Intelligent control method and system for waterproof material production and processing

Abstract

The invention discloses an intelligent control method and system for waterproof material production and processing, relates to the technical field of automatic control, and realizes comprehensive monitoring and regulation of a waterproof material production process, firstly, through real-time data acquisition of an industrial camera and a sensor group, construction of a high-dimensional feature vector Vfeed (t) and application of a classification model, a high-dimensional feature vector Vfeed (t) is obtained; and the defects on the surface of the material can be accurately identified and classified. And then, by generating a feedback control signal Cctrl (Vfeed, S, P), the system can adjust the production speed S and the pressure P in time, and dynamically regulate and control the detected defective material. Moreover, the system ensures the adjustment effectiveness in the production process through an iterative verification mechanism, and avoids the limitations, including low efficiency, misjudgment and missing detection, of the traditional manual inspection and optical detection methods. The production efficiency and the material quality are improved, and the defect rate and the rework rate are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of automation control technology, and in particular to an intelligent control method and system for the production and processing of waterproof materials. Background Art

[0002] The production and processing of waterproof materials involves multiple fields such as material science, chemical engineering and manufacturing technology. Among these fields, the production process of waterproof materials is particularly important because it directly affects the performance and service life of the materials. The manufacturing process of waterproof materials usually includes multiple links such as raw material proportioning, mixing, molding, curing and surface treatment. Each link must ensure high-precision control to produce products with excellent performance.

[0003] In the traditional production process of waterproof materials, the detection of surface defects usually relies on manual inspection or simple optical inspection equipment. Manual inspection is not only inefficient, but also prone to misjudgment due to operator fatigue or lack of experience. Although optical inspection equipment can reduce the manual burden to a certain extent, its detection accuracy and sensitivity are often insufficient, and it is difficult to identify tiny surface defects such as fine bubbles, cracks or uneven surface textures. If these defects are not detected and corrected in time during the production process, it may lead to substandard product quality and even cause production problems in subsequent links, ultimately affecting the overall performance and market competitiveness of waterproof materials. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides an intelligent control method and system for the production and processing of waterproof materials, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent control method for the production and processing of waterproof materials, comprising the following steps:

[0006] S1. Install industrial cameras and sensor groups on the production line to collect real-time image data and surface status data of waterproof materials, and integrate the feature vector V at the collection time t. cj (t);

[0007] S2, for the feature vector V cj (t) Perform image preprocessing and image feature extraction to obtain an image feature set, and fuse it with the non-image data obtained after nonlinear function processing and the image feature set to obtain a high-dimensional feature vector V at acquisition time t feat (t);

[0008] S3, through the obtained high-dimensional feature vector V feat(t) Establish a classification model, identify and classify defects through training of the classification model, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat );

[0009] S4. According to the defect classification results D class (V feat ) and high-dimensional feature vector V feat (t) is fitted and combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal C for regulating the waterproof material processing ctrl (V feat , S, P);

[0010] S5, according to the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to regulate the defective materials specifically, and to trigger the monitoring strategy for the regulated equipment, to iteratively verify the material defect classification results D within a fixed period T class (V feat ) quantity, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result.

[0011] Preferably, step S1 also includes S11 and S12;

[0012] S11, by installing an industrial camera and a sensor group on the production line, real-time shooting of the surface of the waterproof material passing through the production line is performed, and image data of the surface of the waterproof material is collected in real time, and the image data at time t is obtained and recorded as an image matrix I(t), wherein the image matrix I(t) includes image pixel values ​​I i，j (t);

[0013] S12, collect the surface state data of the waterproof material in real time through the sensor group installed on the production line, the sensor group includes a temperature sensor, an optical interference sensor and a stress sensor, and obtains the temperature value T of different areas of the material at time t x (t), surface roughness value S r (t) and material surface stress value Y ext (t);

[0014] Synchronize, image matrix I(t), temperature value T x (t), surface roughness value S r (t) and material surface stress value Y ext (t) is integrated to obtain the feature vector V at acquisition time t cj (t).

