Anti-sinking extrusion molding control method for medium-voltage cross-linked cable

Abstract

The invention belongs to the technical field of intelligent control, and particularly relates to an anti-sinking extrusion molding control method for a medium-voltage cross-linked cable. Aiming at the problems that the existing extrusion control depends on manual voltage regulation, the precision is low, the hysteresis quality is high and the like, the method comprises the following steps: acquiring historical extrusion pressure and cable image data, and carrying out adaptive gray scale calibration, edge reservation denoising, multi-modal feature extraction, regional enhancement and multi-scale fusion optimization preprocessing on an image; matching the enhanced image with the pressure data according to an axial position to construct a data set; an optimal three-layer extrusion molding pressure parameter combination is mined by adopting a genetic algorithm, precise pressure adjustment is realized by improving fuzzy PID control, and meanwhile, historical data is dynamically replaced by real-time data for continuous optimization. The method forms intelligent closed-loop control, effectively reduces recess defects, improves cable quality stability and product qualification rate, and improves production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology, and in particular relates to a method for controlling the extrusion of medium-voltage cross-linked cables to prevent denting. Background Technology

[0002] Medium-voltage cross-linked cables, as core components of power transmission systems, are widely used in urban power distribution networks and industrial power supply. Indentations are a common quality defect in extrusion production, leading to uneven cable surfaces and inconsistent insulation thickness. This not only weakens the uniformity of the electric field distribution but also easily causes breakdown faults due to stress concentration during operation, seriously threatening power transmission safety. Therefore, precise pressure control during the extrusion process is crucial to ensuring cable quality. Existing medium-voltage cross-linked cable extrusion control technology has significant limitations, relying heavily on manual experience to adjust the pressure parameters of the three-layer extrusion unit. Operators must visually observe the cable surface condition and adjust the pressure value based on subjective judgment. This method has many drawbacks: it is difficult to identify subtle indentations with the naked eye, and judgment is easily affected by factors such as light and fatigue, resulting in extremely low accuracy; pressure adjustment has a strong lag, unable to respond in real time to parameter fluctuations during production; differences in experience among different operators lead to a lack of standardized pressure parameter combinations, easily causing interlayer pressure mismatch, which exacerbates indentation defects, resulting in large fluctuations in product qualification rate and low production efficiency. Summary of the Invention

[0003] In view of the technical problems existing in the background art, the present invention proposes a method for controlling the extrusion of medium-voltage cross-linked cables to prevent denting.

[0004] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0005] S1. Collect historical extrusion pressure data of medium-voltage cross-linked cables and obtain cross-linked cable image data;

[0006] S2. Perform enhancement preprocessing on the cross-linked cable image data to obtain the enhanced image. Match the enhanced image with the historical medium-voltage cross-linked cable extrusion pressure data to obtain the corresponding matched cross-linked cable pressure-image feature data.

[0007] S3. Perform multi-dimensional correlation analysis on the matched cross-linked cable pressure-image data to obtain the optimal combination of pressure parameters for the three-layer extrusion device;

[0008] S4. Based on the optimal combination of pressure parameters of the three-layer extrusion device, the pressure is adjusted to the set optimal value by improving the fuzzy PID pressure control.

[0009] S5. Collect real-time optimized control pressure-image data, replace historical data in chronological order based on preset data volume, and re-optimize the pressure of the extrusion device after replacement for adjustment and control.

[0010] Preferably, the specific implementation of enhancing the cross-linked cable image data in step S2 to obtain the enhanced image is as follows:

[0011] S211. Adaptive grayscale calibration and edge-preserving denoising are performed on the original cross-linked cable image. Local grayscale mean normalization is used for adaptive grayscale calibration to eliminate grayscale shifts caused by uneven illumination. ,in, This is an image of the original cross-linked cable. This represents the grayscale value of the original image at pixel (x, y). These represent the mean and standard deviation of gray levels within a sliding window centered at (x, y), respectively. To avoid extremely small values ​​where the denominator is 0, we obtain the pixel values ​​of the image after grayscale calibration. ;

[0012] S212. Adaptive edge-preserving denoising based on improved bilateral filtering preserves the edges of concave defects while suppressing extrusion noise: Where W is the normalized weight used to balance the weights of all pixels within the window, avoiding image brightness shift after weighted averaging. These represent the spatial domain weights and the grayscale domain weights, respectively. m and n are the horizontal and vertical coordinates of any pixel within the sliding window, used to traverse each pixel within the window and calculate its corresponding weight. The pixel values ​​of the image after noise reduction and calibration;

[0013] S213. Extract multimodal feature maps from the denoised calibration image, and suppress feature redundancy through spatial and channel reconstruction convolution to obtain purified multimodal features.

