A computer vision-based early warning device and method for sludge bulking in an sbr reactor

Abstract

The application discloses a kind of early warning device and method of sludge bulking in early stage of SBR reactor based on computer vision.The device uses double visual angle to carry out real-time monitoring to activated sludge of each operation cycle of SBR reactor, one is through collecting sludge apparent image of SBR reactor, the early characteristics of sludge bulking are extracted, then early warning model of sludge bulking in early stage is used to judge the running state of reactor;Another is to collect sludge settling image of SBR reactor in the sedimentation stage, automatically generate sludge settling curve and sludge settling rate curve and carry out information extraction, and then judge the settling performance of sludge in SBR reactor, and alarm sludge bulking soon.This method combines sludge apparent image and settling image information, comprehensively analyzes the early characteristics of sludge bulking, reduces misjudgment, improves recognition accuracy, solves the problems of sewage plant sludge bulking phenomenon burst, high subsequent control cost, long recovery time and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of early warning device and method of SBR reactor sludge expansion in sewage treatment plant, belong to the field of sewage treatment optimization operation. BACKGROUND

[0002] At present, most of the domestic sewage treatment plants use activated sludge method to treat sewage, and this biological treatment technology is widely concerned due to its low cost, stability and high efficiency. However, the incidence of sludge bulking is high when using this method to treat sewage. Sludge bulking is one of the more serious abnormal phenomena in biochemical treatment system, with high incidence, wide range and difficult to suppress. In severe cases, it can cause poor water quality, deteriorated settling performance, sludge loss, and even lead to the collapse of the entire sewage treatment system. In the process of sewage treatment, timely identification of sludge bulking is the premise of implementing sludge bulking suppression. Therefore, it is of great significance to seek effective methods for early warning and suppression of sludge bulking.

[0003] At present, a variety of methods have been developed to prevent and control the occurrence of sludge bulking. The monitoring of conventional indicators such as dissolved oxygen, pH value, temperature and effluent quality provides a basis for early warning and timely prevention of activated sludge bulking in sewage treatment plants. However, manual detection is time-consuming and has a certain lag. Determining sludge settling velocity (Sludge settling velocity, SV) and sludge volume index (Sludge Volume Index, SVI) to judge sludge settling performance often requires time and effort. And SVI can only reflect the settling characteristics of activated sludge after 30 minutes, ignoring the process monitoring of sludge settling performance. The recognition method based on microscope image information can identify sludge bulking under certain conditions, but this method is highly dependent on operators and equipment, is complex to operate, and does not have the ability to identify online. The sludge bulking prediction method based on mechanism and data-driven model can accurately describe and detect sludge bulking, but due to the complex nonlinearity of sludge bulking problem, it has high dependence on data, poor adaptability and slow response time. Although there are many studies on early warning and control of sludge bulking, the lack of key information and the difficulty in capturing early characteristics limit the effectiveness of existing control methods.

[0004] With the advent of the information age, the field of artificial intelligence is rapidly developing, and image recognition technology, as an important branch, is gradually intersecting and integrating with the field of sewage treatment. Further promoting the application of "smart water" in sewage treatment plants can realize intelligent control of sewage treatment plants and improve the treatment capacity and management efficiency of sewage treatment plants. Computer vision, as a new technology, can realize automatic analysis and processing of images, improve work efficiency and reduce labor costs, and has been widely used in sewage treatment monitoring systems in recent years.

[0005] The early warning device and method for sludge bulking based on computer vision technology can replace daily inspection of operation and maintenance personnel, and realize unmanned management. The system alarms when the sludge is about to swell, the image information and conventional data are comprehensively analyzed, operation control suggestions are given, the sludge swelling problem is quickly and efficiently solved, the difficulty of sludge swelling recovery is reduced, and the problems of early discovery and early intervention are solved. Meanwhile, the process monitoring of the sewage treatment plant is realized by using a low-cost camera, which has important significance for improving the monitoring level and operation stability of the sewage treatment plant. SUMMARY

[0006] The present application provides an early warning device and method for sludge bulking in SBR reactor based on computer vision technology. The sludge online early warning system includes model early warning of sludge apparent image and information extraction of sludge settling curve. The comprehensive judgment of the two methods can comprehensively reflect the settling performance of the sludge and accurately identify the early characteristics of sludge bulking.