[0015] Preferably, step S2 also includes S21 and S22;

[0016] S21, for the eigenvector V cj (t) Perform image preprocessing, including denoising preprocessing, contrast enhancement preprocessing, and edge detection preprocessing, and then use a convolutional neural network to extract features to generate a high-dimensional image feature set F inmg (t), including texture, edge shape, and edge inflection points;

[0017] Among them, denoising preprocessing includes using median filtering and Gaussian filtering to remove noise in the image and remove the impact of optical interference in the production environment; contrast enhancement preprocessing includes using histogram equalization technology to adjust the contrast of the image to highlight tiny defects on the surface of the material; edge detection preprocessing includes using Canny and Sobel operators for edge detection to mark edge information of defects in the image.

[0018] Preferably, S22, for the feature vector V cj (t) Non-image data, including temperature value T x (t), surface roughness value S r (t) and material surface stress value Y ext (t), non-linear processing is performed to form non-image data V s (t), adjust the response relationship to the material state, including using polynomial regression and neural network models to establish a nonlinear processing model to extract nonlinear relationships and compare them with the image feature set F inmg (t) is fused, and the fusion includes splicing and weighted averaging to obtain the high-dimensional feature vector V at the acquisition time t. feat (t).

[0019] Preferably, step S3 specifically obtains the high-dimensional feature vector V feat (t) Establishing a classification model, including using CNN technology to establish a classification model, wherein the classification model includes a convolution layer, a pooling layer and a fully connected layer, and performing defect recognition and classification through the training of the classification model, including using a back propagation algorithm to continuously optimize the model parameters and minimize the loss function, so that the classification model can recognize and classify the defect types in the material, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat );

[0020] Among them, the defect classification result D class (V feat ) The recognition calculation output is obtained through steps S311 and S312.

[0021] Preferably, S311, obtain a high-dimensional feature vector V through the output probability vector of the convolutional neural network feat (t) The probability value p of the k-th defect after input into the CNN classification model k (V feat );

[0022] Among them, defects include bubbles, cracks, impurities, color difference, wrinkles, depressions, delamination and roughness anomalies, and the defects are marked with subscripts corresponding to bubbles, cracks, impurities, color difference, wrinkles, depressions, delamination and roughness anomalies, forming a defect type set D target ;

[0023] The CNN classification model specifically includes step S3111 calculation of the convolutional layer, step S3112 calculation of the activation function and the pooling layer, and step S3113 calculation of the fully connected layer;

[0024] S3111, perform a convolution operation by combining each convolution kernel in the convolution layer with the input high-dimensional feature vector V feat (t) Perform convolution operation to obtain feature map Specifically, it represents the feature map value calculated by the qth convolution kernel at position (i, j) in the pth layer;

[0025] S3112, execute the activation function, and convert the feature map after the convolution operation By using the activation function ReLU for nonlinear mapping, the activated feature map value is obtained The activation function ReLU synchronously sets negative values ​​to 0 to retain positive values;

[0026] Perform pooling operations, including using maximum pooling and average pooling on the activated feature map values Downsample, adjust the size of the feature map and extract the display features to obtain the pooled feature values Specifically, by taking the activated feature map value Peak acquisition in the (i, j) area;

[0027] S3113, perform full connection operation, and pool the feature values ​​of all positions Flatten to form a vector X p , synchronized as the input vector X of the fully connected layer p , and with the weight matrix W (fc) and bias b (fc) Fitting is performed and the output features are obtained through the activation function ReLU: high-dimensional feature vector V feat (t) The probability value p of the kth defect after input into the CNN classification model k (V feat );

[0028] S312, by obtaining the probability value p of the k-th defect k (V feat ) to mark the defect type label and obtain the defect classification result label D corresponding to the defect type identified by the material class (V feat ).

[0029] Preferably, step S4 also includes S41 and S42;

[0030] S41, using the multivariate linear regression model to classify the defect results D class (V feat ) and high-dimensional feature vector V feat (t) Fitting, including the defect classification result D class (V feat ) as the corresponding variable, and the high-dimensional feature vector V feat (t) is used as the independent variable and fitted through the linear regression model to obtain the defect classification result D after fitting. class (V feat ).

[0031] Preferably, S42, the defect classification result D after fitting class (V feat ) is combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal for regulating the waterproof material processing. Cctrl (V feat , S, P).

