Dual-vision intelligent monitoring belt conveyor material scattering and cleaning device and control method thereof

Abstract

The application discloses a double-vision intelligent monitoring belt conveyor scattering and cleaning device and a control method thereof, and belongs to the technical field of belt conveyor cleaning, state monitoring and control. The device comprises a modularized closed material collecting box, a chain scraper automatic cleaning mechanism, a roller wear and dust accumulation visual monitoring unit composed of an industrial camera, an image closed-loop dustproof self-cleaning protection subsystem, and an edge controller. The edge controller is configured to: calculate image definition in real time and control blowing and sweeping; run wear identification and dust accumulation identification algorithms; perform collaborative diagnosis, utilize the fact that belt deviation causes roller eccentric wear and material collecting box dust accumulation eccentric load to have spatial consistency, when the conditions are met and the spatial angle is met, indirectly determine belt deviation and output a diagnosis signal and a deviation degree quantitative value; and output a deviation correction instruction according to the diagnosis signal and a preset mapping relationship to actively correct the deviation. Indirect diagnosis and closed-loop active deviation correction under the condition that the belt edge is blocked are realized.

Description

Technical Field

[0001] This invention relates to the field of belt conveyor cleaning, condition monitoring and control technology, specifically to a device and method for conveying concrete solids, construction waste and crushed and recycled materials, integrating return spillage closed cleaning, online monitoring of roller wear, monitoring of dust accumulation inside the box and intelligent control, especially suitable for industrial conveying scenarios with high dust, high wear and high intensity continuous production. Background Technology

[0002] In production lines for solid material crushing, concrete waste recycling, and construction waste resource utilization, crushed particles and fine powders are continuously transported via belt conveyors. Due to the high hardness, high fine ash content, and slight adhesiveness of some materials, fine powders and ash continuously spill from the belt joints and adhered materials on the lower belt during belt operation. This results in high dust concentration, rapid equipment wear, and significant maintenance difficulties. This spillage not only pollutes the working environment and increases cleaning costs but also accelerates the wear of critical components such as return rollers. In severe cases, it can cause belt misalignment or even tearing, threatening the continuous operation of the entire production line.

[0003] The existing technologies in the fields of belt conveyor cleaning, monitoring, and maintenance have developed independently, forming multiple mature solutions in single-function areas. For example, Chinese invention patent CN111891695A discloses a belt conveyor return section material spillage cleaning system, in which the bottom support surface of the belt conveyor corridor has a U-shaped structure, and the spillage collection platform surface is recessed relative to the maintenance walkway surfaces on both sides, to solve the problem of limited space under the belt conveyor and tight installation height for cleaning devices. Chinese utility model patent CN217625952U discloses a belt conveyor return belt material cleaning and drop material collection and conveying device, in which the material collection chute is equipped with a scraper conveying mechanism including a chain, scraper, motor reducer, and transmission sprocket, realizing the mechanical collection and conveying of spilled materials. Chinese invention patent CN117347372B discloses a belt wear detection method, which takes images of the belt surface by installing visible light detection cameras at both ends of the conveyor line, and the controller judges the belt wear information by detecting the belt surface roughness and pit points, providing a visual inspection means for belt maintenance. In addition, a belt conveyor cleaner based on vision detection has been disclosed. It achieves adaptive cleaning through a multimodal data acquisition unit and a cleaning effect feedback unit, reflecting the trend of intelligent development.

[0004] However, the aforementioned existing technologies still suffer from the following core technological deficiencies, which constitute a technological bottleneck restricting the development of this field: (1) Existing enclosed cleaning boxes generally cover the entire width and area of ​​the belt return section. Replacing the return rollers and idlers requires disassembling the main body of the box, resulting in narrow operation space and low efficiency. Although there are U-shaped corridor structures that attempt to solve the maintenance space problem, they are a complete structural modification of the corridor and cannot be used as independent equipment for flexible modification of existing belt conveyors. For a large number of existing belt conveyors that have been put into operation, how to install a material collection and cleaning device without large-scale modification of the civil engineering foundation while maintaining sufficient roller maintenance space is a long-standing unresolved engineering problem.

