Sludge settling performance intelligent diagnosis and early warning system and method based on full-automatic visual analysis

Abstract

The invention discloses a sludge settling performance intelligent diagnosis and early warning system and method based on full-automatic visual analysis, and relates to the technical field of sewage treatment. The system comprises an imaging unit, a laser calibration unit, an image processing module, an image segmentation and tracking module, a data analysis unit and an ultrasonic cleaning device. The method comprises the steps of sampling, laser calibration and image acquisition, preprocessing, segmentation tracking, parameter calculation, early warning and cleaning. According to the method, a pixel millimeter mapping relation is established through 635nm laser calibration, UNet segmentation and an LSTM prediction model are combined, online MLSS data are fused, high-precision interface tracking is realized, Vs, Cc, dynamic SVIt and other parameters are output in real time, and the early warning accuracy is improved 2-6 hours before sludge bulking / disintegration. According to the scheme, the problems that traditional manual detection is large in error and high in hysteresis quality and an existing visual scheme is low in precision are solved, the measurement efficiency is improved, and the method is suitable for AAO, MBR and other processes.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an intelligent diagnostic and early warning system and method for sludge settling performance based on fully automated visual analysis. Background Technology

[0002] In wastewater treatment plant operation, sludge settling performance is a core indicator for assessing the health of the activated sludge system, directly determining the solid-liquid separation effect of the secondary sedimentation tank and the effluent quality. Currently, sludge settling performance testing mainly relies on manual testing of sludge volume index (SVI) and sludge settling ratio (SV30). The procedure involves manually collecting mixed liquor from the aeration tank or secondary sedimentation tank, injecting it into a 1000mL graduated cylinder, allowing it to stand for 30 minutes, reading the SV30, and then calculating the SVI by combining it with the laboratory-measured mixed liquor suspended solids concentration (MLSS). However, this method has significant drawbacks: large subjective error, with measurement errors reaching ±5% due to visual errors, cylinder accuracy, and bubble interference when manually reading the interface height, and readings from different operators differing by more than 10%; strong data lag, as SV30 only provides 30 minutes of single-point data, which cannot reflect dynamic parameters such as free settling velocity (Vs) and compression point concentration (Cc), making it difficult to provide early warning of sludge bulking or disintegration; poor real-time performance, as it can only detect 12 times per day, making it unable to capture instantaneous settling deterioration caused by sudden changes in influent load or toxic shocks; and lack of environmental compensation, as water temperature fluctuations and turbidity changes affect the settling process, but there are no temperature control and turbidity compensation methods.