[0032] Preferably, step S5 also includes S51 and S52;

[0033] S51, according to the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to specifically regulate the defective materials, and to obtain the regulated production speed SE and equipment pressure PE;

[0034] S52. Trigger the monitoring strategy for the control equipment and iteratively verify the material defect classification result D within a fixed period T. class (V feat ) quantity, obtain the defect quantity N of all defect types Dclass , and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result;

[0035] The results of the iterative evaluation of the control are obtained by Matching method acquisition;

[0036] When the return value is 1, the qualified result of the control iteration evaluation is obtained;

[0037] When the return value is 0, the unqualified result of the control iteration evaluation is obtained, and steps S4 and S5 are iteratively executed.

[0038] An intelligent control system for waterproof material production and processing, comprising a data acquisition module, a data processing and extraction module, a data modeling module, a defect fitting module and a feedback control module;

[0039] The data acquisition module collects image data and surface status data of waterproof materials in real time by installing industrial cameras and sensor groups on the production line, and integrates the feature vector V at the acquisition time t. cj (t);

[0040] The data processing and extraction module processes the feature vector V cj (t) Perform image preprocessing and image feature extraction to obtain an image feature set, and fuse it with the non-image data obtained after nonlinear function processing and the image feature set to obtain a high-dimensional feature vector V at acquisition time t feat (t);

[0041] The data modeling module obtains the high-dimensional feature vector V feat (t) Establish a classification model, identify and classify defects through training of the classification model, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat );

[0042] The defect fitting module is based on the defect classification result D class (V feat ) and high-dimensional feature vector V feat (t) is fitted and combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal C for regulating the waterproof material processing ctrl (V feat , S, P);

[0043] The feedback control module is based on the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to regulate the defective materials specifically, and to trigger the monitoring strategy for the regulated equipment, to iteratively verify the material defect classification results D within a fixed period T class (V feat ) quantity, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result.

[0044] The present invention provides an intelligent control method and system for the production and processing of waterproof materials, which has the following beneficial effects:

[0045] (1) Through steps S1 to S5, comprehensive monitoring and control of the waterproof material production process is achieved. First, through real-time data collection by industrial cameras and sensor groups, and high-dimensional feature vector V feat (t) is constructed and the classification model is applied to accurately identify and classify the defects on the material surface. Subsequently, the feedback control signal C is generated ctrl (V feat , S, P), the system can adjust the production speed S and pressure P in time to dynamically control the detected defective materials. In addition, the system ensures the effectiveness of adjustments during the production process through an iterative verification mechanism, avoiding the limitations of traditional manual inspection and optical inspection methods, including inefficiency, misjudgment and missed inspections. It improves production efficiency and material quality, and effectively reduces defect rate and rework rate.

[0046] (2) By installing industrial cameras and a variety of sensors on the production line, the multi-dimensional feature data of the material is captured and integrated in real time. Advanced image preprocessing technology is used to deeply explore the surface features of the material to ensure the high accuracy and comprehensiveness of the detection process. Subsequently, the system uses the constructed convolutional neural network (CNN) classification model to conduct in-depth analysis and accurate classification of possible defects in the material, covering defect types in multiple dimensions from surface roughness, temperature differences to internal stress distribution. It reflects the ability to adapt to complex production environments. Through multi-level and multi-angle feature extraction and classification, more accurate defect detection and real-time correction are achieved, which significantly reduces quality problems caused by material defects and ensures the efficiency and reliability of waterproof material production.

[0047] (3) The defect classification results are fitted with high-dimensional feature vectors through a multivariate linear regression model to accurately grasp the types of defects that may occur in the material during the production process. Combined with the speed and pressure of the current production line, an optimized feedback control signal is generated, which realizes the dynamic adjustment of the production line speed and pressure. An iterative verification mechanism is also used to ensure that every adjustment in the production process is effectively verified and optimized, especially highlighting the adaptive ability in a complex production environment. Through real-time feedback and fine control, the material defect rate is reduced, the operating stability of the production line and product consistency are improved, and energy consumption and raw material waste in the production process are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic diagram of the steps of an intelligent control method for producing and processing waterproof materials according to the present invention;

[0049] Figure 2The present invention is a schematic diagram of a block diagram of an intelligent control system for the production and processing of waterproof materials. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] Example 1

[0052] The present invention provides an intelligent control method and system for the production and processing of waterproof materials. Figure 1 , including the following steps:

[0053] S1. Install industrial cameras and sensor groups on the production line to collect real-time image data and surface status data of waterproof materials, and integrate the feature vector V at the collection time t. cj (t);

