A plate surface cleaning control method and system based on segmented speed regulation

Abstract

The present invention discloses a plate surface cleaning control method and system based on segmented speed regulation. The method comprises: S1, performing trial cleaning on the plate at different conveying speeds, collecting images to establish a data set D1; S2, segmenting data set D1 to obtain strip area images; identifying, labeling, and processing dark spots in the strip area images to form data sets D2 and D3; S3, establishing a detection and recognition model for dark spots on the plate surface and training and deploying it using D2; S4, establishing an index prediction model for dark spots on the plate surface after cleaning and training and deploying it using D3; S5, collecting images of the plate to be cleaned, identifying and obtaining dark spot area information, inputting the index prediction model for dark spots on the cleaned surface, obtaining dark spot index prediction data, and calculating the ideal speed. By combining a segmented speed regulation strategy with a deep learning model, the present invention achieves efficient, low-cost, and highly consistent control of wide plate surface cleaning.

Description

Technical Field

[0001] The present invention belongs to the field of automatic control technology, and in particular relates to a control method and control system for performing segmented real-time adjustment of the conveying speed during surface cleaning of wide plates in a high-pressure water jet surface cleaning device for plates. Background Art

[0002] The cleaning station is the core component of a high-pressure abrasive water jet surface cleaning device. When using a high-pressure abrasive water jet surface cleaning device to clean the surface of a metal strip, a conveyor drives the strip through the cleaning station. The jet nozzles in the cleaning station use a mixture of high-pressure water and abrasive to impact the surface of the strip, removing oxidized impurities on the strip surface and meeting subsequent processing requirements. Metal strips are relatively narrow, and a high-pressure abrasive water jet cleaning device only contains one cleaning station. All jet nozzles are arranged in a single jet box. By setting an open-loop constant conveying speed or closed-loop speed control, the strip is conveyed through the cleaning station to complete the surface cleaning of the steel strip.

[0003] Unlike surface cleaning devices for narrower strips, the scale of water jet surface cleaning devices for cleaning the surfaces of wider metal plates will be significantly increased. The device will contain multiple cleaning stations, each of which will be equipped with a set of nozzles responsible for cleaning only a specific strip area along the conveying direction of the steel plate. The steel plate will not be cleaned until it passes through all cleaning stations in sequence (the steel plate will not be conveyed to the next cleaning station until it has completed cleaning at the previous cleaning station). In this case, if a single constant speed is used to convey the steel plate through all cleaning stations in sequence, the conveying speed setting will tend to be conservative to ensure the consistency of the cleaning effect of all strip areas in the steel plate, resulting in excessively long operation time. This not only increases unnecessary electricity consumption, but also increases the burden on the cleaning equipment itself, and aggravates the potential failure of components such as the jet nozzles and high-pressure pipelines in the cleaning device. The patent with publication number CN118060249A adopts the method of photographing the strip and adjusting the speed of the strip cleaning in real time according to the image processing results. However, this method is not suitable for speed control of wide plates. Compared with the strip, the wide plate is wider and shorter. The cleaning may have been completed when the conveyor belt is used for post-camera collection, and speed control cannot be implemented. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to propose a staged speed control method to address the problems of energy consumption, potential equipment failures and the inapplicability of real-time speed regulation for strip cleaning faced by the constant conveying speed control method when cleaning wide plates. The method controls the speed of the plate passing through each cleaning machine in a manner that predicts the surface cleaning quality of the cleaned plate, so as to achieve the goal of balancing cleaning quality and energy consumption when cleaning the surface of wide plates.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0006] A plate surface cleaning control method based on segmented speed regulation includes the following steps:

[0007] S1, using a cleaning device to test-clean several plates at different conveying speeds, collecting surface images of the plates before and after the test-cleaning, and establishing an unstructured dataset D1;

[0008] S2. Divide the surface images of the plate before and after the test wash in the unstructured dataset D1 into a plurality of portions along the width direction of the plate according to the number of cleaning stations in the cleaning device, thereby obtaining a plurality of strip area images of the plate before cleaning and a plurality of strip area images of the plate after cleaning; and identify, mark, and process dark spots in the plurality of strip area images of the plate before cleaning and the plurality of strip area images of the plate after cleaning to form an image structured dataset D2 and a numerical structured dataset D3.

[0009] S3. Establish a detection and recognition model for dark spots on the surface of plate materials, and use the structured image dataset D2 to train and deploy the detection and recognition model for dark spots on the surface of plate materials;

[0010] S4. Establish an index prediction model for dark spots on the surface of the plate after cleaning, and use the numerical structured dataset D3 to train and deploy the index prediction model for dark spots on the surface of the plate after cleaning;

[0011] S5. Collect the image of the plate to be cleaned before cleaning at the set conveying speed, input the image before cleaning into the detection and recognition model of dark spots on the plate surface, identify and obtain the dark spot area information in the image before cleaning, input the dark spot area information into the index prediction model of the dark spots on the surface after cleaning, obtain the dark spot index prediction data after cleaning, calculate the ideal speed of the plate passing through each cleaning position based on the dark spot index prediction data after cleaning, and control the plate to be cleaned to be cleaned at the ideal speed.