[0003] In recent years, some technologies have attempted to improve detection through machine vision, using cameras to capture the sedimentation process and identify sludge interfaces. However, technical bottlenecks remain: significant perspective errors, non-orthogonal shooting leading to distortion in interface height measurement with errors exceeding ±2mm; severe turbidity interference, making it difficult to clearly identify sludge interfaces in high-turbidity water bodies with NTU>100; lack of high-precision calibration, with most systems not integrating calibration devices, making millimeter-level measurements impossible; isolated data, with the vision system not integrated with online MLSS sensors, and parameter calculations relying on manual input; and low reliability of early warnings, with warnings for abnormal operating conditions relying on empirical thresholds, resulting in a false alarm rate exceeding 30%. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent diagnostic and early warning system and method for sludge settling performance based on fully automated visual analysis. It aims to solve the problems of "low accuracy, poor timeliness, and insufficient automation" in existing sludge settling performance testing. Specifically, it includes: eliminating subjective errors from manual testing and perspective errors from existing visual methods to achieve millimeter-level interface tracking; integrating online MLSS data to output dynamic parameters such as Vs, Cc, and dynamic SVIt in real time; using artificial intelligence models to achieve early warning of abnormal operating conditions, reducing false alarm rates; and integrating self-cleaning functions to ensure long-term stable operation.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A fully automated visual analysis-based intelligent diagnosis and early warning system for sludge settling performance includes: Imaging unit: Used to acquire images of the sludge settling process in the settling tank. It includes a 5-megapixel CMOS camera (IMX586), 30fps frame rate, 650nm high-pass filter and 850nm near-infrared auxiliary light source, 10W power and penetration depth >10cm. It uses orthogonal shooting to acquire the macroscopic trajectory of the interface to avoid perspective distortion. Laser calibration unit: A 635nm laser with 5mW power is used to project dual reference lines on the inner wall of the settling tank with a spacing of 50.0±0.1mm, establishing a mapping relationship between pixels and millimeters; an integrated PID temperature control module is used to stabilize the temperature inside the settling tank at 20±0.5℃; after every 10 tests, the equipment is automatically calibrated by inserting a standard calibration plate with a precision of ±0.01mm through a robotic arm, eliminating equipment drift error; Image processing module: used to preprocess the images acquired by the imaging unit, including dividing the image into 8×8 blocks for illumination equalization, using a 5×5 pixel kernel for Gaussian blurring to suppress noise, superimposing the equalized image and the blurred image with a weight of 1.5:0.5 to achieve sharpening and enhancement, and removing bubbles with a diameter <2mm through a 3×3 kernel morphological opening operation. Image segmentation and tracking module: Based on the UNet model, the input size is 512×512 pixels, and the output is a binary image of the sludge region. The training data consists of 100,000 sedimentation images covering different MLSS and turbidity with an intersection-over-union ratio (mIoU) > 0.95. The highest point height of the sludge interface contour line is extracted, and the height sequence is smoothed by Kalman filtering. Anomalies such as interface breakage, instantaneous bubble occlusion, and sudden changes in illumination are handled by linear interpolation completion, 5-frame window mid-range filtering to remove outliers, and automatic lighting adjustment and re-acquisition. Data Analysis Unit: Communicates with the online MLSS sensor (Hach Solitax SC model) with a data refresh rate of 1Hz to acquire online MLSS data in real time; Based on the height sequence acquired by the imaging unit and laser calibration unit, it calculates parameters such as free settling velocity (Vs), compressibility point concentration (Cc), dynamic SVT, and sludge layer density (ρ30); It integrates an LSTM time-series prediction model to trigger multi-level early warnings (P1-P4 levels) based on parameter changes. Ultrasonic cleaning device: The ultrasonic transmitter with a frequency of 40kHz and a power of 200W is externally coupled and installed on the outer wall of the settling tank to avoid direct contact with the sample; it automatically starts after the test, emits ultrasonic vibration for 30 seconds to remove sludge residue and biofilm from the inner wall, then starts 50℃ heat drying for 10 seconds, and finally performs optical self-inspection to ensure the accuracy of the next test.