[0005] (2) Belt condition inspection (wear, misalignment, etc.) and cleaning effect detection are two independent technical branches that do not overlap. In existing publicly available cleaning devices, roller wear monitoring serves the equipment maintenance department and aims to predict lifespan; while dust accumulation monitoring of the collection bin serves the production cleaning department and aims to determine whether cleaning is necessary. The two belong to different subsystems with vastly different data formats, sampling and analysis frequencies. There is a common technical bias in this field that considers it complex and unnecessary to spatially correlate these two monitoring systems with different functions and purposes in real time. Therefore, no technical solution has ever emerged that effectively integrates them into the same device and achieves data interoperability. This technical bias prevents existing systems from utilizing the synergistic effect of multi-source data to achieve higher-level fault diagnosis.

[0006] (3) Timed purging and timed wipers are both open-loop controls with fixed cycles, which cannot adapt to fluctuations in dust concentration on site. A few solutions that trigger cleaning through additional sensors are costly and complex to install, and the external sensor values ​​deviate from the actual dust accumulation on the lens surface. A complete self-closed-loop protection solution that uses the image clarity parameters output by the vision algorithm itself as the trigger is still lacking in this field. Due to the extremely large instantaneous changes in dust concentration on industrial sites, open-loop control purging strategies often result in either untimely purging causing lens failure or excessive purging causing waste of air source.

[0007] (4) Existing vision-based direct detection methods for belt misalignment heavily rely on clear, unobstructed images of the belt edge taken by the camera. In crushing and screening production lines, the return section of the belt is often covered by guide chutes and dust covers to suppress dust, making it impossible to install or effectively photograph the belt edge, rendering the direct detection solution completely ineffective. The industry lacks an indirect diagnostic solution for belt misalignment that does not rely on direct observation of the belt edge, leaving many belt conveyors operating under harsh conditions in a long-term blind spot for misalignment monitoring.

[0008] Therefore, there is an urgent need for a closed-loop control device and method that can overcome the above-mentioned technical biases, realize indirect diagnosis of belt misalignment under obstructed working conditions, and integrate cleaning and active correction, so as to solve the technical problems in the existing technology where the cleaning, monitoring and correction functions of belt conveyors are independent and lack coordination, especially in the case of belt misalignment diagnosis and closed-loop control that cannot be realized under working conditions where the belt edge is obstructed. Summary of the Invention

[0009] This invention addresses the problems of existing belt conveyor cleaning devices, such as the inability of the material collection structure to balance airtightness and roller maintenance space; the independent and uncoordinated monitoring of roller wear and dust accumulation in the collection box; the open-loop visual dust prevention and self-cleaning control, which cannot adapt to fluctuations in dust concentration; and the single method of belt misalignment detection, which completely fails when the belt edge is obscured. It provides a device and method that integrates modular material collection and cleaning, dual-vision collaborative monitoring, closed-loop image self-cleaning, and three-dimensional linkage active correction.

[0010] To address the aforementioned technical problems, this invention provides a dual-vision intelligent monitoring and cleaning device for conveyor belt spillage with active correction function. This device, through a modular, sunken collection box design, provides ample vertical maintenance space between the top surface of the box and the bottom surface of the return roller, achieving both sealed material collection and convenient maintenance, thus solving the problem of insufficient space in the retrofitting of existing conveyor belts.

[0011] The device includes: a belt conveyor frame; a modular, enclosed collection box, horizontally arranged below the return section of the belt conveyor, the collection box being assembled from multiple standard length units, with a reserved vertical maintenance space between the top surface of the box and the bottom surface of the return roller; and outwardly expanding inclined guide plates on the upper sides of both sides of the collection box, with the upper edge of the guide plates extending to the outer edges of both sides of the belt and the lower edge connecting to the side wall of the box. The outwardly expanding inclined guide plates effectively guide material spilled from the belt edges into the box, compensating for any gaps in the receiving area after the box sinks, achieving complete collection of spilled material.