[0012] The present invention is a cleaning method for wide plates with multiple cleaning positions. It segments and identifies the cleaning effect images of the plates according to the distribution of the cleaning positions in the cleaning device, and constructs corresponding dark spot detection and recognition models and dark spot index prediction models. The above models are trained through the effect images and data of several trial washes. Through the collected images of the plates to be cleaned, the ideal speed of each cleaning position can be directly calculated and output according to the detection, recognition and prediction results, thereby performing segmented control on the plate conveying speed. On the one hand, it can accurately ensure that each cleaning position can play the best effect and obtain the expected cleaning effect. On the other hand, it avoids the conservative unified low speed setting and reduces the energy consumption of the equipment.

[0013] As a further optimization of the above technical solution, in S1, when a plurality of plates are tested at different conveying speeds, the conveying speed range of the plates is set. and speed growth step ,from Start cleaning n plates, then increase the speed , and then clean n pieces of plate until the speed is completed Cleaning of the next n plates.

[0014] As a further preferred embodiment of the above technical solution, in S2, the number of segmentations of the surface image of the plate before and after the trial washing corresponds to the number of cleaning stations in the cleaning device.

[0015] As a further optimization of the above technical solution, in S2, the image structured dataset D2 includes a strip area image of the plate before cleaning with dark spots marked and an image of the strip area after cleaning with dark spots marked; the numerical structured dataset D3 includes the mean dark spot area, the maximum dark spot area, the plate conveying speed, and the mean dark spot area, the maximum dark spot area of ​​the strip area image after cleaning.

[0016] As a further optimization of the above technical solution, in S3, a detection and recognition model for dark spots on the surface of plate materials is established based on the single-stage target detection network YOLOv8n, the image structured dataset D2 is divided into a training set, a validation set and a test set, and the training set is used to train YOLOv8n. During the training process, the validation set is used to evaluate the model indicators, and the hyperparameters are adjusted as needed (the input resolution (imgsz), data enhancement (hsv), learning rate (lr0), loss weight (cls) and other hyperparameters can be adjusted based on the performance of the model evaluated by the validation set, such as the loss curve, mAP and other indicators). When the performance of the detection and recognition model for dark spots on the surface of plate materials on the validation set meets the preset standards, an independent test set is used for a one-time test. When the test results meet the requirements, the model training is terminated and a deployable detection and recognition model for dark spots on the surface of plate materials is generated.

[0017] As a further optimization of the above technical solution, during the training of YOLOv8n, a composite loss function is used for model training. The composite loss function expression is:

[0018] ;

[0019] In the formula 、 and is the weight coefficient, adjusted by experience, is the confidence loss, It is the loss of identification and positioning of oxide dark spots, It is the envelope area loss of the oxide dark spot recognition annotation box; the oxide dark spot recognition annotation box refers to the dark spot area reference box generated manually or by tools during the training process to guide model training and learning; the oxide dark spot recognition prediction box refers to the dark spot recognition box output by YOLOv8n, which includes the predicted position and confidence level;

[0020] in,

[0021] Where, It is the intersection-over-union ratio of the oxide dark spot recognition prediction box and the oxide dark spot true annotation box; the oxide dark spot true annotation box refers to the standard box of the dark spot in the dataset, which is used for IoU and area loss calculation;

[0022] Where, is the true value, is the predicted probability, N is the sample size;

[0023] Where, represents the area of ​​the oxide dark spot recognition prediction box, Represents the area of ​​the oxide dark spot annotation box. Using a composite loss function that combines positioning error (IoU), envelope area error, and confidence error to train the model improves positioning accuracy and enhances robustness.

[0024] As a further optimization of the above technical solution, when the model is evaluated using the validation set, the indicators include the accuracy, recall rate and F1 score of YOLOv8n; the hyperparameters include the initial learning rate, the learning rate decay factor and the number of training samples per batch.

[0025] As a further optimization of the above technical solution, in S4, an index prediction model for dark spots on the surface of the plate after cleaning is established based on the MLP multi-layer network model, the numerical structured data set D3 is divided into a training set, a validation set and a test set, and the MLP multi-layer network model is trained using the training set. After the index evaluation of the validation set and the generalization ability test of the test set, the model training is terminated and a deployable index prediction model for dark spots on the surface after cleaning is generated.