[0054] S2, for the feature vector V cj (t) Perform image preprocessing and image feature extraction to obtain an image feature set, and fuse it with the non-image data obtained after nonlinear function processing and the image feature set to obtain a high-dimensional feature vector V at acquisition time t feat (t);

[0055] S3, through the obtained high-dimensional feature vector V feat (t) Establish a classification model, identify and classify defects through training of the classification model, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat );

[0056] S4. According to the defect classification results D class (V feat ) and high-dimensional feature vector V feat (t) is fitted and combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal C for regulating the waterproof material processing ctrl (V feat , S, P);

[0057] S5, according to the feedback control signal C ctrl (V feat, S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to regulate the defective materials specifically, and to trigger the monitoring strategy for the regulated equipment, to iteratively verify the material defect classification results D within a fixed period T class (V feat ) quantity, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result.

[0058] In this embodiment, through steps S1 to S5, comprehensive monitoring and control of the waterproof material production process is achieved. First, through real-time data collection by industrial cameras and sensor groups, and high-dimensional feature vector V feat (t) is constructed and the classification model is applied to accurately identify and classify the defects on the material surface. Subsequently, the feedback control signal C is generated ctrl (V feat , S, P), the system can adjust the production speed S and pressure P in time to dynamically control the detected defective materials. In addition, the system ensures the effectiveness of adjustments during the production process through an iterative verification mechanism, avoiding the limitations of traditional manual inspection and optical inspection methods, including inefficiency, misjudgment and missed inspections. It improves production efficiency and material quality, and effectively reduces defect rate and rework rate.

[0059] Example 2

[0060] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: step S1 also includes S11 and S12;

[0061] S11, by installing an industrial camera and a sensor group on the production line, real-time shooting of the surface of the waterproof material passing through the production line is performed, and image data of the surface of the waterproof material is collected in real time, and the image data at time t is obtained and recorded as an image matrix I(t), wherein the image matrix I(t) includes image pixel values ​​I i，j (t);

[0062] S12, collect the surface state data of the waterproof material in real time through the sensor group installed on the production line, the sensor group includes a temperature sensor, an optical interference sensor and a stress sensor, and obtains the temperature value T of different areas of the material at time t x (t), surface roughness value S r (t) and material surface stress value Y ext (t);

[0063] Synchronize, image matrix I(t), temperature value T x (t), surface roughness value S r (t) and material surface stress value Y ext (t) is integrated to obtain the feature vector V at acquisition time tcj (t).

[0064] Step S2 also includes S21 and S22;

[0065] S21, for the eigenvector V cj (t) Perform image preprocessing, including denoising preprocessing, contrast enhancement preprocessing, and edge detection preprocessing, and then use a convolutional neural network to extract features to generate a high-dimensional image feature set F inmg (t), including texture, edge shape, and edge inflection points;

[0066] Among them, denoising preprocessing includes using median filtering and Gaussian filtering to remove noise in the image and remove the impact of optical interference in the production environment; contrast enhancement preprocessing includes using histogram equalization technology to adjust the contrast of the image to highlight tiny defects on the surface of the material; edge detection preprocessing includes using Canny and Sobel operators for edge detection to mark edge information of defects in the image.

[0067] S22, for the feature vector V cj (t) Non-image data, including temperature value T x (t), surface roughness value S r (t) and material surface stress value Y ext (t), non-linear processing is performed to form non-image data V s (t), adjust the response relationship to the material state, including using polynomial regression and neural network models to establish a nonlinear processing model to extract nonlinear relationships and compare them with the image feature set F inmg (t) is fused, and the fusion includes splicing and weighted averaging to obtain the high-dimensional feature vector V at the acquisition time t. feat (t);

[0068] Among them, the high-dimensional feature vector V feat (t) is V feat (t) = {F inmg (t), V s (t)}.

[0069] Step S3 specifically obtains the high-dimensional feature vector V feat (t) Establishing a classification model, including using CNN technology to establish a classification model, wherein the classification model includes a convolution layer, a pooling layer and a fully connected layer, and performing defect recognition and classification through the training of the classification model, including using a back propagation algorithm to continuously optimize the model parameters and minimize the loss function, so that the classification model can recognize and classify the defect types in the material, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label.class (V feat );

[0070] Among them, the defect classification result D class (V feat ) The recognition calculation output is obtained through steps S311 and S312.