[0007] The purpose of the present application is achieved by the following technical solutions:

[0008] The present application is based on the traditional activated sludge process, and a new type of early warning system for sludge bulking is developed. The system realizes real-time data acquisition, processing and graphical presentation by installing display (1), industrial computer (2), alarm (3), data transmission line (4), overhead near-infrared / visible light binocular camera (5), side-view near-infrared / visible light binocular camera (6), telescopic support (7), tripod (8), LED fill light (9), luminous ruler (10), pH probe (11), dissolved oxygen probe (12), COD probe (13), photovoltaic cell panel (14), storage battery (15) and other hardware in the SBR reactor. Finally, real-time early warning and optimization control of early sludge bulking in sewage plant are realized.

[0009] As shown in Figure 1 The device is divided into three modules according to the function, including image acquisition module, data transmission and processing module, and data visualization and alarm module.

[0010] 1. Image acquisition module:

[0011] (1) For real-time monitoring of whether the sludge in the SBR reactor is about to swell, an overhead near-infrared / visible light binocular camera (5) for collecting sludge apparent image is installed at the top of the SBR reactor. The industrial computer (2) sends instructions to call the overhead near-infrared / visible light binocular camera (5) to collect sludge apparent image at high frequency during the aerobic, anoxic and sedimentation stages of the SBR reactor. Whether the LED fill light (9) is turned on or not can be selected according to the light conditions.

[0012] (2) For real-time monitoring of the settling performance of the sludge in the SBR reactor, a light-emitting ruler (10) is installed inside the SBR reactor to measure the position of the sludge-water interface, and a side-view near-infrared / visible light binocular camera (6) is installed to capture images of the sludge settling, and a industrial computer (2) is used to issue instructions to call the side-view near-infrared / visible light binocular camera (6) to capture images of the sludge settling during the settling stage of the SBR reactor.

[0013] 2. Data transmission and processing module:

[0014] The data transmission and processing module includes an industrial computer (2), a data transmission line (4), a pH probe (11), a dissolved oxygen probe (12), and a COD probe (13). The apparent images of the sludge captured by the overhead near-infrared / visible light binocular camera (5) and the sludge settling images captured by the side-view near-infrared / visible light binocular camera (6) are transmitted to the industrial computer (2) through the data transmission line (4). At the same time, the industrial computer (2) is connected to the alarm (3), the LED light supplement lamp (9), the pH probe (11), the dissolved oxygen probe (12), and the COD probe (13) to realize the control of each device and obtain real-time data. Among them, the pH probe (11), the dissolved oxygen probe (12), and the COD probe (13) are used to obtain the operating parameters of the reactor as input of the conventional detection parameters.

[0015] (1) Recognition of sludge apparent images: The overhead near-infrared / visible light binocular camera (5) is used to capture the apparent images of the sludge in the SBR reactor, and the OpenCV module responsible for image processing in the Python software pre-installed in the industrial computer (2) is used to preprocess the captured apparent images of the sludge. First, the non-main area is cropped and the image features are enhanced, and computer vision technology is used to extract features from the apparent images of the sludge in the aerobic, anoxic, and settling stages, including color, line, contour, or more abstract high-level semantic features, and the accuracy of model recognition is tested; secondly, the pre-constructed early warning model for sludge bulking is called to recognize the apparent images of the sludge in the reactor; finally, the recognition results are compared with the normal state to determine whether the SBR reactor is about to bulge, and real-time alarm is given to the system that is about to bulge. In addition, the model can also realize the range prediction of SVI.

[0016] (2) Recognition of sludge settling images: The sludge settling images during the settling stage of the SBR reactor are collected using a side-view near-infrared / visible binocular camera (6). The OpenCV module in the Python software pre-installed on the industrial control computer (2) is used to perform image preprocessing such as bilateral filtering, histogram equalization, and Otsu algorithm binarization to reduce image noise and improve image quality. Then, the image is refined by the connected component labeling method to identify the sludge-water interface and generate sludge settling curves and sludge settling rate curves in real time. Finally, the information of the settling curves is extracted by the industrial control computer (2) to determine the sludge settling performance in the SBR reactor and to provide real-time alarm for the system that is about to experience sludge expansion.