[0012] The device also includes an automatic cleaning mechanism with a chain scraper, which is installed inside the collection box. This cleaning mechanism is used to continuously push the powder collected in the box to the discharge port, realizing automatic cleaning and material recovery, replacing the traditional manual cleaning operation.

[0013] The device also includes: a first visual monitoring unit, comprising a first industrial camera mounted on the side of the conveyor frame, with its lens aimed at the working surface of the return roller, for acquiring images of the roller surface throughout the entire process to achieve online monitoring of wear status; and a second visual monitoring unit, comprising a second industrial camera mounted on the upper part of the collection box, with its lens vertically downward aimed at the bottom collection area of ​​the box, for continuously acquiring images of the powder accumulation state inside the box to achieve accurate quantification of the amount of ash. The coordinated deployment of the two visual monitoring units enables the invention to simultaneously acquire information on both roller wear and ash distribution, laying the hardware foundation for cross-validation diagnosis.

[0014] The device also includes an image closed-loop dustproof self-cleaning protection subsystem, comprising compressed air pipelines, electromagnetic control valves, and air nozzles. The air nozzles are directed at the lens surfaces of the first and second visual monitoring units. This subsystem automatically blows away dust from the camera lenses and viewing window surfaces to ensure image clarity.

[0015] The device also includes an edge controller, which is connected to the electromagnetic control valve of the first visual monitoring unit, the second visual monitoring unit, the drive module of the chain scraper automatic cleaning mechanism, the image closed-loop dustproof self-cleaning protection subsystem, and the drive mechanism of a belt correction device. By integrating all monitoring, control, and communication functions with a single edge controller, it achieves high hardware integration, simple wiring, and low modification costs.

[0016] The edge controller is configured to implement the core intelligent closed-loop control logic of this invention. Specifically, it calculates the comprehensive sharpness evaluation parameter of the current frame image from the first or second visual monitoring unit in real time. When the comprehensive sharpness evaluation parameter is lower than a preset threshold, it controls the electromagnetic control valve to trigger purging to restore sharpness. The purpose of this configuration is to achieve intelligent judgment of the purging timing, using the image quality output by the visual algorithm itself as the triggering basis to accurately match the actual dust accumulation state of the lens, thus solving the defect of traditional timed purging that cannot adapt to fluctuations in dust concentration.

[0017] The edge controller is further configured to: run a first wear recognition algorithm, process the images captured by the first industrial camera to quantify wear parameters at different circumferential angles of the roller, and calculate the roller wear coefficient. and the angle of the wear position The second ash accumulation recognition algorithm is run to process the images captured by the second industrial camera to calculate the ash accumulation volume on the left and right sides of the box, and to calculate the ash accumulation off-center loading coefficient. And the direction of off-center loading. By running two sets of vision algorithms in parallel, the edge controller can simultaneously acquire key parameters of off-center wear and off-center loading, providing data support for subsequent collaborative diagnosis.

[0018] The edge controller is also configured to perform a collaborative diagnostic that utilizes the spatially consistent causal relationship between belt misalignment causing uneven roller wear and uneven dust accumulation in the collection bin, by verifying the position angle of the uneven wear. Whether the off-center loading direction is spatially consistent with the belt misalignment is used to indirectly determine belt misalignment. When the following conditions are met simultaneously, a belt misalignment diagnostic signal and a quantified value of the misalignment degree are generated and output: a) The dust accumulation off-center loading coefficient The absolute value is greater than the first threshold. b) The roller wear coefficient Greater than the second threshold c) The angle of the wear position The absolute value of the spatial angle difference between the load direction and the off-center load direction is less than the third threshold. The core innovation of this invention lies in its collaborative diagnostic logic. It overcomes the technical bias in the prior art that treats dust accumulation monitoring and roller wear monitoring as independent systems. It utilizes the causal relationship that belt misalignment causes two corresponding effects in space: one-sided roller wear and one-sided dust accumulation in the collection box. By verifying the spatial consistency of these two secondary phenomena, it achieves a high-confidence indirect determination of belt misalignment without the need for direct observation of the belt edge. This is especially suitable for situations where the belt edge is blocked by the guide chute or dust cover, making it impossible to install a direct detection camera.