[0014] S214. Combining the obtained purified multimodal feature map, perform region-specific differential enhancement on the denoised calibration image to obtain the region-enhanced image. ;

[0015] S215. Perform multi-scale fusion and global optimization on the region-enhanced image to generate multi-scale feature maps. Downsampling with scale factors s=1, 2, and 4 respectively yields multi-scale images: ,in, To perform floor function calculations, the sharpness weights of images at each scale are calculated. The calculation method is as follows: ,in, These represent the width and height of the image at scale s, respectively.

[0016] S216. Perform weighted fusion using sharpness weights to obtain the fused image: ;

[0017] S217. Global optimization based on multi-scale reconstruction loss yields the loss function: ,in For content loss, ,in The images after fusion are respectively Width and height, This represents the absolute value of the grayscale difference between corresponding pixels in the two images. This is the loss balance coefficient. For multi-scale reconstruction loss, Where F() is the feature mapping function, This is the ideal feature enhancement map at scale s. This represents the absolute value of the difference between the enhanced image features and the ideal features at different scales.

[0018] S218. The final enhanced image is obtained through fine-tuning using gradient descent: ,in, This is the gradient of the total loss function.

[0019] Preferably, the spatial domain weights in step S212 Grayscale weight ,in, The standard deviation of the spatial domain, , is the standard deviation of the grayscale of the target pixel within the sliding window.

[0020] Preferably, step S213 involves extracting multimodal feature maps from the denoised calibration image and suppressing feature redundancy through spatial and channel reconstruction convolution to obtain purified multimodal features.

[0021] S2131. Extract grayscale modal feature maps Characterizes the gray-scale distribution features of the region: ,in, The mean and standard deviation of gray levels within a 3×3 sub-block are given; edge modal feature maps are extracted. To characterize the defect edge contour, the Sobel operator is used to convolve the denoised and calibrated image to obtain the horizontal and vertical gradients. The gradient magnitudes are then calculated as edge modal features. Texture modal feature maps are extracted. The surface texture of the cable is characterized by a grayscale matrix. ,in, It is a 16-level gray-level co-occurrence matrix with a distance of 2 and an angle of 0.

[0022] S2132. Perform spatial reconstruction on the three types of modal features respectively, and separate the effective spatial features from redundant features: , , Where F represents a certain type of modal feature of the input. To train the weights and biases, These are the preliminary spatial features after linear transformation. For group normalization operation, Choose a mask for the space. For element-wise multiplication, Features for spatial reconstruction;

[0023] S2133, Reconstructing spatial features Split into two branches based on channel dimension. and Grouped depthwise convolution and pointwise convolution are performed on the two branches respectively to extract cross-channel feature maps: , ,in, For grouped convolution, To perform pointwise convolution, average pooling and channel attention are applied to the fused branch features to obtain the final channel reconstructed feature map: .

[0024] Preferably, step S214 combines the obtained purified multimodal feature map to perform regional differential enhancement on the denoised calibration image, resulting in a regional enhanced image. The implementation involves reconstructing the grayscale feature map for each pixel by reconstructing the channel. Channel reconstruction edge feature map and channel reconstruction texture feature map A weighted summation method is used to calculate the comprehensive score of defect features, which is then compared with a set threshold. If the score is greater than or equal to the threshold, the corresponding area is identified as a suspected defect area, and its edges and contrast are enhanced. If the score is less than the threshold, the corresponding area is identified as a normal texture area, and its texture is preserved and enhanced. Finally, the data are integrated to obtain a region-specific enhanced image. .

[0025] Preferably, the specific implementation of matching the enhanced image obtained in step S2 with the historical medium-voltage cross-linked cable extrusion pressure data to obtain the corresponding matched cross-linked cable pressure-image feature data is as follows:

[0026] S221. Mark the historical extrusion pressure data and cable image data with timestamps respectively;

[0027] S222. Based on the constant traction speed of the cable extrusion production line, convert the time interval of pressure acquisition into the axial length position of the cable.

