjusting the hyperparameters of the CNN-LSTM hybrid model, increasing the amount of training data, introducing regularization methods, and through repeated iterative optimization, finally obtaining a CNN-LSTM hybrid model with excellent performance.

7. The method for optimizing and controlling an active deployable terminal device based on a deep learning model as claimed in claim 1, characterized in that: Step 6: The trained CNN-LSTM model is converted into a format suitable for real-time control and embedded in the control unit of the slider active control system of the active deployment end device to receive sensor data in real time, generate and execute optimal displacement and speed control instructions, and achieve high synchronization between slider movement and sample deformation. The process of real-time control is as follows: First, the control unit of the slider active control system receives various sensor data in real time, and after data synchronization and data preprocessing, ensures the consistency of the data input into the CNN-LSTM hybrid model; Secondly, the preprocessed data is input into the CNN-LSTM hybrid model for real-time reasoning to generate the optimal displacement and speed control instructions of the slider; Then, the control unit sends the optimal displacement and speed control instructions output by the CNN-LSTM hybrid model to the servo motor. The servo motor adjusts its output torque to drive the slider to move accordingly, thereby adjusting the movement path and speed of the slider to ensure that the movement of the slider is highly synchronized with the deformation process of the sample. Finally, the information of the slider movement is collected in real time by the sensor and transmitted to the control unit of the slider active control system to form a closed-loop control. The CNN-LSTM hybrid model continuously adjusts the control instructions to ensure the timeliness and accuracy of the response.

Images