[0019] The edge controller is further configured to: respond to the output belt misalignment diagnostic signal, generate and output a correction control command to the drive mechanism of the belt correction device based on the quantized value of the misalignment degree and a preset mapping relationship between the quantized value and the correction command. The command includes the correction direction and correction amount to actively correct the belt misalignment. The purpose of this configuration is to directly use the diagnostic results to drive the correction device, forming a complete closed loop from monitoring to diagnosis to active correction, thus achieving unattended intelligent operation and maintenance.

[0020] Furthermore, the standard length unit is 1.5 meters, the vertical maintenance space is 200-300 mm, and the angle between the outwardly expanding inclined guide plate and the horizontal plane is 30°-60°.

[0021] Furthermore, the automatic cleaning mechanism of the chain scraper includes a drive motor, a reduction transmission assembly, a main shaft, an annular transmission chain, wear-resistant scrapers, and an adjustable tension bracket; sprockets are fixedly installed at both ends of the main shaft, and the annular chain surrounds the sprockets to form a closed-loop transmission structure, with multiple sets of wear-resistant scrapers evenly fixed on the outside of the chain; the adjustable tension bracket is used to adjust the installation position of the main shaft to fine-tune the chain tension.

[0022] Furthermore, the first visual monitoring unit also includes a dustproof supplementary lighting source and a sealed protective housing; the second visual monitoring unit also includes a sealed protective cover, a built-in supplementary lighting module, and a wear-resistant, non-stick tempered glass viewing window.

[0023] Furthermore, the image closed-loop dustproof self-cleaning protection subsystem also includes an air pressure monitoring sensor; the edge controller is also configured to: periodically control the electromagnetic control valve to perform purging; immediately recheck the clarity after a single purging, and if it is not restored, purge and recheck again; when it still does not meet the standard after 3 consecutive purgings, trigger a hardware fault alarm.

[0024] Furthermore, the first threshold The second threshold is 0.3. The third threshold is 0.2. It is 30°.

[0025] Furthermore, the collaborative diagnostics performed by the edge controller is an indirect determination that does not rely on directly capturing and identifying images of the belt edge.

[0026] On the other hand, the present invention also provides an intelligent operation and maintenance and active deviation correction control method for a belt conveyor system based on the above-mentioned device. This method correspondingly implements the various functions of the device, protecting the core algorithm and control logic of the present invention at the method level.

[0027] The method includes the following steps: S1, continuously acquiring a time-series image sequence of the surface of the belt return roller using a first industrial camera, extracting the working surface area of ​​the roller, identifying and quantifying wear parameters at different circumferential angle positions of the roller, including the wear area. and equivalent wear depth ,in For circumferential angle, The time is used to calculate the roller wear coefficient. and the angle of the wear position ,in This represents the maximum circumferential wear depth. This represents the minimum circumferential wear depth. This represents the original wall thickness of the roller. The purpose of this step is to quantify the uneven wear condition of the roller using visual algorithms, providing precise data on the location and extent of uneven wear for subsequent diagnosis.

[0028] S2. Continuously acquire image sequences of the inside of the collection bin using a second industrial camera. Use multi-frame dynamic filtering to distinguish between suspended dust and stationary powder, identify the powder accumulation boundary inside the bin, and calculate the dust accumulation volume on the left side. and the volume of dust accumulation on the right side Based on this, the ash accumulation eccentric load factor is calculated. And the direction of off-center loading. The purpose of this step is to quantify the off-center loading status of dust accumulation in the box using visual algorithms, providing accurate data on the direction and degree of off-center loading for subsequent diagnosis.

