A method and system for intelligently detecting the service life state of filter cloth of a filter press

Abstract

The application discloses a filter press filter cloth life state intelligent detection method and system, the method comprises the following steps: obtaining the filter cloth image of a plurality of filter cloths in the filter press; for each filter cloth, the filter cloth image corresponding to the filter cloth is subjected to damage identification, the damage degree of the filter cloth is obtained, and the damage level is determined; and the filter cloth image corresponding to the filter cloth is subjected to multi-dimensional feature extraction, the image features are obtained, and the pollution level of the filter cloth is determined according to the image features; and the damage level and the pollution level are fused to obtain the comprehensive health index of the filter cloth; based on the comprehensive health index, the damage level, the image features and the pollution level, the remaining service life of the filter cloth is determined, and the filter press is provided with a filter cloth replacement suggestion based on the remaining service life. In this way, through machine vision and multi-source data fusion, the health state of the filter cloth of the filter press can be rapidly, accurately and comprehensively detected automatically, and a scientific basis is provided for predictive maintenance.

Description

Technical Field

[0001] This application relates to the field of solid-liquid separation equipment condition monitoring technology, and in particular to an intelligent detection method and system for the life status of filter cloth in a filter press. Background Technology

[0002] Filter presses are widely used in industries such as chemical, pharmaceutical, food, metallurgy, and environmental protection, serving as core equipment for solid-liquid separation. Taking the chemical industry as an example, a filter press typically consists of dozens or even hundreds of filter plates, each covered with a filter cloth. The filter cloth separates solid particles from the liquid through physical interception. During continuous operation, the filter cloth is subjected to high pressure, chemical corrosion, and material abrasion, making it prone to damage, clogging, and aging. Once the filter cloth is damaged, solid particles can penetrate it with the liquid, leading to turbid filtrate, reduced filtration efficiency, and in severe cases, damage to the filter plates and shutdown of the entire machine, resulting in significant economic losses.

[0003] However, the current technologies for detecting the condition of filter cloth in filter presses typically employ manual visual inspection, indirect pressure / flow monitoring, offline laboratory testing, and detection methods based on simple visual sensors. These methods cannot provide a fast, accurate, and comprehensive automated detection of the health status of the filter cloth. Summary of the Invention

[0004] This application provides an intelligent detection method and system for the life status of filter cloth in a filter press, which solves the following technical problem: how to achieve rapid, accurate and comprehensive automated detection of the health status of filter cloth in a filter press.

[0005] In a first aspect, embodiments of this application provide an intelligent detection method for the lifespan status of filter cloth in a filter press, the method comprising: Obtain images of multiple filter cloths in the filter press; For each of the filter cloths, damage identification is performed on the corresponding filter cloth image to obtain the degree of damage of the filter cloth, and the damage level of the filter cloth is determined based on the degree of damage of the filter cloth; For each of the filter cloths, feature extraction is performed on the filter cloth image corresponding to the filter cloth to obtain image features of at least one dimension corresponding to the filter cloth image, and the pollution level of the filter cloth is determined based on the image features of at least one dimension corresponding to the filter cloth image. For each of the filter cloths, the damage level of the filter cloth and the contamination level of the filter cloth are combined to obtain the comprehensive health index of the filter cloth; For each of the filter cloths, the remaining service life of the filter cloth is determined based on its comprehensive health index, damage level, image features, and contamination level, so as to provide filter cloth replacement recommendations for the filter press based on the remaining service life of each of the filter cloths.

[0006] Secondly, embodiments of this application also provide an intelligent detection system for the life status of filter cloth in a filter press. The system includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the method as described above.

[0007] Thirdly, embodiments of this application also provide a computer storage medium storing computer-executable instructions, which, when executed, implement the method described in any of the above claims.

[0008] The intelligent detection method and system for the life status of filter cloth in a filter press provided in this application has the following beneficial effects: First, by acquiring images of multiple filter cloths from a filter press and identifying damage to each cloth, the degree of damage and damage level can be calculated. This provides a direct quantitative basis for predicting filter cloth lifespan and assessing physical degradation. Second, multi-dimensional feature extraction from the filter cloth images determines the contamination level, effectively overcoming the limitations of judging filter cloth health solely based on damaged area. This comprehensively reflects the decline in filter cloth permeability from both clogging and contamination perspectives. Third, the damage level and contamination level are integrated to obtain a comprehensive health index, achieving unified quantification of physical damage and functional performance decline, thus avoiding the bias of single-indicator assessment. Finally, the remaining service life is determined by combining the comprehensive health index, damage level, image features, and contamination level, and replacement recommendations are provided. Collaborative analysis using multi-dimensional historical data and status characteristics makes lifespan prediction more accurate and reliable, scientifically guiding preventative filter cloth replacement and preventing production interruptions or product quality degradation due to filter cloth failure. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of an intelligent detection method for the life status of filter cloth in a filter press, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the internal structure of an intelligent detection system for the life status of filter cloth in a filter press, provided in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] It is understood that in the embodiments of this application, data related to user information (such as user accounts) is involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0013] In the following description, the terms “first, second, ...” are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that “first, second, ...” may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0014] Filter presses are widely used in industries such as chemical, pharmaceutical, food, metallurgy, and environmental protection, serving as core equipment for solid-liquid separation. Taking the chemical industry as an example, a filter press typically consists of dozens or even hundreds of filter plates, each covered with a filter cloth. The filter cloth separates solid particles from the liquid through physical interception. During continuous operation, the filter cloth is subjected to high pressure, chemical corrosion, and material abrasion, making it prone to damage, clogging, and aging. Once the filter cloth is damaged, solid particles can penetrate it with the liquid, leading to turbid filtrate, reduced filtration efficiency, and in severe cases, damage to the filter plates and shutdown of the entire machine, resulting in significant economic losses.

