Automatic control system and method of calender

Abstract

The invention relates to the field of intelligent control of calendaring machines, and particularly discloses an automatic control system and method of a calendaring machine. The method comprises the following steps: firstly, acquiring a surface monitoring video of a roller in a to-be-monitored calender, temperature values of the roller in the to-be-monitored calender at a plurality of preset time points and product quality monitoring images of a to-be-monitored product at a plurality of preset time points; performing convolutional coding on the surface monitoring video, the temperature values at the plurality of preset time points and the product quality monitoring images at the plurality of preset time points through a convolutional neural network model in deep learning to obtain roller surface state feature vectors; the method comprises the following steps: acquiring a roller surface state feature vector, a roller temperature change feature vector and a product quality feature vector, and performing feature fusion on the roller surface state feature vector, the roller temperature change feature vector and the product quality feature vector to obtain a classification result used for representing whether the working state of the roller in the calender to be monitored needs to be adjusted or not.

Description

Technical Field

[0001] This application relates to the field of intelligent control of calenders, and more specifically, to an automated control system and method for calenders. Background Technology

[0002] A rolling mill is a type of machinery used in metal processing, primarily for rolling metal materials into products such as sheets, strips, and bars of various shapes and sizes. It works by passing the metal material through a series of rollers, causing it to undergo plastic deformation, thereby changing its cross-sectional shape and reducing its thickness. During the operation of the rolling mill, the working condition of the rollers affects the quality of the rolled products. For example, the rollers endure enormous pressure and friction during rolling, and will experience wear after prolonged use. Uneven roller wear can lead to inaccurate product dimensions or thickness. Roller wear and deformation can cause the gaps to widen or become uneven; incorrect roller gaps can result in uneven product thickness or inaccurate dimensions. Furthermore, in some special rolling applications, inaccurate roller temperature control can lead to poor product surface quality, such as burns or overheating.

[0003] Therefore, there is a need for an automated control system and method for a calender, based on deep learning technology, to maintain stable product quality by controlling the working state of the rollers in the calender (including roller wear, gap size, and surface temperature). Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an automated control system and method for a calender. First, it acquires surface monitoring video of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points, and product quality monitoring images of the product to be monitored at multiple predetermined time points. Then, it uses a convolutional neural network model in deep learning to perform convolutional encoding on the surface monitoring video, the temperature values ​​at the multiple predetermined time points, and the product quality monitoring images at the multiple predetermined time points to obtain roller surface state feature vectors, roller temperature change feature vectors, and product quality feature vectors. Finally, it fuses these feature vectors to obtain a classification result indicating whether the working state of the rollers in the calender to be monitored needs adjustment.

[0005] According to a first aspect of this application, an automated control system for a calender is provided, comprising:

[0006] The calender data acquisition module is used to acquire surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points in the calender to be monitored, and product quality monitoring images of the product to be monitored at multiple predetermined time points.

[0007] The calender data preprocessing module is used to preprocess the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor.

[0008] The calender data feature extraction module is used to extract feature information from the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor to obtain roller surface state feature vector, roller temperature change feature vector, and product quality feature vector.

[0009] The classification result generation module is used to obtain classification results based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

[0010] According to a second aspect of this application, an automated control method for a calendering mill is provided, comprising:

[0011] Collect surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points in the calender to be monitored, and product quality monitoring images of the product to be monitored at multiple predetermined time points;

[0012] The surface monitoring video, the temperature values ​​at the multiple predetermined time points, and the product quality monitoring images at the multiple predetermined time points are preprocessed to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor.

[0013] The feature information of the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor are extracted respectively to obtain the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector;

[0014] The classification result is obtained based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

[0015] Compared with the prior art, the automated control system and method for a calender provided in this application first acquires surface monitoring video of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points, and product quality monitoring images of the product to be monitored at multiple predetermined time points. Then, a convolutional neural network model in deep learning is used to perform convolutional encoding on the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points to obtain roller surface state feature vectors, roller temperature change feature vectors, and product quality feature vectors. Finally, the roller surface state feature vectors, roller temperature change feature vectors, and product quality feature vectors are fused to obtain a classification result indicating whether the working state of the rollers in the calender to be monitored needs to be adjusted. Attached Figure Description

[0016] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 The figure shows a schematic block diagram of an automated control system for a calender according to an embodiment of the present application.