[0071] S311. Obtain a high-dimensional feature vector V through the output probability vector of the convolutional neural network feat (t) The probability value p of the kth defect after input into the CNN classification model k (V feat ):

[0072] The probability value p k (V feat ) is obtained by the following calculation formula:

[0073] p k (V feat )=CNN(V feat (t)) k ;

[0074] In the formula, p k (V feat ) represents the high-dimensional feature vector V feat (t) The probability value of the k-th defect after input into the CNN classification model, CNN (V feat (t)) k Represents the probability value output by the CNN classification model for the k-th type of defect;

[0075] Among them, defects include bubbles, cracks, impurities, color difference, wrinkles, depressions, delamination and roughness anomalies, and the defects are marked with subscripts corresponding to bubbles, cracks, impurities, color difference, wrinkles, depressions, delamination and roughness anomalies, forming a defect type set D target ;

[0076] The CNN classification model specifically includes step S3111 calculation of the convolutional layer, step S3112 calculation of the activation function and the pooling layer, and step S3113 calculation of the fully connected layer;

[0077] S3111, perform a convolution operation by combining each convolution kernel in the convolution layer with the input high-dimensional feature vector V feat (t) Perform convolution operation to obtain feature map Specifically, it represents the feature map value calculated by the qth convolution kernel at position (i, j) in the pth layer;

[0078] S3112, execute the activation function, and convert the feature map after the convolution operation By using the activation function ReLU for nonlinear mapping, the activated feature map value is obtained The activation function ReLU synchronously sets negative values ​​to 0 to retain positive values;

[0079] Perform pooling operations, including using maximum pooling and average pooling on the activated feature map values Downsample, adjust the size of the feature map and extract the display features to obtain the pooled feature values Specifically, by taking the activated feature map value Peak acquisition in the (i, j) area;

[0080] S3113, perform full connection operation, and pool the feature values ​​of all positions Flatten to form a vector X p , synchronized as the input vector X of the fully connected layer p , and with the weight matrix W (fc) and bias b (fc) Fitting is performed and the output features are obtained through the activation function ReLU: high-dimensional feature vector V feat (t) The probability value p of the kth defect after input into the CNN classification model k (V feat );

[0081] S312, by obtaining the probability value p of the k-th defect k (V feat ) to mark the defect type label and obtain the defect classification result label D corresponding to the defect type identified by the material class (V feat ):

[0082] The defect classification result label D class (V feat ) is obtained by the following calculation formula:

[0083] D class (V feat ) = argmax k (p k (V feat ));

[0084] Where D class (V feat ) represents the defect classification result label, specifically the defect type label subscript corresponding to the defect type identified by the material, argmax k (p k (V feat )) represents the corresponding defect type label subscript of the output peak probability value.

[0085] In this embodiment, by installing industrial cameras and a variety of sensors on the production line, the multi-dimensional feature data of the material is captured and integrated in real time, and advanced image preprocessing technology is used to deeply explore the surface features of the material to ensure the high accuracy and comprehensiveness of the detection process. Subsequently, the system uses the constructed convolutional neural network CNN classification model to conduct in-depth analysis and accurate classification of possible defects in the material, covering defect types in multiple dimensions from surface roughness, temperature differences to internal stress distribution. It reflects the ability to adapt to complex production environments. Through multi-level and multi-angle feature extraction and classification, more accurate defect detection and real-time correction are achieved, which significantly reduces quality problems caused by material defects and ensures the efficiency and reliability of waterproof material production.

[0086] Example 3

[0087] This embodiment is explained in Example 2. Please refer to Figure 1 , specifically: step S4 also includes S41 and S42;

[0088] S41, using the multivariate linear regression model to classify the defect results D class (V feat ) and high-dimensional feature vector V feat (t) Fitting, including the defect classification result D class (V feat ) as the corresponding variable, and the high-dimensional feature vector V feat (t) is used as the independent variable and fitted through the linear regression model to obtain the defect classification result D after fitting. class (V feat );

[0089] The defect classification result D class (V feat ) is obtained by the following fitting formula:

[0090]

[0091] In the formula, β0 represents the intercept term of the multiple linear regression model, β i Represents the regression coefficient, specifically represents the high-dimensional feature vector V feat The influence of the i-th vector in (t), V feat，i (t) represents the i-th vector in the high-dimensional feature vector, and n represents the high-dimensional feature vector V feat (t) is the vector dimension number, specifically representing the high-dimensional feature vector V feat (t) length;

[0092] S42, defect classification result after fitting D class (V feat) is combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal for regulating the waterproof material processing. Cctrl (V feat , S, P).