[0015] However, the current technologies for detecting the condition of filter cloth in filter presses typically employ manual visual inspection, indirect pressure / flow monitoring, offline laboratory testing, and detection methods based on simple visual sensors. These methods cannot provide a fast, accurate, and comprehensive automated detection of the health status of the filter cloth.

[0016] Based on this, this application provides an intelligent detection method for the life status of filter cloth in a filter press. By integrating machine vision with multi-source data, the health status of the filter cloth in the filter press can be quickly, accurately, and comprehensively detected automatically, providing a scientific basis for predictive maintenance.

[0017] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0018] Figure 1 This application provides a flowchart of an intelligent detection method for the lifespan status of filter cloth in a filter press. This method can be applied to various solid-liquid separation equipment status monitoring scenarios. For example: in mining and metallurgy, this solution detects the damage and contamination of filter cloth during the filtration of slurry, tailings, etc., allowing for timely replacement and ensuring production continuity; in chemical and pharmaceutical settings, it detects the lifespan status of filter cloths used for filtering chemical raw materials and pharmaceuticals, preventing product quality degradation or production accidents due to filter cloth failure; in food and beverage processing, it detects the hygiene and damage level of filter cloths during solid-liquid separation to ensure food safety; in environmental wastewater treatment, it monitors the filter cloth status in real time for sludge dewatering filter presses, optimizing equipment maintenance cycles and reducing operating costs; and in industrial automation and intelligent operation and maintenance scenarios, it uses image recognition technology to automate filter cloth lifespan detection and early warning, improving equipment management efficiency. Certain input parameters or intermediate results in the process can be manually adjusted to improve accuracy.

[0019] This application provides an intelligent detection method for the lifespan status of filter cloth in a filter press. It should be noted that the executing entity in these embodiments can be a server or any terminal device with data processing capabilities. For example, the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal device can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, in-vehicle terminal, etc., but is not limited to these.

[0020] like Figure 1 As shown in the figure, the intelligent detection method for the life status of filter cloth in a filter press provided in this application embodiment specifically includes the following steps: Step 101: Obtain filter cloth images of multiple filter cloths in the filter press.

[0021] It should be noted that filter cloth images can be acquired by installing high-definition industrial cameras at specific stations on the filter press and using a programmable logic controller (PLC) to control the system to trigger timed or fixed-point shooting; alternatively, line scan or area scan cameras can be integrated into the filter press production line, along with strobe light sources and sensors, to perform online scanning imaging during filter cloth operation; or tracked robots, robotic arms, or drones equipped with camera equipment can be used to perform multi-angle, full-coverage moving images of the inside of the filter press or large filter plates; or infrared thermal imagers can be used to acquire thermal distribution images of the filter cloth to help determine the clogging or damage of the filter cloth; when acquiring filter cloth images, each filter cloth can correspond to one filter cloth image or multiple filter cloth images. No specific limitations are made here.

[0022] As an example, taking the automated production line scenario of a large chemical filter press workshop, when the filter press completes a filtration cycle and begins automatic plate unloading, the control system (PLC) detects the start signal of the filter plate movement. At this time, the high-definition industrial camera mounted on the fixed bracket on the side of the filter press is triggered. Subsequently, as the filter plates are pulled open one by one, the camera captures each exposed filter cloth frame by frame. Here, to eliminate interference from complex lighting conditions (such as overhead light flicker and window light), the camera is usually used in conjunction with a high-brightness LED strobe fill light to ensure that the surface texture of the filter cloth is clear and shadow-free in the captured images. Afterwards, for some large filter presses, if a single angle cannot cover the entire filter cloth, a movable industrial camera (e.g., mounted on a robotic arm or linear module) may be used. After the filter plates are pulled open, the camera moves along the track to dynamically scan the filter cloth, thereby acquiring a set of continuous multi-angle images. Finally, after the images are captured, the image data is transmitted in real time to the central processing unit or edge computing industrial control computer via an industrial gigabit network port or fiber optic cable for subsequent damage identification and feature extraction.

[0023] Step 102: For each of the filter cloths, perform damage identification on the filter cloth image corresponding to the filter cloth to obtain the degree of damage of the filter cloth.

[0024] It should be noted that the aforementioned damage identification can be achieved through traditional image processing threshold segmentation methods. For example, firstly, the filter cloth image is screened to remove images with obviously large areas of residual filter cake. Then, grayscale conversion, Gaussian filtering for noise reduction, and binarization are performed to simplify the image to black and white pixels. Morphological operations (such as erosion and dilation) are used to remove small noise points and fill holes. By calculating the area ratio of white (or black) connected regions, the area ratio of the damaged area relative to the entire filter cloth image is directly quantified to determine the current degree of damage to the filter cloth. Alternatively, it can be achieved through deep learning object detection and segmentation methods. For example, a dedicated damage identification model can be trained using a convolutional neural network (CNN) and a mask can be used. Semantic segmentation algorithms such as R-CNN or U-Net allow the model to learn pixel-level features of damaged areas (e.g., cracks, holes), thereby accurately calculating the actual damaged area to determine the degree of damage to the current filter cloth. Contour extraction via edge detection, such as using edge operators like Canny or Sobel to extract edge information from images, can also be used. Since filter cloth damage is often accompanied by texture breaks or discontinuous edges, analyzing the degree of edge breakage and the area within the closed contour can determine whether damage exists and calculate its size, thus determining the degree of damage to the current filter cloth. The degree of filter cloth damage refers to the degree of damage to the filter cloth fibers, which can be quantified by the damaged area, the number of tears, or the location of the tears; no specific limitations are made here.