[0018] Figure 2 The figure shows a schematic block diagram of a calender data preprocessing module in an automated control system for a calender according to an embodiment of the present application.

[0019] Figure 3 The figure shows a schematic block diagram of a video preprocessing unit in the data preprocessing module of a calender in an automated control system of a calender according to an embodiment of the present application.

[0020] Figure 4 The figure shows a schematic block diagram of a calender data feature extraction module in an automated control system for a calender according to an embodiment of the present application.

[0021] Figure 5 The figure shows a schematic block diagram of the target detection unit in the data feature extraction module of the calender in the automated control system of the calender according to an embodiment of the present application.

[0022] Figure 6 The figure shows a schematic block diagram of a classification result generation module in an automated control system for a calender according to an embodiment of the present application.

[0023] Figure 7The diagram illustrates the architecture of an automated control system for a calender according to an embodiment of this application.

[0024] Figure 8 The illustration shows a flowchart of an automated control method for a calender according to an embodiment of this application. Detailed Implementation

[0025] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0026] Exemplary System

[0027] Figure 1 The illustration shows a schematic block diagram of an automated control system for a calender according to an embodiment of this application. Figure 1 As shown, the automated control system 100 of the calender according to an embodiment of this application includes: a calender data acquisition module 110, used to acquire surface monitoring video of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points in the calender to be monitored, and product quality monitoring images of the product at multiple predetermined time points; a calender data preprocessing module 120, used to preprocess the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor; a calender data feature extraction module 130, used to extract feature information from the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor to obtain a roller surface state feature vector, a roller temperature change feature vector, and a product quality feature vector; and a classification result generation module 140, used to obtain a classification result based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

[0028] A rolling mill is a device used to roll metal materials into desired shapes and sizes, and the rollers are one of the core components of the rolling mill. Rollers play a crucial role in the operation of the rolling mill. Their main functions include: pressing and plastic deformation, controlling thickness and dimensions, improving material properties, and transmitting power.

[0029] As described in the background section, the working condition of the rollers during the operation of a calender affects the quality of the calendered product. For example, rollers endure enormous pressure and friction during calendering, and wear will occur after prolonged use. Uneven roller wear can lead to inaccurate product dimensions or uneven thickness. Roller wear and deformation may cause the gap to widen or become uneven; incorrect roller gaps can result in uneven product thickness or inaccurate dimensions. Furthermore, in some special calendering applications, inaccurate roller temperature control can lead to poor product surface quality, such as burns or overheating. Therefore, an automated control system and method for a calender, based on deep learning technology, is desired to maintain stable product quality by controlling the working condition of the rollers in the calender (including roller wear, gap size, and surface temperature).

[0030] Specifically, deep learning is a branch of machine learning that simulates the learning and reasoning processes of the human brain by building and training multi-layered neural networks. Deep learning technology offers advantages in image, data, and video processing, including automatic feature extraction, highly abstract representation, large-scale data processing, powerful generalization capabilities, multimodal data processing, and end-to-end learning. These advantages make deep learning a powerful tool for handling complex tasks and large-scale data, driving many important advancements in the fields of image, data, and video processing. Therefore, in embodiments of this application, deep learning technology is used to perform feature analysis on the working state of the rollers in a calender and the quality of the product.

[0031] In this embodiment, the calender data acquisition module 110 is used to acquire surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points, and product quality monitoring images of the product to be monitored at multiple predetermined time points. It should be understood that the rollers are key components in the calender used to apply pressure. The rollers contact the material and apply pressure, causing the material to undergo plastic deformation, thus changing its shape and size. Furthermore, the temperature of the rollers can affect the plastic deformation properties, surface quality, and dimensional stability of the material. That is, the surface quality and temperature of the rollers have a significant impact on the quality of the product and the processing effect. Therefore, it is necessary to control the working state of the rollers in order to maintain stable quality of the products processed by the calender. In addition, considering that whether the working state of the rollers needs adjustment can also be reflected in the quality of the product, the working state of the rollers can also be adjusted by acquiring monitoring images of the product.

