A mud scraper chain recognition method based on a quadruped robot

Abstract

The application discloses a kind of based on four-footed robot's mud scraper chain identification method, comprising the following steps: step S1, inspection path planning and image acquisition;Step S2, image pre-processing and instance segmentation;Step S3, geometric feature calculation and abnormality determination;If abnormal, then enter step S4;If normal, then return step S1 and continue image acquisition;Step S4, time sequence cross verification;Step S5, result output and alarm.The application solves the deficiencies in the prior art, realizes early detection, early warning, early disposal of chain link shedding, gap anomaly and the like.

Description

Technical Field

[0001] This invention relates to intelligent inspection and machine vision fault detection technology for wastewater treatment equipment, and more specifically, to a method for identifying the chain of a sludge scraper based on a quadruped robot. Background Technology

[0002] The scraper chain is the core transmission component of the sedimentation tank sludge collection system. It drives the scraper to collect sludge from the bottom of the tank into the sludge hopper, ensuring effective sludge-water separation. Regular chain inspection is crucial for timely detection of elongation, cracks, and derailment caused by long-term wear, wastewater corrosion, or tension changes, preventing sudden breakage. If defects are detected, the tensioning device must be adjusted or worn links or the entire chain replaced, depending on the severity. If left untreated, the chain may suddenly break, causing the scraper to jam, the motor to burn out, leading to anaerobic sludge accumulation at the bottom of the tank, and a large amount of sludge floating to the surface. Ultimately, this can cause the sedimentation tank to shut down, the biological treatment system to be impacted, and the effluent to exceed suspended solids standards, potentially resulting in environmental penalties and high costs for tank cleaning and repair.

[0003] Traditional wastewater treatment plant inspections primarily rely on manual on-site inspections, fixed-point camera monitoring, and contact sensor monitoring. However, manual inspections suffer from low frequency, incomplete coverage, high safety risks associated with working at heights, delayed defect detection, reliance on personnel experience for judgment, and difficulty in quantifying and tracing data. Fixed cameras have fixed positions and limited viewing angles, resulting in blind spots and obstructions, and are susceptible to interference from light, water mist, and sludge, leading to poor recognition performance. Contact sensors are exposed to humid, corrosive, and high-vibration environments, making them prone to damage and failure, with short lifespans and high maintenance costs. Furthermore, they can only monitor the overall operating status and cannot identify minor local defects such as loose links, detachments, or deformations. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method for identifying the chain of a sludge scraper based on a quadruped robot, thereby overcoming the deficiencies of existing technologies and enabling early detection, early warning, and early handling of faults such as chain link detachment and abnormal gaps.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for identifying the chain of a sludge scraper based on a quadruped robot includes the following steps:

[0007] Step S1: Inspection path planning and image acquisition;

[0008] Step S2, image preprocessing and instance segmentation;

[0009] Step S3: Geometric feature calculation and anomaly detection; if an anomaly is found, proceed to step S4; if normal, return to step S1 to continue image acquisition.

[0010] Step S4, timing cross-validation;

[0011] Step S5: Output the results and issue alarms.

[0012] Preferably, step S1 specifically includes:

[0013] The inspection path of the quadruped robot is pre-calibrated, and the shooting parameters of the sludge scraper monitoring points are set.

[0014] When the quadruped robot reaches the monitoring point of the sludge scraper, it automatically adjusts its posture and takes images of the chain.

[0015] Preferably, the shooting parameters include shooting angle, shooting distance, and exposure value;

[0016] The quadruped robot takes at least three images from different angles.

[0017] Preferably, step S2 specifically includes:

[0018] The captured image is input into the YOLOv8 instance segmentation model, which outputs the mask region, class label, and confidence score of each link in the chain.

[0019] Preferably, the calculation of the YOLOv8 instance segmentation model is as follows:

[0020] Input preprocessing involves scaling the chain images acquired with optimized shooting parameters to the fixed input size of the model, and then normalizing and augmenting them.