[0025] In some embodiments, the damage identification of the filter cloth image corresponding to the filter cloth in step 102 above, to obtain the degree of damage of the filter cloth, can be achieved in the following way: The filter cloth image is encoded layer by layer through N cascaded encoding networks to obtain multi-scale encoding results, where N is a positive integer greater than 1; the encoding result of the Nth layer encoding network is decoded through the first layer decoding network to obtain the decoding result of the first layer decoding network; when 1 < i ≤ N, the following process is repeatedly performed: attention fusion is performed between the decoding result of the (i-1)th layer decoding network and the encoding result of the (N-i+1)th layer encoding network. The fusion result is obtained, and the fusion result is decoded through the i-th layer decoding network to obtain the decoding result of the i-th layer decoding network, where i is a positive integer and is incremented by one after each execution. The decoding result of the N-th layer decoding network is classified into pixels to obtain the pixel type of each pixel in the filter cloth image. Based on the pixel type of each pixel in the filter cloth image, the number of pixels of the damaged type in the filter cloth image is counted. Based on the spatial resolution of the filter cloth image, the number of pixels of the damaged type is converted into area to obtain the damaged area of ​​the filter cloth, and the damaged area of ​​the filter cloth is used as the degree of damage of the filter cloth.

[0026] Thus, by adopting an improved autoencoder structure (U-Net variant) and introducing an attention fusion mechanism, the accuracy and robustness of filter cloth damage identification can be significantly improved. Furthermore, through multi-scale encoding, rich features from texture details to structural contours can be extracted layer by layer. At the same time, by using skip connections in conjunction with the attention mechanism, the problems of low resolution and loss of details in deep network feature maps can be effectively solved. This allows the model to accurately focus on the damaged area and suppress background noise interference. Finally, through pixel-level classification and spatial resolution conversion, accurate mapping from micro-pixels to macro-physical areas can be achieved, providing a highly reliable quantitative data foundation for subsequent life assessment.

[0027] It should be noted that the encoding network is used to extract and compress features from the input image layer by layer; the decoding network is used to gradually restore the abstract features output by the encoding network to the target output.

[0028] As an example, taking the automated quality inspection scenario of a large-scale chemical filter press workshop, the system inputs the high-definition image of the filter cloth into the first-layer encoding network. The encoding network extracts low-level features (such as edges and textures) of the image through convolution operations and downsamples (reduces the size) before passing it to the second layer. Subsequently, the second layer extracts higher-level semantic features (then local damage shapes) and continues to pass them, layer by layer until the Nth layer, finally obtaining a deep feature map containing global contextual information (i.e., multi-scale encoding results). Then, the first-layer decoding network receives the encoding result of the Nth-layer encoding network and begins upsampling to initially restore the spatial information. After that, the system... The decoding result of the layer decoding network is fused with the encoding result of the (N-1)th layer encoding network to obtain fused features. The fused features are then decoded by the second layer decoding network to obtain the decoding result of the second layer decoding network. This process is repeated until the Nth layer decoding network is completed, resulting in a feature map with the same size as the original image. Each pixel on the feature map is classified to determine the pixel type (e.g., damaged or intact). Finally, by counting the number of damaged pixels and combining the camera calibration parameters (spatial resolution), the area of ​​the damaged pixels is converted to obtain the actual damaged area of ​​the filter cloth image, i.e., the degree of damage to the filter cloth.

[0029] In some embodiments, the above-described attention fusion of the decoding result of the (i-1)th layer decoding network and the encoding result of the (N-i+1)th layer encoding network to obtain the fusion result can be achieved as follows: Global average pooling is performed on each channel of the encoding result of the (N-i+1)th layer encoding network to obtain a channel description vector; the channel description vector is dimensionality-reduced through a first fully connected layer to obtain a dimensionality-reduced vector; the dimensionality of the dimensionality-reduced vector is restored through a second fully connected layer to obtain an original weight vector, and the original weight vector is activated to obtain a channel attention weight vector; the channel attention vector is multiplied by the encoding result channel by channel to obtain a weighted encoding feature; the decoding result of the (i-1)th layer decoding network is concatenated with the weighted encoding feature to obtain the fusion result.

[0030] Thus, by intelligently and adaptively refining and focusing the raw features extracted from the encoder, the core performance of the entire deep learning model in complex industrial scenarios can be significantly improved. Furthermore, this design forces the model to no longer treat all feature channels equally when fusing detailed features from the encoder. Instead, it uses a learnable feedforward network to automatically evaluate and strengthen feature channels that are more critical for identifying filter cloth damage, while suppressing irrelevant or interfering channels that primarily respond to normal background textures, wrinkles, or uniform stains on the filter cloth. Feature selection allows the subsequent decoder to obtain cleaner and more targeted detailed information, thereby more accurately delineating the boundaries of minor damage in pixel-level segmentation tasks. This effectively reduces the probability of misclassifying complex textures as damage or missing detections due to background noise. This solution directly translates into higher damage recognition accuracy, stronger algorithm robustness, and better generalization ability for filter cloths of different materials and ages, laying a reliable technical foundation for subsequent accurate quantization, grade determination, and lifespan prediction.

[0031] As an example, taking the automated quality inspection scenario of a large chemical filter press workshop as an example, the total number of layers in the encoding and decoding networks is 5. Assuming that the current processing is the 3rd layer decoding network, firstly, global average pooling is performed on each channel (assuming there are 256 channels, each channel can be regarded as a feature detector) of the encoded feature map E3 (encoding result) output by the 3rd layer encoding network. For example, the 101st channel may specifically detect horizontal edges. After pooling, a scalar value is obtained, representing the average activity level of horizontal edges in the overall filter cloth image. Performing this operation on all 256 channels can obtain a 256-dimensional channel description. The 256-dimensional channel description vector is then input into the first fully connected layer, compressing it to a lower dimension (e.g., 16 dimensions) and passing it through the ReLU activation function to obtain a dimensionality-reduced vector. Next, the 16-dimensional intermediate vector is input into the second fully connected layer, restoring its dimension back to the original 256 dimensions. Then, a Sigmoid activation function is used to map each value to a range of 0 to 1, resulting in a 256-dimensional channel attention weight vector (the original weight vector). Finally, the obtained channel attention weight vector is multiplied channel-by-channel with the original encoded feature map E3 to obtain the weighted encoded feature E3. · Finally, the decoded feature map D3 (decoding result) output by the second-layer decoding network is compared with the weighted encoded feature E3. · The concatenation is performed along the channel dimension to obtain the fusion result, which serves as the input feature for the third-layer decoding network.