[0032] More specifically, when acquiring monitoring video of the rollers in the calender under monitoring, devices such as cameras or sensors can be used. These acquisition devices can be installed at suitable locations near the rollers of the calender to acquire the working status of the rollers in real time. Simultaneously, temperature sensors can be installed on the rollers of the calender to acquire the temperature values ​​of the roller surface in real time. These temperature sensors can transmit temperature data to the monitoring system for recording and analysis via analog or digital signals. Furthermore, high-resolution cameras or sensors can be installed at appropriate locations on the calender to acquire image data of the product in real time.

[0033] In this embodiment, the calender data preprocessing module 120 is used to preprocess the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor. Considering that the surface monitoring video consists of a series of frames, it needs to be converted into a form suitable for deep learning model processing, such as tensor form, before processing it using deep learning technology. Similarly, the temperature values ​​at multiple predetermined time points are usually recorded in time series form, while the product quality monitoring images at multiple predetermined time points are usually multiple two-dimensional images. Before processing both using deep learning technology, they need to be converted into an input form suitable for model processing. Therefore, before performing deep learning-based data processing and feature analysis on the above input data, preprocessing operations are first required on the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points.

[0034] Figure 2 The illustration shows a schematic block diagram of a calender data preprocessing module in an automated control system for a calender according to an embodiment of this application. For example... Figure 2 As shown, the calender data preprocessing module 120 includes: a video preprocessing unit 121, used to collect key information from the surface monitoring video to obtain the three-dimensional roller surface state video tensor; a temperature value preprocessing unit 122, used to arrange the temperature values ​​at multiple predetermined time points into the one-dimensional roller surface temperature input vector; and an image preprocessing unit 123, used to arrange the product quality monitoring images at multiple predetermined time points into the three-dimensional product quality input tensor.

[0035] It is understood that the surface monitoring video may contain a large amount of redundant information, which may affect the accuracy of data processing and feature extraction. Therefore, in order to obtain key information from the surface monitoring video and improve the accuracy of data acquisition, key frames are first acquired from the surface monitoring video. This reduces the dimensionality of the video data and extracts useful information. Furthermore, considering that the acquired key frames are multiple two-dimensional image data, there are implicit temporal features among these data. Further, to better obtain the temporal features in the surface monitoring video, the acquired key frames need to be arranged according to the time dimension.

[0036] Figure 3 The illustration shows a schematic block diagram of a video preprocessing unit in the data preprocessing module of a calender in an automated control system according to an embodiment of this application. For example... Figure 3 As shown, the video preprocessing unit 121 includes: a video keyframe acquisition subunit 121-1, used to acquire multiple roller surface state video keyframes from the surface monitoring video; and a keyframe arrangement subunit 121-2, used to arrange the multiple roller surface state video keyframes into the three-dimensional roller surface state video tensor according to the time dimension.

[0037] Specifically, the video keyframe acquisition subunit 121-1 operates as follows: the surface monitoring video is loaded into the computer, and necessary video format conversion, decoding, and adjustment of video resolution, frame rate, etc. are performed; keyframes are extracted from the surface monitoring video using a keyframe extraction algorithm; and keyframes related to the roller surface state are selected based on the content and features of the keyframes.

[0038] In this embodiment, the calender data feature extraction module 130 is used to extract feature information from the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor to obtain roller surface state feature vectors, roller temperature change feature vectors, and product quality feature vectors. It should be understood that after preprocessing the input data, deep learning technology can be used to extract features from it. By extracting these feature vectors, we can better understand the roller surface state, temperature changes, and product quality, and use them for further analysis and decision-making.

[0039] Figure 4 The illustration shows a schematic block diagram of a calender data feature extraction module in an automated control system for a calender according to an embodiment of this application. For example... Figure 4As shown, the calender data feature extraction module 130 includes: a target detection unit 131, used to acquire feature information of the roller surface state in the three-dimensional roller surface state video tensor to obtain the roller surface state feature vector; a temperature feature extraction unit 132, used to pass the one-dimensional roller surface temperature input vector through a roller surface temperature feature extraction module based on a time encoder to obtain the roller temperature change feature vector; and a product quality feature extraction unit 133, used to pass the three-dimensional product quality input tensor through a convolutional neural network model as a product quality feature extractor to obtain the product quality feature vector.