[0026] As a further optimization of the above technical solution, when using the training set to train the MLP multi-layer network model, the feature vector of the strip area before the plate is cleaned is used. As input value, the feature vector of the strip area after the plate is cleaned As the output value, multi-layer fully connected neurons are used as hidden layers, ReLU function is used as activation function, and mean square error is used as loss function. The gradient is calculated using the backpropagation algorithm, and the Adam optimizer is used for parameter updates. Using the mean squared error (MSE) as the loss function minimizes the error between the predicted value and the true value, and using the Adam optimizer for parameter updates ensures rapid model convergence.

[0027] As a further optimization of the above technical solution, the feature vector Where, is the average area of ​​dark spots in the strip area before the plate is cleaned, It is the maximum dark spot area in the strip area before the plate is cleaned. is the sheet conveying speed; the eigenvector Where, is the average area of ​​dark spots in the strip area after the plate is cleaned, It is the maximum area of ​​dark spots in the strip area after the plate is cleaned.

[0028] As a further optimization of the above technical solution, the loss function Where, 、 It is the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area in the strip area after the plate is cleaned.

[0029] As a further optimization of the above technical solution, in S5, after collecting the image of the plate to be cleaned before cleaning at the set conveying speed, the image before cleaning is divided into several rectangular strips according to the method of S2, and then each of the rectangular strips is input into the detection and recognition model of dark spots on the surface of the plate to identify and obtain the dark spot area information in the image before cleaning, and the dark spot area information includes the mean dark spot area, the maximum dark spot area and the conveying speed of the plate.

[0030] As a further optimization of the above technical solution, the dark spot area information is input into the index prediction model of the dark spots on the surface after cleaning to obtain the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area. The ideal speed of the plate passing through each cleaning position is calculated in combination with the conveying speed of the plate and the expected cleaning effect.

[0031] As a further optimization of the above technical solution, the ideal speed The calculation method is:

[0032] Where, is the conveying speed of the plate, is the predicted value of the mean dark spot area, is the predicted value of the maximum dark spot area, is the expected value of the mean dark spot area, is the expected value of the maximum dark spot area, which is determined by operating experience and actual needs. and is the weight coefficient, which is adjusted by experience (based on the analysis of image data, the variation range of the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area, combined with the speed adjustment formula, first determine that the adjusted speed is within a reasonable process range, and then consider the contribution rate of the mean dark spot area and the maximum dark spot area respectively).

[0033] Based on the same technical concept, the present invention also provides a plate surface cleaning control system based on segmented speed regulation, including a visual subsystem, a plate surface dark spot detection and recognition subsystem, a plate surface dark spot index prediction subsystem, a speed automatic control subsystem and a calculation and processing device;

[0034] The visual subsystem includes a front camera assembly arranged in front of the first cleaning position of the cleaning device and a rear camera assembly arranged behind the last cleaning position of the cleaning device, for acquiring a surface image of the plate;

[0035] The plate surface dark spot detection and recognition subsystem is used to generate and deploy a plate surface dark spot detection and recognition model, which is used to recognize the plate surface image obtained by the visual subsystem and output dark spot area information in the surface image, including the dark spot area mean and dark spot area maximum;

[0036] The plate surface dark spot index prediction subsystem is used to generate and deploy an index prediction model for dark spots on the plate surface after cleaning. The model is used to process the acquired dark spot area information. The input is the dark spot area mean, the dark spot area maximum, and the plate conveying speed. The output is the predicted value of the dark spot area mean and the predicted value of the dark spot area maximum after cleaning.

[0037] The speed automatic control subsystem is equipped with a detection and recognition model for dark spots on the plate surface and an index prediction model for dark spots on the plate surface after cleaning, which is used to calculate the ideal speed of the plate passing through each cleaning station and control the conveying speed of the plate in the cleaning device;

[0038] The computing and processing device consists of a central processing unit, memory, graphics card and storage, and is used to realize the storage and processing of data sets, the training and deployment of the plate surface dark spot detection and identification subsystem, the training and deployment of the plate surface dark spot index prediction subsystem and the related computing and processing functions of the speed automatic control subsystem.

[0039] As a further preference of the above technical solution, it also includes a plate position detection device, which includes a number of sensors arranged at the entrance of each cleaning position of the cleaning device, for determining the cleaning position where the plate is located and sending a signal to the speed automatic control subsystem, to assist the speed automatic control subsystem in adjusting and controlling the conveying speed of the plate.

