Material uniformizing control system and method for sludge dryer

Abstract

The invention relates to a material uniformizing control system and method for a sludge dryer. The material uniformizing control system of the sludge dryer comprises a data acquisition module which is provided with a first detection unit and a second detection unit and is used for acquiring sludge thickness data and sludge image data; the data processing module is used for carrying out feature extraction, fusion and dimensionality reduction on the sludge data so as to output a dimensionality-reduced feature vector; the control module is used for constructing a material uniformizing decision model and outputting a control strategy; and the material homogenizing module is used for executing the control strategy to homogenize the sludge. According to the material uniformizing control system and control method for the sludge dryer, the thickness feature, the texture feature and the shape feature can be fused together more effectively, meanwhile, the dimension of data is reduced, better input features are provided for a subsequent material uniformizing decision model, accurate material uniformizing of the sludge in the dryer is achieved, and the material uniformizing efficiency of the sludge dryer is improved. Sludge distribution is more uniform, and the problem that the drying effect is poor due to uneven material distribution is solved.

Description

Technical Field

[0001] The invention relates to the technical field of sludge treatment, and in particular to a sludge dryer material uniformity control system and method. Background Art

[0002] With the continuous growth and improvement of my country's sewage treatment capacity, sludge treatment and disposal has become a new difficulty and challenge in the comprehensive environmental governance work. Sludge disposal often adopts the "drying + separate incineration" sludge disposal process. In this sludge disposal process, the moisture content of the sludge after drying treatment must be reduced to meet the standard before the next step of incineration treatment can be carried out. Therefore, it is very important to effectively dry the sludge. At present, sludge dryers are widely used in sludge drying scenarios because they are energy-saving and environmentally friendly.

[0003] During the drying process of the sludge in the drying oven, uneven distribution is likely to occur, and the material on the mesh belt needs to be balanced. Insufficient equalization will lead to too high a moisture content in the sludge, which will directly affect the drying effect of the sludge; while too frequent equalization will cause a lot of dust, causing the filter equipment to quickly clog and directly reduce the service life of the sludge dryer. Therefore, it is very important to balance the material effectively and accurately and reduce unnecessary equalization times. Summary of the invention

[0004] Based on this, the purpose of the present invention is to provide a sludge dryer material distribution control system and control method, which can more effectively integrate thickness features, texture features and shape features, while reducing the dimension of the data, highlighting the main information in the data, and providing better input features for the material distribution decision model, thereby improving the accuracy and efficiency of the model, and realizing accurate material distribution of the sludge on the mesh belt inside the drying oven of the dryer, so that the sludge is more evenly distributed, avoiding the problem of poor drying effect caused by uneven distribution. At the same time, it effectively reduces the dust generated by excessive material distribution, reduces the risk of dust blocking the filter system, and extends the service life of the filter equipment.

[0005] A sludge dryer material uniformity control system, comprising:

[0006] A data acquisition module, used for collecting sludge thickness data and sludge image data;

[0007] A data processing module is used to perform feature extraction, feature fusion and dimension reduction on the sludge thickness data and image data after preprocessing, so as to output a feature vector after dimension reduction;

[0008] A control module is used to construct a material distribution decision model, receive the feature vector after dimension reduction and input it into the material distribution decision model, and output a control strategy;

[0009] The material leveling module is arranged in the oven of the sludge dryer and is used to execute the control strategy to level the sludge.

[0010] Furthermore, the data acquisition module includes:

[0011] The first detection unit is provided with a distance sensor and a temperature sensor, and performs thickness compensation according to the original thickness of the sludge obtained by the distance sensor and the temperature obtained by the temperature sensor to obtain sludge thickness data;

[0012] The second detection unit is provided with an industrial camera for taking sludge images and outputting sludge image data.

[0013] Furthermore, the data processing module includes:

[0014] A preprocessing unit, used for denoising the sludge thickness data output by the first detection unit, and for graying, normalizing and filtering the image data output by the second detection unit;

[0015] A feature extraction unit, used for extracting sludge thickness features and sludge image features;

[0016] The feature fusion unit is used to fuse and reduce the dimension of the feature vector composed of the extracted thickness feature and image feature.

[0017] Furthermore, the sludge thickness characteristics include average thickness, thickness standard deviation and thickness gradient; the sludge image characteristics include texture characteristics and shape characteristics, the texture characteristics include contrast, correlation, energy and entropy; the shape characteristics include area, perimeter and circularity.