[0093] The feedback control signal C ctrl (V feat , S, P) are obtained by the following calculation formula:

[0094] C ctrl (V feat , S, P) = α·D class (V feat )+γ·S+δ·P;

[0095] Where α, γ and δ represent the defect classification results after fitting, respectively. class (V feat ), preset weight values ​​of production speed S and equipment pressure P, where 0≤α≤1, 0≤γ≤1, 0≤δ≤1, and α+γ+δ=1.

[0096] Step S5 also includes S51 and S52;

[0097] S51, according to the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to specifically regulate the defective materials, and to obtain the regulated production speed SE and equipment pressure PE;

[0098] The production speed SE and equipment pressure PE are obtained by the following calculation formula:

[0099] SE=S+λs·C ctrl (V feat , S, P);

[0100] In the formula, λs represents the speed control gain coefficient, which specifically indicates the influence ratio of the control feedback signal on the production speed;

[0101] PE=P+λp·C ctrl (V feat , S, P);

[0102] In the formula, λp represents the pressure control gain coefficient, which specifically indicates the influence ratio of the control feedback signal on the equipment pressure;

[0103] S52. Trigger the monitoring strategy for the control equipment and iteratively verify the material defect classification result D within a fixed period T. class (V feat ) quantity, obtain the defect quantity N of all defect types Dclass, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result;

[0104] The number of defects N Dclass Obtained through the following calculation formula:

[0105]

[0106] In the formula, ||(D class (V feat (t))∈D target ) in ||() represents an exponential function. When ||(D class (V feat (t))∈D target ) is true, the value returns 1, when ||(D class (V feat (t))∈D target ) is not true, the value returns 0, D target Represents a set of defect types;

[0107] The results of the iterative evaluation of the control are obtained by Matching method acquisition;

[0108] When the return value is 1, the qualified result of the control iteration evaluation is obtained;

[0109] When the return value is 0, the unqualified result of the control iteration evaluation is obtained, and steps S4 and S5 are iteratively executed.

[0110] In this embodiment, the defect classification results are fitted with high-dimensional feature vectors through a multivariate linear regression model to accurately grasp the types of defects that may occur in the material during the production process, and generate optimized feedback control signals in combination with the speed and pressure of the current production line, thereby realizing dynamic adjustment of the production line speed and pressure. An iterative verification mechanism is also used to ensure that every adjustment in the production process is effectively verified and optimized, and the adaptive ability in a complex production environment is particularly highlighted. Through real-time feedback and fine control, the material defect rate is reduced, the operating stability of the production line and product consistency are improved, thereby reducing energy consumption and waste of raw materials in the production process.

[0111] Example 4

[0112] An intelligent control method and system for the production and processing of waterproof materials, please refer to Figure 2 ,Specifically: including data acquisition module, data processing and extraction module, data modeling module, defect fitting module and feedback control module;

[0113] The data acquisition module collects image data and surface status data of waterproof materials in real time by installing industrial cameras and sensor groups on the production line, and integrates the feature vector V at the acquisition time t. cj (t);

[0114] The data processing and extraction module processes the feature vector V cj (t) Perform image preprocessing and image feature extraction to obtain an image feature set, and fuse it with the non-image data obtained after nonlinear function processing and the image feature set to obtain a high-dimensional feature vector V at acquisition time t feat (t);

[0115] The data modeling module obtains the high-dimensional feature vector V feat (t) Establish a classification model, identify and classify defects through training of the classification model, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat );

[0116] The defect fitting module is based on the defect classification result D class (V feat ) and high-dimensional feature vector V feat (t) is fitted and combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal C for regulating the waterproof material processing ctrl (V feat , S, P);

[0117] The feedback control module is based on the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to regulate the defective materials specifically, and to trigger the monitoring strategy for the regulated equipment, to iteratively verify the material defect classification results D within a fixed period T class (V feat ) quantity, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result.