[0040] It should be understood that the 3D roller surface state video tensor not only contains surface state information during roller operation but may also contain other information, such as the operating environment and noise. However, the goal of extracting roller surface state features is to extract key features of the roller surface. Therefore, a target detection network is used to extract key features of the roller from the 3D roller surface state video tensor to obtain a roller surface state feature map. The target detection network helps to locate and identify the roller surface. That is, through the target detection network, the roller surface can be accurately detected and located, ensuring that the feature extraction module extracts features from the correct area and avoids mis-extraction or omission of important surface state information.

[0041] Furthermore, considering that the roller surface state feature map is typically high-dimensional, containing a large amount of spatial information, it requires significant computational and storage space. Therefore, a pooling operation is performed on the roller surface state feature map. Pooling reduces the size of the roller surface state feature map, thereby reducing the dimensionality of the data. This helps reduce computational and storage requirements and simplifies subsequent feature extraction and analysis processes.

[0042] Figure 5 The illustration shows a schematic block diagram of the target detection unit in the data feature extraction module of a calender in an automated control system of a calender according to an embodiment of this application. For example... Figure 5 As shown, the target detection unit 131 includes: a target detection subunit 131-1, used to extract features from the three-dimensional roller surface state video tensor using a roller surface state feature extraction module based on a target detection network to obtain a roller surface state feature map; and a feature map dimensionality reduction subunit 131-2, used to perform pooling operations on the roller surface state feature map to obtain a roller surface state feature vector. The target detection network is an anchor window-based target detection network.

[0043] Specifically, the target detection subunit 131-1 is used to: process the three-dimensional roller surface state video tensor using the roller surface state feature extraction module based on the target detection network with the following target detection formula to obtain the roller surface state feature map; wherein, the target detection formula is:

[0044] ROI=H(ψ det ,B)=(cls(ψ det ,B),Regr(ψ) det ,B))

[0045] Where, ψ det Let B be the 3D video tensor of the roller surface state, B be the anchor frame, ROI be the feature map of the roller surface state, and cls(ψ) be the ROI. det B) represents the classification, Regr(ψ) det B) represents regression.

[0046] Next, considering that the roller surface temperature is a time-varying sequence of data, containing temporal correlations and dependencies, a time-dependent feature extraction module based on a time encoder is used to further extract implicit correlation features from the one-dimensional roller surface temperature input vector in order to better extract the temporal dependency features. In this way, by using the time encoder-based temperature feature extraction module, the temporal patterns and trends in the temperature sequence can be better captured, thereby extracting important features of roller temperature changes.

[0047] In one specific embodiment of this application, the temperature feature extraction unit 132 is configured to: use the fully connected layer of the timing encoder to perform fully connected encoding on the one-dimensional roller surface temperature input vector using the following formula to extract high-dimensional hidden features of the feature values ​​at each position in the one-dimensional roller surface temperature input vector, wherein the formula is: Where X is the one-dimensional roller surface temperature input vector, Y is the output vector, W is the weight matrix, and B is the bias vector. This represents matrix multiplication; and a one-dimensional convolutional layer using a temporal encoder performs one-dimensional convolutional encoding on the one-dimensional roller surface temperature input vector according to the following formula to extract high-dimensional implicit correlation features between feature values ​​at various positions in the one-dimensional roller surface temperature input vector, wherein the formula is:

[0048]

[0049] Where a is the width of the first convolution kernel in the x direction, F(a) is the parameter vector of the first convolution kernel, G(xa) is the local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, X represents the one-dimensional roller surface temperature input vector, and Cov(X) represents the one-dimensional convolution encoding of the one-dimensional roller surface temperature input vector.