[0018] Furthermore, the control module comprises:

[0019] A sample data acquisition unit is used to acquire sludge thickness data and sludge image data from the data acquisition module, as well as operation data from the material equalization module;

[0020] The model building unit builds a material balancing decision model based on a multi-layer perceptron;

[0021] The decision output unit is used to input the reduced-dimensional feature vector into the trained average material decision model and output the control strategy.

[0022] Furthermore, the material balancing decision model includes an input layer and an output layer; the number of nodes in the input layer is equal to the dimension of the feature vector after dimensionality reduction; and the output nodes of the output layer include at least material balancing decisions and material balancing strengths.

[0023] The beneficial effects of the present invention are:

[0024] (1) Comprehensively utilize the thickness data of the laser triangulation sensor and the image data of the industrial camera to give full play to the advantages of different sensors, describe the distribution status of the sludge from multiple angles, improve the accuracy of the control strategy, and make the sludge evenly distributed on the mesh belt through precise material balancing, thereby improving the uniformity of drying, thereby improving the drying efficiency of the entire dryer and reducing drying time and energy consumption;

[0025] (2) In order to solve the problem that the distance measuring sensor is affected by temperature, the measured thickness data is corrected through the temperature compensation formula to improve the accuracy of the sludge thickness data;

[0026] (3) Grayscale, normalize and filter the image data to reduce the impact of environmental factors such as lighting and noise on image feature extraction;

[0027] (4) Extract sludge thickness features, texture features, and shape features, perform feature fusion and dimensionality reduction, reduce data dimensions while retaining key information, provide better input features for the material balancing decision model, and improve the efficiency and accuracy of the material balancing decision model;

[0028] (5) Based on the neural network, the material balancing decision model learns the complex relationship between the feature vector and the control decision under different working conditions, so as to accurately output the material balancing decision and material balancing intensity according to the real-time collected data, control the material balancing module to balance the material, and realize closed-loop control;

[0029] (6) The entire material distribution control system realizes automatic and intelligent material distribution control, reduces manual intervention, improves the intelligence level of the production process, and reduces labor costs and the possibility of operational errors.

[0030] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of a sludge dryer material uniformity control system provided in the embodiment of the present application Figure 1 ;

[0032] Figure 2 Schematic diagram of a sludge dryer material uniformity control system provided in the embodiment of the present application Figure 2 ;

[0033] Figure 3 A flow chart of a material distribution control method for a sludge dryer provided in an embodiment of the present application;

[0034] Figure 4 A flowchart of the steps of sludge feature fusion and dimensionality reduction provided in an embodiment of the present application.

[0035] In the figure: 10-data acquisition module; 11-first detection unit; 12-second detection unit; 20-data processing module; 21-preprocessing unit; 22-feature extraction unit; 23-feature fusion unit; 30-control module; 31-sample data acquisition unit; 32-model building unit; 33-decision output unit; 40-material equalization module. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] During the drying process of the sludge in the drying oven, uneven distribution is likely to occur, and the material on the mesh belt needs to be balanced. Insufficient equalization will lead to too high a moisture content in the sludge, which will directly affect the drying effect of the sludge; while too frequent equalization will cause a lot of dust, causing the filter equipment to quickly clog and directly reduce the service life of the sludge dryer. Therefore, it is very important to balance the material effectively and accurately and reduce unnecessary equalization times.

[0038] In the prior art, the sludge thickness in the sludge dryer is measured by a distance measuring sensor, and the sludge thickness data is used to control whether the sludge is evenly distributed. However, this even distribution control method based on sludge thickness measurement has a single characteristic dimension, which leads to low model accuracy, resulting in inaccurate even distribution control, and also unable to control even distribution intensity, and unable to achieve efficient even distribution.

[0039] Based on this, the embodiment of the present application extracts thickness features, texture features and shape features from multiple angles and merges them together, while reducing the dimension of the data, highlighting the main information in the data, providing better input features for the material distribution decision model, improving the accuracy and efficiency of the model, and achieving accurate material distribution of the sludge on the mesh belt inside the dryer oven, making the sludge distribution more even, and avoiding the problem of poor drying effect caused by uneven distribution.