[0021] Feature extraction: Multi-scale feature maps of chain images are extracted through the CSPDarkNet backbone network in the YOLOv8 instance segmentation model to capture chain segment contours, textures and local anomaly details.

[0022] Feature fusion: The neck network fuses multi-scale feature maps of different scales to enhance the ability to express small targets and edge information;

[0023] Predicted output

[0024] Simultaneous prediction of the detection head branch:

[0025] The bounding box and category of each chain instance;

[0026] A corresponding set of prototype mask coefficients is linearly combined with the prototype mask to generate a pixel-level binary mask for each instance;

[0027] Post-processing:

[0028] Redundant detection is filtered by nonmaximum suppression, and the precise segmentation mask, confidence level, and position of each chain instance are output for quantification of pitch and wear area.

[0029] Preferably, step S3 specifically includes:

[0030] For each link, the mask area is fitted with the minimum bounding rectangle, and the length, spacing between adjacent links, and alignment angle of each link are calculated. The calculation results are compared with preset standard parameter thresholds. If the result exceeds the threshold range, it is judged as abnormal.

[0031] Preferably, the length of each link, the spacing between adjacent links, and the link alignment angle are calculated as follows:

[0032] (1) Minimum bounding rectangle fitting

[0033] For each link in the binary mask region, extract the set of pixel coordinates of all its contour points. The minimum bounding rectangle is obtained by using the rotating caliper method, and the center of the rectangle is then obtained. Length of the longer side Length of the shorter side principal axis direction angle ;

[0034] (2) Link length

[0035]

[0036] Among them, the The minimum bounding rectangle of each link has a long side pixel length. After calibration, the actual length is converted. The unit is mm;

[0037] (3) Spacing between adjacent chain segments

[0038] Calculate the first The and the first Euclidean distance between the center points of each link:

[0039]

[0040] in, For the first The center coordinates of the smallest bounding rectangle of each link;

[0041] (4) Link alignment angle

[0042] Calculate the angular deviation of the main axis direction between adjacent links:

[0043]

[0044] To eliminate directional periodicity, the process is as follows:

[0045]

[0046] in, For the first The main axis direction angle of each link;

[0047] If any parameter is abnormal, proceed to step S4.

[0048] Preferably, step S4 specifically includes:

[0049] Retrieve the historical identification results of the last three inspections, including this one, at the current sludge scraper monitoring point, and compare the trend of the chain status at the same location. If the same anomaly is detected multiple times in a row and the confidence level exceeds verify_conf_thresh, then the anomaly is confirmed.

[0050] Preferably, step S5 specifically includes:

[0051] Generate JSON format data containing all recognition results. If any anomalies are found, trigger an alarm and push it synchronously to the operation and maintenance management platform, along with the labeled image and fault handling suggestions.

[0052] The test data will be stored in a historical database for wear trend analysis.

[0053] This invention provides a method for identifying sludge scraper chains based on a quadruped robot. Existing sludge scraper chain detection primarily relies on manual inspection and sensor monitoring. Manual inspection suffers from low frequency, high-risk high-altitude work, delayed defect detection, and reliance on personnel experience. Contact sensors are susceptible to sewage corrosion and vibration, resulting in high failure rates, and can only monitor the overall operating status, failing to identify localized chain link defects. This invention uses a quadruped robot to replace manual inspection, adapting to the complex ground environment of sewage treatment plants. It can flexibly reach any monitoring point around the sludge scraper, achieving high-frequency, full-coverage inspections and avoiding the risks of high-altitude work for personnel. Through instance segmentation technology, it achieves fine-grained defect identification at the chain link level, identifying obvious and early-stage detachment faults, shifting the fault handling point forward, and reducing the risk of unplanned downtime. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the method for identifying the chain of a sludge scraper according to the present invention;

[0055] Figure 2 This is a schematic diagram illustrating the principle of measuring the geometric parameters of a chain link;

[0056] Figure 3 This is a flowchart illustrating step S4 in the sludge scraper chain identification method of the present invention. Detailed Implementation

[0057] To better understand the above-mentioned technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0058] Combination Figure 1 As shown, this invention provides a method for identifying the chain of a sludge scraper based on a quadruped robot, comprising the following steps:

[0059] Step S1, inspection path planning and image acquisition, specifically includes:

[0060] The inspection path of the quadruped robot is pre-calibrated, and the shooting parameters of the sludge scraper monitoring points are set.