[0040] Compared with the prior art, the advantages of the present invention are:

[0041] This invention achieves efficient, low-energy, and highly consistent control of wide plate surface cleaning by combining a phased speed regulation strategy with a deep learning model. Using the plate's surface dark spot index and conveyor speed before cleaning, the system predicts the plate's surface dark spot index after passing through the current cleaning station. This serves as the basis for calculation and control decisions, dynamically adjusting the conveyor speed of each station. This avoids the energy waste associated with traditional constant low-speed modes, significantly reducing equipment operating energy consumption and unnecessary electrical energy consumption while ensuring that cleaning quality meets standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 This is a flow chart of the plate surface cleaning control method based on segmented speed regulation of Example 1;

[0044] Figure 2 This is a schematic structural diagram of a plate surface cleaning control system based on segmented speed regulation according to Example 2;

[0045] Figure 3 This is a schematic diagram of the arrangement of the plate surface cleaning control system based on segmented speed regulation on the cleaning device of Example 2;

[0046] Figure 4 This is a schematic diagram of the segmentation of the steel plate image in the plate surface cleaning control method based on segmented speed regulation in Example 1. DETAILED DESCRIPTION

[0047] To facilitate understanding of the present invention, the present invention will be described in more comprehensive and detailed form below in conjunction with the accompanying drawings and preferred embodiments. However, the protection scope of the present invention is not limited to the following specific embodiments.

[0048] Unless otherwise defined, all technical terms used hereinafter have the same meanings as those generally understood by those skilled in the art. The technical terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the scope of protection of the present invention.

[0049] Unless otherwise specified, various instruments and equipment used in the present invention can be purchased from the market or obtained through existing methods.

[0050] Example 1:

[0051] The plate surface cleaning control method based on segmented speed regulation of this embodiment is used to clean the steel plate, such as Figure 1 As shown, it includes the following steps:

[0052] S1, test cleaning the steel plate at different conveying speeds, collect surface images of the steel plate before and after cleaning, and establish an unstructured data set D1. The specific process is: according to the operating experience of the cleaning device, set the range of steel plate conveying speed and speed increase step length , from the lower speed limit Start cleaning the wide steel plates, complete the cleaning of n steel plates, and then increase the speed , and then complete the cleaning of n steel plates until the speed is completed The cleaning of n steel plates under the following conditions is carried out. During the trial cleaning process of each steel plate, the speed of each cleaning position is constant. During this process, a high-resolution camera is used to shoot the steel plates before and after cleaning to establish an unstructured dataset D1 for steel plate surface cleaning. Each data record of the dataset consists of three elements: the image of each steel plate before cleaning, the image after cleaning, and the conveying speed of the corresponding steel plate. Specifically, in this embodiment, the range of the steel plate conveying speed determined by experience is 6 meters / minute to 18 meters / minute, and the speed increase step is set to 0.2 meters / minute. Therefore, the speed is divided into 60 levels. Starting from a speed of 6 meters / minute, 10 steel plates are cleaned. Similarly, 10 steel plates are cleaned for each increase in speed. During the cleaning process of each steel plate, the speed level of each cleaning position is kept constant. The camera is used to shoot the surface of the steel plate before and after cleaning. A total of 1,200 steel plate images are obtained to form the dataset D1. Each record of the dataset includes the image of the steel plate before cleaning, the image of the steel plate after cleaning, and the corresponding conveying speed.

[0053] S2. Divide the aforementioned unstructured data set D1 into structured data sets corresponding to the steel plate surface oxide dark spot detection and recognition model and the steel plate surface oxide dark spot index prediction model after cleaning. The specific process is: perform equidistant strip area segmentation on the pre-cleaning image and the post-cleaning image of each steel plate in the D1 data set along the width direction of the steel plate (which is also the conveying direction) to convert them into several steel plate strip area images. The number of segmentations is equal to the number of cleaning positions in the surface cleaning equipment. Mark the positions of the oxide dark spots in the steel plate strip area images, and calculate the mean and maximum values ​​of the dark spot area in each steel plate strip area image to form structured data sets D2 and D3. Where D2 is an image data set, which consists of an image of a strip area of ​​a steel plate before cleaning and an image of a strip area of ​​a steel plate after cleaning, with oxide dark spots marked on the steel plate. D3 is a numerical data set, which consists of five data: the mean area of ​​the dark spots in the strip area of ​​the steel plate before cleaning, the maximum area of ​​the dark spots, the steel plate conveying speed, and the mean area of ​​the dark spots in the strip area after cleaning, and the maximum area of ​​the dark spots in the strip area after cleaning. Specifically, in this embodiment, according to Figure 4 The segmentation method shown divides the unstructured dataset D1 into four steel plate strip area images equidistantly along the width of the steel plate (corresponding to the number and positions of the cleaning stations in the cleaning device). The labeling tool LabelImg is used to label the location of the oxide dark spots in each steel plate strip area image, and the average and maximum area of ​​the dark spots in each steel plate strip area image are calculated to form structured datasets D2 and D3. The D2 and D3 datasets each contain 4800 records.