[0040] Example 1

[0041] See also Figure 1 and Figure 2, an embodiment of the present application provides a sludge dryer material distribution control system, including: a data acquisition module 10, a data processing module 20, a control module 30 and a material distribution module 40. The data acquisition module 10 is arranged in the oven of the sludge dryer, and is used to collect sludge data. Specifically, the data acquisition module 10 is provided with a first detection unit 11 and a second detection unit 12, wherein the first detection unit 11 is used to collect sludge thickness data, and the second detection unit 12 is used to collect sludge image data. The data processing module 20 is used to perform feature extraction, feature fusion and dimension reduction on the sludge thickness data and image data after preprocessing, so as to output the feature vector after dimension reduction. The control module 30 is used to construct a material distribution decision model, and receive the feature vector after dimension reduction and input it into the material distribution decision model, and output the control strategy. The material distribution module 40 is arranged in the oven of the sludge dryer, and is used to receive the control strategy to distribute the sludge.

[0042] Furthermore, the first detection unit 11 is provided with a distance measuring sensor, which is arranged in the oven of the sludge dryer. It is understandable that the distance measuring sensor is a sensor that detects distance based on optical principles, and an infrared distance measuring sensor, an ultrasonic sensor or a laser distance measuring sensor can be used. In the embodiment of the present application, the distance measuring sensor adopts a laser triangulation sensor, which is arranged at different positions above the mesh belt in the oven. During operation, multiple laser triangulation sensors continuously emit laser beams to the sludge surface and receive reflected light. Each laser triangulation sensor measures its distance from the sludge surface according to the change in the angle of the reflected light, and calculates the original thickness data of the sludge.

[0043] Furthermore, the first detection unit 11 also includes a temperature sensor arranged in conjunction with the ranging sensor, which is used to test the real-time temperature. In order to address the problem that the laser triangulation sensor is affected by temperature, thickness compensation is performed based on the original sludge thickness data obtained by the ranging sensor test and the temperature obtained by the temperature sensor test to obtain the sludge thickness data. The measured original sludge thickness data is corrected through temperature compensation to improve the accuracy of the sludge thickness data.

[0044] Specifically, the sludge thickness data after temperature compensation is T compensated for:

[0045] T compensated =T raw ×(1+k×(T env -T ref ));

[0046] Among them, T raw is the original thickness data of sludge; T env is the ambient temperature; the reference temperature T ref is the reference temperature, which is set to 25°C; k is the temperature compensation coefficient, which is obtained through calibration experiments.

[0047] As an example, at a certain moment, a distance sensor measures T raw =25.3mm, T env =60℃, then the sludge thickness after compensation is:

[0048] T compensated =25.3×((1+0.001×(60-25))=25.6mm.

[0049] Furthermore, the second detection unit 12 is provided with an industrial camera for capturing sludge images and outputting sludge image data. In the embodiment of the present application, a plurality of industrial cameras are arranged on the top of the oven, with their lenses vertically directed downward toward the mesh belt, and the industrial cameras capture images at certain time intervals to obtain image data of the sludge on the mesh belt. Specifically, the image data includes information such as the texture and shape of the sludge.

[0050] Furthermore, in some specific embodiments, the data processing module 20 includes: a preprocessing unit 21, a feature extraction unit 22 and a feature fusion unit 23, wherein the preprocessing unit 21 is used to denoise the sludge thickness data output by the first detection unit 11, and to grayscale, normalize and filter the image data output by the second detection unit 12; the feature extraction unit 22 is used to extract sludge thickness features and sludge image features; the feature fusion unit 23 is used to fuse and reduce the dimension of the feature vector composed of the extracted thickness features and image features.

[0051] Furthermore, the preprocessing unit 21 includes a thickness filtering subunit for the sludge thickness data, which performs wavelet transform filtering on the temperature compensated sludge thickness data to remove noise interference. The thickness signal after filtering is:

[0052]

[0053] Among them, w(a,b) is the wavelet coefficient, is the wavelet basis function, A and B are the scale and displacement parameters of the wavelet transform.

[0054] Furthermore, in an embodiment of the present application, the preprocessing unit 21 also includes a grayscale value calculation subunit, a normalization processing subunit and a filtering processing subunit for sludge image data, which are used to grayscale, normalize and filter the image data to reduce the impact of environmental factors such as light and noise on image feature extraction.