[0017] 3. Data visualization and alarm module:

[0018] The data visualization and alarm module includes an electronic display (1) and an alarm (3). The final processing results of the sludge data are presented on the display (1) in the form of a user interface, showing the sludge bulking status, sludge settling performance and basic operating parameters of the SBR reactor in real time. Based on the early appearance image and settling curve image characteristics of sludge bulking, the alarm (3) sounds an alarm when the sludge is about to expand, reminding the operation and maintenance personnel to intervene in time. At the same time, the industrial control computer (2) combines the prior knowledge of sewage treatment to conduct a comprehensive analysis of the data from the pH probe (11), dissolved oxygen probe (12) and COD probe (13), makes a preliminary judgment on the cause of the sludge bulking and gives operation and control suggestions, thereby realizing early warning and control of sludge bulking.

[0019] In addition, to save energy and reduce costs, an external photovoltaic panel (14) and a storage battery (15) are used to power the entire system. The photovoltaic panel (14) is used to directly convert solar energy into electrical energy, and the storage battery (15) is used to store and convert electrical energy to provide energy for the system.

[0020] Compared with existing early warning methods, the present invention has the following advantages:

[0021] 1) This method combines sludge appearance image and sedimentation image information to comprehensively analyze the early characteristics of sludge bulking, reducing misjudgments and improving the recognition accuracy. It is expected to solve the problems of sudden sludge bulking in sewage treatment plants, high subsequent control costs, and long recovery time.

[0022] 2) This invention can automatically generate sludge settling curves and extract information from them, making up for the lack of key information during the settling process, enabling earlier warning of sludge bulking, reducing the cost of manual monitoring while increasing the monitoring density.

[0023] 3) This device achieves timely early warning and control of sludge bulking through automated intelligent analysis. It is easy to operate and simple to control. Compared with traditional manual inspection and routine testing, it has lower cost, higher identification accuracy, stronger real-time performance, and does not require manual operation, which is in line with the development trend of "smart water management".

[0024] 4) The entire system is powered by a combination of external photovoltaic panels (14) and batteries (15), which saves energy and reduces costs. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the device structure of the present invention;

[0026] Figure 2 A flowchart illustrating the workflow of an early warning device for sludge bulking.

[0027] Figure 3 This is a flowchart of the sludge appearance image processing.

[0028] Figure 4 This is a flowchart of the sludge settling image processing.

[0029] Figure 5 This is a heatmap showing the prediction results of the early warning model for sludge bulking.

[0030] Figure 6 Loss diagram for training and validation of early warning model for sludge bulking.

[0031] Figure 7 The training and validation accuracy of the early warning model for sludge bulking is shown in the figure.

[0032] Figure 8 The graph shows the settlement curve and settlement rate change over 30 minutes. Detailed Implementation

[0033] The operation of this device is divided into the following stages.

[0034] I. Construction of an early-stage experimental platform for sludge bulking

[0035] Reference Figure 1The system's top-view near-infrared / visible light binocular camera (5) and side-view near-infrared / visible light binocular camera (6) are respectively installed on the top and side of the SBR reactor. An LED supplementary light (9) is installed on the top as a supplementary light source, and a built-in luminous scale (10) is used to measure the sludge-water interface position. The top-view near-infrared / visible light binocular camera (5) is fixed at the center position 50cm above the liquid surface of the SBR reactor via a telescopic bracket (7) to collect images of the sludge surface of the SBR reactor; similarly, the side-view near-infrared / visible light binocular camera (6) is installed on the side of the SBR reactor via a tripod (8), 100cm from the reactor wall. A built-in WTW3420 portable multi-parameter analyzer monitors the DO and pH values ​​in the reactor in real time, and a COD probe (13) is used to detect the organic matter content. The overall operation of the reactor is controlled by a wastewater treatment intelligent control device.