[0061] When the quadruped robot reaches the monitoring point of the sludge scraper, it automatically adjusts its posture and takes images of the chain.

[0062] A normal chain image should show complete chain link outlines, uniform and clear pitch, no abnormal dark areas or overexposed spots on the surface, and be able to visually distinguish slight rust or wear marks. Therefore, shooting parameters include shooting angle, shooting distance, exposure value, etc.

[0063] The quadruped robot takes at least three high-resolution images from different angles.

[0064] The quadruped robot moves along a pre-set inspection path around the sludge scraper, adaptively adjusting the shooting angle to obtain multi-angle high-definition images of the chain on the sludge scraper, overcoming the problem of limited field of view in fixed-point monitoring.

[0065] Step S2, image preprocessing and instance segmentation, specifically includes:

[0066] Images captured by a quadruped robot are input into the YOLOv8 instance segmentation model, which outputs the mask region, class label, and confidence score for each link in the chain.

[0067] The YOLOv8 instance segmentation model is calculated as follows:

[0068] Input preprocessing involves scaling the chain images acquired with optimized shooting parameters to a fixed input size for the model (e.g., 640×640), and then normalizing and data augmenting (optional).

[0069] Feature extraction: Multi-scale feature maps of chain images are extracted through the CSPDarkNet backbone network in the YOLOv8 instance segmentation model to capture chain segment contours, textures and local anomaly details.

[0070] Feature fusion: The neck network (PAN-FPN structure) fuses multi-scale feature maps of different scales to enhance the ability to express small targets (such as micro cracks) and edge information;

[0071] Predicted output

[0072] Simultaneous prediction of the detection head branch:

[0073] The bounding box and category (normal / wear / cracked, etc.) of each chain instance.

[0074] A corresponding set of prototype mask coefficients is linearly combined with the prototype mask to generate a pixel-level binary mask for each instance;

[0075] Post-processing:

[0076] Redundant detection is filtered by nonmaximum suppression (NMS), and the precise segmentation mask, confidence level, and position of each chain instance are output for quantification of pitch and wear area.

[0077] The YOLOv8 instance segmentation model is used to process chain images, accurately locate and segment chain regions, chain links and abnormal shapes (detachment), and achieve fine-grained defect classification.

[0078] Category labels include normal links and chain detachment.

[0079] Step S3, geometric feature calculation and anomaly detection, specifically includes:

[0080] For each link, the mask area is fitted with the minimum bounding rectangle, and parameters such as the length of each link, the distance between adjacent links, and the alignment angle of the links are calculated. The calculation results are compared with the preset standard parameter thresholds. If the result exceeds the threshold range, it is judged as abnormal.

[0081] The length of each link, the spacing between adjacent links, and the link alignment angle are calculated as follows:

[0082] (1) Minimum bounding rectangle fitting

[0083] like Figure 2 As shown, for each link in the binary mask region, the set of pixel coordinates of all contour points is extracted. The minimum bounding rectangle is obtained by using the rotating caliper method, and the center of the rectangle is then obtained. Length of the longer side (Link length), Short side length (Link width), spindle direction angle (Angle between the longer side and the horizontal axis, ranging from 0° to 180°);

[0084] (2) Link length

[0085]

[0086] Among them, the The minimum bounding rectangle of each link has a long side pixel length. After calibration, the actual length is converted. The unit is mm;

[0087] (3) Spacing between adjacent chain segments

[0088] Calculate the first The and the first Euclidean distance between the center points of each link:

[0089]

[0090] in, For the first The center coordinates (pixel coordinates, later converted to actual distance) of the smallest bounding rectangle of each link; if a more precise "net distance between links" (i.e. the gap between the back end of the front link and the front end of the rear link) is required, the sum of half lengths of the two links in the direction of the line can be subtracted, but usually the center distance is used directly to reflect whether the pitch is uniform.

