Cutting system and method for pre-judging thickness of semi-conductive layer of cable by fusing mechanics and vision

Abstract

The invention relates to a cutting system and method for pre-judging the thickness of a cable semi-conductive layer through mechanics and vision fusion. The cutting system comprises a cutting module, an active profiling module, a vision module and a closed-loop control module. The cutting module is used for receiving a control signal of the upper computer to cut the surface of the cable; the active profiling module is used for achieving the active profiling function through cooperation of a profiling block and a pressure motor in the cutting module. The visual module is used for positioning black residues on the surface of the white insulating layer of the cable and verifying the cutting effect after processing; and the closed-loop control module is used for processing and calculating data and controlling the feed amount of the cutting module. The cutting depth can be accurately controlled.

Description

Technical Field

[0001] The present invention belongs to the technical field of cable cutting, and relates to a system and method for pre-determining the thickness of a cable semi-conductive layer, in particular to a system and method for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision. Background Art

[0002] Cables are generally composed of conductors, outer sheaths, aluminum sheaths, cable insulation shielding layers, cable insulation layers and copper cores. Cables are produced in fixed lengths during production. Therefore, during the actual installation of the cables, each section of the cable needs to be connected through an intermediate joint so that the cable length meets the requirements for long-distance transmission.

[0003] Before installing, connecting and using power cables, the joints of power cables need to be preprocessed, mainly including removing the cable insulation layer and the semi-conductive layer, laying a good foundation for the installation of cable accessories, ensuring smooth conduction between cables and ensuring the normal use of prefabricated power cable products.

[0004] Traditional cutting methods have certain limitations in dealing with the complex shapes, sizes, and different material properties of cable connectors. In order to ensure the quality and performance of cable connectors, high precision and high stability requirements are put forward for the processing process.

[0005] The main disadvantages of traditional cutting methods are: it is easy to cause mechanical damage to the cable insulation layer, weaken the insulation performance of the cable, and increase the risk of electrical failure; it is difficult to ensure that the thickness of the semi-conductive layer is removed at a reasonable level, and there may be residual semi-conductive layer or the removal depth is too deep. This will lead to uneven electric field distribution and shorten the service life of the cable; the traditional cutting process is complicated to operate, requiring high technology and deep experience, which increases the workload and training costs of operators, and the manual operation time is long and the efficiency is low; the traditional cutting process is difficult to accurately control the cutting depth, which is easy to cause over-cutting or under-cutting. Over-cutting will damage the insulation layer, and under-cutting may not completely remove the semi-conductive layer, affecting the performance of the cable joint.

[0006] After searching, no prior art documents identical or similar to the present invention were found. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and to propose a cutting system and method for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision, which can accurately control the cutting depth.

[0008] The present invention solves the practical problem by adopting the following technical solutions:

[0009] A cutting system for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision, comprising: a cutting module, an active profiling module, a vision module and a closed-loop control module;

[0010] The cutting module is used to receive the control signal of the host computer to cut the cable surface;

[0011] The active profiling module is used to realize the active profiling function through the cooperation of the profiling block and the pressure motor in the cutting module;

[0012] The visual module is used to locate the black residue on the surface of the white insulation layer of the cable and verify the cutting effect after processing;

[0013] The closed-loop control module is used to process and calculate data and control the feed amount of the cutting module.

[0014] Moreover, the cutting module includes a pressure motor, a tool holder, a slide rail and a cutting tool; the input end of the pressure motor is connected to the host computer for receiving a control signal from the host computer; the output end of the pressure motor is connected to the tool through the tool holder and the slide rail, and the cable surface is cut using the tool according to the control signal received from the host computer;

[0015] Moreover, the active profiling module includes a profiling block; the profiling block is fixed to the cutting module and can slide and retract on the cable surface; the upper computer controls the cutting module to apply a constant pressure to the cable joint through the profiling block; the hardness difference of different layers of the cable will cause the profiling block to produce different expansion and contraction amounts; the pressure motor transmits the expansion and contraction amount signal to the upper computer; the upper computer pre-determines the thickness change of the semi-conductive layer through the neural network algorithm, and transmits the feed amount control signal back to the cutting module, thereby realizing a dynamic and real-time detection of the value of the semi-conductive thickness of the cable;