[0050] It is worth mentioning that, besides the time-encoder-based method, there is another possible implementation for extracting the roller surface temperature feature vector. This alternative implementation involves: preprocessing the one-dimensional roller surface temperature sequence as needed, which may include noise removal, data smoothing, and filling in missing values; extracting features from the roller surface temperature sequence using statistical feature extraction methods, such as mean, variance, maximum, minimum, slope, and energy; dividing the roller surface temperature sequence into multiple time windows, each containing continuous temperature values, with the window size selectable based on specific circumstances, such as a fixed window size or an adaptive window size based on the data; applying statistical features to the temperature values ​​within each time window, for example, calculating the mean, variance, and other statistical features of the temperature values ​​within the window; and combining the statistical features of each time window into a feature vector, which can be formed by arranging each statistical feature sequentially.

[0051] Furthermore, it should be understood that the three-dimensional product quality input tensor contains the shape, structure, and geometric information of the object in three-dimensional space. Based on this, when extracting features from it, convolutional neural network models, which have excellent performance in the image domain, can be used for convolutional encoding as product quality feature extractors. Through the convolutional neural network model as a product quality feature extractor, spatial information related to product quality can be extracted, enhancing the model's invariance and robustness, reducing dimensionality and extracting abstract features, and learning feature representations suitable for product quality analysis tasks. This results in more compact and representative product quality feature vectors, providing more effective input for subsequent analysis and decision-making.

[0052] In one specific embodiment of this application, the product quality feature extraction unit 133 is configured to: use each layer of the convolutional neural network model to perform the following operations on the input data during the forward propagation of the layers: perform convolution processing on the input data based on the convolution kernel to generate a convolutional feature map; perform global mean pooling processing on the convolutional feature map based on the feature matrix to generate a pooled feature map; and perform nonlinear activation on the feature values ​​at each position in the pooled feature map to generate an activation feature map; wherein, the output of the last layer of the convolutional neural network model is the product quality feature vector, the input of the second to last layers of the convolutional neural network model is the output of the previous layer, and the input of the convolutional neural network model is the three-dimensional product quality input tensor.

[0053] In this embodiment of the application, the classification result generation module 140 is used to obtain a classification result based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

[0054] Figure 6 The illustration shows a schematic block diagram of a classification result generation module in the automated control system of a calender according to an embodiment of this application. Figure 6 As shown, the classification result generation module 140 includes: a roller feature combination unit 141, used to fuse the roller surface state feature vector and the roller temperature change feature vector to obtain a roller working state feature vector; a calender feature fusion unit 142, used to fuse the roller working state feature vector and the product quality feature vector to obtain a calender classification feature vector; a probabilistic unit 143, used to input the calender classification feature vector into a Sigmoid function to obtain a probabilistic calender classification feature vector; a calender feature optimization unit 144, used to perform parametric geometric relationship transition prior feature rigidification on the probabilistic calender classification feature vector to obtain an optimized calender classification feature vector; and a classification unit 145, used to perform feature classification on the optimized calender classification feature vector through a classifier to obtain a classification result indicating whether the working state of the rollers in the calender to be monitored needs to be adjusted.

[0055] It is understandable that roller surface condition, temperature changes, and product quality are important factors affecting roller performance and quality. Combining these different types of feature vectors can integrate multiple information sources to more comprehensively describe the roller's condition and quality characteristics. By fusing different types of feature vectors together, more accurate and comprehensive input can be provided, which helps to better determine and classify the roller's condition.

[0056] In one specific embodiment of this application, the roller feature combining unit 141 is configured to: use a cascaded function to fuse the roller surface state feature vector and the roller temperature change feature vector to obtain the roller working state feature vector, wherein the cascaded function is expressed by the formula:

[0057] f(X i ,X j = Relu(W) f [θ(X i ),φ(X j )])

[0058] Among them, W f ,θ(X i ) and φ(X j ) indicates pointwise convolution of the input, ReLU is the activation function, [] indicates concatenation operation, X i X represents the feature value at each position in the feature vector of the roller surface state. j This represents the feature value at each position in the feature vector of the roller temperature change.

[0059] In one specific embodiment of this application, the calender feature fusion unit 142 is used to: fuse the features of the roller working state feature vector and the product quality feature vector through a weighted summation formula to obtain the calender classification feature vector, wherein the weighted summation formula is: f0 = αf + βf1, where f0 is the calender classification feature vector, f is the roller working state feature vector, f1 is the product quality feature vector, "+" indicates that the elements at corresponding positions of the roller working state feature vector and the product quality feature vector are added together, and α and β are weighting parameters used to control the balance between the roller working state feature vector and the product quality feature vector in the calender classification feature vector.