[0006] This invention also provides a method for intelligent diagnosis and early warning of sludge settling performance based on fully automated visual analysis, using the intelligent diagnosis and early warning system described above, including the following steps: S1: Sampling: A telescopic titanium alloy pipette with a diameter of 8mm, a stroke of 0.52m, and an integrated ceramic anti-clogging nozzle at the end is used to insert into the sewage treatment tank to a depth of 50cm below the liquid surface; the mixed liquid is extracted at a rate of 200mL / s using a high-precision peristaltic pump with a flow rate error of ±1%, and closed-loop control with an electromagnetic flowmeter error of ±3mL is used to ensure that the extracted volume is 1000±3mL; the electromagnetic flowmeter verifies the volume in real time, and triggers supplementary sampling if an abnormality is found; S2: Laser calibration and image acquisition: The sampled mixture is injected into a quartz glass settling tank with a thickness of 5mm and a light transmittance of >92%; the laser calibration unit is activated to project a 635nm dual reference line; the PID temperature control module is activated to stabilize the temperature in the settling tank to 20±0.5℃ within 5 minutes; the CMOS camera of the imaging unit is simultaneously turned on at 30fps with an 850nm near-infrared auxiliary light source to acquire images of the sludge settling process. S3: Image preprocessing: Preprocess the acquired image, including 8×8 small block illumination equalization to avoid local overexposure or underexposure, 3×3 kernel morphological opening operation to remove bubbles with a diameter <2mm, 5×5 pixel kernel Gaussian blur to suppress noise, and 1.5:0.5 weight superposition equalization of the image and the blurred image to achieve sharpening and enhancement. S4: Interface Segmentation and Tracking: The preprocessed image is input into the UNet model, and a binary image of the sludge region is output. The height of the highest point of the sludge interface contour line in the binary image is extracted. Combined with the pixel-millimeter mapping relationship of the laser calibration unit, the actual height sequence is obtained. The height sequence is smoothed by Kalman filtering. For interface breakage, instantaneous bubble occlusion, and sudden change in illumination, linear interpolation, 5-frame window mid-range filtering, and automatic illumination adjustment and re-acquisition are used respectively. S5: Parameter Calculation: Dynamic parameters are calculated based on the height sequence and online MLSS data: ① Free settling velocity (Vs): After removing the initial disturbance data of the first 30 seconds, linear regression is performed on the height sequence of 0.5-5 minutes, and the result is calculated according to Vs=ΔH / Δt, with an accuracy of ±0.03m / h; ② Compressibility point concentration (Cc): The second derivative of the height sequence is calculated, and the inflection point of the second derivative from positive to negative is located at time tc. The height Hc at this time point is taken, and the result is calculated according to Cc=MLSS online×(Ho / Hc), where Ho is the initial height at t=0, with an accuracy of ±0.15g / L; ③ Dynamic SVI: The result is calculated according to SVI=(Ht / Ho×1000) / MLSS online, and the SVI values ​​at t=5, 10, 20, and 30 minutes are output; ④ Sludge layer density (ρ30): The result is calculated according to ρ30=MLSS online / (H30 / Ho), where H30 is the sludge layer height at t=30 minutes, with an accuracy of ±2%; S6: Early Warning: Analyze parameter change trends using an LSTM time-series prediction model to trigger multi-level early warnings: ① P1 (Emergency): When Vs < 2 m / h and Cc < 6 g / L, send a notification: "Severe risk of bulking! Immediately: ① Add sodium hypochlorite 10 mg / L ② Adjust SRT"; ② P2 (High Risk): When SVI5 / SVI30 > 1.5 and ρ30 < 25 g / L, send a notification: "Sludge disintegration! Recommendation: ① Reduce aeration by 50% ② Add PAM 2 mg / L"; ③ P3 (Warning): When dynamic SVI fluctuation > 20%, send a notification: "Settling unstable! Check: ① Influent load ② Return ratio"; ④ P4 (Reminder): When daily ρ30 decrease > 5%, send a notification: "Dewatering performance declines! Optimize: ① Sludge discharge frequency ② Conditioner addition"; Early warning response time < 1 hour; S7: Ultrasonic Cleaning: After the warning is completed, start the ultrasonic cleaning device. The 40kHz ultrasonic transmitter vibrates for 30 seconds, followed by 50℃ hot drying for 10 seconds. Finally, perform optical self-inspection to ensure that there are no residues on the inner wall of the settling tank.

[0007] Compared with the prior art, the present invention has the following beneficial effects: 1. Significantly improved accuracy: Through 635nm laser calibration and orthogonal imaging, the interface tracking accuracy reaches ±0.1mm, far exceeding the ±5% error of traditional manual methods and the >±2mm error of existing vision solutions.

[0008] 2. Enhanced real-time performance and dynamism: Real-time monitoring is achieved at a frame rate of 30fps, and dynamic parameters such as Vs, Cc, and dynamic SVIt can be output, solving the shortcomings of traditional methods that "only provide single-point data and lack dynamic processes".

[0009] 3. Reliable early warning: The LSTM model can provide early warning 2-6 hours before sludge bulking / disintegration, with an accuracy rate of 95% and a false alarm rate of less than 5%, which is a significant improvement over the existing empirical threshold warning false alarm rate of >30%.

[0010] 4. High degree of automation: It integrates fully automatic sampling, laser calibration, image analysis and self-cleaning functions, requiring no manual intervention and improving measurement efficiency compared to traditional methods.

[0011] 5. Strong environmental adaptability: PID temperature control at 20±0.5℃ and near-infrared light source penetration depth >10cm can offset interference from water temperature and turbidity, making it suitable for various wastewater treatment processes such as AAO and MBR. Detailed Implementation

[0012] The technical solutions of the present invention will be further described below with reference to the embodiments.