[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent control method for the production and processing of waterproof materials, characterized in that: The following steps are involved: S1. Install industrial cameras and sensor groups on the production line to collect real-time image data and surface status data of waterproof materials, and integrate the feature vector V at the collection time t. cj (t); S2, for the feature vector V cj (t) Perform image preprocessing and image feature extraction to obtain an image feature set, and fuse it with the non-image data obtained after nonlinear function processing and the image feature set to obtain a high-dimensional feature vector V at acquisition time t feat (t); S3, through the obtained high-dimensional feature vector V feat (t) Establish a classification model, identify and classify defects through training of the classification model, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat ); S4. According to the defect classification results D class (V feat ) and high-dimensional feature vector V feat (t) is fitted and combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal C for regulating the waterproof material processing ctrl (V feat , S, P); S5, according to the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to regulate the defective materials specifically, and to trigger the monitoring strategy for the regulated equipment, to iteratively verify the material defect classification results D within a fixed period T class (V feat ) quantity, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result.

2. The intelligent control method for waterproof material production and processing according to claim 1, characterized in that: S1 also includes S11 and S12; S11, by installing an industrial camera and a sensor group on the production line, real-time shooting of the surface of the waterproof material passing through the production line is performed, and image data of the surface of the waterproof material is collected in real time, and the image data at time t is obtained and recorded as an image matrix I(t), wherein the image matrix I(t) includes image pixel values ​​I i，j (t); S12, collect the surface state data of the waterproof material in real time through the sensor group installed on the production line, the sensor group includes a temperature sensor, an optical interference sensor and a stress sensor, and obtains the temperature value T of different areas of the material at time t x (t), surface roughness value S r (t) and material surface stress value Y ext (t); Synchronize, image matrix I(t), temperature value T x (t), surface roughness value S r (t) and material surface stress value Y ext (t) is integrated to obtain the feature vector V at acquisition time t cj (t).

3. The intelligent control method for waterproof material production and processing according to claim 1 is characterized in that: S2 also includes S21 and S22; S21, for the eigenvector V cj (t) Perform image preprocessing, including denoising preprocessing, contrast enhancement preprocessing, and edge detection preprocessing, and then use a convolutional neural network to extract features to generate a high-dimensional image feature set F inmg (t), including texture, edge shape, and edge inflection points; Among them, denoising preprocessing includes using median filtering and Gaussian filtering to remove noise in the image and remove the impact of optical interference in the production environment; contrast enhancement preprocessing includes using histogram equalization technology to adjust the contrast of the image to highlight tiny defects on the surface of the material; edge detection preprocessing includes using Canny and Sobel operators for edge detection to mark edge information of defects in the image.

4. The intelligent control method for waterproof material production and processing according to claim 3 is characterized by: S22, for the feature vector V cj (t) Non-image data, including temperature value T x (t), surface roughness value S r (t) and material surface stress value Y ext (t), non-linear processing is performed to form non-image data V s (t), adjust the response relationship to the material state, including using polynomial regression and neural network models to establish a nonlinear processing model to extract nonlinear relationships and compare them with the image feature set F inmg (t) is fused, and the fusion includes using splicing and weighted averaging to obtain the high-dimensional feature vector V at the acquisition time t. feat (t).

5. The intelligent control method for waterproof material production and processing according to claim 1 is characterized by: S3 specifically obtains the high-dimensional feature vector V feat (t) Establishing a classification model, including using CNN technology to establish a classification model, wherein the classification model includes a convolution layer, a pooling layer and a fully connected layer, and performing defect recognition and classification through the training of the classification model, including using a back propagation algorithm to continuously optimize the model parameters and minimize the loss function, so that the classification model can recognize and classify the defect types in the material, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat ); Among them, the defect classification result D class (V feat ) The recognition calculation output is obtained through S311 and S312.