[0055] Specifically, the gray value calculation subunit is used to grayscale the sludge image data and convert the color image into a grayscale image using the weighted average method. The red, green, and blue channel values ​​of the pixels in the image are defined as R, G, and B, respectively, and the gray value Gray is:

[0056] Gray=0.299×R+0.587×G+0.114×B.

[0057] Specifically, the normalization processing subunit is used to scale the value ranges of different features to similar intervals to eliminate the dimensional differences between different variables so that different features have similar numerical ranges, that is, to map the gray value Gray to the interval [0,1] so that the mean of the input signal of all samples is close to 0 or very small compared to its mean square error; this is conducive to improving the convergence of the subsequent average material decision model.

[0058] The gray value Normalized_Gray after normalization is:

[0059]

[0060] Among them, Gray max and Gray min are the maximum and minimum grayscale values ​​in the image respectively.

[0061] Specifically, in the embodiment of the present application, the filtering processing subunit removes the noise in the image by median filtering, sets the gray value of each pixel point to the median of the gray values ​​of all pixels in a certain neighborhood window of the point, and makes the surrounding pixel values ​​close to the true value, thereby eliminating isolated noise points, which can protect the edge of the signal from being blurred. The original image pixel value is I(x,y), and the filtering window is W, then the pixel value of the filtered image at the coordinate (x,y) is:

[0062] Median_Filerted_Image(x,y)=Median{I(x+i,y+j),(i,j)∈W}.

[0063] Furthermore, the feature extraction unit 22 includes a thickness feature extraction subunit and an image feature extraction subunit. In the embodiment of the present application, the thickness feature extracted by the thickness feature extraction subunit includes the average thickness Thickness standard deviation σ T and thickness gradient

[0064] Specifically, the average thickness Based on the sludge thickness data after temperature compensation The average sludge thickness is calculated and calculated by the following formula:

[0065]

[0066] Where N is the number of measurement points.

[0067] Specifically, the thickness standard deviation σ TBased on average thickness Sludge thickness data after temperature compensation It is used to measure the discreteness of thickness. The calculation formula is:

[0068]

[0069] Specifically, the thickness gradient The thickness change rate of adjacent measuring points is calculated by the difference method to capture the local thickness change trend.

[0070] Furthermore, the image feature extraction subunit extracts the texture features and shape features of the sludge. Specifically, in the embodiment of the present application, the grayscale co-occurrence matrix is ​​used to extract the texture features of the sludge, which can accurately reflect the texture roughness and repetition direction. The grayscale co-occurrence matrix reflects the texture features of the image by counting the grayscale co-occurrence frequency of pixel pairs, and then extracts texture features such as contrast, correlation, energy and entropy. It can be understood that the grayscale co-occurrence matrix is ​​only an effective method for extracting sludge texture features. In some other embodiments, Tamura texture features, autoregressive texture models, wavelet transforms and other methods may also be used, but are not limited to this.

[0071] Specifically, in the embodiment of the present application, sludge lumps are identified by an image segmentation algorithm, and the shape characteristics of the sludge, such as its area, perimeter, and circularity, are calculated.

[0072] Furthermore, the feature fusion unit 23 includes a normalization subunit, a feature vector acquisition subunit, and a dimension reduction processing subunit.

[0073] In the embodiment of the present application, the standardization subunit is used to standardize the thickness characteristics, shape characteristics and texture characteristics of the extracted sludge so that they have the same mean and standard deviation, thereby obtaining a standardized feature matrix.

[0074] Specifically, in the embodiment of the present application, based on the thickness characteristics, such as the average thickness Thickness standard deviation σ T , thickness gradient Texture features, such as contrast C, correlation R, energy E, entropy H; and shape features, such as area A, perimeter P, circularity C r , combined into an original feature matrix X, the original feature matrix X is standardized so that each feature dimension has a mean and standard deviation.

[0075] Corresponding to the feature element x in the i-th row and j-th column of the original feature matrix X ij , calculate its mean μ j and standard deviation σ j :

[0076]

[0077] Where m is the number of samples.

[0078] Furthermore, based on the mean μ j and standard deviation σ j , calculate the standardized feature matrix X std Characteristic elements

[0079] In the embodiment of the present application, the feature vector acquisition subunit is used to construct the feature vector. In the embodiment of the present application, the covariance matrix of the standardized feature matrix is ​​calculated, and the eigenvalue decomposition and eigenvector calculation are performed to obtain the feature vector.