[0036] II. Construction of an Early-Stage Intelligent Sensing System for Sludge Bulking

[0037] 1. Acquisition of sludge image data

[0038] (1) Acquisition of sludge appearance images:

[0039] Using the Python programming language integrated development environment PyCharm pre-installed on the industrial control computer (2), a command was sent to turn on the LED supplementary light (9). The OpenCV module was used to call the top-view near-infrared / visible light binocular camera (5) fixed on the top of the SBR reactor. The image acquisition frequency was 1 frame / second, and the sludge appearance images of the SBR reactor in each cycle of the anoxic stage, aerobic stage and sedimentation stage were collected at a high frequency.

[0040] (2) Acquisition of sludge settling images:

[0041] The pre-installed Python program on the industrial control computer (2) was used to write an automatic image acquisition program in advance. The side-viewing near-infrared / visible light binocular camera (6) was used to acquire sludge settling images during each settling stage of the SBR reactor. The image acquisition frequency was set to 1 image / second, and the total acquisition time was 30 minutes. The number of cycles was set to ensure that the total number of images acquired was 1800. The images were named according to the time of capture. After 30 minutes, the acquisition of 1800 settling images was completed. The camera was then disconnected, and the program terminated automatically.

[0042] 2. Data transmission and processing

[0043] The images of sludge appearance and sludge settling acquired by the top-view near-infrared / visible light binocular camera (5) and the side-view near-infrared / visible light binocular camera (6) are transmitted in real time to the designated folder of the industrial control computer (2) via the data transmission line (4) to achieve automatic image acquisition and saving. The images are named according to the time of capture, and the data is stored in the set storage path. The image name "24-03-02_09-25-24" indicates that the image was acquired at 9:25:24 on March 2, 2024, which is convenient for maintenance personnel to call later. At the same time, the online monitoring data of process water quality variables are also transmitted to the industrial control computer (2) to achieve early feature extraction and timely early warning of sludge bulking through comprehensive analysis.

[0044] (1) Sludge appearance image processing workflow: Refer to Figure 3 The acquired sludge appearance images were preprocessed using the OpenCV module, which is responsible for image processing, in the Python software pre-installed in the industrial control computer (2). Each dataset includes appearance images of four different sludge expansion stages, with 1800 images in each category, divided into training and testing sets at a ratio of 4:1. An early warning model for sludge expansion was established using the VGG16 deep learning network. Finally, the appearance image of sludge in the SBR reactor was input into the model, and the output was the sludge expansion stage category, thereby providing real-time alarm for sludge expansion.

[0045] a. Dataset preprocessing: The size of the collected appearance image is 840*840. Modify [0:1000,0:1000] in the input layer to [120:810,730:1440], where [120:810] represents the vertical dimension of the image and [730:1440] represents the horizontal dimension of the image. The sludge appearance image is then cropped to a similar size.

[0046] b. Dataset Splitting: Based on the SVI changes during SBR reactor operation, four sludge bulking stages were defined: "normal, impending bulking, mild bulking, and severe bulking," with 1800 images for each stage. The training and validation sets were split in a 4:1 ratio, with 1440 images per stage used for model training and 360 images per stage used to validate model accuracy. To evaluate the model training effect, the average accuracy (Avg_acc) was used to measure the model's prediction accuracy, calculated as follows:

[0047]

[0048] Where ns is the total number of sample categories (there are 4 sludge bulking categories, ns = 4); i is the sample category label i∈[1,4]; ni represents the total number of samples in category i; nii represents the total number of samples in category i with prediction result i, that is, the number of correctly predicted samples in each category.

[0049] c. Model Construction: Based on the VGG16 deep learning network, the network structure parameters were adjusted, and metrics such as the number of image classification categories, number of iterations, learning rate, and batch size were set. The model file was named, and the storage location of the model after training was specified. Model training and validation were then performed. This model will automatically extract features from sludge appearance images, including color, lines, contours, bubbles, or more abstract high-level semantic features, and then establish an early warning model for sludge bulking.

[0050] d. Model Application: A pre-built model is used to identify the appearance of sludge within the reactor. The identification results are compared with those of a normal state to determine whether the SBR reactor is about to expand, thus providing a real-time alarm for systems where sludge bulking is imminent. Furthermore, the model can also predict the range of SVI (Sludge Vibration Intensity).