[0054] S3. Establish a detection and recognition model for dark spots on plate surfaces. Use the structured dataset D2 to train and deploy this model. The specific process is as follows: Based on the single-stage target detection network YOLOv8n, establish a detection and recognition model for dark spots on plate surfaces. In this embodiment, D2 is divided into a training set, a validation set, and a test set in a ratio of 7:2:1. YOLOv8n is trained using the training set, and a composite loss function of positioning error IoU, envelope area error, and confidence error is used for model training. To optimize network parameters, the Adam optimizer is used to reduce the loss. During training, the model is evaluated using the validation set, and hyperparameters are adjusted as needed. When the model's performance on the validation set meets the preset standards, a one-time test is performed using an independent test set to evaluate the model's generalization ability. If the test results meet the requirements, the model training is terminated and a deployable detection and recognition model is generated. In the Python environment, using the functions provided by the ultralytics library, you can complete the processes of loading the YOLOv8n model, configuring training parameters, model training, and verifying model performance. If the test meets the predetermined indicators, the detection and recognition model for dark spots on the surface of the plate built using YOLOv8n can be exported as a deployable model in ONNX format.

[0055] In this step, the composite loss function expression is:

[0056] The composite loss function expression is:

[0057] ;

[0058] In the formula 、 and is the weight coefficient, is the confidence loss, It is the loss of identification and positioning of oxide dark spots, It is the area loss of the oxide dark spot identification mark box envelope;

[0059] in,

[0060] Where, It is the intersection-over-union ratio of the oxide dark spot recognition prediction box and the oxide dark spot true annotation box;

[0061] Where, is the true value, is the predicted probability, N is the sample size;

[0062] Where, represents the area of ​​the oxide dark spot recognition prediction box, Represents the area of ​​the oxide dark spot marking box.

[0063] In this step, the validation set is used to evaluate the model. The indicators include YOLOv8n's accuracy, recall rate, and F1 score. Hyperparameters such as the initial learning rate, learning rate decay factor, and the number of training samples per batch are optimized and adjusted according to the requirements of the indicators.

[0064] In this step, if the results on the test set do not meet the requirements, the speed in S1 is further reduced and the step size is increased. Alternatively, the number of cleanups n at each speed may be increased, and image enhancement technology and other methods may be used to further expand the size of the data set D2, and then this step may be repeated.

[0065] S4. Establish an index prediction model for dark spots on the surface of the plate after cleaning. Use the structured data set D3 to train and deploy the index prediction model for dark spots on the surface of the plate after cleaning. The specific process is: Construct an MLP multi-layer network model to predict the mean area and maximum area of ​​oxide dark spots after the steel plate is cleaned. In this embodiment, D3 is divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set is used to train the MLP. The input of the model is the feature vector of the strip area of ​​the steel plate before cleaning. , the output is the feature vector of the cleaned steel strip area The hidden layer is a multi-layer fully connected neuron, the activation function uses the ReLU function, and the mean square error (MSE) is used as the loss function Minimize the error between the predicted and true values, calculate gradients using the backpropagation algorithm, update weights and biases, and use the Adam optimizer for parameter updates to ensure rapid model convergence. After passing the validation set's metrics evaluation and the test set's generalization test, model training is completed and a deployable prediction model is generated. Functions provided by the PyTorch framework allow you to define the MLP network, define the loss function and optimizer, perform training, and evaluate the model. If the test meets the predetermined metrics, the MLP-based prediction model for dark spots on the cleaned plate surface, built using the MLP, is exported as a deployable model in ONNX format.

[0066] In this step, the feature vector Where, is the average area of ​​dark spots in the strip area before the plate is cleaned, It is the maximum dark spot area in the strip area before the plate is cleaned. is the sheet conveying speed; the eigenvector Where, is the average area of ​​dark spots in the strip area after the plate is cleaned, It is the maximum area of ​​dark spots in the strip area after the plate is cleaned.

[0067] In this step, the loss function Where, 、 It is the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area in the strip area after the plate is cleaned.

[0068] S5, the implementation stage of segmented speed control, collects the image of the steel plate to be cleaned before cleaning at the set conveying speed, inputs the image before cleaning into the detection and recognition model of dark spots on the plate surface, identifies the dark spot area information in the image before cleaning, inputs the dark spot area information into the index prediction model of the dark spots on the surface after cleaning, obtains the dark spot index prediction data after cleaning, calculates the ideal speed of the plate passing through each cleaning position based on the dark spot index prediction data after cleaning, and adjusts the current conveying speed to the ideal speed to pass through the current cleaning position, specifically: at the preset steel plate conveying speed Before the steel plate is transported to the first cleaning station, an image of the steel plate before cleaning is obtained, and the image is divided into several rectangular strips at equal intervals according to the conveying direction of the steel plate (the segmentation method of this embodiment is also as shown in FIG. Figure 4 As shown in the figure, it is divided into 4 rectangular strips), and each rectangular strip image is input into the plate surface dark spot detection and recognition model deployed by S3 to detect the oxide dark spot area in the strip image and calculate the average area of ​​the oxide dark spot. , maximum dark spot area , combined with the steel plate conveying speed , construct the input feature variables of the indicator prediction model of the dark spots on the surface after cleaning deployed by S4 The output of the model is used to obtain the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area in each rectangular strip after the steel plate is cleaned. , according to the predicted value and As the basis for control decision-making, the actual speed required for the steel plate to pass through each cleaning position is obtained through the following formula:

[0069]

[0070] Where, is the preset conveying speed of the plate, is the predicted value of the mean dark spot area, is the predicted value of the maximum dark spot area, is the expected value of the mean dark spot area, is the expected value of the maximum dark spot area, which is determined by operating experience and actual needs. and is the weight coefficient, which is adjusted by experience.