[0016] Moreover, the visual module includes a color sensor; the color sensor is used to perceive the condition of the cable surface after processing in real time, detect whether a black semi-conductive layer remains after cutting, transmit the visual signal to the host computer, fuse it with the expansion and contraction data transmitted by the pressure motor, and establish a corresponding relationship with the feed amount;

[0017] Moreover, the closed-loop control module includes a host computer; the host computer combines and analyzes the signals of the active profiling module and the visual module, combines feature fusion with the neural network prediction algorithm, and sends a control signal to the pressure motor in the cutting module to control the timing of feed and retract and the feed and retract amount.

[0018] A cutting method for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision comprises the following steps:

[0019] Step 1, host computer initialization;

[0020] Step 2, determine the expansion and contraction amount benchmark of the pressure motor;

[0021] Step 3, cutting the black semi-conductive layer on the surface of the cable.

[0022] Moreover, the specific method of step 1 is:

[0023] First, the parameters need to be initialized in the host computer, and the constant force applied to the cable surface by the profiling module. At the same time, the trained LSTM and ResNet neural network models are loaded in the host computer to prepare to receive data and perform predictive control.

[0024] Moreover, the data processing process of the LSTM and ResNet neural networks in step 1 on the profiling module and the visual module is as follows:

[0025] (1) The mechanical motor inputs the expansion and contraction data under constant force. The data is a time-varying sequence. The time series data is input into the LSTM network, and the network calculation result is:

[0026] F = lstm(x)#(1)

[0027] Where F represents the result of the LSTM network, lstm(·) is the entire long short-term memory recurrent neural network;

[0028] For the time series prediction of the expansion amount, the mean square error (MSE) is used as the loss function:

[0029]

[0030] Where N represents the number of sampling points of the expansion and contraction amount, y i represents the true value, Represents the predicted value of the network output; by training the LSTM network to minimize the MSE, fix the network structure, and complete the initialization of the LSTM network;

[0031] (2) The visual sensor inputs the visual image data, and firstly grayscales each pixel of the image using the following formula:

[0032] Gray=0.2989×R+0.5870×G+0.1140×B#(3)

[0033] The pixel value is calculated in the range of [0,255]. Gray represents the grayscale pixel value after conversion, and R, G, and B represent the pixel values ​​of the three channels of the image respectively.

[0034] The data is passed into the ResNet network to locate the black semi-conductive layer:

[0035] The ResNet network contains multiple residual connections. After the image data X is passed in, the linear superposition of the residual connections is first performed:

[0036] V 1 =f(X)+X#(4)

[0037] Among them, f(X) usually contains two convolutional layers, batch normalization and ReLU activation function;

[0038] Then global average pooling is performed.

[0039] G=GlobalAveragePooling(A)#(5)

[0040] Among them, A is the result after the data passes through multiple residual blocks;

[0041] The final output of ResNet is:

[0042] V=W fc G+b fc #(6)

[0043] Among them, W fc is the weight matrix of the fully connected layer, b fc for deviation;

[0044] For ResNet, the cross entropy loss function is used for evaluation:

[0045]

[0046] Where C is the number of categories, y c is the true label, is the predicted probability;

[0047] By training the ResNet network to minimize CE, fix the network structure, and complete the initialization of the ResNet network;

[0048] (3) Data fusion of the expansion and contraction time series data and the image data is performed, and the mapping relationship between the feed amount D and the above data is obtained by using the weighted average fusion method:

[0049] D=αF+(1-α)V#(8)

[0050] Among them, α is a weight coefficient between 0 and 1, F represents the prediction result of the stretch feedback, and V is the prediction result of the visual data.

[0051] Moreover, the specific method of step 2 is:

[0052] Fix the cable to be cut and manually adjust the distance between the profiling module and the cable surface. The profiling module needs to contact the white insulation layer first, and the expansion and contraction amount of the pressure motor at this time is used as the basis for the subsequent expansion and contraction amount.