[0051] (2) Sludge settling image processing workflow: Refer to Figure 4 The sludge settling images collected in the first 30 minutes of the sedimentation stage were preprocessed using the OpenCV module responsible for image processing in the Python software pre-installed in the industrial control computer (2). First, irrelevant backgrounds were removed by image cropping and the image of the area where the reactor's luminous scale was located was obtained. The reactor position was fixed, and the pixel range of the target area in each batch of images was fixed when the position of the camera for image acquisition remained unchanged. Therefore, the cropping position could remain unchanged after being set by the operator. Then, noise was suppressed and the image quality was improved by performing fine processing such as bilateral filtering, histogram equalization, and Otsu algorithm binarization on the image. Finally, the positional relationship between the mud-water interface and the luminous scale (10) was compared by the connected region marker to obtain the specific value of the mud-water interface, and the scale was automatically normalized to generate sludge settling curves and sludge settling rate curves in real time.

[0052] a. Bilateral filtering: Calculates the spatial distance between the pixel and its surrounding pixels, as well as the grayscale similarity between pixels. The spatial distance weight and the pixel similarity weight are combined to obtain the final filtering weights. For example, in a window centered at q, the weight of a point p in the bilateral filtering process is calculated as follows:

[0053]

[0054] Among them, W q G is the sum of weights for each pixel value within the filtering window, used for weight normalization; r G represents the pixel value weight; s This represents the spatial distance weight.

[0055] b. Histogram Equalization: First, the image is converted to grayscale. Then, the histogram of the grayscale image is calculated, and the frequency of each grayscale level is counted. For each grayscale level in the histogram, its cumulative distribution function (CDF) is calculated to obtain the proportion of pixels with a grayscale level lower than or equal to that level. The formula for calculating the CDF for a grayscale value i is as follows:

[0056]

[0057] Where P(j) represents the frequency of a pixel with gray value j appearing in the image.

[0058] Based on the cumulative distribution function, each gray level is mapped to a new gray level, thus diffusing the densely distributed gray levels in the original image's gray-level distribution histogram into a uniform distribution globally. The gray value of each pixel in the image is then replaced with its corresponding mapped gray value, resulting in the equalized image.

[0059] c. Otsu's Binarization Algorithm: Calculate the grayscale histogram of the grayscale image and count the number of pixels at each grayscale level. For each possible threshold t, divide the image into two categories: foreground (target) and background. Then calculate the mean of the pixels in the two categories:

[0060] σ 2 (t)=w1(t)·w2(t)·[m1(t)-m2(t)] 2

[0061] Where w1(t) and w2(t) represent the proportion of pixels in the two categories to the total number of pixels; m1(t) and m2(t) represent the average gray values ​​of pixels in the two categories.

[0062] d. Connected Component Labeling: Scan each pixel in the image from top left to bottom right, dividing it into background pixels with a grayscale value of 0, foreground pixels with a non-zero grayscale value, and labeled pixels in the foreground that have been assigned identifiers. By labeling the target pixels in the reactor's binary image, each individual connected component forms a labeled block. The data structure records the position and number of tick marks.

[0063] e. Settling Curve Generation: After preprocessing and refining the sludge settling image of the reactor, the number of scale lines above the sludge-water interface is obtained. Based on the actual reactor volume and the interval between each scale line, the position of the sludge-water interface at the current moment is obtained by subtracting the product of the number of scale lines and the unit range from the total range. The scale is then automatically normalized. Finally, a sludge settling curve is plotted with the settling time as the x-axis and the sludge-water interface position as the y-axis.

[0064] Finally, the industrial control computer (2) extracts information from the settling curve in the following two ways. Then, it judges the settling performance of the sludge in the SBR reactor and issues a real-time alarm for the system that is about to experience sludge bulking.

[0065] a. The program automatically calculates the integral area and change of the sludge settling curve and saves them as parameters in real time. Then, it calculates the increment ratio of the integral area of ​​the sludge settling curve in adjacent operating cycles and compares it with the set value.

[0066] b. The industrial control computer (2) automatically generates the sludge settling rate curve and extracts the maximum settling rate for comparison with the set value.