[0032] In some embodiments, the above-mentioned damage recognition is achieved through a trained damage recognition model. Before performing step 102, the following processing may also be performed: using the initialized damage recognition model, damage recognition is performed on the sample image to obtain the predicted damage probability of each pixel in the sample image and the predicted damage area in the sample image; based on the difference between the predicted damage probability of each pixel in the sample image and the actual damage probability of each pixel in the sample image, a first loss is determined; based on the overlap between the predicted damage area in the sample image and the actual damage area in the sample image, a second loss is determined; based on a preset first weight, the first loss and the second loss are fused to obtain a third loss, and the initialized damage recognition model is updated based on the third loss to obtain the trained damage recognition model.

[0033] Thus, by designing an efficient and highly directional model optimization mechanism, instead of using a single, ordinary loss function that is sensitive to the distribution of pixel numbers, a weighted fusion of two loss functions with different optimization objectives is adopted. The first loss is responsible for constraining the model from the overall probability distribution, ensuring that the classification tendency of each pixel is basically correct. The second loss directly focuses on the spatial overlap between the predicted damaged area and the actual damaged area, which are two sets of foreground targets. Its calculation is not sensitive to the difference in the number of pixels between the foreground and the background. This ensures that the model not only correctly classifies most background pixels during training, but also strives to improve the recall and segmentation accuracy of sparse damaged pixels, so as to ensure that the model focuses on sparse but critical damaged pixels. The model trained by this scheme has extremely high sensitivity and segmentation accuracy for millimeter-level micro-damage against the background of complex filter cloth texture, which can significantly reduce the false negative rate. At the same time, the constraint of the loss function can also avoid generating too many false positives by blindly pursuing overlap, so as to ensure high accuracy and high robustness of damage recognition.

[0034] As an example, during model training, a sample image is selected from the training set. The image consists of a filter cloth with minor tears and two small holes. First, the sample image is input into the initialized damage recognition model. After encoding-decoding and attention fusion, the model outputs a predicted damage probability for each pixel of the sample image. Based on the predicted probability of damage for each pixel in the sample image The true probability of damage to each pixel in the sample image Calculate the first loss based on the difference between them. (Refer to formula (1)); Subsequently, according to a preset probability threshold (e.g., 0.5), pixels with a predicted damage probability greater than the probability threshold are identified as damaged, thereby obtaining the predicted damaged area in the sample image. Based on the predicted damaged area in the sample image Compared with the actual damaged area in the sample image The degree of overlap is used to calculate the second loss. (Refer to formula (2)); then, according to the preset first weight The first loss calculated Second loss We perform a weighted summation to obtain the third loss. That is, the total loss of the model; finally, according to the third loss Backpropagation updates the parameters of the damage identification model, thus obtaining the trained damage identification model.

[0035] (1) (2) Where N is the total number of pixels in the sample image. This represents the true probability of damage to the i-th pixel. This represents the predicted probability of damage for the i-th pixel. The predicted damaged area in the sample image. This represents the actual damaged area in the sample image. This indicates the number of pixels in the set. This represents the intersection of two sets, specifically the overlapping portion between the predicted damaged area and the actual damaged area. For a very small smoothing constant (e.g.) (), used to prevent the denominator from being zero.

[0036] Here's an explanation of backpropagation: The sample image is input into the input layer of the neural network model (damage recognition model), passes through the hidden layer, and finally reaches the output layer to output the result. This is the forward propagation process of the neural network model. Since there is an error between the output result of the neural network model and the actual result, the error between the result and the actual value is calculated and propagated back from the output layer to the hidden layer until it reaches the input layer. During the backpropagation process, the values ​​of the model parameters are adjusted according to the error. The above process is iterated until convergence.

[0037] The loss function in the embodiments of this application For example, the server determines the model loss based on the loss function, backpropagates the model loss from the output layer of the damage identification model, and backpropagates the model loss layer by layer. When the model loss reaches each layer, the gradient (that is, the partial derivative of the loss function with respect to the parameters of each layer) is solved by combining the propagated model loss, and the corresponding gradient value of the parameters of each layer is updated.

[0038] Step 103: Determine the damage level of the filter cloth based on the degree of damage.

[0039] It should be noted that the damage level can be determined directly based on the preset range into which the damage falls. For example, taking the damaged area as an example, if the damaged area is <5... For minor damage, taking the number of damaged items as an example, if the number of damaged items is less than 2, it is considered minor damage. Alternatively, a comprehensive judgment can be made based on the damaged area, combined with other quantitative indicators such as the maximum damaged size and the number of damaged items. No specific limitation is made here.

[0040] In some embodiments, step 103 described above can be implemented as follows: in response to the degree of damage being less than a first threshold, the damage level of the filter cloth is determined to be slightly damaged; in response to the degree of damage being greater than or equal to the first threshold and the degree of damage being less than a second threshold, the damage level of the filter cloth is determined to be moderately damaged; in response to the degree of damage being greater than or equal to the second threshold, the damage level of the filter cloth is determined to be severely damaged.

[0041] In this way, the continuous and precise quantitative physical indicators output by deep learning models can be efficiently and objectively converted into discrete engineering decision-making instructions that can be directly used to guide maintenance actions without human intervention. This fundamentally eliminates the subjectivity and inconsistency of traditional manual inspection, ensuring the uniformity and reproducibility of judgment standards. Furthermore, the thresholding and grading method can transform complex visual inspection results into clear language that can be quickly understood and executed, allowing them to be directly linked to differentiated maintenance strategies and providing a reliable data foundation for subsequent maintenance decisions. This solution not only improves the efficiency and scientific nature of maintenance decisions, avoiding over-maintenance or under-maintenance caused by ambiguous judgments, but also provides standardized and structured key inputs for the subsequent calculation of a comprehensive health index by fusing damage levels and contamination levels, and for achieving predictive maintenance.