[0091] (4) Link alignment angle

[0092] Calculate the angular deviation of the main axis direction between adjacent links:

[0093]

[0094] To eliminate directional periodicity (e.g., 1° and 179°, which actually differ by 2°), the process is as follows:

[0095]

[0096] in, For the first The principal axis direction angle of each link (unit: degrees);

[0097] If any parameter is abnormal, proceed to step S4.

[0098] Step S4, timing cross-validation, such as Figure 3 As shown, it specifically includes:

[0099] Retrieve the historical identification results of the last three inspections, including this one, at the current sludge scraper monitoring point, and compare the trend of the chain status at the same location. If the same anomaly is detected three times in a row and the confidence level exceeds verify_conf_thresh (confidence threshold), then the anomaly is confirmed.

[0100] Step S4 is a necessary step after step S3 reports an anomaly; that is, based on the current detection and combined with the results of the two most recent inspections, an anomaly is determined when the threshold is exceeded for three consecutive times.

[0101] A timing verification logic is introduced to cross-verify the recognition results of multiple consecutive frames in the same chain. The secondary confirmation of abnormal results is controlled by the parameters chain_verify_switch (timing verification switch), verify_frame_num (verification frame number), and verify_conf_thresh (confidence threshold), which avoids misjudgment caused by occasional shooting shake.

[0102] Step S5, result output and alarm, specifically includes:

[0103] Generate JSON format data containing all recognition results. If any anomalies are found, trigger an alarm and push it synchronously to the operation and maintenance management platform, along with the labeled image and fault handling suggestions.

[0104] The test data will be stored in a historical database for wear trend analysis.

[0105] Output a structured recognition report, which includes information such as anomaly type, anomaly location, and suggested handling methods. Different types of defects can be marked on the image with different colors to intuitively display the fault location.

[0106] In summary, this invention is the first to combine a quadruped robot equipped with an industrial camera with the YOLOv8 instance segmentation model for link-level fault identification in sludge scraper chains at wastewater treatment plants. High-precision detection is achieved through multi-angle adaptive image acquisition, quantitative comparison of chain geometric features, and temporal multi-frame cross-validation. Addressing the pain points of complex wastewater treatment environments, high risks associated with manual inspections, limited field of view in fixed monitoring systems, sensor corrosion and failure, and high misjudgment rates in traditional identification methods, this invention employs a technical concept combining autonomous quadruped robot inspection with fine-grained link-level detection. The dual mechanism of geometric feature judgment and temporal verification improves identification accuracy. It can be directly deployed without structural modifications to existing sludge scraper equipment, enabling automated, high-frequency, and comprehensive chain status monitoring. This effectively reduces labor costs and safety risks, is feasible to manufacture and use, and is suitable for large-scale promotion in the wastewater treatment industry.

[0107] This invention also makes the inspection of sewage treatment plants safer and more efficient: quadruped robots replace manual labor to enter complex / high-altitude areas, eliminating safety hazards and enabling 24-hour uninterrupted high-frequency inspections.

[0108] Higher recognition accuracy: Based on the YOLOv8 instance segmentation model, fine-grained recognition at the chain link level is achieved, which can detect early and minute detachment defects.

[0109] Extremely low false positive rate: dual verification of geometric feature comparison and temporal multi-frame check significantly reduces false alarms caused by environmental jitter and light interference.

[0110] Non-contact and non-destructive: Purely visual inspection, does not contact the chain, does not affect equipment operation, and causes no corrosion or wear.

[0111] High level of intelligence: It automatically outputs structured reports, locates faults, pushes alarms, and supports historical data trend analysis, which facilitates operation and maintenance decision-making.

[0112] Highly adaptable to the environment: The quadruped robot can adapt to muddy, uneven, and slippery surfaces in sewage treatment plants and reach monitoring points that are difficult to reach by humans and traditional equipment.