[0071] In this embodiment, if the cleaning preset conveying speed of a certain plate is Before arriving at the No. 2 cleaning station, according to the predicted value , predicted value , expected value , expected value , , , then the ideal speed for clearing position 2 is , indicating that the speed needs to be reduced to improve the cleaning quality. After analyzing the images collected by the rear camera component, the actual value of the average dark spot area after the steel plate passes through the No. 2 cleaning position after the speed is reduced. , the actual value of the maximum dark spot area after cleaning , all achieved the approach to the expected value, improved the cleaning quality, and improved the cleaning consistency;

[0072] For example, the conveying speed of a steel plate Before reaching the No. 2 cleaning station, the predicted value , predicted value , expected value , expected value , , , then the ideal speed , indicating that the speed can be increased and energy consumption can be reduced. After analyzing the images collected by the rear camera component, the actual value of the average dark spot area after the steel plate passes through the No. 2 cleaning position after the speed is increased , the actual value of the maximum dark spot area after cleaning While ensuring the cleaning quality, it reduces the cleaning time in the No. 2 machine position and reduces the energy consumption of the equipment.

[0073] Example 2:

[0074] like Figure 2 and Figure 3 As shown ( Figure 3 In the figure, 1 is a front camera assembly, 2 is a rear camera assembly at the cleaning outlet, 3 is a plate position detection device, 4 is a computing and processing device, 5 is a conveyor roller, 6 is a steel plate before cleaning, 7 is a steel plate after cleaning, and 8 is a cleaning position in the cleaning device). The plate surface cleaning control system based on segmented speed regulation of this embodiment includes a visual subsystem, a plate surface dark spot detection and recognition subsystem, a plate surface dark spot index prediction subsystem, an automatic speed control subsystem, a plate position detection device, and a computing and processing device.

[0075] Among them, the visual subsystem includes a front camera assembly arranged in front of the first cleaning position of the cleaning device and a rear camera assembly arranged behind the last cleaning position of the cleaning device, which are used to obtain the surface image of the plate. The obtained image is used for the plate surface dark spot detection and identification subsystem and the plate surface dark spot index prediction subsystem.

[0076] The plate surface dark spot detection and identification subsystem is used to generate and deploy a detection and identification model for dark spots on the plate surface. The model is used to identify the surface image of the plate obtained by the vision subsystem and output the dark spot area information in the surface image, including the mean dark spot area and the maximum dark spot area. During the implementation phase of the segmented speed control, this subsystem is offline, but it still accepts steel plate surface images taken by the vision subsystem to expand the data set and facilitate subsequent network tuning.

[0077] The plate surface dark spot index prediction subsystem is used to generate and deploy an index prediction model for dark spots on the plate surface after cleaning. The model is used to process the acquired dark spot area information. The input is the mean dark spot area, the maximum dark spot area, and the plate conveying speed. The output is the predicted value of the mean dark spot area after cleaning the plate and the predicted value of the maximum dark spot area after cleaning the plate. During the implementation stage of the segmented speed control, this subsystem is offline, but it still accepts feature data input from the plate surface dark spot detection and identification subsystem to expand the data set and facilitate subsequent network tuning.

[0078] Plate in-position detection device: includes several sensors arranged at the entrance of each cleaning position of the cleaning device. When the sensor detects that the transported steel plate to be cleaned arrives at the entrance of the cleaning position, it sends a corresponding in-position signal. The speed automatic control subsystem can know which cleaning position the steel plate has been transported to based on the corresponding signal.

[0079] Automatic speed control subsystem: This system deploys a detection and recognition model for dark spots on the plate surface in ONNX format and an index prediction model for dark spots on the plate surface after cleaning. The system calculates the ideal speed of the steel plate passing through each cleaning position based on the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area at the current conveying speed output by the plate surface dark spot index prediction model. The system then controls the conveying speed of the steel plate in the cleaning device based on the signal from the plate placement detection device and the ideal speed.

[0080] Computing and processing device: It consists of a central processing unit (CPU), memory (SDRAM), graphics card (GPU), and storage. The storage and processing of data sets, the training and deployment of the plate surface dark spot detection and identification subsystem, the training and deployment of the plate surface dark spot index prediction subsystem, and the related computing and processing of the speed automatic control subsystem are all implemented in the computing and processing device.