[0013] A fully automated visual analysis-based intelligent diagnosis and early warning system for sludge settling performance includes: Imaging unit: The camera is a 5-megapixel IMX586 CMOS camera with a frame rate of 30fps and a 650nm high-pass filter, paired with a 10W 850nm near-infrared LED array with a penetration depth of >10cm. The camera is installed directly above the laser calibration settling tank and uses an orthogonal shooting method to ensure that the optical axis of the lens is perpendicular to the liquid surface of the settling tank, avoiding perspective errors and acquiring macroscopic trajectory images of the sludge interface.

[0014] Laser calibration unit: A 5mW 635nm laser is used. The laser is installed on the inner wall of one side of the settling tank and projects two parallel reference lines with a spacing of 50.0±0.1mm, which are orthogonal to the camera's field of view, establishing a "pixel-millimeter" mapping relationship. The temperature control module uses a PID algorithm and controls the heating element and cooling fan through an embedded controller (STM32F407) to stabilize the temperature in the settling tank at 20±0.5℃. After every 10 tests, a UR5 robotic arm automatically inserts a ceramic standard calibration plate with a precision of ±0.01mm into the settling tank to calibrate the laser reference line spacing and eliminate equipment drift.

[0015] Image processing module: Built on an ARM Cortex-A72 processor (Raspberry Pi 4B), running the OpenCV image processing library; the image acquired by the camera is first divided into an 8×8 grid, and the contrast is adjusted block by block to achieve illumination equalization; then, morphological opening operation with 3×3 cores is used to remove bubbles with a diameter <2mm by erosion and then dilation; subsequently, a Gaussian blur algorithm with 5×5 pixel cores is used to smooth image noise; finally, the equalized image and the blurred image are superimposed with a pixel value weight of 1.5:0.5 to highlight the edge of the sludge interface.

[0016] Image segmentation and tracking module: A UNet model is built based on the PyTorch framework. The input image size is adjusted to 512×512 pixels. The training dataset consists of 100,000 labeled images covering scenes with MLSS 1-8 g / L and turbidity 50-500 NTU. After training, the model's mIoU > 0.95. The upper edge contour of the sludge region is obtained using the contour extraction algorithm cv2.findContours. The pixel coordinates of the highest point of the contour are taken and converted into the actual height using the mapping relationship of laser calibration. The state equation of the Kalman filter is: Hk+1=Hk+Vs×Δt; The observation equation is Zk=Hk+ωk, which is a smooth height sequence. When a discontinuous interface fracture contour is detected, it is completed by linear interpolation of the height of the preceding and following 5 frames. When there is a sudden change in height >5mm due to instantaneous bubble occlusion, outliers are removed by median filtering of 5 frames. When the average gray level of the image changes by >20% due to a sudden change in illumination, the camera is automatically adjusted to adjust the power of the near-infrared light source and re-acquire the current frame.

[0017] Data Analysis Unit: Communicates with the Hach Solitax SC online MLSS sensor via RS485 bus, acquiring online MLSS data at a 1Hz frequency; the parameter calculation program is written in C++, calculating parameters such as Vs, Cc, and dynamic SVI10 in real time; the LSTM early warning model is built using the TensorFlow framework, with the input being the parameter sequence Vs, Cc, SVI5, and SVI10 from the past 30 minutes, and the output being the predicted parameter values ​​for the next 6 hours. When the predicted value triggers the early warning threshold, the early warning information is pushed to the wastewater treatment plant's SCADA system via Ethernet.

[0018] The ultrasonic cleaning device uses a 40kHz frequency, 200W piezoelectric ceramic transducer (model CSB40-200), which is installed on the bottom of the outer wall of the settling tank via silicone grease coupling to avoid direct contact with the sludge sample. After the test is completed, the controller STM32F407 outputs a trigger signal, and the transducer vibrates for 30 seconds, using the ultrasonic cavitation effect to remove sludge residue and biofilm from the inner wall. Then, the heating belt wrapped around the outer wall of the settling tank with a power of 50W is activated to raise the tank temperature to 50℃ and maintain it for 10 seconds to achieve thermal drying. Finally, a blank settling tank image is taken with a camera. If the average grayscale value of the image changes by less than 5%, the cleaning is deemed qualified and the optical self-test is passed; otherwise, the cleaning process is restarted.