[0028] S223. Bind the pressure value of the corresponding production period at the corresponding position to the feature data of the enhanced image of the axial position segment to obtain the pressure-image feature dataset.

[0029] Preferably, step S3 involves performing a multi-dimensional correlation analysis on the matched cross-linked cable pressure-image data to obtain the optimal combination of pressure parameters for the three-layer extrusion device:

[0030] S31. Initialize the population using a genetic algorithm, with each individual in the population corresponding to a set of three-layer pressure parameter combinations;

[0031] S32. The objective function is the comprehensive score of the defect features of all pixels in the cable image corresponding to the pressure value, and the adaptive function is the reciprocal of the objective function.

[0032] S33. Perform selection crossover and mutation operations;

[0033] S34. Set the number of iterations. When the maximum number of iterations is reached, the iteration terminates. Select the pressure parameter combination corresponding to the individual with the highest fitness in the population as the optimal pressure parameter combination for the three-layer extrusion device.

[0034] Preferably, in step S4, the adjustment of pressure to the set optimal value based on the optimal combination of pressure parameters of the three-layer extrusion device is specifically implemented by improving fuzzy PID pressure control as follows:

[0035] S41. Collect the actual pressure value at the current moment and calculate the pressure deviation. and pressure deviation change rate The pressure deviation is the difference between the actual pressure value at the current moment and the set optimal value, and the pressure deviation change rate is obtained by dividing the difference between the pressure deviation at the current moment and the pressure deviation at the previous moment by the time interval.

[0036] S42, to and Fuzzification is performed using a Gaussian membership function. ,in, Represent and , Let be the membership degree of fuzzy subset A, and let a be the center value of fuzzy subset A. The width parameter of the fuzzy subset A;

[0037] S43. Preset a fuzzy rule base, and match the current rule from the rule base. and All rules corresponding to the membership degree are weighted and integrated to obtain a fuzzy set of PID parameter correction values.

[0038] S44. Using the weighted average method, the fuzzy set is transformed into precise PID parameter correction values. The defuzzified correction values ​​are then superimposed with the initial PID parameters corresponding to the deviation level to obtain the final control parameters.

[0039] S45. Finally, control is achieved by calculating the control quantity using the discrete PID formula based on the final control parameters.

[0040] Compared with existing technologies, the advantages and positive effects of this invention are as follows: it overcomes the limitations of manual pressure adjustment based on experience, accurately captures subtle indentation features through preprocessing such as adaptive grayscale calibration and improved bilateral filtering, and enhances defect identification accuracy by combining multimodal feature extraction and regional enhancement; it achieves precise matching of pressure and image data through time-axis position mapping, mines the optimal combination of three-layer extrusion pressure parameters using a genetic algorithm, and then achieves real-time precise pressure adjustment through improved fuzzy PID control, which can dynamically update historical data to optimize the control strategy. This forms an intelligent closed loop from defect perception and parameter optimization to real-time control, effectively solving problems such as low accuracy, strong lag, and inconsistent parameters in manual pressure adjustment, significantly improving the stability of cable surface quality and product qualification rate, and increasing production efficiency. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram illustrating the implementation process of the anti-denting extrusion control method for medium-voltage cross-linked cables. Detailed Implementation

[0043] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0044] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0045] In existing medium-voltage cross-linked cable extrusion processes, surface depression defects often occur due to pressure mismatch. Pressure adjustment mainly relies on manual experience, leading to adjustment lag and poor consistency. Therefore, this invention proposes a method for preventing depressions during the extrusion process of medium-voltage cross-linked cables. By constructing a dynamic mapping relationship between cable surface condition, image features, extrusion pressure parameters, and control feedback, a fully intelligent closed-loop process from defect perception and parameter optimization to real-time control is achieved. This significantly improves the stability and consistency of the appearance quality of medium-voltage cross-linked cables while ensuring production continuity. The specific implementation process is as follows: Figure 1 As shown.

[0046] To comprehensively depict the surface condition changes of cables under different extrusion pressures and address the issues of insufficient single-condition samples and poor generalization ability of empirical parameters, extrusion pressure data of medium-voltage cross-linked cables and corresponding cross-linked cable image data were collected from historical data. Then, the cross-linked cable image data underwent enhancement preprocessing to obtain enhanced images.