[0053] Moreover, the specific method of step 3 is:

[0054] The active profiling module cooperates with the vision module to transmit the multimodal data of the cable back to the host computer. The host computer predicts and controls the feed amount and accurately processes the black semi-conductive layer on the cable surface.

[0055] Advantages and beneficial effects of the present invention:

[0056] 1. The present invention proposes a system and method for cutting the thickness of a cable semi-conductive layer by integrating mechanics and vision. When cutting the thickness of the black semiconductor layer on the surface of the cable, the thickness of the semi-conductive layer is predicted in real time by the expansion and contraction amount of the pressure motor (screw motor with pressure detection function) while ensuring consistent pressure, providing data for the feed amount of removing the semi-conductive layer, and then verifying whether the thickness of the semi-conductive removal is reasonable by the visual module. During the exploitation process, the pressure motor of the cutting module generates a constant output torque. Due to the difference in hardness between the white insulating layer and the black semi-conductive layer of the cable connector, the expansion and contraction amount of the motor under constant pressure will change, and the data will be transmitted to the upper computer neural network to predict the thickness of the semi-conductive layer. At the same time, the color sensor senses the surface of the processed cable in real time to assist in the change of the feed amount.

[0057] 2. The present invention uses active profiling technology to provide a new solution to the problem of difficult to accurately control cutting depth in traditional cutting processes. The present invention is based on real-time feedback adaptive control technology, which can dynamically adjust cutting parameters (such as feed amount, retract amount, etc.) according to the specific conditions of the cable connector surface, thereby improving the efficiency, accuracy and reliability of the cable cutting process. In addition, combined with the neural network control algorithm in the host computer and forming a closed-loop control with other modules, it can overcome some unknown interferences in the processing process and further improve the stability of the processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a closed-loop control block diagram of the present invention;

[0059] Figure 2 It is a schematic diagram of the process of active profiling in the present invention;

[0060] Figure 3 It is a flow chart of the control algorithm in the present invention. DETAILED DESCRIPTION

[0061] The following further describes the embodiments of the present invention:

[0062] A cutting system for predicting the thickness of a cable semi-conductive layer by integrating mechanics and vision, such as Figure 1 As shown, it includes: a cutting module, an active profiling module, a vision module and a closed-loop control module;

[0063] A cutting module, including a pressure motor, a tool holder, a slide rail and a cutting tool;

[0064] The input end of the pressure motor is connected to the host computer to receive the control signal of the host computer; the output end of the pressure motor is connected to the tool through the tool holder and the slide rail, and the tool is used to cut the cable surface according to the control signal received from the host computer;

[0065] An active profiling module, including a profiling block;

[0066] like Figure 2 As shown, the profiling block is fixed to the cutting module and can slide and retract on the cable surface. The upper computer controls the cutting module to apply constant pressure to the cable joint through the profiling block. The hardness difference of different layers of the cable will cause the profiling block to produce different expansion and contraction amounts. The pressure motor transmits the expansion and contraction amount signal to the upper computer. The upper computer pre-determines the thickness change of the semi-conductive layer through the neural network algorithm and sends the feed amount control signal back to the cutting module, thereby realizing a dynamic and real-time detection of the semi-conductive thickness value of the cable.

[0067] Vision module, including color sensor;

[0068] like Figure 2 As shown, the color sensor is used to perceive the condition of the cable surface after processing in real time, detect whether the black semi-conductive layer is still left after cutting, transmit the visual signal to the host computer, merge it with the expansion and contraction data transmitted by the pressure motor, and establish the corresponding relationship with the feed amount;

[0069] Closed-loop control module, including a host computer;

[0070] like Figure 3 As shown, the host computer combines and analyzes the signals of the active profiling module and the visual module, combines feature fusion with the neural network prediction algorithm, and sends a control signal to the pressure motor in the cutting module to control the timing of feed and retract and the feed and retract amount.

[0071] The functions of each module are further described below:

[0072] The invention provides a cutting system and method for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision, comprising a cutting module, an active profiling module, a vision module and a closed-loop control module.