[0019] In addition to the core parameters, the CMOS camera of the imaging unit is equipped with an adjustable polarization filter with an angle of 0-180°. The angle of the filter is adjusted every 15° by rotating it to eliminate the interference of water surface reflection on the image. The near-infrared auxiliary light source adopts an array of 6 LEDs, each with a power of 1.67W, which are evenly distributed around the camera lens to ensure uniform illumination in the settling tank and avoid local shadows affecting interface recognition.

[0020] The dual reference lines of the laser calibration unit are projected in an "upper and lower" distribution. The upper reference line is 50 mm from the top of the settling tank, and the lower reference line is 50.0 ± 0.1 mm from the upper reference line. The standard calibration plate is made of ceramic material and has 10 scale lines with a spacing of 10 mm and an accuracy of ± 0.01 mm. When the robotic arm is inserted, it ensures that the calibration plate is parallel to the laser reference line. The camera identifies the overlap between the scale lines and the laser lines and automatically corrects the "pixel-millimeter" mapping coefficient.

[0021] The LSTM model of the intelligent analysis and early warning unit contains 3 hidden layers with 128 neurons per layer. It adopts the Adam optimizer and MSE loss function, with a training set to test set ratio of 8:2 and a model prediction error of <3%. The thresholds for multi-level early warnings can be manually adjusted through the SCADA system to adapt to the process differences of different wastewater treatment plants. In addition to text push, the early warning information also includes parameter change curves over the past 30 minutes, such as the Vs time series curve and the SVIt curve, to facilitate operation and maintenance personnel in judging the operating conditions.

[0022] The automatic diagnosis and early warning method based on the above system includes the following steps: S1 Sampling: The drive mechanism of the telescopic titanium alloy straw uses a stepper motor (model 28BYJ-48) and achieves a 0.52m stroke extension / retraction via a lead screw drive; the ceramic anti-clogging suction head has a 3mm orifice diameter to prevent large particles from clogging the straw; the peristaltic pump (model BT100-2J) uses a corrosion-resistant silicone tube with an inner diameter of 6mm; the electromagnetic flowmeter (model LDG-15) is installed in the pipeline between the peristaltic pump and the settling tank to provide real-time flow data feedback; when a volume deviation of >3mL is detected, the controller triggers the peristaltic pump to re-extract.

[0023] S2 Laser Calibration and Image Acquisition: The settling tank is made of quartz glass, measuring Φ100mm×500mm and 5mm thick, with a light transmittance >92%, ensuring the penetration of laser and near-infrared light; the temperature sensor of the temperature control module is a DS18B20, inserted into the side wall of the settling tank, with an accuracy of ±0.1℃, providing real-time temperature data feedback, and the PID controller adjusts the heating power according to the deviation; image acquisition and laser calibration are started synchronously, and each frame of the image includes the laser baseline to ensure the accuracy of height calculation.

[0024] S3 Image Preprocessing: Illumination equalization uses the "CLAHE" algorithm for contrast-limited adaptive histogram equalization to avoid local overexposure; the morphological opening operation uses a 3×3 rectangular kernel as the structuring element; the standard deviation of Gaussian blur is set to 1.5 to preserve interface details while denoising; sharpening uses the "Laplacian operator" superimposed on the original image to further highlight the interface edges.

[0025] S4 Interface Segmentation and Tracking: The output layer of the UNet model uses the "Sigmoid" activation function to map pixel values ​​to the 0-1 range, and generates a binary image with a threshold of 0.5; the contour extraction uses the "CHAIN_APPROX_SIMPLE" algorithm to simplify the number of contour points; the sampling frequency of the height sequence is consistent with the camera frame rate of 30Hz to ensure that no dynamic process is missed.