[0042] As an example, assuming a filter cloth inspection scenario in a filter press, taking the damaged area as an example, the first threshold is 5. The second threshold is 20. The system then begins processing the test results for each filter cloth. Taking three filter cloths in different states (filter cloth #45, filter cloth #78, and filter cloth #102) as an example; for filter cloth #45, the system analyzes it and finds the damage level to be 3.2. Less than the first threshold 5 The damage level of filter cloth #45 can be determined to be minor; for filter cloth #78, the system analysis shows a damage level of 12.7. Greater than the first threshold 5 And less than the second threshold of 20 The damage level of filter cloth #78 can be determined to be moderate; for filter cloth #102, the system analysis shows a damage level of 35.5. Greater than the second threshold 20 It can be determined that the damage level of filter cloth #102 is severe damage.

[0043] Step 104: For each of the filter cloths, perform feature extraction on the filter cloth image corresponding to the filter cloth to obtain image features of at least one dimension for each filter cloth image.

[0044] It should be noted that feature extraction of filter cloth images can include features in at least one dimension such as color, texture, shape, spatial relationship, deep learning, and frequency domain, without specific limitations here.

[0045] In some embodiments, at least one dimension of image features includes texture features, grayscale features, and frequency domain features. The feature extraction of the filter cloth image corresponding to the filter cloth in step 104 above, to obtain at least one dimension of image features for each filter cloth image, can be achieved as follows: determining the grayscale co-occurrence matrix corresponding to the filter cloth image, and extracting the texture features of the filter cloth image from the grayscale co-occurrence matrix, the texture features including energy, contrast, entropy, and correlation; performing statistical analysis on the pixel intensity of the filter cloth image to obtain the grayscale features of the filter cloth image, the grayscale features including average grayscale, grayscale standard deviation, and grayscale histogram skewness; performing a two-dimensional Fourier transform on the filter cloth image to obtain the spectrum corresponding to the filter cloth image, and extracting the frequency domain features of the filter cloth image from the spectrum, the frequency domain features including high-frequency energy proportion and spectral center shift.

[0046] In this way, texture features can sensitively capture the changes in surface structure from complex and random to simple and uniform due to blockage. Grayscale features can effectively reflect the statistical changes in overall brightness and contrast caused by contaminant adhesion. Frequency domain features can focus on the attenuation of image details caused by microscopic pore blockage. Through multi-dimensional feature fusion, the detection accuracy and sensitivity of blockage and pollution of different degrees and types can be significantly improved, ensuring that each feature has a clear physical or statistical interpretation. This makes the entire evaluation process more transparent, understandable, and debuggable, greatly enhancing the system's adaptability and reliability in complex and variable scenarios. It provides a solid and reliable data foundation for subsequent accurate calculation of pollution index, level determination, and feature fusion.

[0047] As an example, consider an image of a specific filter cloth (e.g., the filter cloth located on filter plate number 35) captured during a shutdown inspection of a chemical filter press. First, the system calculates the gray-level co-occurrence matrix (GLCM) of the filter cloth image. This GLCM quantifies the joint probability distribution of gray values ​​between any two points in the image at a given direction and distance. Four key texture features are extracted from the matrix: energy (quantifying texture uniformity; energy increases if the filter cloth surface becomes smoother due to clogging); contrast (reflecting texture clarity; severe clogging may decrease contrast); entropy (characterizing texture randomness and complexity; normal filter cloth has rich texture and high entropy, while severe clogging leads to a decrease in entropy); and correlation (describing the regularity of texture in a direction; clogging may enhance texture regularity, increasing correlation). Second, the system performs statistical analysis on the pixel gray values ​​of the image, obtaining three gray-level features: average gray value, ... The system extracts the following features: the overall average brightness of the image; the grayscale standard deviation, reflecting the strength of image contrast; and the grayscale histogram skewness, describing the asymmetry of grayscale distribution (uneven adhesion of pollutants may alter this distribution). Finally, the system performs a two-dimensional Fourier transform on the filter cloth image, converting it from the spatial domain to the frequency domain to obtain the corresponding spectrum. Two frequency domain features are extracted from the spectrum: high-frequency energy percentage (the proportion of high-frequency components in the total energy; when the filter cloth micropores are blocked by fine particles, image detail loss leads to a significant attenuation of high-frequency energy); and spectrum center offset, calculating the offset distance of the spectrum centroid relative to the geometric center of the image, used to capture changes in the dominant texture direction caused by pollution. Through these feature extractions, comprehensive feature extraction of filter cloth No. 35 from texture and grayscale to the frequency domain is achieved. The resulting feature vector will serve as the basic quantitative data for subsequently calculating the comprehensive pollution index of the filter cloth and determining its pollution level.

[0048] Step 105: Determine the contamination level of the filter cloth based on image features corresponding to at least one dimension of the filter cloth image.

[0049] It should be noted that the contamination level of the filter cloth can be determined by calculating and summarizing the image features to obtain the degree of clogging of the filter cloth mesh; alternatively, the contamination level of the filter cloth can be directly determined by using a model to predict based on image features; no specific limitation is made here.

[0050] In some embodiments, step 105 described above can be implemented as follows: for each dimension of image features, the ratio between the image features of the dimension and the baseline features of the dimension is used as the normalization index of the dimension; based on a preset second weight, the normalization indices of each dimension are fused to obtain the comprehensive contamination index of the filter cloth corresponding to the filter cloth image; feature similarity is calculated between the image features corresponding to the filter cloth and the image features corresponding to the filter cloth in the best state to obtain the horizontal comparison similarity, and the standardized deviation between the image features corresponding to the filter cloth and the mean of the historical image features of the filter cloth is used as the vertical comparison similarity; based on the comprehensive contamination index of the filter cloth, the horizontal comparison similarity, and the vertical comparison similarity, the contamination level of the filter cloth is matched to obtain the contamination level of the filter cloth.