[0073] The main function of the cutting module is to control the feed amount and cut the semi-conductive layer of the cable. The cutting module mainly consists of a pressure motor, a tool holder, a slide rail and a cutting tool. The tool holder plays a fixing role. The pressure motor receives the control signal from the host computer to control the up and down movement of the tool, thereby performing high-precision cutting on the cable, and connecting the profiling block in the profiling module to apply constant force to the cable surface and detect the change in the expansion and contraction amount.

[0074] like Figure 2 As shown in the figure, the main function of the active force-controlled profiling module is to realize an active profiling function similar to "touch" through the cooperation of the profiling block and the pressure motor in the cutting module. The module mainly includes a profiling block. The pressure motor in the cutting module is connected to the profiling block, contacts the cable surface through the profiling block, and applies a constant force to the cable surface. Because of the difference in hardness between the black semiconductor layer and the white insulating layer on the cable surface, when the cable is cut abnormally by applying a constant force through the profiling block, the expansion and contraction amount of the pressure motor will change due to the appearance of the black semiconductive layer and the change in its thickness. Taking the expansion and contraction amount of the pure white insulating layer as the benchmark, under the same pressure value, the expansion and contraction amount change signal of the pressure motor is transmitted to the host computer for display, and the feed amount in the cutting module is controlled by the prediction algorithm in the host computer. Based on this, the thickness of the black semiconductive layer on the surface of the cable is accurately and real-time judged to provide data for controlling the feed amount.

[0075] like Figure 2 As shown in the figure, the visual module mainly includes a color sensor, whose main function is to accurately locate the black residue on the surface of the white insulation layer of the cable and verify the rationality of the cutting effect after processing. By accurately locating the position of the black residue, the timing of feeding and retracting can be accurately controlled. The data transmitted by the visual module and the above-mentioned active profiling module are integrated and analyzed by the neural network algorithm in the host computer, which assists the host computer to control the feed amount in the cutting module and accurately process the cable surface.

[0076] like Figure 3 As shown in the figure, the closed-loop control module mainly includes a host computer, which processes and calculates data. In the host computer, there are two input parts. One part of the input is the feedback signal of the expansion and contraction amount sent back by the active profiling module, which is time series data. The data is passed to the LSTM network for prediction analysis; the other part of the input is the visual data of the cable surface, which is grayed and passed to the ResNet network for recognition analysis. Then, the analysis results of the two are feature fused, and a mapping relationship between the feed amount is established and sent to the pressure motor in the cutting module. The pressure motor controls the feed, thereby achieving high-precision and high-stability cutting processing of the cable surface.

[0077] A cutting method for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision comprises the following steps:

[0078] Step 1, host computer initialization;

[0079] The specific method of step 1 is:

[0080] First, the parameters need to be initialized in the host computer, and the constant force applied to the cable surface by the profiling module. At the same time, the trained LSTM and ResNet neural network models are loaded in the host computer to prepare to receive data and perform predictive control.

[0081] The data processing process of the LSTM and ResNet neural networks in step 1 on the profiling module and the visual module is as follows:

[0082] (1) The mechanical motor inputs the expansion and contraction data under constant force. The data is a time-varying sequence. The time series data is input into the LSTM network, and the network calculation result is:

[0083] F = lstm(x)#(1)

[0084] Among them, F represents the result of the LSTM network, and lstm(·) is the entire long short-term memory recurrent neural network.

[0085] For the time series prediction of the expansion amount, the mean square error (MSE) is used as the loss function:

[0086]

[0087] Where N represents the number of sampling points of the expansion and contraction amount, y i represents the true value, Represents the predicted value of the network output. By training the LSTM network to minimize the MSE, fix the network structure, and complete the initialization of the LSTM network.

[0088] (2) The visual sensor inputs the visual image data, and firstly grayscales each pixel of the image using the following formula:

[0089] Gray=0.2989×R+0.5870×G+0.1140×B#(3)

[0090] What is calculated is the pixel value, the value range is [0,255], Gray represents the grayscale pixel value after conversion, R, G, B represent the pixel values ​​of the three channels of the image respectively.

[0091] The data is passed into the ResNet network to locate the black semi-conductive layer:

[0092] The ResNet network contains multiple residual connections. After the image data X is passed in, the linear superposition of the residual connections is first performed:

[0093] V 1=f(X)+X#(4)

[0094] Among them, f(X) usually contains two convolutional layers, batch normalization and ReLU activation function.