[0029] S3. Perform a collaborative diagnosis that utilizes the causal relationship between belt misalignment causing uneven roller wear and uneven ash accumulation in the collection bin, which are spatially consistent, and determines the position and angle of the uneven wear. Mapped to the same spatial coordinate system as the off-center loading direction, when the following three conditions are met simultaneously, belt misalignment is indirectly determined to be the root cause, and a belt misalignment diagnostic signal and a quantified value of the misalignment degree are output: a) The dust accumulation off-center loading coefficient The absolute value is greater than the first preset threshold. b) The roller wear coefficient Greater than the second preset threshold c) The angle of the wear position The absolute value of the spatial angle between the angle corresponding to the off-center loading direction and the angle is less than a preset angle threshold. This step is the core diagnostic logic of this invention. It achieves indirect judgment through spatial consistency verification. Its technical effect is that it can provide a high-confidence deviation diagnosis even when the belt edge cannot be directly photographed.

[0030] S4. In response to the output belt misalignment diagnostic signal, based on the quantized value of the misalignment degree and the preset mapping relationship between the quantized value and the correction command, a correction control command is generated and output to the drive mechanism of a belt correction device. The command includes the correction direction and correction amount to actively correct the belt misalignment. This step transforms the diagnostic result into a control action, realizing a closed loop of "diagnosis equals correction".

[0031] S5. The maximum wear depth accumulated over time. Fitting the roller wear rate curve The system also collects real-time motor current data driving the chain scraper automatic cleaning mechanism and uses the current change rate to correct the wear rate curve. To predict the remaining service life of the roller This step improves the accuracy of lifespan prediction by correcting wear rate using fused current data, providing a basis for decision-making regarding condition-based maintenance.

[0032] S6. Based on the roller wear coefficient The absolute value of the ash accumulation eccentric load coefficient Based on the aforementioned deviation diagnosis results, calculate the overall health score of the equipment. ,in The preset wear coefficient warning threshold, As a penalty factor for deviation, when a deviation diagnosis signal exists. When there is no deviation diagnosis signal , Based on the comprehensive health score, a tiered maintenance recommendation is output. This step integrates multi-dimensional monitoring data into an intuitive health indicator, allowing maintenance personnel to quickly grasp the overall status of the equipment.

[0033] Furthermore, the quantification value of the deviation degree mentioned in step S3 is a deviation index. The index is based on the aforementioned wear coefficient. And ash accumulation eccentric load factor The absolute value weighted calculation is obtained. ,in and The preset weighting coefficients; the mapping relationship between the preset quantization value and the correction command mentioned in step S4 includes the deviation index. The corresponding correction direction and correction level are determined based on the threshold range in which the error occurs.

[0034] Furthermore, the determination in step S3 utilizes the causal relationship and consistency between the dust accumulation and uneven load caused by belt misalignment and the uneven wear of the rollers in the spatial distribution, and is an indirect diagnosis that does not rely on direct imaging of the belt edge.

[0035] Furthermore, the method for distinguishing between suspended dust and stationary accumulated powder by multi-frame dynamic filtering in step S2 is as follows: suspended dust exhibits random motion characteristics in continuous frames, while accumulated powder exhibits stationary or slow growth characteristics. Motion feature vectors are extracted through optical flow field analysis, and the categories of dust and materials are determined based on the temporal consistency of motion features.

[0036] Furthermore, the preset angle threshold mentioned in step S3 The angle of the wear position is 30°. The absolute value of the angle difference between the eccentric loading direction and the roller circumferential coordinate system is less than 30°.

[0037] Furthermore, the first preset threshold The value is 0.3, the second preset threshold. The preset wear coefficient warning threshold is 0.2. It is 0.5.

[0038] Furthermore, it also includes a closed-loop image dustproof self-cleaning protection step: a preset timed purging cycle is used to periodically spray compressed air onto the camera lens and viewing window; the comprehensive image sharpness evaluation parameters are calculated in real time, and multi-dimensional indicators such as grayscale variance, edge gradient and texture contrast are integrated. When the sharpness is lower than the preset threshold, purging is automatically triggered; after a single purging is completed, a new image is acquired for sharpness re-check. If the sharpness recovers to above the threshold, the closed loop ends. If it still does not meet the standard, purging is performed again and re-checked. If it still does not meet the standard after 3 consecutive purgings, a hardware fault alarm is triggered.