[0051] Thus, by calculating the ratio of image features in each dimension to their baseline features, a normalized index is generated and fused into a single comprehensive pollution index based on preset weights. This effectively eliminates systematic biases caused by differences in filter cloth batches, materials, or imaging conditions, ensuring consistency and comparability of evaluation results across different scenarios. Furthermore, the introduction of horizontal comparison similarity calculation, using the best-performing filter cloth in the same filter press as a reference, allows for the precise identification of deviations of individual filter cloths from the overall health level through feature similarity measurement. This provides strong robustness against common interferences such as lighting and environment, significantly improving the stability of on-site detection. Simultaneously, vertical comparison similarity is combined, calculating the current feature's similarity to its own... The standardized deviation of historical characteristic means can effectively capture the cumulative degradation trend of filter cloth performance over time, enabling early warning of minor contamination and slow performance degradation, and providing a key basis for predictive maintenance. In addition, by inputting the comprehensive contamination index and horizontal and vertical comparison results into preset decision rules for contamination level matching, a refined hierarchical decision-making based on multi-source information fusion can be achieved. Through this application, not only can the accuracy and reliability of contamination detection be significantly improved, but the assessment can also be upgraded from a single static judgment to dynamic monitoring that combines horizontal comparison and historical trend analysis, thereby providing strong data support for the scientific maintenance and replacement decisions of filter cloth and the optimization of overall equipment operating efficiency.

[0052] As an example, suppose a filter press with 140 filter cloths is being inspected. The system has extracted feature vectors in three dimensions—texture, grayscale, and frequency domain—from each filter cloth image. Taking the pollution level of filter cloth number 57 as an example, firstly, the comprehensive pollution index of the filter cloth is calculated. The system retrieves the corresponding feature values ​​of the new filter cloth from the database as the baseline features for each dimension, and calculates the ratio of each feature value of the current filter cloth to the baseline value to obtain the texture normalization index, grayscale normalization index, and frequency domain normalization index. Subsequently, the system performs a weighted summation of the three normalization indices according to preset weights (e.g., texture weight 0.4, grayscale weight 0.3, frequency domain weight 0.3) to obtain the comprehensive pollution index (CI) of filter cloth number 57, for example, CI = 0.65. After that, the system performs a dual-benchmark comparison. In the horizontal comparison, the system selects the filter cloth from the 140 filter cloths inspected that has the closest feature vectors in all dimensions to the baseline state (i.e., the best comprehensive state) as the baseline. Using a reference filter cloth, the Gaussian weighted similarity between the feature vector of filter cloth No. 57 and the feature vector of the reference filter cloth is calculated to obtain the horizontal comparison similarity (Si), for example, Si=0.60. In the vertical comparison, the system retrieves the historical feature data of the past five tests of filter cloth No. 57 and calculates the standardized deviation between the current feature value and its historical mean to obtain the vertical comparison similarity (Di), for example, Di=2.5 (meaning that the current feature deviates from the historical mean by 2.5 standard deviations). Finally, the calculated comprehensive pollution index (CI=0.65), the horizontal comparison similarity (Si=0.60), and the significant anomaly indicator of the vertical comparison (Di>2 can be considered significant) are input into the preset pollution level decision table for matching. According to the decision table rules, if the CI value is between 0.50 and 0.70 and the Si value is between 0.55 and 0.70, and the significant anomaly shown in the vertical comparison (Di>2), the pollution level of the current filter cloth is "moderate pollution".

[0053] Step 106: For each filter cloth, combine the damage level of the filter cloth with the pollution level of the filter cloth to obtain the comprehensive health index of the filter cloth.

[0054] In some embodiments, the step 106 above, which integrates the damage level of the filter cloth with the contamination level of the filter cloth to obtain the comprehensive health index of the filter cloth, can be achieved in the following way: based on a preset third weight, the damage level of the filter cloth and the contamination level of the filter cloth are weighted and summed to obtain a weighted summation result, and the theoretical worst-case damage state value of the filter cloth is determined; the value 1 is subtracted from the ratio of the weighted summation result to the worst-case damage state value to obtain the comprehensive health index of the filter cloth.

[0055] Thus, by weighting and summing discrete damage and contamination levels using a preset third weight, the comprehensive assessment can be flexibly adjusted according to actual process requirements, greatly enhancing the method's adaptability and engineering practicality. Furthermore, by introducing the theoretical worst-case damage state value to normalize the weighted summation result, the health index is successfully mapped to a continuous interval of [0, 1], giving the assessment results an intuitive and unified dimension, facilitating comparisons of different filter cloths and at different times, and enabling rapid understanding by on-site personnel. Simultaneously, the construction of subtracting the ratio from the value 1 shows a negative correlation between the health index and the degree of filter cloth deterioration, providing a clear quantitative basis for subsequent maintenance decisions. This application not only enables comprehensive quantification of the health status of individual filter cloths but also provides accurate and reliable input for subsequent life prediction models and priority-based replacement decisions, thereby opening up a key link from multi-dimensional state detection to predictive maintenance decisions, effectively supporting the scientific and refined nature of maintenance decisions.

[0056] As an example, assuming a filter press used in the fine chemical industry, the damage level has a weight of 0.7, and the contamination level has a weight of 0.3. First, the system identifies the filter cloth as moderately damaged (value 2) by performing damage identification and threshold determination. Simultaneously, through multi-dimensional feature extraction and comparative evaluation, the system determines the current filter cloth's contamination level as slightly contaminated (value 1). Subsequently, the system performs a weighted summation of the current filter cloth's damage level and contamination level according to preset weights, resulting in a weighted summation of 1.7. Then, the maximum value of the damage level and contamination level is 3, yielding a theoretical worst-case value of 3. Finally, subtracting the value 1 from the ratio of the weighted summation to the theoretical worst-case value, the system obtains the current filter cloth's comprehensive health index as 0.433.