[0095] Then global average pooling is performed.

[0096] G=GlobalAveragePooling(A)#(5)

[0097] Among them, A is the result after the data passes through multiple residual blocks.

[0098] The final output of ResNet is:

[0099] V=W fc G+b fc #(6)

[0100] Among them, W fc is the weight matrix of the fully connected layer, b fc For deviation.

[0101] For ResNet, the cross entropy loss function is used for evaluation:

[0102]

[0103] Where C is the number of categories, y c is the true label, is the predicted probability.

[0104] By training the ResNet network to minimize CE and fix the network structure, the ResNet network is initialized.

[0105] (3) Data fusion of the expansion and contraction time series data and the image data is performed, and the mapping relationship between the feed amount D and the above data is obtained by using the weighted average fusion method:

[0106] D=αF+(1-α)V#(8)

[0107] Among them, α is a weight coefficient between 0 and 1, F represents the prediction result of the stretch feedback, and V is the prediction result of the visual data.

[0108] Step 2, determine the expansion and contraction amount benchmark of the pressure motor;

[0109] The specific method of step 2 is:

[0110] Fix the cable to be cut and manually adjust the distance between the profiling module and the cable surface. The profiling module needs to contact the white insulation layer first, and the expansion and contraction amount of the pressure motor at this time is used as the basis for the subsequent expansion and contraction amount.

[0111] Step 3, cutting the black semi-conductive layer on the surface of the cable.

[0112] The specific method of step 3 is:

[0113] The active profiling module cooperates with the vision module to transmit the multimodal data of the cable back to the host computer. The host computer predicts and controls the feed amount and accurately processes the black semi-conductive layer on the cable surface.

[0114] In the specific implementation, first, the cutting module is responsible for controlling the forward and backward movement of the cutting tool to cut the cable surface. The specific operation includes: the host computer sends a control signal to the pressure motor, the pressure motor controls the movement of the tool in the vertical direction, the tool holder fixes the pressure motor and the tool, and the slide rail is used for the up and down movement of the tool, thereby realizing the cutting process.

[0115] Secondly, the active profiling module realizes the active profiling function through the cooperation of the profiling block and the pressure motor. The operation process includes: the pressure motor drives the profiling block to apply constant pressure to the cable surface. During the cutting process, the feedback signal of the expansion and contraction amount is transmitted to the host computer to accurately and in real time judge the thickness of the black semi-conductive layer on the surface of the cable, and predict the thickness of the black semi-conductive layer through the neural network to adjust the cutting parameters in time.

[0116] Next, the color sensor is used to verify the rationality of the cutting effect and locate the residual black semi-conductive layer on the cable surface. The collected data is uploaded to the host computer to assist the host computer in controlling the cutting timing. The operation process is as follows: the color sensor scans the cable surface and accurately locates the position and size of the black semi-conductive layer.

[0117] Finally, the neural network predictive control algorithm in the closed-loop control module processes the data sent back and outputs precise control signals to adjust the cutting parameters and achieve high-precision cutting of the cable surface. The input of the neural network includes the feedback signal of the pressure motor expansion and contraction and the positioning data of the vision module. The output is the adjusted cutting parameters, which mainly include the timing of advance and retreat and the amount of advance or retreat. The pressure motor receives the control signal and drives the cutting tool to move. In practical applications, after the neural network is trained, real-time data is input into the network to generate corresponding control signals to achieve precise processing.

[0118] The working principle of the present invention is:

[0119] When cutting the cable, due to processing and process errors in the cable processing process, the cylindrical cross-section of the cable is non-circular, and the thickness of the semi-conductive layer on the surface of the same cable has certain differences. Therefore, in the process of realizing the processing and production of cable joints, the accurate stripping of the black semi-conductive layer is a difficulty in automated cable stripping. There are often problems of incomplete stripping or excessive stripping, and it is difficult to realize automated removal and achieve processing consistency. Therefore, accurate and real-time judgment of the thickness of the black semi-conductive layer on the surface of the cable is crucial to controlling the feed amount.