Claims

1. A plate surface cleaning control method based on segmented speed regulation, characterized in that: The following steps are involved: S1, using a cleaning device to test-clean several plates at different conveying speeds, collecting surface images of the plates before and after the test-cleaning, and establishing an unstructured dataset D1; S2. Dividing the surface images of the plate before and after the trial washing in the unstructured dataset D1 into a plurality of portions along the width direction of the plate according to the number of cleaning stations in the cleaning device, thereby obtaining a plurality of strip area images of the plate before cleaning and a plurality of strip area images of the plate after cleaning; identifying, labeling, and processing dark spots in the plurality of strip area images of the plate before cleaning and the plurality of strip area images of the plate after cleaning, thereby forming an image structured dataset D2 and a numerical structured dataset D3; S3. Establish a detection and recognition model for dark spots on the surface of plate materials, and use the structured image dataset D2 to train and deploy the detection and recognition model for dark spots on the surface of plate materials; S4. Establish an index prediction model for dark spots on the surface of the plate after cleaning, and use the numerical structured dataset D3 to train and deploy the index prediction model for dark spots on the surface of the plate after cleaning; S5. Collect the image of the plate to be cleaned before cleaning at the set conveying speed, input the image before cleaning into the detection and recognition model of the dark spots on the surface of the plate, obtain the dark spot area information in the image before cleaning, input the dark spot area information into the index prediction model of the dark spots on the surface after cleaning, obtain the dark spot index prediction data after cleaning, calculate the ideal speed of the plate passing through each cleaning position according to the dark spot index prediction data after cleaning, and control the plate to be cleaned to be cleaned at the ideal speed.

2. The plate surface cleaning control method based on segmented speed regulation according to claim 1 is characterized in that: In S1, when testing a number of plates at different conveying speeds, set the conveying speed range of the plates. and speed growth step ,from Start cleaning n plates, then increase the speed , and then clean n pieces of plate until the speed is completed Cleaning of the next n plates.

3. The plate surface cleaning control method based on segmented speed regulation according to claim 1 is characterized in that: In S2, the number of segmentations of the surface image of the plate before and after the trial washing corresponds to the number of cleaning stations in the cleaning device.

4. The plate surface cleaning control method based on segmented speed regulation according to any one of claims 1 to 3, characterized in that: In S2, the image structured dataset D2 includes a strip area image of the plate before cleaning with dark spots marked and an image of the strip area after cleaning with dark spots marked; the numerical structured dataset D3 includes the mean dark spot area, the maximum dark spot area, the plate conveying speed, and the mean dark spot area, the maximum dark spot area of ​​the strip area image after cleaning.

5. The plate surface cleaning control method based on segmented speed regulation according to claim 4 is characterized in that: In S3, a detection and recognition model for dark spots on the surface of plate materials is established based on the single-stage object detection network YOLOv8n. The image structured dataset D2 is divided into a training set, a validation set, and a test set. YOLOv8n is trained using the training set. During the training process, the validation set is used to evaluate the model's indicators and hyperparameters are adjusted as needed. When the performance of the detection and recognition model for dark spots on the surface of plate materials on the validation set meets the preset standards, an independent test set is used for a one-time test. When the test results meet the requirements, the model training is terminated and a deployable detection and recognition model for dark spots on the surface of plate materials is generated.

6. The plate surface cleaning control method based on segmented speed regulation according to claim 5 is characterized in that: During the training of YOLOv8n, a composite loss function is used for model training. The composite loss function expression is: ； In the formula 、 and is the weight coefficient, is the confidence loss, It is the loss of identification and positioning of oxide dark spots, It is the area loss of the oxide dark spot identification mark box envelope; in, Where, It is the intersection-over-union ratio of the oxide dark spot recognition prediction box and the oxide dark spot true annotation box; Where, is the true value, is the predicted probability, N is the sample size; Where, represents the area of ​​the oxide dark spot recognition prediction box, Indicates the area of ​​the oxide dark spot marking box; Among them, the oxide dark spot identification annotation box refers to the dark spot area reference box generated manually or by tools during the training process to guide model training and learning; the oxide dark spot identification prediction box refers to the dark spot identification box output by YOLOv8n, which includes the predicted position and confidence; the oxide dark spot real annotation box is the standard box of the dark spots in the dataset, which is used for IoU and area loss calculation.

7. The plate surface cleaning control method based on segmented speed regulation according to claim 5 is characterized in that: When the model is evaluated using the validation set, the indicators include the accuracy, recall, and F1 score of YOLOv8n; the hyperparameters include the initial learning rate, the learning rate decay factor, and the number of training samples per batch.