[0057] Step 107: For each of the filter cloths, determine the remaining service life of the filter cloth based on the comprehensive health index, damage level, image features and contamination level of the filter cloth, and provide filter cloth replacement recommendations for the filter press based on the remaining service life of each of the filter cloths.

[0058] In some embodiments, determining the remaining service life of the filter cloth based on its comprehensive health index, damage level, image features, and contamination level in step 107 above can be achieved by fusing the comprehensive health index, damage level, image features, and contamination level of the filter cloth to obtain fused features; and predicting the remaining service life of the filter cloth based on the fused features to obtain the remaining service life of the filter cloth.

[0059] Thus, by deeply fusing a comprehensive health index reflecting the immediate health status of the filter cloth, the damage and contamination levels quantifying specific failure modes, and original image features containing rich surface details, a feature vector comprehensively and multidimensionally describes the degradation state of the filter cloth is constructed. This overcomes the limitations of a single data source, enabling the prediction model to simultaneously perceive the current health status, historical evolution trend, and operating environment of the filter cloth, thereby greatly improving the accuracy and reliability of lifespan prediction. Furthermore, by introducing the comprehensive health index and its rate of change as key features, the model can sensitively capture the dynamic decay rate of filter cloth performance, achieving... This represents an evolution from static assessment to dynamic prognostic early warning. Furthermore, based on the prediction results of integrated features, the filter cloth maintenance strategy has been completely transformed from traditional periodic replacement or replacement after damage to scientific predictive replacement. This allows users to accurately grasp the remaining effective lifespan of each filter cloth and formulate on-demand, orderly maintenance plans accordingly. This not only maximizes the use value of the filter cloth and avoids the waste of resources caused by premature replacement, but also effectively prevents unplanned downtime caused by sudden filter cloth failure. As a result, it ensures continuous production, improves filtration efficiency, optimizes spare parts inventory, and significantly reduces overall maintenance costs, providing core decision support for the intelligent and lean operation and maintenance of filter presses.

[0060] As an example, taking a filter press used for municipal sludge dewatering, its operating conditions focus on filtration flux. During a shutdown inspection, the system needs to assess the remaining lifespan of filter cloth #102. First, the system obtains the filter cloth's comprehensive health index (HI) value of 0.62, damage level (D_damage) value of 1 (minor damage), and contamination level (D_clog) value of 3 (severe contamination) from the visual inspection and decision module. It also obtains image feature vectors from the image processing unit, including texture features (energy, contrast, etc.) extracted from the gray-level co-occurrence matrix, statistically derived gray-level features (average gray level, standard deviation, etc.), and frequency domain features (high-frequency energy percentage, etc.) extracted from the spectrum diagram. Simultaneously, it retrieves the filter cloth's HI values ​​from the historical database for the past three times and calculates the health index change rate (ΔHI). Furthermore, it obtains operating condition features from the equipment monitoring system, including the filter cloth's cumulative operating time (1500 hours), average working pressure (0.8 MPa), and average solids content of the processed material. Subsequently, all the above features are used to determine the immediate health status (HI, D_damage, ...). The fused feature vector is obtained by fusing the following data: D_clog), state change trend (ΔHI), visual details (image features), and operating load (operating condition features). This fused feature vector is then input into a pre-trained random forest regression prediction model. The model processes the fused feature vector based on patterns learned from a large amount of data on the entire life cycle of similar filter cloths. The final output shows that the remaining service life of filter cloth No. 102 is 200 hours. The result is also converted into understandable engineering language, such as "It is expected to continue to operate safely for about 200 hours". This information, along with the filter cloth number and health index, is output as the core basis for generating priority replacement recommendations and overall equipment maintenance plans.

[0061] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide an intelligent detection system for the life status of filter cloth in a filter press, the structure of which is as follows: Figure 2 As shown.

[0062] Figure 2 This is a schematic diagram of the internal structure of an intelligent detection system for the lifespan status of filter cloth in a filter press, provided as an embodiment of this application. Figure 2 As shown, the system includes: At least one processor 201; And a memory 202 that is communicatively connected to at least one processor; The memory 202 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 201 to enable at least one processor 201 to perform the steps of the method corresponding to any of the above embodiments.

[0063] It should be noted that the system provided in this application embodiment can be implemented as various types of terminals such as laptops, tablets, desktop computers, set-top boxes, smartphones, smart speakers, smartwatches, smart TVs, and vehicle terminals, or it can be implemented as a server.

[0064] In some embodiments, in addition to including at least one processor 201 and a memory 202 communicatively connected to at least one processor, the system may also include a vision-based intelligent detection system that shares data with at least one processor 201, the intelligent detection system being used to acquire filter cloth images of multiple filter cloths in a filter press in real time.

[0065] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium stores computer-executable instructions configured to perform the steps of the method corresponding to any of the above embodiments.

[0066] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0067] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0068] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0073] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0074] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0075] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0076] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for intelligently detecting the lifespan status of filter cloth in a filter press, characterized in that, The method includes: Obtain images of multiple filter cloths in the filter press; For each of the filter cloths, damage identification is performed on the corresponding filter cloth image to obtain the degree of damage of the filter cloth, and the damage level of the filter cloth is determined based on the degree of damage of the filter cloth; For each of the filter cloths, feature extraction is performed on the filter cloth image corresponding to the filter cloth to obtain image features of at least one dimension corresponding to the filter cloth image, and the pollution level of the filter cloth is determined based on the image features of at least one dimension corresponding to the filter cloth image. For each of the filter cloths, the damage level of the filter cloth and the contamination level of the filter cloth are combined to obtain the comprehensive health index of the filter cloth; For each of the filter cloths, the remaining service life of the filter cloth is determined based on its comprehensive health index, damage level, image features, and contamination level, so as to provide filter cloth replacement recommendations for the filter press based on the remaining service life of each of the filter cloths.