[0120] The present invention utilizes an active profiling module to provide feedback on the amount of expansion and contraction, cooperates with a visual sensor, and adopts a neural network prediction algorithm to predict the thickness of the black semiconductive layer, thereby controlling the amount of feed and achieving adaptive cutting.

[0121] Specifically, due to the different hardness of the white insulating layer and the black semi-conductive layer of the cable joint, when the pressure motor is used to drive the cutting tool module to cut the cable, the pressure of the force control profiling module remains unchanged. When black residue appears or the thickness suddenly changes, due to the difference in hardness between the black semi-conductive layer and the white insulating layer, the same pressure will cause a change in the value of the pressure motor expansion and contraction. Through this change, the change in the thickness of the residue can be inferred. At the same time, the position of the black residue can be accurately located using a color sensor. The time series data and visual data are respectively passed to the trained LSTM network and ResNet network prediction model to obtain the relationship between the change in expansion and contraction during the cutting process and the change in the thickness of the black semi-conductive layer. The output results of the two deep learning networks are feature fused, and the control signal is output to the pressure motor in the cutting module to control the timing of the advance and retreat of the tool and the advance and retreat of the tool during the cutting process, thereby realizing the precise processing of the cable joint. After processing, the data obtained by the visual module can also be used to verify the rationality of the cutting thickness and complete the closed-loop control. Compared with the traditional manual cutting method, this technology can obtain more comprehensive cable surface data faster, improve the processing positioning efficiency, and achieve more efficient and more accurate cable surface semi-conductive layer cutting tasks. For cables of different shapes and materials, high-precision processing can be achieved by adjusting parameters. In addition, the neural network control algorithm in the host computer forms a closed-loop control with other modules, which can overcome some unknown interferences in the processing process and further improve the stability of processing. Active force control profiling technology can adaptively adjust the cutting strategy and process route according to the roughness and contour characteristics of the cable surface to achieve efficient processing of irregular surfaces. It can automatically identify and adjust the processing requirements of irregular cable surfaces, reduce dependence on operators and human intervention, improve production efficiency and reduce processing errors.

[0122] It should be emphasized that the embodiments of the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation modes. Any other implementation modes derived by those skilled in the art based on the technical solutions of the present invention also fall within the scope of protection of the present invention.

Claims

1. A mechanical and visual fusion pre-determination cable semi-conductive layer thickness cutting system, characterized by: include: Cutting module, active profiling module, vision module and closed-loop control module; The cutting module is used to receive the control signal of the host computer to cut the cable surface; The active profiling module is used to realize the active profiling function through the cooperation of the profiling block and the pressure motor in the cutting module; The visual module is used to locate the black residue on the surface of the white insulation layer of the cable and verify the cutting effect after processing; The closed-loop control module is used to process and calculate data and control the feed amount of the cutting module.

2. According to claim 1, a mechanical and visual fusion pre-determination cable semi-conductive layer thickness cutting system is characterized by: The cutting module includes a pressure motor, a tool holder, a slide rail and a cutting tool; the input end of the pressure motor is connected to a host computer for receiving a control signal from the host computer; the output end of the pressure motor is connected to the tool through the tool holder and the slide rail, and the cable surface is cut using the tool according to the control signal received from the host computer.

3. According to claim 1, a mechanical and visual fusion pre-determination cable semi-conductive layer thickness cutting system is characterized by: The active profiling module includes a profiling block; the profiling block is fixed to the cutting module and can extend and slide on the cable surface. The upper computer controls the cutting module to apply constant pressure to the cable joint through the profiling block. The hardness difference of different layers of the cable will cause the profiling block to produce different extension amounts. The pressure motor transmits the extension amount signal to the upper computer. The upper computer pre-determines the thickness change of the semi-conductive layer through a neural network algorithm, and sends the feed amount control signal back to the cutting module, thereby realizing a dynamic and real-time detection of the cable semi-conductive thickness value.

4. According to claim 1, a mechanical and visual fusion pre-determination cable semi-conductive layer thickness cutting system is characterized by: The visual module includes a color sensor; the color sensor is used to perceive the condition of the cable surface after processing in real time, detect whether a black semi-conductive layer still remains after cutting, transmit the visual signal to the host computer, merge it with the expansion and contraction data transmitted by the pressure motor, and establish a corresponding relationship with the feed amount.