[0080] Specifically, the standardized feature matrix X std The element c of the covariance matrix C is pq for:

[0081]

[0082] Where p,q=1,2···, n,n is the total number of features, i.e., the sum of thickness features, texture features and shape features.

[0083] Then, the covariance matrix C is decomposed into eigenvalues:

[0084] C Vk =λ k υ k (k=1,2,···,n);

[0085] Get the eigenvalues ​​λ1,λ2,···λ n and the corresponding eigenvectors υ1,υ2,···,υ n .

[0086] In the embodiment of the present application, the dimensionality reduction processing subunit uses principal component analysis (PCA) to reduce the dimensionality of the feature vector to obtain the reduced dimensionality feature vector F PCA , reducing data dimensions while retaining key information, improving the efficiency and accuracy of the material averaging decision model.

[0087] Specifically, the eigenvectors are sorted according to the size of the eigenvalue. The larger the eigenvalue, the larger the data variance in the direction corresponding to the eigenvector, which means that this direction contains more information. The eigenvectors corresponding to the first k largest eigenvalues ​​are selected as the principal components to form the transformation matrix W, where the selection of the k value is based on the cumulative variance contribution rate (CVR). The calculation formula for the cumulative variance contribution rate (CVR) is:

[0088]

[0089] As an example, a k value is selected when the cumulative variance contribution rate reaches 90%.

[0090] Furthermore, principal component analysis (PCA) is used to calculate the eigenvector F after dimensionality reduction. PCA :

[0091] F PCA =W T X std

[0092] Where W T is the transposed matrix of W.

[0093] The embodiment of the present application fuses and reduces the dimension of the feature vector composed of the extracted thickness features and image features, thereby more effectively fusing the thickness features, texture features and shape features of the sludge, while highlighting the main information in the data, providing better input features for the subsequent material balancing decision model, and improving the accuracy and efficiency of the model.

[0094] Further, in the embodiment of the present application, the control module 30 includes a sample data acquisition unit 31, a model construction unit 32 and a decision output unit 33, wherein the sample data acquisition unit 31 is used to obtain the sludge thickness data and sludge image data of the data acquisition module 10, as well as the operation data of the material leveling module 40; the model construction unit 32 uses a multi-layer perceptron (MLP) as a neural network structure to construct a material leveling decision model; the decision output unit 33 is used to input the reduced dimension feature vector into the trained material leveling decision model and output the control strategy. The control strategy includes at least a material leveling decision D and a material leveling intensity S, wherein the material leveling decision is represented by or 1, wherein 0 indicates that material leveling is not required and 1 indicates that material leveling is required; the value range of the material leveling intensity S is between [0, 1], reflecting the operating parameters of the material leveling module. As an example, the model output D=1, S=0.6, indicating that a material leveling operation is required and the material leveling intensity is of medium level.

[0095] Specifically, the sample data acquisition unit 31 collects a large amount of sludge thickness data and sludge image data collected by the first detection unit 11 and the second detection unit 12 under different working conditions, as well as the sludge feed speed, humidity, mesh belt running speed, etc. of the sludge dryer under different working conditions. For each set of data, the values ​​of the material balancing decision D and the material balancing strength S are manually annotated, and these annotated data are divided into a training set, a validation set, and a test set.

[0096] Specifically, the model building unit 32 uses a multi-layer perceptron (MLP) as a neural network structure. The neural network structure includes an input layer, a number of hidden layers, and an output layer. The number of nodes in the input layer is equal to the feature vector F after dimensionality reduction. PCAThe output layer outputs the control strategy. In the embodiment of the present application, the output layer includes two nodes, namely, the material distribution decision D and the material distribution intensity S. It can be understood that in some other embodiments, the output layer can be provided with more nodes, such as the drying temperature, which is used to control the working temperature of the sludge dryer oven. It can be understood that the number of layers and nodes of the hidden layer are determined by experiments and cross-validation. As an example, the number of layers of the hidden layer is set to 2-3 layers, and the number of nodes in each layer is between 10-50.

[0097] Based on the constructed average feed decision model, it is trained using the training set data. As an example, during the training process, the cross entropy loss function is used for the output of the average feed decision D, and the mean square error loss function is used for the output of the average feed strength S. At the same time, an optimization algorithm is used to adjust the weights of the neural network to minimize the value of the loss function. As an example, the optimization algorithm can be a stochastic gradient descent algorithm and its variants Adam, Adagrad, etc., which will not be described in detail here. During the training process, the performance of the average feed decision model is regularly evaluated on the validation set, and the training is stopped when the performance on the validation set no longer improves to prevent overfitting.