[0060] Specifically, in the technical solution of this application, it is considered that during the data acquisition process, the uneven sampling of key frames in the roller surface monitoring video or the discontinuous acquisition time points of temperature values ​​may lead to an unbalanced distribution of feature sampling. This unbalanced distribution may result in a lack of constraints on the probability distribution of feature values, with some values ​​appearing too frequently or too infrequently. Due to the discontinuity of features, data at some time points may be missing or unavailable; for example, the intervals in temperature value monitoring may result in missing temperature values ​​at certain time points. Missing data may lead to uncertainty in the probability distribution of feature values, resulting in a lack of constraints on the feature values ​​in the classification feature matrix. Noise or errors may be introduced during feature extraction, especially in the extraction of roller surface state features and temperature change features. These noises and errors may cause the probability distribution of feature values ​​to be unstable, exhibiting large fluctuations or outliers, thus leading to a lack of constraints on feature values. The lack of constraints on the probability distribution of features may affect the training and decision-making process of the classification model. For example, during model training, the model may have difficulty distinguishing which feature values ​​are important for the classification task, thus affecting the learning of feature weights. During the decision-making process, the model may not be able to make reliable classification judgments accurately based on the probability distribution of feature values, leading to a decrease in the accuracy and reliability of the classification results.

[0061] To address this issue, the technical solution of this application involves rigidly unifying the parameterized geometric relationship transition prior features of the probabilistic rolling mill classification feature vector to impose bidirectional constraints on the information entropy dimension of each feature value sample in the probabilistic rolling mill classification feature vector.

[0062] Specifically, firstly, a transitional prior feature is established by parameterizing the geometric relationships of feature value samples at various positions in the classification feature vector of the rolling mill. This feature reflects the distribution and variation of feature value samples across different categories. Then, a rigid uniformity mechanism can be used to adjust and optimize the transitional prior feature, making it more consistent with the inherent structure and logic of the data. This allows for a two-way constraint on the feature value samples in terms of information entropy, maintaining both high entropy within each category and low entropy between categories, thereby improving the discriminative and expressive power of the feature value samples.

[0063] In one specific embodiment of this application, the calender feature optimization unit 144 is used to: rigidly unify the parameterized geometric relationship transition prior features of the probabilistic calender classification feature vector using the following formula to obtain the optimized calender classification feature vector; wherein, the formula is:

[0064]

[0065] Among them, V ijLet λ represent the feature value at position (i,j) in the probabilistic calender classification feature vector, λ represent the predetermined hyperparameter, log represent the logarithmic function value to the base 2, and V represent the feature value. ij ′ represents the feature value at position (i,j) in the optimized calender classification feature vector.

[0066] Furthermore, considering that the classifier can quickly process input data and provide classification results, it has a high processing speed. Therefore, the classifier is used to perform feature classification on the optimized calender classification feature vector to obtain the classification result. Further, by performing feature classification on the optimized calender classification feature vector using the classifier, the optimized calender classification feature vector can be mapped to two different classification labels: the working state of the rollers in the calender to be monitored needs adjustment, and the working state of the rollers in the calender to be monitored does not need adjustment. In this way, by controlling the working state of the rollers, the processing quality and efficiency of the insulating material can be ensured.

[0067] In one specific embodiment of this application, the classification unit 145 is used to: pass the optimized calender classification feature vector through a classifier to obtain a classification result, the classification result being used to indicate whether the working state of the rollers in the calender to be monitored needs to be adjusted.