5. According to claim 1, a mechanical and visual fusion pre-determination cable semi-conductive layer thickness cutting system is characterized by: The closed-loop control module includes a host computer; the host computer combines and analyzes the signals of the active profiling module and the visual module, combines feature fusion with a neural network prediction algorithm, and sends a control signal to the pressure motor in the cutting module to control the timing of feed and retract and the feed and retract amount.

6. A cutting method for pre-determining the thickness of a cable semi-conductive layer by integrating mechanics and vision, characterized in that: The following steps are involved: Step 1: Initialize the host computer; Step 2, determine the expansion and contraction amount benchmark of the pressure motor; Step 3: Cut the black semi-conductive layer on the cable surface.

7. The method for cutting the thickness of a cable semi-conductive layer by integrating mechanics and vision according to claim 6 is characterized in that: The specific method of step 1 is: First, the parameters need to be initialized in the host computer, and the constant force applied to the cable surface by the profiling module. At the same time, the trained LSTM and ResNet neural network models are loaded in the host computer to prepare to receive data and perform predictive control.

8. The method for cutting the thickness of a cable semi-conductive layer by integrating mechanics and vision according to claim 7 is characterized in that: The data processing process of the LSTM and ResNet neural networks in step 1 on the profiling module and the visual module is as follows: (1) The mechanical motor inputs the expansion and contraction data under constant force. The data is a time-varying sequence. The time series data is input into the LSTM network, and the network calculation result is: F = lstm(x)#(1) Where F represents the result of the LSTM network, lstm(·) is the entire long short-term memory recurrent neural network; For the time series prediction of the expansion amount, the mean square error (MSE) is used as the loss function: Where N represents the number of sampling points of the expansion and contraction amount, y i represents the true value, Represents the predicted value of the network output; by training the LSTM network to minimize the MSE, fix the network structure, and complete the initialization of the LSTM network; (2) The visual sensor inputs the visual image data, and firstly grayscales each pixel of the image using the following formula: Gray=0.2989×R+0.5870×G+0.1140×B#(3) The pixel value is calculated in the range of [0,255]. Gray represents the grayscale pixel value after conversion, and R, G, and B represent the pixel values ​​of the three channels of the image respectively. The data is passed into the ResNet network to locate the black semi-conductive layer: The ResNet network contains multiple residual connections. After the image data X is passed in, the linear superposition of the residual connections is first performed: V1=f(X)+X#(4) Among them, f(X) usually contains two convolutional layers, batch normalization and ReLU activation function; Then global average pooling is performed. G=GlobalAveragePooling(A)#(5) Among them, A is the result after the data passes through multiple residual blocks; The final output of ResNet is: V=W fc G+b fc #(6) Among them, W fc is the weight matrix of the fully connected layer, b fc for deviation; For ResNet, the cross entropy loss function is used for evaluation: Where C is the number of categories, y c is the true label, is the predicted probability; By training the ResNet network to minimize CE, fix the network structure, and complete the initialization of the ResNet network; (3) Data fusion of the expansion and contraction time series data and the image data is performed, and the mapping relationship between the feed amount D and the above data is obtained by using the weighted average fusion method: D=αF+(1-α)V#(8) Among them, α is a weight coefficient between 0 and 1, F represents the prediction result of the stretch feedback, and V is the prediction result of the visual data.

9. The method for cutting the thickness of a cable semi-conductive layer by integrating mechanics and vision according to claim 6 is characterized by: The specific method of step 2 is: Fix the cable to be cut and manually adjust the distance between the profiling module and the cable surface. The profiling module needs to contact the white insulation layer first, and the expansion and contraction amount of the pressure motor at this time is used as the basis for the subsequent expansion and contraction amount.

10. The method for cutting the thickness of a cable semi-conductive layer by integrating mechanics and vision according to claim 6, characterized in that: The specific method of step 3 is: The active profiling module cooperates with the vision module to transmit the multimodal data of the cable back to the host computer. The host computer predicts and controls the feed amount and accurately processes the black semi-conductive layer on the cable surface.

Images