[0047] Specifically, local grayscale statistics are performed on the original cross-linked cable image using a sliding window method. The mean and standard deviation of the grayscale within the window are used to calibrate the center pixel to obtain the pixel value of the grayscale-calibrated image. This eliminates the overall grayscale drift caused by uneven illumination. Specifically, adaptive grayscale calibration and edge-preserving denoising are performed on the original cross-linked cable image. Local grayscale mean normalization is used for adaptive grayscale calibration to eliminate grayscale shift caused by uneven illumination. ,in, This is an image of the original cross-linked cable. This represents the grayscale value of the original image at pixel (x, y). These represent the mean and standard deviation of gray levels within a sliding window centered at (x, y), respectively. To avoid extremely small values ​​where the denominator is 0, we obtain the pixel values ​​of the image after grayscale calibration. .

[0048] Subsequently, an improved bilateral filter combining spatial domain weights and gray-level domain weights is introduced to smooth random noise while maintaining the geometric continuity of concave edges, thus obtaining the pixel values ​​of the denoised and calibrated image. That is, adaptive denoising with edge preservation based on improved bilateral filtering, which preserves the edges of concave defects while suppressing extrusion noise: Where W is the normalized weight used to balance the weights of all pixels within the window, avoiding image brightness shift after weighted averaging. These represent the spatial domain weights and the grayscale domain weights, respectively. m and n are the horizontal and vertical coordinates of any pixel within the sliding window, used to traverse each pixel within the window and calculate its corresponding weight. These are the pixel values ​​of the denoised and calibrated image. Among them, the spatial domain weights... Grayscale weight ,in, The standard deviation of the spatial domain, , is the standard deviation of the grayscale of the target pixel within the sliding window.

[0049] To comprehensively characterize the minute indentation features on the cable surface from different information dimensions and address the insufficient expressive power of single features for complex defects, a joint modeling scheme of grayscale, edge, and texture modal features is adopted. Feature redundancy is suppressed through spatial and channel reconstruction. Multimodal feature extraction is performed on the denoised calibration image, and feature redundancy is further suppressed through spatial and channel reconstruction convolution, resulting in purified multimodal feature maps. Specifically, grayscale modal feature maps are extracted from the enhanced image to reflect local brightness distribution changes, edge modal feature maps are extracted to characterize the abrupt changes in indentation contours, and texture modal feature maps are extracted to describe the material structure of the cable surface. Subsequently, spatial reconstruction is performed on each modal feature map to highlight the effective spatial region; then, channel reconstruction and attention mechanisms are used to strengthen key channels and suppress redundant information, resulting in a purified set of multimodal feature maps. Specifically, S2131, extracting grayscale modal feature maps. Characterizes the gray-scale distribution features of the region: ,in, The mean and standard deviation of gray levels within a 3×3 sub-block are given; edge modal feature maps are extracted. To characterize the defect edge contour, the Sobel operator is used to convolve the denoised and calibrated image to obtain the horizontal and vertical gradients. The gradient magnitudes are then calculated as edge modal features. Texture modal feature maps are extracted. The surface texture of the cable is characterized by a grayscale matrix. ,in, The gray-level co-occurrence matrix is ​​16 levels, with a distance of 2 and an angle of 0. Spatial reconstruction is performed on the three types of modal features to separate effective spatial features from redundant features. , , Where F represents a certain type of modal feature of the input. To train the weights and biases, These are the preliminary spatial features after linear transformation. For group normalization operation, Choose a mask for the space. For element-wise multiplication, For spatial reconstruction features; to reconstruct spatial features Split into two branches based on channel dimension. and Grouped depthwise convolution and pointwise convolution are performed on the two branches respectively to extract cross-channel feature maps: , ,in, For grouped convolution, To perform pointwise convolution, average pooling and channel attention are applied to the fused branch features to obtain the final channel reconstructed feature map: . represent , as well as Any one of them.

[0050] To enhance the representation of defects in suspected concave areas and maintain texture consistency in normal areas, and to address the issue of noise amplification caused by global uniform enhancement, a region-specific differentiated enhancement scheme based on multimodal feature scores is adopted. This scheme, combined with the purified multimodal feature map, is used to perform region-specific differentiated enhancement on the denoised calibration image, resulting in a region-specific enhanced image. For each pixel, reconstruct the grayscale feature map from the channels. Channel reconstruction edge feature map and channel reconstruction texture feature map A weighted summation method is used to calculate the comprehensive score of defect features, which is then compared with a set threshold. If the score is greater than or equal to the set threshold, the corresponding area is identified as a suspected defect area, and the edges and contrast are enhanced. in, The edge enhancement coefficient, This is the grayscale stretching factor. The function is a contrast stretching function. If the contrast is less than a set threshold, the corresponding area is considered a normal texture area, and the texture is preserved with slight enhancement. ,in, Enhancement coefficients are preserved for texture, and finally, the images are integrated to obtain region-enhanced images. .