[0067] 3. Data visualization and online early warning

[0068] The sludge online early warning system includes model-based early warning of sludge appearance images and information extraction from sludge settling curves. The combined assessment of these two methods can comprehensively reflect the settling performance of the sludge and accurately identify early characteristics of sludge bulking. When either of the following two methods triggers an alarm, maintenance personnel are alerted to intervene and take appropriate measures.

[0069] (1) Early warning of sludge appearance images: The VGG16 model performs well in predicting the appearance images of sludge in SBR reactors. The accuracy rate reaches over 99% after 10 iterations, and it can accurately identify the image features of early sludge bulking. That is, when the sludge appearance image is input in real time, the model output "warning" indicates that the sludge is about to expand, and the system starts to alarm.

[0070] (2) Early warning of sludge settling images: The industrial control computer (2) extracts information from the settling curve in the following two ways. Then, it judges the settling performance of the sludge in the SBR reactor and issues a real-time alarm for the system that is about to experience sludge bulking. That is, the system starts to alarm when the settling curve meets one of the following two conditions.

[0071] a. If the increase in the integral area of ​​the settling curve is greater than 5 times within 30 minutes in adjacent operating cycles, it indicates that the sludge settling curve has changed significantly and the sludge is about to expand, and the system will automatically alarm.

[0072] b. When the maximum settlement rate of the settlement curve within 30 minutes is less than 2 cm 3 ·s -1 When this occurs, it indicates that the sludge settling performance has deteriorated and the sludge is about to expand, triggering an automatic alarm from the system.

[0073] The final processing results of the sludge data are displayed in real time on the monitor (1) in the form of a user interface, which is a visualization interface of sludge bulking status, sludge settling performance and basic operating parameters. By comparing the identified image with the normal image, the model can successfully identify the current sludge bulking status. For example, during normal operation, the user interface displays "Normal, sludge settling is good, SVI: 60-90"; when the sludge is about to bulk, the system alarms and provides operation prompts based on the online monitoring data of process water quality variables. The user interface displays "About to expand, SVI: 90-120, appearance image / settling curve abnormal, dissolved oxygen is low", reminding the operation and maintenance personnel to intervene in time; when the sludge has already bulked, the interface prompts can help the operation and maintenance personnel to troubleshoot the abnormality and provide relevant information on the subsequent sludge bulking recovery.

[0074] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0075] III. Operation and Maintenance of the Early Warning and Control System for Sludge Bulking

[0076] To ensure the long-term stable operation of the early intelligent sensing system for sludge bulking, staff need to conduct periodic inspections and maintenance of the device. In case of extreme weather (heavy rain, typhoon, hail, etc.), the equipment must be inspected and repaired. Wastewater treatment plant operation and maintenance personnel need to regularly check the integrity and operating status of the system equipment and fill in the operation and maintenance records accurately. The photovoltaic panels (14) need to be cleaned regularly by maintenance personnel to ensure the stability of the system's power supply. The working performance and operating status of the industrial control computer (2) should be checked monthly, and data backup and cleanup should be done in a timely manner.