[0098] Specifically, the decision output unit 33 inputs the feature vector after dimension reduction into F based on the trained average material decision model. PCA Into the material balancing decision model, the material balancing decision model outputs the material balancing decision D and the material balancing intensity S.

[0099] Furthermore, the material leveling module 40 is arranged in the oven of the sludge dryer. The material leveling module 40 adjusts parameters such as the moving speed and swing amplitude of the scraper according to the material leveling decision and the material leveling strength to level the sludge on the mesh belt. As an example, the material leveling strength S=0.6 output by the material leveling decision model, the material leveling module 40 adjusts the moving speed of the scraper to 60% of the system preset speed, and the swing amplitude to 1.2 times the system preset amplitude. It should be noted that the specific parameter adjustment relationship of the material leveling module 40 is determined according to the equipment characteristics and experiments, and will not be described in detail here. After the material leveling is completed, the material leveling control system enters the data collection stage again, repeats the above process, and continuously monitors and adjusts the distribution state of the sludge.

[0100] The sludge dryer material distribution control system provided in the embodiment of the present application realizes automatic and intelligent material distribution control, reduces manual intervention, improves the intelligence level of the production process, and reduces labor costs and the possibility of operational errors.

[0101] Example 2

[0102] See also Figure 3 and Figure 4 Based on the above-mentioned sludge dryer material distribution control system, the embodiment of the present application also provides a sludge dryer material distribution control method, including the following steps:

[0103] S1, obtaining sludge thickness data and sludge image data of sludge on the mesh belt;

[0104] S2, preprocessing the sludge thickness data and extracting the sludge thickness features; at the same time, preprocessing the sludge image data and extracting the sludge image features;

[0105] S3, obtaining an original feature matrix based on the combination of sludge thickness features and sludge image features, fusing the original feature matrix to obtain a feature vector, and performing dimensionality reduction processing on the feature vector;

[0106] S4, constructing a material balancing decision model based on feature vectors;

[0107] S5, inputting the reduced-dimensional feature vector into the average material decision model to obtain the control strategy;

[0108] S6, execute control strategy.

[0109] Furthermore, in step S1, the sludge thickness data is sludge thickness data obtained after temperature compensation based on original sludge thickness data tested by a distance measuring sensor.

[0110] Furthermore, in step S2, based on the temperature-compensated sludge thickness data, preprocessing is performed on it, and then feature extraction is performed, including the following steps:

[0111] S211, wavelet transform filtering is performed on the temperature compensated sludge thickness data to separate the wavelet transform of the signal from the wavelet transform of the noise to remove the noise. The thickness signal after filtering is:

[0112]

[0113] Among them, w(a,b) is the wavelet coefficient, is the wavelet basis function, A and B are the scale and displacement parameters of the wavelet transform.

[0114] S212, based on the pre-treated sludge thickness data, average thickness Thickness standard deviation σ T and thickness gradient The specific steps for extraction are:

[0115] First, based on the temperature-compensated sludge thickness data Calculate the average thickness

[0116]

[0117] Where N is the number of measurement points.

[0118] Then, based on the average thickness Sludge thickness data after temperature compensation Calculate the thickness standard deviation σ T :

[0119]

[0120] Next, the thickness gradient is calculated by calculating the thickness change rate of adjacent measurement points using the differential method.

[0121] Furthermore, in step S2, the sludge image data is preprocessed and sludge image features are extracted, including the following steps:

[0122] S221, grayscale processing, normalization processing and filtering processing are performed on the sludge image data to reduce the influence of environmental factors such as light and noise on image feature extraction. The specific steps are as follows:

[0123] First, the color image is converted into a grayscale image using the weighted average method, and the grayscale value is calculated;

[0124] Then, the grayscale value is mapped to the [0,1] interval to obtain the normalized grayscale value;

[0125] Next, the noise in the image is removed by median filtering. The grayscale value of each pixel is set to the median of the grayscale values ​​of all pixels in a neighborhood window of the point, so that the surrounding pixel values ​​are close to the true value, thereby eliminating isolated noise points.

[0126] S222, based on the pre-processed sludge image data, the texture features are calculated by the gray level co-occurrence matrix, and the shape features are calculated by the image segmentation algorithm. Specifically, the extracted texture features include contrast, correlation, energy and entropy; the extracted shape features include area, perimeter and circularity.