[0051] Next, multi-scale fusion and global optimization are performed on the region-enhanced images to generate multi-scale feature maps. Downsampling with scale factors s=1, 2, and 4 respectively yields multi-scale images: ,in, To perform floor function calculations, the sharpness weights of images at each scale are calculated. The calculation method is as follows: ,in, Here, represents the width and height of the image at scale s, respectively; weighted fusion is performed using sharpness as the weight, resulting in the fused image: Global optimization based on multi-scale reconstruction loss yields the following loss function: ,in For content loss, ,in The images after fusion are respectively Width and height, This represents the absolute value of the grayscale difference between corresponding pixels in the two images. This is the loss balance coefficient. For multi-scale reconstruction loss, Where F() is the feature mapping function, This is the ideal feature enhancement map at scale s. The values ​​represent the absolute differences between the enhanced image features and the ideal features at different scales; finally, the enhanced image is obtained through fine-tuning using gradient descent. ,in, This is the gradient of the total loss function.

[0052] To establish a precise correspondence between the cable surface condition and the corresponding extrusion process parameters, the enhanced images are matched with historical medium-voltage cross-linked cable extrusion pressure data to obtain corresponding matched cross-linked cable pressure-image feature data. This involves timestamping the historical extrusion pressure data and cable image data; converting the pressure acquisition time interval into the axial length position of the cable based on the constant traction speed of the cable extrusion production line; and binding the pressure value of the corresponding production period with the feature data of the enhanced image of that axial position segment to obtain a pressure-image feature dataset. Specifically, based on the constant traction speed of the production line, time information is converted into the axial length position of the cable. Specifically, a certain moment is taken as the axial zero point, the timestamp of that moment is recorded, and the cumulative displacement of the cable in the extrusion direction at any subsequent moment is calculated based on the traction speed, thus establishing a "timestamp-axial position" mapping function. For continuously acquired pressure data, its corresponding time interval is mapped to a continuous axial position interval; for discretely acquired image data, each frame is mapped to its corresponding central axial position. Secondly, after completing the axial position mapping, the cable axial direction is segmented, and the extrusion pressure data within the same axial position segment are statistically summarized. A range averaging or weighted averaging method is preferred to obtain the representative pressure value for that position segment. Simultaneously, the enhanced image feature data corresponding to that axial position segment is extracted. Finally, the representative pressure values ​​obtained within the same axial position segment are bound one-to-one with the corresponding images to construct pressure-image feature data pairs, thus forming a complete pressure-image feature dataset. Through this method, precise matching of pressure parameters and cable surface conditions at the physical space level is achieved, providing a highly consistent data foundation for subsequent multidimensional correlation analysis and optimal control parameter solving.