6. The intelligent control method for waterproof material production and processing according to claim 5 is characterized by: S311. Obtain a high-dimensional feature vector V through the output probability vector of the convolutional neural network feat (t) The probability value p of the k-th defect after input into the CNN classification model k (V feat ); Among them, defects include bubbles, cracks, impurities, color difference, wrinkles, depressions, delamination and roughness anomalies, and the defects are marked with subscripts corresponding to bubbles, cracks, impurities, color difference, wrinkles, depressions, delamination and roughness anomalies, forming a defect type set D target ; The CNN classification model specifically includes the calculation of the S3111 convolutional layer, the calculation of the S3112 activation function and pooling layer, and the calculation of the S3113 fully connected layer; S3111, perform a convolution operation by combining each convolution kernel in the convolution layer with the input high-dimensional feature vector V feat (t) Perform convolution operation to obtain feature map Specifically, it represents the feature map value calculated by the qth convolution kernel at position (i, j) in the pth layer; S3112, execute the activation function, and convert the feature map after the convolution operation By using the activation function ReLU for nonlinear mapping, the activated feature map value is obtained The activation function ReLU synchronously sets negative values ​​to 0 to retain positive values; Perform pooling operations, including using maximum pooling and average pooling on the activated feature map values Downsample, adjust the size of the feature map and extract the display features to obtain the pooled feature values Specifically, by taking the activated feature map value Peak acquisition in the (i, j) area; S3113, perform full connection operation, and pool the feature values ​​of all positions Flatten to form a vector X p , synchronized as the input vector X of the fully connected layer p , and with the weight matrix W( fc ) and bias b( fc ) is fitted, and the output feature is obtained through the activation function ReLU: the high-dimensional feature vector V feat (t) The probability value p of the k-th defect after input into the CNN classification model k (V feat ); S312, by obtaining the probability value p of the k-th defect k (V feat ) to mark the defect type label and obtain the defect classification result label D corresponding to the defect type identified by the material class (V feat ).

7. The intelligent control method for waterproof material production and processing according to claim 1 is characterized by: S4 also includes S41 and S42; S41, using the multivariate linear regression model to classify the defect results D class (V feat ) and high-dimensional feature vector V feat (t) Fitting, including the defect classification result D class (V feat ) as the corresponding variable, and the high-dimensional feature vector V feat (t) is used as the independent variable and fitted through the linear regression model to obtain the defect classification result D after fitting. class (V feat ).

8. The intelligent control method for waterproof material production and processing according to claim 7 is characterized by: S42, defect classification result after fitting D class (V feat ) is combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal for regulating the waterproof material processing. Cctrl (V feat , S, P).

9. The intelligent control method for waterproof material production and processing according to claim 6, characterized in that: S5 also includes S51 and S52; S51, according to the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to specifically regulate the defective materials, and to obtain the production speed SE and equipment pressure PE after regulation; S52. Trigger the monitoring strategy for the control equipment and iteratively verify the material defect classification result D within a fixed period T. class (V feat ) quantity, obtain the defect quantity N of all defect types Dclass , and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result; The results of the iterative evaluation of the control are obtained by Matching method acquisition; When the return value is 1, the qualified result of the control iteration evaluation is obtained; When the return value is 0, the unqualified result of the control iteration evaluation is obtained, and S4 and S5 are executed iteratively.

10. An intelligent control system for waterproof material production and processing, comprising an intelligent control method for waterproof material production and processing according to any one of claims 1 to 9, characterized in that: It includes data acquisition module, data processing and extraction module, data modeling module, defect fitting module and feedback control module; The data acquisition module collects image data and surface status data of waterproof materials in real time by installing industrial cameras and sensor groups on the production line, and integrates the feature vector V at the acquisition time t. cj (t); The data processing and extraction module processes the feature vector V cj (t) Perform image preprocessing and image feature extraction to obtain an image feature set, and fuse it with the non-image data obtained after nonlinear function processing and the image feature set to obtain a high-dimensional feature vector V at acquisition time t feat (t); The data modeling module obtains the high-dimensional feature vector V feat (t) Establish a classification model, identify and classify defects through training of the classification model, obtain the output defect type label of the classification model, and reflect the defect classification result of the current inspection material through the defect classification label. class (V feat ); The defect fitting module is based on the defect classification result D class (V feat ) and high-dimensional feature vector V feat (t) is fitted and combined with the production speed S and equipment pressure P of the current material production and processing equipment to generate a feedback control signal C for regulating the waterproof material processing ctrl (V feat , S, P); The feedback control module is based on the feedback control signal C ctrl (V feat , S, P) to regulate the production speed S and equipment pressure P of the material processing production line, to regulate the defective materials specifically, and to trigger the monitoring strategy for the regulated equipment, to iteratively verify the material defect classification results D within a fixed period T class (V feat ) quantity, and matches it with the preset iterative verification evaluation threshold DZ to obtain the control iterative evaluation result.

Images