[0127] Furthermore, in step S3, an original feature matrix is ​​obtained based on the combination of the sludge thickness feature and the sludge image feature, and the original feature matrix is ​​fused to obtain a feature vector, and then the feature vector is subjected to dimensionality reduction processing, including the following steps:

[0128] S31, standardize the feature data of the original feature matrix to obtain a standardized feature matrix X std ;

[0129] S32, based on the standardized feature matrix X std , calculate its covariance matrix C;

[0130] S33, perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalue λ i and the corresponding eigenvector υi ;

[0131] S34, sorting the eigenvectors according to the size of the eigenvalues, selecting the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components, and forming the transformation matrix W; the selection of the k value is based on the cumulative variance contribution rate CVR, and the k value when the CVR reaches the set ratio is selected, and the calculation formula is:

[0132]

[0133] S35, calculate the feature vector F after dimensionality reduction processing PCA :

[0134] F PCA =W T X std ;

[0135] Among them, W T is the transposed matrix of W.

[0136] In the above steps, steps S31-S33 are specific steps for fusing the original feature matrix to obtain the eigenvalues ​​and eigenvectors; steps S34-S35 are specific steps for reducing the dimensionality of the eigenvectors to obtain the eigenvector F after dimensionality reduction. PCA .

[0137] Furthermore, in step S4, the specific steps of constructing a material distribution decision model based on a feature vector are as follows:

[0138] S41, collects a large amount of sludge thickness data and sludge image data collected under different working conditions, as well as the sludge feed speed, humidity, mesh belt running speed and other conditions of the sludge dryer under different working conditions; for each set of data, manually annotates the value of the material balancing decision D and the material balancing strength S, and divides these annotated data into training set, verification set and test set.

[0139] S42, using a multi-layer perceptron as a neural network structure, and establishing a material decision model based on the collected data and parameters.

[0140] S43, based on the constructed average feed decision model, use the training set data to train it; during the training process, regularly evaluate the performance of the average feed decision model on the validation set, and stop training when the performance on the validation set no longer improves.

[0141] Furthermore, the reduced feature vector F PCA Input into the material balancing decision model, and the material balancing decision model outputs the material balancing decision D and material balancing intensity S.

[0142] Furthermore, according to the material balancing decision D and material balancing intensity S output by the material balancing decision model, parameters such as the moving speed and swing amplitude of the scraper are adjusted to balance the sludge on the mesh belt.

[0143] The material distribution control method of the sludge dryer provided in the embodiment of the present application realizes accurate material distribution of the sludge on the mesh belt in the drying oven of the dryer, making the sludge thickness distribution more uniform, avoiding the problem of poor drying effect caused by uneven distribution. At the same time, accurate material distribution control helps to maintain the stable operation of the entire drying system, reduce equipment failures and unstable operation problems caused by uneven sludge distribution, and improve equipment reliability and production continuity.

[0144] Compared with the prior art, the beneficial effects of the present invention are:

[0145] (1) Comprehensively utilize the thickness data of the laser triangulation sensor and the image data of the industrial camera to give full play to the advantages of different sensors, describe the distribution status of the sludge from multiple angles, improve the accuracy of the control strategy, and make the sludge evenly distributed on the mesh belt through precise material balancing, thereby improving the uniformity of drying, thereby improving the drying efficiency of the entire dryer and reducing drying time and energy consumption;

[0146] (2) In order to solve the problem that the distance measuring sensor is affected by temperature, the measured thickness data is corrected through the temperature compensation formula to improve the accuracy of the sludge thickness data;

[0147] (3) Grayscale, normalize and filter the image data to reduce the impact of environmental factors such as lighting and noise on image feature extraction;

[0148] (4) Extract sludge thickness features, texture features, and shape features, perform feature fusion and dimensionality reduction, reduce data dimensions while retaining key information, provide better input features for the material balancing decision model, and improve the efficiency and accuracy of the material balancing decision model;

[0149] (5) Based on the neural network, the material balancing decision model learns the complex relationship between the feature vector and the control decision under different working conditions, so as to accurately output the material balancing decision and material balancing intensity according to the real-time collected data, control the material balancing module to balance the material, and realize closed-loop control;

[0150] (6) The entire material distribution control system realizes automatic and intelligent material distribution control, reduces manual intervention, improves the intelligence level of the production process, and reduces labor costs and the possibility of operational errors.