8. The plate surface cleaning control method based on segmented speed regulation according to claim 4 is characterized in that: In S4, an indicator prediction model for dark spots on the surface of the plate after cleaning is established based on the MLP multi-layer network model. The numerical structured dataset D3 is divided into a training set, a validation set, and a test set. The MLP multi-layer network model is trained using the training set. After the indicator evaluation of the validation set and the generalization ability test of the test set are passed, the model training is completed and a deployable indicator prediction model for dark spots on the surface after cleaning is generated.

9. The plate surface cleaning control method based on segmented speed regulation according to claim 8 is characterized in that: When using the training set to train the MLP multi-layer network model, the feature vector of the strip area before the plate is cleaned is used. As input value, the feature vector of the strip area after the plate is cleaned As the output value, multi-layer fully connected neurons are used as hidden layers, ReLU function is used as activation function, and mean square error is used as loss function. , the gradient is calculated by the back-propagation algorithm, and the Adam optimizer is used to update the parameters.

10. The plate surface cleaning control method based on segmented speed regulation according to claim 9, characterized in that: The feature vector Where, is the average area of ​​dark spots in the strip area before the plate is cleaned, It is the maximum dark spot area in the strip area before the plate is cleaned. is the sheet conveying speed; the eigenvector Where, is the average area of ​​dark spots in the strip area after the plate is cleaned, It is the maximum area of ​​dark spots in the strip area after the plate is cleaned.

11. The plate surface cleaning control method based on segmented speed regulation according to claim 9, characterized in that: The loss function The expression is Where, 、 It is the predicted value of the mean dark spot area and the predicted value of the maximum dark spot area in the strip area after the plate is cleaned.

12. The plate surface cleaning control method based on segmented speed regulation according to claim 4, characterized in that: In S5, after collecting the pre-cleaning image of the plate to be cleaned at the set conveying speed, the pre-cleaning image is divided into several rectangular strips according to the method of S2, and then each of the rectangular strips is input into the detection and recognition model of dark spots on the surface of the plate to identify and obtain the dark spot area information in the pre-cleaning image, and the dark spot area information includes the mean dark spot area, the maximum dark spot area and the conveying speed of the plate.

13. The plate surface cleaning control method based on segmented speed regulation according to claim 4, characterized in that: The dark spot area information is input into the index prediction model of the dark spots on the surface after cleaning to obtain the predicted values ​​of the mean dark spot area and the maximum dark spot area. Combined with the conveying speed of the plate and the expected cleaning effect, the ideal speed of the plate passing through each cleaning position is calculated.

14. The plate surface cleaning control method based on segmented speed regulation according to claim 13, characterized in that: The ideal speed The calculation method is: Where, is the conveying speed of the plate, is the predicted value of the mean dark spot area, is the predicted value of the maximum dark spot area, is the expected value of the mean dark spot area, is the expected value of the maximum dark spot area, and is the weight coefficient.

15. A plate surface cleaning control system based on segmented speed regulation, characterized in that: Used to implement the plate surface cleaning control method based on segmented speed regulation as described in any one of claims 1 to 14, comprising a visual subsystem, a plate surface dark spot detection and recognition subsystem, a plate surface dark spot index prediction subsystem, a speed automatic control subsystem and a computing and processing device; The visual subsystem includes a front camera assembly arranged in front of the first cleaning position of the cleaning device and a rear camera assembly arranged behind the last cleaning position of the cleaning device, for acquiring a surface image of the plate; The plate surface dark spot detection and recognition subsystem is used to generate and deploy a plate surface dark spot detection and recognition model. The plate surface dark spot detection and recognition model is used to recognize the surface image of the plate obtained by the visual subsystem and output dark spot area information in the surface image, including the mean dark spot area and the maximum dark spot area. The plate surface dark spot index prediction subsystem is used to generate and deploy an index prediction model for dark spots on the plate surface after cleaning. The index prediction model for dark spots on the plate surface after cleaning is used to process the acquired dark spot area information and output a predicted value for the mean area of ​​dark spots on the plate after cleaning and a predicted value for the maximum area of ​​dark spots on the plate after cleaning. The speed automatic control subsystem is equipped with a detection and recognition model for dark spots on the plate surface and an index prediction model for dark spots on the plate surface after cleaning, which is used to calculate the ideal speed of the plate passing through each cleaning station and control the conveying speed of the plate in the cleaning device; The computing and processing device consists of a central processing unit, memory, graphics card and storage, and is used to realize the storage and processing of data sets, the training and deployment of the plate surface dark spot detection and identification subsystem, the training and deployment of the plate surface dark spot index prediction subsystem and the related computing and processing functions of the speed automatic control subsystem.

16. The plate surface cleaning control system based on segmented speed regulation according to claim 15, characterized in that: It also includes a plate position detection device, which includes a number of sensors arranged at the entrance of each cleaning position of the cleaning device, for determining the cleaning position where the plate is located and sending a signal to the speed automatic control subsystem to assist the speed automatic control subsystem in adjusting and controlling the conveying speed of the plate.

Images