[0039] Furthermore, the graded maintenance recommendations are based on a three-level graded handling strategy: Level 1 warnings are triggered when there is slight wear, dust accumulation is close to the upper limit, or there is a slight current fluctuation, with only local pop-up notifications and log recordings; Level 2 alarms are triggered when there is moderate wear, dust accumulation reaches the upper limit, or operating parameters are abnormal, with local audible and visual alarms and automatic activation of the cleaning mechanism; Level 3 emergency interlocks are triggered when there is roller perforation, dust overflow risk, or motor jamming and overload, with continuous audible and visual alarms and push emergency fault signals; the three-level graded handling strategy supports dual-mode switching between production priority mode and equipment protection mode; and the active correction function has the highest priority, and its execution is not constrained by the three-level graded handling strategy.

[0040] Furthermore, the edge controller locally caches more than 7 days of full operating data, image data, and fault records. All monitoring and control functions operate normally when the network is offline, and the cached data is automatically re-uploaded to the cloud platform after the network is restored.

[0041] The beneficial effects of this invention compared to the prior art are: 1. This invention, through a modular sunken box design, ensures fully enclosed aggregate while reserving 200-300mm of unobstructed vertical maintenance space below the rollers, completely resolving the contradiction between aggregate airtightness and maintenance space in existing equipment retrofitting. As verified in Example 1, the roller replacement operation time is reduced from 2.5 hours in the traditional solution to 25 minutes, improving maintenance efficiency by 83%.

[0042] 2. This invention creatively solves the technical challenge of directly diagnosing belt misalignment when the belt edge is obstructed by the guide chute and dust cover by spatially verifying the spatial consistency of two secondary phenomena caused by belt misalignment: dust accumulation and uneven load. It also achieves active closed-loop control from diagnosis to correction. In Example 2, the system's correction success rate exceeds 98%, significantly reducing the risk of belt edge tearing.

[0043] 3. This invention solves the problem of traditional timed cleaning systems being unable to respond to instantaneous changes in dust concentration by using a closed-loop self-cleaning scheme that uses the image clarity parameters output by the vision algorithm itself as the sole basis for triggering and verification. Based on data from Examples 1 to 3, the camera lens cleaning and maintenance cycle is generally extended from approximately 5 days in the traditional scheme to 28-35 days, ensuring the long-term availability of the vision system under harsh working conditions.

[0044] 4. This invention achieves on-demand start / stop control of the cleaning mechanism by real-time quantification of the dust accumulation volume inside the bin based on a second visual monitoring unit, solving the problems of high energy consumption and rapid wear of vulnerable parts caused by continuous operation of traditional cleaning mechanisms. In Example 1, the cleaning mechanism achieves a comprehensive energy saving of 38%, and the scraper chain service life is extended by 55%.

[0045] 5. This invention incorporates multiple anomalies such as roller wear, dust accumulation, and mechanical failures into a unified three-level hierarchical handling strategy and provides a dual-mode switching of "production priority / equipment protection". This solves the contradiction that a single fault handling logic cannot simultaneously take into account the timely protection of emergency faults and the production continuity of minor anomalies, making the system flexibly adaptable to the management needs of different factories. Attached Figure Description

[0046] The present invention will now be further described with reference to the accompanying drawings.

[0047] Figure 1 This is a flowchart of the intelligent control process of the dual-vision algorithm in this invention; Figure 2 This is a schematic cross-sectional view of the chain scraper cleaning mechanism in this invention; Figure 3 This is a block diagram of the three-level hierarchical dual-mode control logic in this invention. Detailed Implementation

[0048] The following is in conjunction with the appendix Figures 1 to 3 The invention will be further illustrated by specific embodiments.