[0151] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and the present invention is also intended to include these modifications and modifications.

Claims

1. A sludge dryer material uniformity control system, characterized in that: include: A data acquisition module (10), used for acquiring sludge thickness data and sludge image data; A data processing module (20) is used to perform feature extraction, feature fusion and dimension reduction on the sludge thickness data and image data after preprocessing, so as to output a feature vector after dimension reduction; A control module (30) is used to construct a material distribution decision model, receive the feature vector after dimension reduction and input it into the material distribution decision model, and output a control strategy; A material leveling module (40) is arranged in the oven of the sludge dryer and is used to execute a control strategy to level the sludge.

2. A sludge dryer material uniformity control system according to claim 1, characterized in that: The data acquisition module (10) comprises: The first detection unit (11) is provided with a distance sensor and a temperature sensor, and performs thickness compensation according to the original thickness of the sludge obtained by the distance sensor and the temperature obtained by the temperature sensor to obtain sludge thickness data; The second detection unit (12) is provided with an industrial camera for taking sludge images and outputting sludge image data.

3. A sludge dryer material uniformity control system according to claim 2, characterized in that: The data processing module (20) comprises: A preprocessing unit (21) is used to perform denoising on the sludge thickness data output by the first detection unit (11), and to perform grayscale, normalization and filtering processing on the image data output by the second detection unit (12); A feature extraction unit (22), used for extracting sludge thickness features and sludge image features; The feature fusion unit (23) is used to fuse and reduce the dimension of the feature vector composed of the extracted thickness feature and image feature.

4. A sludge dryer material uniformity control system according to claim 3, characterized in that: The sludge thickness features include average thickness, thickness standard deviation and thickness gradient; the sludge image features include texture features and shape features, the texture features include contrast, correlation, energy and entropy; the shape features include area, perimeter and circularity.

5. A sludge dryer material uniformity control system according to claim 1, characterized in that: The control module (30) comprises: A sample data acquisition unit (31) is used to acquire sludge thickness data and sludge image data from the data acquisition module (10), as well as operation data of the material equalization module (40); A model building unit (32), used for building a material balancing decision model; The decision output unit (33) is used to input the feature vector after dimension reduction into the trained average material decision model and output the control strategy.

6. A sludge dryer material uniformity control system according to claim 5, characterized in that: The material balancing decision model includes an input layer and an output layer; the number of nodes in the input layer is equal to the dimension of the feature vector after dimensionality reduction; the output nodes of the output layer at least include material balancing decisions and material balancing strength.

7. A material distribution control method for a sludge dryer, characterized in that: The steps include: S1, obtaining sludge thickness data and sludge image data of sludge on the mesh belt; S2, preprocessing the sludge thickness data and extracting the sludge thickness features; at the same time, preprocessing the sludge image data and extracting the sludge image features; S3, obtaining an original feature matrix based on the combination of sludge thickness features and sludge image features, fusing the original feature matrix to obtain a feature vector, and performing dimensionality reduction processing on the feature vector; S4, constructing a material balancing decision model based on feature vectors; S5, inputting the reduced-dimensional feature vector into the average material decision model to obtain the control strategy; S6, execute control strategy.

8. The material distribution control method of a sludge dryer according to claim 7, characterized in that: In step S2, the sludge thickness feature is a temperature-compensated thickness feature; and the sludge image feature includes a texture feature and a shape feature.

9. A material distribution control method for a sludge dryer according to claim 8, characterized in that: In step S3, the specific steps of fusing the original feature matrix are: S31, standardize the feature data of the original feature matrix to obtain a standardized feature matrix X std ; S32, based on the standardized feature matrix X std , calculate its covariance matrix C; S33, perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalue λ i and the corresponding eigenvector υ i .

10. A material distribution control method for a sludge dryer according to claim 9, characterized in that: In step S3, the specific steps of performing dimensionality reduction processing on the feature vector are: Sort the eigenvectors according to the size of the eigenvalues, select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components, and form the transformation matrix W, where the k value is selected based on the cumulative variance contribution rate CVR, and the k value is selected when the CVR reaches the set ratio; The eigenvector F after dimensionality reduction PCA for, F PCA =W T X std ; Among them, W T is the transposed matrix of W.

Images