[0053] Furthermore, to extract the optimal three-layer extrusion pressure combination from historical samples and address the lack of global optimality in manual pressure adjustment, a multi-dimensional correlation analysis was performed on the matched cross-linked cable pressure-image data to obtain the optimal pressure parameter combination for the three-layer extrusion device. This was achieved by initializing the population using a genetic algorithm, with each individual in the population corresponding to a set of three-layer pressure parameter combinations; using the comprehensive score of defect features of all pixels in the cable image corresponding to the pressure value as the objective function, and the reciprocal of the objective function as the fitness function; performing selection, crossover, and mutation operations; setting the number of iterations, and terminating the iteration when the maximum number of iterations is reached; and selecting the pressure parameter combination corresponding to the individual with the highest fitness in the population as the optimal pressure parameter combination for the three-layer extrusion device. Specifically, based on the structural characteristics of the three-layer extrusion device, the extrusion pressure parameters of the inner, middle, and outer layers were set as continuous decision variables, and upper and lower limits for each pressure parameter were given according to the safe operating range of the equipment. Genetic algorithm individuals are constructed using real-number encoding. Each individual corresponds to a set of three-layer extrusion pressure parameter combinations. Multiple individuals form the initial population, which is preferably generated through random uniform sampling to enhance search space coverage. Based on the constructed pressure-image feature dataset, the fitness of each individual in the population is evaluated. According to the three-layer pressure parameter combination corresponding to the individual, cable image data produced within that pressure range is extracted, and the comprehensive defect feature score of all pixels in the image is calculated. This comprehensive score reflects the severity of surface dents in the cable. The comprehensive defect feature score is used as the objective function value, and its reciprocal is used as the fitness function of the individual, ensuring that pressure combinations with milder defects correspond to higher fitness. Individuals in the population are selected based on their fitness, preferably using a roulette wheel or tournament selection strategy to ensure that high-fitness individuals have a higher probability of entering the next generation. Subsequently, a crossover operation is performed on the selected individuals, generating new offspring individuals through linear recombination along the three-layer pressure parameter dimension, thus achieving the combination and inheritance of superior genes. Simultaneously, a low-probability mutation operation is introduced to randomly perturb one or more layers of pressure parameters in an individual, in order to maintain population diversity and avoid premature convergence of the algorithm. Finally, the fitness evaluation, selection, crossover, and mutation processes are repeated until the preset maximum number of iterations is reached. After the iteration terminates, the individual with the highest fitness is selected from the final population, and its corresponding three-layer extrusion pressure parameter combination is determined as the optimal pressure parameter combination.

[0054] To achieve rapid and stable adjustment of the actual extrusion pressure to the optimal value during production, and to address the issues of slow response and insufficient robustness of traditional PID control to changes in operating conditions, an improved fuzzy PID pressure control method is used, based on the optimal pressure parameter combination of the three-layer extruder, to adjust the pressure to the set optimal value. This is implemented by collecting the actual pressure value at the current moment and calculating the pressure deviation. and pressure deviation change rate The pressure deviation is the difference between the actual pressure value at the current moment and the set optimal value, and the pressure deviation change rate is obtained by dividing the difference between the pressure deviation at the current moment and the pressure deviation at the previous moment by the time interval; and Fuzzification is performed using a Gaussian membership function. ,in, Represent and , Let be the membership degree of fuzzy subset A, and let a be the center value of fuzzy subset A. The width parameter of the fuzzy subset A; a preset fuzzy rule base, from which the current... and The membership degrees of all rules are weighted and integrated to obtain a fuzzy set of PID parameter correction values. A weighted average method is used to transform the fuzzy set into precise PID parameter correction values. These defuzzified correction values ​​are then superimposed with the initial PID parameters corresponding to the deviation levels to obtain the final control parameters. Finally, the control quantity is calculated using the discrete PID formula based on the final control parameters to achieve control. Specifically, continuous... and The mapping is represented as a fuzzy set, providing input for rule matching. Both are divided into five fuzzy subsets: negative large (NB), negative small (NS), zero (ZO), positive small (PS), and positive large (PB). Each subset corresponds to a specific physical meaning. Fuzzification is achieved using a Gaussian membership function to obtain the membership degree of each value to the five fuzzy subsets. The sum of the membership degrees must meet the normalization requirement. Finally, the fuzzy vectors of both are output. Then, a fuzzy rule base based on process experience and experimental data is constructed, covering... and The combination of all fuzzy subsets, each rule corresponds to the correction direction and magnitude of PID parameters (Kp, Ki, Kd). Rule formulation follows the extrusion pressure control logic: when the deviation is large, Kp is preferentially increased to quickly eliminate the deviation, Ki is decreased to avoid integral saturation, and Kd is adjusted to suppress overshoot; when the deviation is small, Kp is decreased, Ki is increased to eliminate steady-state error, and Kd is optimized to ensure stability; when the deviation is 0, the parameters are fine-tuned to maintain stable pressure. In the rule matching stage, based on the output... and The fuzzy vector is filtered to select fuzzy subsets with a membership degree greater than 0.1. Corresponding rules are retrieved from the rule base to obtain the fuzzy output. Then, a weighted integration is performed, with the weights calculated as follows: Subset membership degree × Subset membership. The weighted sums of ΔKp, ΔKi, and ΔKd of all valid rules are used to form three fuzzy sets of PID parameter corrections, outputting the parameter corrections ΔKp, ΔKi, and ΔKd. The weighted average method is then used to defuzzify the PID parameter corrections to obtain the precise correction values. Where x is any one of p, i, and d, To determine the membership degree of each fuzzy subset, the precise correction value is superimposed with the initial PID parameters corresponding to the deviation level to generate the final PID control parameters for the current moment. Finally, based on the final PID parameters, the control quantity u(t) is calculated using the discrete PID formula to achieve precise control of the pressure in the three-layer extrusion process.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting, characterized in that, Includes the following steps: S1. Collect historical extrusion pressure data of medium-voltage cross-linked cables and obtain cross-linked cable image data; S2. Perform enhancement preprocessing on the cross-linked cable image data to obtain the enhanced image. Match the enhanced image with the historical medium-voltage cross-linked cable extrusion pressure data to obtain the corresponding matched cross-linked cable pressure-image feature data. S3. Perform multi-dimensional correlation analysis on the matched cross-linked cable pressure-image data to obtain the optimal combination of pressure parameters for the three-layer extrusion device; S4. Based on the optimal combination of pressure parameters of the three-layer extrusion device, the pressure is adjusted to the set optimal value by improving the fuzzy PID pressure control. S5. Collect real-time optimized control pressure-image data, replace historical data in chronological order based on preset data volume, and re-optimize the pressure of the extrusion device after replacement for adjustment and control.

2. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 1, characterized in that, The specific implementation of enhancing the cross-linked cable image data in step S2 to obtain the enhanced image is as follows: S211. Adaptive grayscale calibration and edge-preserving denoising are performed on the original cross-linked cable image. Local grayscale mean normalization is used for adaptive grayscale calibration to eliminate grayscale shifts caused by uneven illumination. ,in, This is an image of the original cross-linked cable. This represents the grayscale value of the original image at pixel (x, y). These represent the mean and standard deviation of gray levels within a sliding window centered at (x, y), respectively. To avoid extremely small values ​​where the denominator is 0, we obtain the pixel values ​​of the image after grayscale calibration. ; S212. Adaptive edge-preserving denoising based on improved bilateral filtering preserves the edges of concave defects while suppressing extrusion noise: Where W is the normalized weight used to balance the weights of all pixels within the window, avoiding image brightness shift after weighted averaging. These represent the spatial domain weights and the grayscale domain weights, respectively. m and n are the horizontal and vertical coordinates of any pixel within the sliding window, used to traverse each pixel within the window and calculate its corresponding weight. The pixel values ​​of the image after noise reduction and calibration; S213. Extract multimodal feature maps from the denoised calibration image, and suppress feature redundancy through spatial and channel reconstruction convolution to obtain purified multimodal features. S214. Combining the obtained purified multimodal feature map, perform region-specific differential enhancement on the denoised calibration image to obtain the region-enhanced image. ; S215. Perform multi-scale fusion and global optimization on the region-enhanced image to generate multi-scale feature maps. Downsampling with scale factors s=1, 2, and 4 respectively yields multi-scale images: ,in, To perform floor function calculations, the sharpness weights of images at each scale are calculated. The calculation method is as follows: ,in, These represent the width and height of the image at scale s, respectively. S216. Perform weighted fusion using sharpness weights to obtain the fused image: ; S217. Global optimization based on multi-scale reconstruction loss yields the loss function: ,in For content loss, ,in The images after fusion are respectively Width and height, This represents the absolute value of the grayscale difference between corresponding pixels in the two images. This is the loss balance coefficient. For multi-scale reconstruction loss, Where F() is the feature mapping function, This is the ideal feature enhancement map at scale s. This represents the absolute value of the difference between the enhanced image features and the ideal features at different scales. S218. The final enhanced image is obtained through fine-tuning using gradient descent: ,in, This is the gradient of the total loss function.

3. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 2, characterized in that, Spatial domain weights in step S212 Grayscale weight ,in, The standard deviation of the spatial domain, , is the standard deviation of the grayscale of the target pixel within the sliding window.

4. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 2, characterized in that, Step S213 involves extracting multimodal feature maps from the denoised calibration image and suppressing feature redundancy through spatial and channel reconstruction convolution to obtain purified multimodal features. S2131. Extract grayscale modal feature maps Characterizes the gray-scale distribution features of the region: ,in, The mean and standard deviation of gray levels within a 3×3 sub-block are given; edge modal feature maps are extracted. To characterize the defect edge contour, the Sobel operator is used to convolve the denoised and calibrated image to obtain the horizontal and vertical gradients. The gradient magnitudes are then calculated as edge modal features. Texture modal feature maps are extracted. The surface texture of the cable is characterized by a grayscale matrix. ,in, It is a 16-level gray-level co-occurrence matrix with a distance of 2 and an angle of 0. S2132. Perform spatial reconstruction on the three types of modal features respectively, and separate the effective spatial features from redundant features: , , Where F represents a certain type of modal feature of the input. To train the weights and biases, These are the preliminary spatial features after linear transformation. For group normalization operation, Choose a mask for the space. For element-wise multiplication, Features for spatial reconstruction; S2133, Reconstructing spatial features Split into two branches based on channel dimension. and Grouped depthwise convolution and pointwise convolution are performed on the two branches respectively to extract cross-channel feature maps: , ,in, For grouped convolution, To perform pointwise convolution, average pooling and channel attention are applied to the fused branch features to obtain the final channel reconstructed feature map: .

5. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 2, characterized in that, Step S214 combines the obtained purified multimodal feature map to perform regional differential enhancement on the denoised calibration image, resulting in a regional enhanced image. The implementation involves reconstructing the grayscale feature map for each pixel by reconstructing the channel. Channel reconstruction edge feature map and channel reconstruction texture feature map A weighted summation method is used to calculate the comprehensive score of defect features, which is then compared with a set threshold. If the score is greater than or equal to the threshold, the corresponding area is identified as a suspected defect area, and its edges and contrast are enhanced. If the score is less than the threshold, the corresponding area is identified as a normal texture area, and its texture is preserved and enhanced. Finally, the data are integrated to obtain a region-specific enhanced image. .

6. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 1, characterized in that, In step S2, the enhanced image obtained is matched with historical medium-voltage cross-linked cable extrusion pressure data to obtain corresponding matched cross-linked cable pressure-image feature data. The specific implementation of this method is as follows: S221. Mark the historical extrusion pressure data and cable image data with timestamps respectively; S222. Based on the constant traction speed of the cable extrusion production line, convert the time interval of pressure acquisition into the axial length position of the cable. S223. Bind the pressure value of the corresponding production period at the corresponding position to the feature data of the enhanced image of the axial position segment to obtain the pressure-image feature dataset.

7. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 1, characterized in that, Step S3 involves performing a multi-dimensional correlation analysis on the matched cross-linked cable pressure-image data to obtain the optimal combination of pressure parameters for the three-layer extrusion device: S31. Initialize the population using a genetic algorithm, with each individual in the population corresponding to a set of three-layer pressure parameter combinations; S32. The objective function is the comprehensive score of the defect features of all pixels in the cable image corresponding to the pressure value, and the adaptive function is the reciprocal of the objective function. S33. Perform selection crossover and mutation operations; S34. Set the number of iterations. When the maximum number of iterations is reached, the iteration terminates. Select the pressure parameter combination corresponding to the individual with the highest fitness in the population as the optimal pressure parameter combination for the three-layer extrusion device.

8. The method for controlling the extrusion molding of medium-voltage cross-linked cables to prevent denting according to claim 1, characterized in that, In step S4, the specific implementation of adjusting the pressure to the set optimal value based on the optimal combination of pressure parameters of the three-layer extrusion device through improved fuzzy PID pressure control is as follows: S41. Collect the actual pressure value at the current moment and calculate the pressure deviation. and pressure deviation change rate The pressure deviation is the difference between the actual pressure value at the current moment and the set optimal value, and the pressure deviation change rate is obtained by dividing the difference between the pressure deviation at the current moment and the pressure deviation at the previous moment by the time interval. S42, to and Fuzzification is performed using a Gaussian membership function. ,in, Represent and , Let be the membership degree of fuzzy subset A, and let a be the center value of fuzzy subset A. The width parameter of the fuzzy subset A; S43. Preset a fuzzy rule base, and match the current rule from the rule base. and All rules corresponding to the membership degree are weighted and integrated to obtain a fuzzy set of PID parameter correction values. S44. Using the weighted average method, the fuzzy set is transformed into precise PID parameter correction values. The defuzzified correction values ​​are then superimposed with the initial PID parameters corresponding to the deviation level to obtain the final control parameters. S45. Finally, control is achieved by calculating the control quantity using the discrete PID formula based on the final control parameters.

Images