2. The method according to claim 1, characterized in that, The step of identifying the damage to the filter cloth image corresponding to the filter cloth to obtain the degree of damage to the filter cloth includes: The filter cloth image is encoded layer by layer through N cascaded coding networks to obtain multi-scale coding results, where N is a positive integer greater than 1. The encoding result of the Nth layer encoding network is decoded through the first layer decoding network to obtain the decoding result of the first layer decoding network. When 1 < i ≤ N, the following process is repeatedly performed: the decoding result of the (i-1)th layer decoding network and the encoding result of the (N-i+1)th layer encoding network are fused by attention to obtain a fused result, and the fused result is decoded by the i-th layer decoding network to obtain the decoding result of the i-th layer decoding network, where i is a positive integer and i is incremented by one after each execution. The decoding results of the Nth layer decoding network are classified by pixels to obtain the pixel type of each pixel in the filter cloth image; Based on the pixel type of each pixel in the filter cloth image, count the number of pixels of the damaged type in the filter cloth image; Based on the spatial resolution of the filter cloth image, the number of pixels of the damage type is converted into area to obtain the damaged area of ​​the filter cloth, and the damaged area of ​​the filter cloth is used as the degree of damage of the filter cloth.

3. The method according to claim 2, characterized in that, The process of performing attention fusion between the decoding result of the (i-1)th layer decoding network and the encoding result of the (N-i+1)th layer encoding network to obtain the fusion result includes: Global average pooling is performed on each channel of the encoding result of the (N-i+1)th layer coding network to obtain the channel description vector; The channel description vector is reduced in dimension by using the first fully connected layer to obtain a dimension-reduced vector. The second fully connected layer performs dimension restoration processing on the dimensionality-reduced vector to obtain the original weight vector, and then performs activation processing on the original weight vector to obtain the channel attention weight vector. The channel attention vector is multiplied with the encoding result channel by channel to obtain the weighted encoding features; The decoding result of the (i-1)th layer decoding network is concatenated with the weighted encoded features to obtain the fusion result.

4. The method according to claim 1, characterized in that, The damage identification is achieved through a trained damage identification model; Before performing damage identification on the filter cloth image corresponding to the filter cloth to obtain the degree of damage to the filter cloth, the method further includes: By using the initialized damage recognition model, damage recognition is performed on the sample image to obtain the predicted damage probability of each pixel in the sample image and the predicted damage area in the sample image. The first loss is determined based on the difference between the predicted damage probability of each pixel in the sample image and the actual damage probability of each pixel in the sample image; The second loss is determined based on the overlap between the predicted damaged area in the sample image and the actual damaged area in the sample image; Based on a preset first weight, the first loss and the second loss are fused to obtain a third loss, and the initialized damage recognition model is updated based on the third loss to obtain the trained damage recognition model.

5. The method according to claim 1, characterized in that, Determining the damage level of the filter cloth based on its degree of damage includes: In response to the damage level being less than a first threshold, the damage level of the filter cloth is determined to be minor damage; In response to the damage level being greater than or equal to the first threshold and the damage level being less than the second threshold, the damage level of the filter cloth is determined to be moderate damage; In response to the degree of damage being greater than or equal to the second threshold, the damage level of the filter cloth is determined to be severe damage.

6. The method according to claim 1, characterized in that, The image features of at least one dimension include texture features, grayscale features, and frequency domain features; The step of extracting features from the filter cloth image corresponding to the filter cloth to obtain image features of at least one dimension corresponding to the filter cloth image includes: Determine the gray-level co-occurrence matrix corresponding to the filter cloth image, and extract the texture features of the filter cloth image from the gray-level co-occurrence matrix. The texture features include energy, contrast, entropy value, and correlation. Statistical analysis is performed on the pixel intensity of the filter cloth image to obtain the grayscale features of the filter cloth image, which include the average grayscale, grayscale standard deviation, and grayscale histogram skewness. A two-dimensional Fourier transform is performed on the filter cloth image to obtain the corresponding spectrum diagram, and the frequency domain features of the filter cloth image are extracted from the spectrum diagram. The frequency domain features include the proportion of high-frequency energy and the frequency center shift.

7. The method according to claim 1, characterized in that, Determining the contamination level of the filter cloth based on image features corresponding to at least one dimension of the filter cloth image includes: For each dimension of image features, the ratio between the image features of that dimension and the baseline features of that dimension is used as the normalization index of that dimension. Based on the preset second weight, the normalized index of each dimension is fused to obtain the comprehensive pollution index of the filter cloth corresponding to the filter cloth image; The image features corresponding to the filter cloth and the image features corresponding to the filter cloth in the best state are calculated to obtain the horizontal comparison similarity. The standardized deviation of the image features corresponding to the filter cloth and the historical image features of the filter cloth is used as the vertical comparison similarity. Based on the comprehensive pollution index of the filter cloth, the horizontal comparison similarity, and the vertical comparison similarity, the pollution level of the filter cloth is matched to obtain the pollution level of the filter cloth.

8. The method according to claim 1, characterized in that, The step of combining the damage level and contamination level of the filter cloth to obtain the comprehensive health index of the filter cloth includes: Based on a preset third weight, the damage level of the filter cloth and the contamination level of the filter cloth are weighted and summed to obtain a weighted sum result, and the theoretical worst-case damage state value of the filter cloth is determined. The comprehensive health index of the filter cloth is obtained by subtracting the value 1 from the ratio of the weighted summation result to the worst damage state value.

9. The method according to claim 1, characterized in that, The determination of the remaining service life of the filter cloth based on its comprehensive health index, damage level, image features, and contamination level includes: The comprehensive health index, damage level, image features, and pollution level of the filter cloth are fused to obtain the fused features; Based on the fusion features, the remaining service life of the filter cloth is predicted to obtain the remaining service life of the filter cloth.

10. An intelligent detection system for the lifespan status of filter cloth in a filter press, characterized in that, The system includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-9.

Images