[0113] Those skilled in the art should recognize that the above embodiments are merely illustrative of the present invention and are not intended to limit the present invention. Any variations or modifications to the above embodiments that are within the spirit and essence of the present invention will fall within the scope of the claims of the present invention.

Claims

1. A method for identifying the chain of a sludge scraper based on a quadruped robot, characterized in that, Includes the following steps: Step S1: Inspection path planning and image acquisition; Step S2, image preprocessing and instance segmentation; Step S3: Geometric feature calculation and anomaly detection; If an abnormality is detected, proceed to step S4; if normal, return to step S1 to continue image acquisition. Step S4, timing cross-validation; Step S5: Output the results and issue alarms.

2. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 1, characterized in that, Step S1 specifically includes: The inspection path of the quadruped robot is pre-calibrated, and the shooting parameters of the sludge scraper monitoring points are set. When the quadruped robot reaches the monitoring point of the sludge scraper, it automatically adjusts its posture and takes images of the chain.

3. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 2, characterized in that: The shooting parameters include shooting angle, shooting distance, and exposure value; The quadruped robot takes at least three images from different angles.

4. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 2, characterized in that, Step S2 specifically includes: The captured image is input into the YOLOv8 instance segmentation model, which outputs the mask region, class label, and confidence score of each link in the chain.

5. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 4, characterized in that, The calculation of the YOLOv8 instance segmentation model is as follows: Input preprocessing involves scaling the chain images acquired with optimized shooting parameters to the fixed input size of the model, and then normalizing and augmenting them. Feature extraction: Multi-scale feature maps of chain images are extracted through the CSPDarkNet backbone network in the YOLOv8 instance segmentation model to capture chain segment contours, textures and local anomaly details. Feature fusion: The neck network fuses multi-scale feature maps of different scales to enhance the ability to express small targets and edge information; Predicted output Simultaneous prediction of the detection head branch: The bounding box and category of each chain instance; A corresponding set of prototype mask coefficients is linearly combined with the prototype mask to generate a pixel-level binary mask for each instance; Post-processing: Redundant detection is filtered by nonmaximum suppression, and the precise segmentation mask, confidence level, and position of each chain instance are output for quantification of pitch and wear area.

6. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 4, characterized in that, Step S3 specifically includes: For each link, the mask area is fitted with the minimum bounding rectangle, and the length, spacing between adjacent links, and alignment angle of each link are calculated. The calculation results are compared with preset standard parameter thresholds. If the result exceeds the threshold range, it is judged as abnormal.

7. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 6, characterized in that, The length of each link, the spacing between adjacent links, and the link alignment angle are calculated as follows: (1) Minimum bounding rectangle fitting For each link in the binary mask region, extract the set of pixel coordinates of all its contour points. The minimum bounding rectangle is obtained by using the rotating caliper method, and the center of the rectangle is then obtained. Length of the longer side Length of the shorter side principal axis direction angle ; (2) Link length ； Among them, the The minimum bounding rectangle of each link has a long side pixel length. After calibration, the actual length is converted. The unit is mm; (3) Spacing between adjacent chain segments Calculate the first The and the first Euclidean distance between the center points of each link: ； in, For the first The center coordinates of the smallest bounding rectangle of each link; (4) Link alignment angle Calculate the angular deviation of the main axis direction between adjacent links: ； To eliminate directional periodicity, the process is as follows: ； in, For the first The main axis direction angle of each link; If any parameter is abnormal, proceed to step S4.

8. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 6, characterized in that, Step S4 specifically includes: Retrieve the historical identification results of the last three inspections, including this one, at the current sludge scraper monitoring point, and compare the trend of the chain status at the same location. If the same anomaly is detected multiple times in a row and the confidence level exceeds verify_conf_thresh, then the anomaly is confirmed.

9. The method for identifying the chain of a sludge scraper based on a quadruped robot according to claim 8, characterized in that, Step S5 specifically includes: Generate JSON format data containing all recognition results. If any anomalies are found, trigger an alarm and push it synchronously to the operation and maintenance management platform, along with the labeled image and fault handling suggestions. The test data will be stored in a historical database for wear trend analysis.

Images