Claims

1. A computer vision-based method for early warning of sludge bulking in an SBR reactor, the apparatus used in this method comprising: Display (1), Industrial computer (2), Alarm (3), Data transmission line (4), Top-view near-infrared / visible light binocular camera (5), Side-view near-infrared / visible light binocular camera (6), Telescopic bracket (7), Tripod (8), LED supplementary light (9), Illuminated ruler (10), pH probe (11), Dissolved oxygen probe (12), COD probe (13), Photovoltaic panel (14), Storage battery (15); A dual-view system was used to collect real-time images of the activated sludge in each operating cycle of the SBR reactor. A top-view near-infrared / visible binocular camera (5) was fixed at the center of the SBR reactor 50 cm above the liquid surface via a telescopic bracket (7) to collect images of the sludge surface of the SBR reactor. A side-view near-infrared / visible binocular camera (6) was mounted on the side of the reactor via a tripod (8) 100 cm away from the reactor wall to obtain the settling performance of the sludge inside the SBR reactor. An LED supplementary light (9) was used to provide auxiliary light when the lighting conditions were poor. A luminous scale (10) was used to provide a reference for the height of the sludge-water interface. Both the top-view near-infrared / visible binocular camera (5) and the side-view near-infrared / visible binocular camera (6) were controlled by an industrial control computer (2), and the collected images were transmitted to the industrial control computer (2) in real time via a data transmission line (4). The device is divided into three main modules according to its functions: image acquisition module, data transmission and processing module, and data visualization and alarm module. Image acquisition module To monitor the properties of activated sludge in the SBR reactor in real time, a top-view near-infrared / visible light binocular camera (5) for collecting images of the sludge appearance and a side-view near-infrared / visible light binocular camera (6) for collecting images of sludge settling are installed on the top and sides of the SBR reactor, respectively. The industrial control computer (2) issues a command to call the top-view near-infrared / visible light binocular camera (5) to collect images of the sludge appearance during the aerobic, anoxic and sedimentation stages of the SBR reactor at a high frequency. The LED supplementary light (9) can be turned on or off depending on the lighting conditions. At the same time, the side-view near-infrared / visible light binocular camera (6) is called to collect images of sludge settling during the sedimentation stage of the SBR reactor. Data transmission and processing module The data transmission and processing module includes an industrial computer (2), a data transmission line (4), a pH probe (11), a dissolved oxygen probe (12), and a COD probe (13). The data transmission line (4) is used to transmit the sludge appearance image collected by the top-view near-infrared / visible light binocular camera (5) and the sludge settling image collected by the side-view near-infrared / visible light binocular camera (6) to the industrial computer (2). At the same time, the industrial computer (2) is connected to the alarm (3), LED supplementary light (9), pH probe (11), dissolved oxygen probe (12), and COD probe (13) to realize the control of each device and data acquisition. Among them, the pH probe (11), dissolved oxygen probe (12), and COD probe (13) are used to acquire reactor operating parameters as input for routine detection parameters. Sludge appearance image recognition The image of the sludge was preprocessed using the OpenCV module in the Python software pre-installed on the industrial control computer (2). First, non-major areas were cropped and image feature enhancement was performed. Computer vision technology was used to extract features from the sludge images of the aerobic, anoxic, and sedimentation stages, and the accuracy of the model recognition was tested. Second, the pre-built early warning model for sludge bulking was called to recognize the image of the sludge in the reactor. Finally, the recognition results were compared with the normal state to determine whether the SBR reactor was about to expand. Then, the system that was about to experience sludge bulking was given a real-time alarm, i.e., the sludge image was input in real time. When the model output was "warning", it indicated that the sludge was about to expand and the system started to alarm. The model was the VGG16 model. Sludge settling image recognition The image quality of the collected sludge settling images is improved by using the OpenCV module in the Python software pre-installed on the industrial control computer (2) to perform image filtering, histogram equalization, and Otsu algorithm binarization. Then, the image is refined by the connected component labeling method to identify the sludge-water interface and generate sludge settling curves and sludge settling rate curves in real time. Finally, the information of the settling curves is extracted by the industrial control computer (2) to determine the settling performance of sludge in the SBR reactor and to provide real-time alarm for the system that is about to experience sludge expansion. The system will trigger an alarm when the settlement curve meets the following conditions: When the maximum settlement rate of the settlement curve within 30 minutes is less than 2 cm 3 ·s -1 When this occurs, it indicates that the sludge settling performance has deteriorated and the sludge is about to expand, and the system will automatically alarm. Data visualization and alarm module The data visualization and alarm module includes an alarm (3) and an electronic display (1); the final processing results of sludge data are presented on the display (1) in the form of a user interface, and the sludge expansion status, sludge settling performance and basic operating parameters of the SBR reactor are displayed in real time; the alarm (3) will sound an alarm when the sludge is about to expand; at the same time, the industrial control computer (2) combines the prior knowledge of sewage treatment to conduct a comprehensive analysis of the data from the pH probe (11), dissolved oxygen probe (12) and COD probe (13), make a preliminary judgment on the reasons for the sludge expansion and give operation and control suggestions, so as to realize early warning and control of sludge expansion.

Images