[0026] S5 parameter calculation: When calculating Vs, the least squares method is used for linear regression, and the fitted results with R² < 0.95 are considered as initial disturbances not yet resolved; when calculating Cc, the second derivative is calculated using the "central difference method". d²H / dt²=(Hk+1-2Hk+Hk-1) / Δt²; Inflection point location is achieved by finding the first time point at which the second derivative changes from positive to negative; the calculation results of dynamic SVIt and ρ30 are stored in real time on an SD card with a capacity of 32GB, retaining 30 days of historical data.

[0027] S6 Alerts: Alert response time is reduced to less than 0.5 seconds per inference by optimizing the inference speed of the LSTM model. Combined with data acquisition and processing time, the total response time is less than 1 hour. Alert information push supports both SMS and SCADA pop-up notifications to ensure timely receipt by maintenance personnel.

[0028] Ultrasonic cleaning is initiated after the early warning information is pushed out, to avoid affecting parameter calculations during the cleaning process. During the cleaning process, the controller monitors the transducer's vibration current in real time. The normal range is 1-1.5A. If the current is abnormal (>2A or <0.5A), the cleaning is stopped and a "cleaning fault" prompt is pushed out. The temperature of the heat drying is monitored in real time by the DS18B20 sensor to prevent the temperature from being too high and damaging the settling tank.

[0029] During sampling, the insertion depth of the telescopic pipette is precisely controlled by the number of pulses of the stepper motor, with each pulse corresponding to a stroke of 0.1 mm, ensuring that it extends 50 cm below the liquid surface; the pumping rate of the peristaltic pump is adjusted by adjusting the motor speed to achieve 200 mL / s corresponding to a rotation speed of 1500 rpm; the electromagnetic flowmeter is calibrated once a day using a standard volumetric flask with a volumetric accuracy of ±0.1 mL, ensuring that the volume measurement error is <3 mL.

[0030] All image preprocessing operations are accelerated on the GPU of Raspberry Pi 4B using the CUDA version of OpenCV, with a single frame processing time of <10ms; the process noise covariance Q of Kalman filtering is set to 0.01, and the observation noise covariance R is set to 0.1, balancing the smoothing effect and response speed; the triggering conditions for anomaly handling are implemented through "threshold judgment", for example, the criterion for "interface breakage" is "the number of consecutive pixels on the upper edge of the contour is <80% of the total width".

[0031] The calculation accuracy of Vs is ensured by automatic calibration through laser calibration, which is calibrated once every 10 tests, and the long-term accuracy is stable at ±0.03m / h. The calculation accuracy of Cc is optimized by "second derivative smoothing" using a 5-frame moving average to avoid instantaneous fluctuations affecting the inflection point location. The warning threshold can be "self-learned and adjusted" based on the historical data of the sewage treatment plant. For example, when the normal Vs range of a certain plant is 3-5m / h, the system can automatically adjust the Vs threshold of the P1 level warning to 3m / h to improve the targeting of the warning.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A fully automated visual analysis-based intelligent diagnosis and early warning system for sludge settling performance, characterized in that, include: The imaging unit is used to acquire images of the sludge settling process in the settling tank, and an orthogonal shooting method is used to avoid perspective error. The laser calibration unit is used to project a baseline to establish the mapping relationship between pixels and millimeters, and to control the temperature inside the settling tank. The image processing module is used to perform illumination equalization, noise suppression, sharpening enhancement, and bubble filtering on the images acquired by the imaging unit. The image segmentation and tracking module outputs a binary map of the sludge region based on the UNet model, extracts the height of the highest point of the sludge interface, and smooths the height sequence through Kalman filtering to handle abnormal working conditions. The data analysis unit communicates with the online MLSS sensor, calculates dynamic parameters based on altitude sequence and online MLSS data, and integrates an LSTM time series prediction model to trigger multi-level early warnings.

2. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 1, characterized in that, It also includes an ultrasonic cleaning device; the ultrasonic transmitter of the ultrasonic cleaning device is externally coupled and installed on the outer wall of the settling tank. After the test is completed, it automatically starts vibration to remove the residue on the inner wall, and then performs heat drying and completes optical self-inspection.

3. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 1, characterized in that, The imaging unit includes a CMOS camera, a 30fps frame rate, a 650nm high-pass filter, and an 850nm near-infrared auxiliary light source; the CMOS camera is mounted directly above the settling tank, and the optical axis of the lens is perpendicular to the liquid surface of the settling tank.

4. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 1, characterized in that, The laser calibration unit uses a 635nm laser to project dual reference lines with a spacing of 50.0±0.1mm. The laser calibration unit also includes a PID temperature control module to stabilize the temperature in the settling tank at 20±0.5℃, and automatically calibrates the system after every 10 tests by inserting a standard calibration plate with a precision of ±0.01mm using a robotic arm.

5. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 1, characterized in that, The image segmentation and tracking module uses a UNet model with an input size of 512×512 pixels and 100,000 labeled subsidence images as training data, with an intersection-over-union ratio (mIoU) > 0.

95. The data analysis unit has a multi-level early warning system, including P1 (emergency), P2 (high risk), P3 (warning), and P4 (prompt), with an early warning response time of < 1 hour.

6. A method for intelligent diagnosis and early warning of sludge settling performance based on fully automated visual analysis, characterized in that, Using the intelligent diagnosis and early warning system as described in any one of claims 1 to 5 includes the following steps: Step S1: Sampling. A set volume of mixed liquid is extracted from the sewage treatment tank using a telescopic suction tube in conjunction with a peristaltic pump and an electromagnetic flow meter. Step S2: Laser calibration and image acquisition. Inject the mixed liquid into the settling tank, start the laser calibration unit and imaging unit, and acquire images of the sludge settling process. Step S3: Image preprocessing, including illumination equalization, bubble filtering, noise suppression, and sharpening enhancement of the acquired image; Step S4: Interface segmentation and tracking. Input the preprocessed image into the UNet model to obtain a binary map of the sludge region. Extract the height of the highest point of the interface and smooth it to handle abnormal working conditions. Step S5: Parameter calculation: Calculate free settling velocity, compression point concentration, dynamic SVT and sludge layer density based on height sequence and online MLSS data; Step S6: Early warning. Analyze parameter trends using the LSTM model, trigger multi-level early warnings, and push early warning information.

7. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 6, characterized in that, It also includes step S7: ultrasonic cleaning; step S7 is performed after step S6, starting the externally installed ultrasonic transmitter to vibrate for 30 seconds, followed by heat drying at 50°C for 10 seconds, and finally performing optical self-test.

8. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 6, characterized in that, In step S1, the telescopic pipette is made of titanium alloy and has an integrated ceramic anti-clogging nozzle at the end. It is inserted 50cm below the liquid surface and draws 1000±3mL of the mixture at a rate of 200mL / s. The peristaltic pump has a flow error of ±1%, and the electromagnetic flowmeter has an error of ±3mL. In case of abnormality, supplementary sampling is triggered.

9. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 6, characterized in that, The image preprocessing in step S3 includes: dividing the image into 8×8 blocks for illumination equalization, using 3×3 kernel morphological opening to remove bubbles with a diameter <2mm, suppressing noise with 5×5 pixel kernel Gaussian blur, and superimposing the equalized image and the blurred image with a weight of 1.5:0.5 to achieve sharpening; the anomaly handling in step S4 includes: linear interpolation completion when the interface is broken, median filtering of 5 frames when there is instantaneous bubble occlusion, and automatic lighting replenishment and re-acquisition when there is a sudden change in illumination.

10. The intelligent diagnosis and early warning system for sludge settling performance based on fully automated visual analysis as described in claim 6, characterized in that, The parameter calculation in step S5 includes: removing the first 30 seconds of data from the free settling velocity, linearly fitting the 0.5-5 minute height sequence with an accuracy of ±0.03 m / h; calculating the compression point concentration by locating the inflection point using the second derivative with an accuracy of ±0.15 g / L; outputting the values ​​of dynamic SVIt at t=5, 10, 20, and 30 minutes; and calculating the sludge layer compaction with an accuracy of ±2%. The early warning in step S6 includes: P1 level: Vs < 2m / h and Cc < 6g / L; Grade P2: SVI5 / SVI30 > 1.5 and ρ30 < 25 g / L; Level P3: Dynamic SVI fluctuation >20%; Level P4: ρ30-day decrease >5%.