[0068] Furthermore, Figure 7 The diagram illustrates the architecture of an automated control system for a calender according to an embodiment of this application. Figure 7As shown, firstly, surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points, and product quality monitoring images of the product to be monitored at multiple predetermined time points are acquired. Next, multiple keyframes of the roller surface state video are obtained from the surface monitoring videos of the rollers in the calender to be monitored. These keyframes are arranged into a three-dimensional roller surface state video tensor and then processed by a tube surface state feature extraction module based on a target detection network to obtain a roller surface state feature map. Then, the roller surface state feature map is subjected to global mean pooling to obtain a roller surface state feature vector. Simultaneously, the temperature values ​​at the multiple predetermined time points are arranged according to the time dimension into a one-dimensional roller surface temperature input vector and then processed by a roller surface temperature feature extraction module based on a time encoder to obtain a roller temperature change feature vector. Finally, the roller surface state feature vector and the roller temperature change feature vector are fused to obtain a roller operating state feature vector. Furthermore, the product quality monitoring images at multiple predetermined time points are arranged into a three-dimensional product quality input tensor and then processed by a convolutional neural network model as a product quality feature extractor to obtain a product quality feature vector. Next, the roller operating state feature vector and the product quality feature vector are fused to obtain a calender classification feature vector. Finally, the calender classification feature vector is processed by a classifier to obtain a classification result, which indicates whether the operating state of the rollers in the calender under monitoring needs adjustment.

[0069] In summary, the automated control system 100 of the calender according to the embodiments of this application is explained. It first acquires surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points, and product quality monitoring images of the product to be monitored at multiple predetermined time points. Then, it performs convolutional encoding on the surface monitoring videos, the temperature values ​​at the multiple predetermined time points, and the product quality monitoring images at the multiple predetermined time points using a convolutional neural network model in deep learning to obtain roller surface state feature vectors, roller temperature change feature vectors, and product quality feature vectors. Finally, it fuses the roller surface state feature vectors, the roller temperature change feature vectors, and the product quality feature vectors to obtain a classification result indicating whether the working state of the rollers in the calender to be monitored needs adjustment.

[0070] As described above, the automated control system 100 for the calender according to the embodiments of this application can be implemented in various terminal systems, such as servers deploying automated control algorithms for the calender. In one example, the automated control system 100 for the calender can be integrated into the terminal system as a software module and / or a hardware module. For example, the automated control system 100 for the calender can be a software module in the operating system of the terminal system, or it can be an application developed for the terminal system; of course, the automated control system 100 for the calender can also be one of many hardware modules of the terminal system.

[0071] Alternatively, in another example, the automatic control system 100 of the calender and the terminal system can also be separate systems, and the automatic control system 100 of the calender can be connected to the terminal system via wired and / or wireless networks and transmit interactive information in accordance with an agreed data format.

[0072] Exemplary methods

[0073] Figure 8 The illustration shows a flowchart of an automated control method for a calender according to an embodiment of this application. For example... Figure 8 As shown, the automated control method for a calender according to an embodiment of this application includes: S110, acquiring surface monitoring video of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points, and product quality monitoring images of the product to be monitored at multiple predetermined time points; S120, preprocessing the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor; S130, extracting feature information from the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor to obtain a roller surface state feature vector, a roller temperature change feature vector, and a product quality feature vector; S140, obtaining a classification result based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

[0074] Here, those skilled in the art will understand that the specific functions and operations of each step in the above-described automated control method for the calender have been referenced above. Figure 1 The description of the automated control system of the calender is detailed here, and therefore, its repeated description will be omitted.

[0075] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0076] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, apparatuses, or units, or they may be electrical, mechanical, or other forms of connection.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0078] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware, or, as clearly understood by those skilled in the art through the above description of the embodiments, the present invention can be implemented in hardware, firmware, or a combination thereof. When implemented in software, the above functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that is accessible to a computer. For example, but not limited to, computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer. Furthermore, any connection can suitably be a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the fixing of the medium.

[0079] In summary, the above description is merely a preferred embodiment of the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automated control system for a calender, characterized in that, include: The calender data acquisition module is used to acquire surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points in the calender to be monitored, and product quality monitoring images of the product to be monitored at multiple predetermined time points. The calender data preprocessing module is used to preprocess the surface monitoring video, the temperature values ​​at multiple predetermined time points, and the product quality monitoring images at multiple predetermined time points to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor. The calender data feature extraction module is used to extract feature information from the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor to obtain roller surface state feature vector, roller temperature change feature vector, and product quality feature vector. The classification result generation module is used to obtain classification results based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

2. The automated control system for the calender according to claim 1, characterized in that, The calender data preprocessing module includes: A video preprocessing unit is used to collect key information from the surface monitoring video to obtain the three-dimensional roller surface state video tensor. A temperature value preprocessing unit is used to arrange the temperature values ​​at the multiple predetermined time points into a one-dimensional roller surface temperature input vector. An image preprocessing unit is used to arrange the product quality monitoring images at the multiple predetermined time points into the three-dimensional product quality input tensor.

3. The automated control system for the calender according to claim 2, characterized in that, The video preprocessing unit includes: The video keyframe acquisition subunit is used to acquire multiple keyframes of the roller surface status video from the surface monitoring video. The keyframe arrangement subunit is used to arrange the multiple roller surface state video keyframes into the three-dimensional roller surface state video tensor according to the time dimension.

4. The automated control system for the calender according to claim 3, characterized in that, The calender data feature extraction module includes: The target detection unit is used to obtain the feature information of the roller surface state in the three-dimensional roller surface state video tensor to obtain the roller surface state feature vector; A temperature feature extraction unit is used to extract the roller temperature change feature vector by passing the one-dimensional roller surface temperature input vector through a roller surface temperature feature extraction module based on a time encoder. The product quality feature extraction unit is used to pass the three-dimensional product quality input tensor through a convolutional neural network model, which serves as a product quality feature extractor, to obtain the product quality feature vector.

5. The automated control system for the calender according to claim 4, characterized in that, The target detection unit includes: The target detection subunit is used to extract features from the three-dimensional roller surface state video tensor through the roller surface state feature extraction module based on the target detection network to obtain a roller surface state feature map. The feature map dimensionality reduction sub-unit is used to perform pooling operations on the roller surface state feature map to obtain the roller surface state feature vector.

6. The automated control system for the calender according to claim 5, characterized in that, The target detection network is an anchor window-based target detection network.

7. The automated control system for the calender according to claim 6, characterized in that, The temporal encoder comprises a fully connected layer and a one-dimensional convolutional layer.

8. The automated control system for the calender according to claim 7, characterized in that, The classification result generation module includes: A roller feature combining unit is used to fuse the roller surface state feature vector and the roller temperature change feature vector to obtain a roller working state feature vector; The calender feature fusion unit is used to fuse the roller working state feature vector and the product quality feature vector to obtain the calender classification feature vector; A probabilistic unit is used to input the calender classification feature vector into the Sigmoid function to obtain a probabilistic calender classification feature vector. The calender feature optimization unit is used to perform parametric geometric relationship transition prior feature rigidification on the probabilistic calender classification feature vector to obtain an optimized calender classification feature vector; The classification unit is used to perform feature classification on the optimized calender classification feature vector through a classifier to obtain a classification result indicating whether the working state of the rollers in the calender to be monitored needs to be adjusted.

9. The automated control system for the calender according to claim 8, characterized in that, The calender feature optimization unit is used to: rigidly unify the parameterized geometric relationship transition prior features of the probabilistic calender classification feature vector using the following formula to obtain the optimized calender classification feature vector; The formula is as follows: Among them, V ij Let λ represent the feature value at position (i,j) in the probabilistic calender classification feature vector, λ represent the predetermined hyperparameter, log represent the logarithmic function value to the base 2, and V represent the feature value. ij ′ represents the feature value at position (i,j) in the optimized calender classification feature vector.

10. An automated control method for a calender, characterized in that, include: Collect surface monitoring videos of the rollers in the calender to be monitored, temperature values ​​of the rollers at multiple predetermined time points in the calender to be monitored, and product quality monitoring images of the product to be monitored at multiple predetermined time points; The surface monitoring video, the temperature values ​​at the multiple predetermined time points, and the product quality monitoring images at the multiple predetermined time points are preprocessed to obtain a three-dimensional roller surface state video tensor, a one-dimensional roller surface temperature input vector, and a three-dimensional product quality input tensor. The feature information of the three-dimensional roller surface state video tensor, the one-dimensional roller surface temperature input vector, and the three-dimensional product quality input tensor are extracted respectively to obtain the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector; The classification result is obtained based on the roller surface state feature vector, the roller temperature change feature vector, and the product quality feature vector.

Images