[0049] The device of this invention comprises: a belt conveyor frame, a modular sealed material collection box, an automatic chain scraper cleaning mechanism, a roller wear visual monitoring unit, a dust accumulation visual monitoring unit inside the box, an image closed-loop dust prevention and self-cleaning protection subsystem, an edge controller, an execution drive module, an audible and visual alarm module, a human-machine interface terminal, and a cloud-based operation and maintenance platform. The overall structure consists of four layers: a mechanical execution layer, a perception and monitoring layer, an edge control layer, and a cloud-based operation and maintenance layer. A single edge controller simultaneously carries two sets of visual algorithms and all control logic, achieving high hardware integration, simple wiring, and low retrofit costs. It is suitable for both new equipment as a complete set and for retrofitting existing belt conveyors. Its core components and preferred technical parameters are shown in the table below: The collection box is horizontally arranged below the return section of the belt conveyor. It is an open-top, closed trough structure, directly opposite the return belt, used to collect fine dust and debris spilled during belt operation. The box adopts a modular, segmented design, with a standard segment length of 1.5 meters. It can be freely spliced ​​and combined according to the actual length of the belt conveyor, adapting to various belt width models such as B800, B1000, and B1200, achieving flexible adaptation to conveyors of different lengths and reducing customization costs. Adjacent box segments are bolted together by flanges welded at the ends. An oil-resistant rubber gasket is sandwiched between the flange faces to ensure dust sealing at the joints and prevent dust leakage. The bottom support plate of the box is bolted to the belt conveyor frame columns or beams on both sides by L-shaped feet, allowing for quick installation and easy disassembly. The box is formed by splicing together a bottom support plate and two side protective plates, forming a fully enclosed material collection chamber, which effectively prevents fine dust from overflowing and completely solves the problem of dust leakage in open dust collection and protective net dust collection.

[0050] The overall installation height of the collection box is lower than the bottom of the return roller. A 200-300mm vertical maintenance space is reserved between the top surface of the box and the bottom surface of the return roller, completely eliminating the traditional protective net and the full-area obstruction structure of conventional boxes, thus creating an unobstructed vertical maintenance channel under the rollers. Maintenance personnel can directly disassemble and replace worn rollers and idlers from below without disassembling the collection box and its associated cleaning structure, significantly improving maintenance efficiency.

[0051] To compensate for the gap in the material receiving area on the top surface after the box body sinks, outwardly expanding inclined guide plates are added to the upper part of both sides of the box body. The upper edge of the guide plate extends to the outer edge of both sides of the belt, and the lower edge connects to the side wall of the box body, so that the material spilled from the edge of the belt slides into the box body along the guide plate. The angle between the guide plate and the horizontal plane is 30°-60°. The purpose of this angle design is to use the material's own weight to achieve smooth sliding and avoid material accumulation.

[0052] The chain scraper automatic cleaning mechanism is fully assembled inside the collection box and is a continuous cleaning structure used to continuously push the powder collected in the box to the discharge port, achieving fully automatic unattended cleaning. This mechanism includes a drive motor, a reduction transmission assembly, a main shaft, a UCP206 bearing housing, an adjustable tension bracket, a ring transmission chain, and wear-resistant scrapers. Sprockets are fixedly installed at both ends of the main shaft, and the ring chain surrounds the sprockets to form a closed-loop transmission structure. Multiple sets of wear-resistant scrapers are evenly fixed on the outside of the chain, forming a continuous scraping surface. The adjustable tension bracket specifically includes a bearing housing base plate with an elongated hole, which is bolted to a mounting base on the side wall of the box. An adjusting block is fixed to the side wall of the box, and an adjusting screw is screwed onto the adjusting block. The front end of the adjusting screw abuts against the side edge of the bearing housing base plate. By turning the adjusting screw, the bearing housing base plate can be moved along the direction of the elongated hole, thereby fine-tuning the installation position of the main shaft and achieving dynamic compensation of the chain tension. The purpose of setting up this adjustable tension bracket is to dynamically compensate for the tensile wear caused by the long-term operation of the chain, to always ensure transmission stability, and to avoid the defects of existing chain cleaning mechanisms such as easy loosening, chain skipping, and uneven cleaning.

[0053] Working principle: After the drive motor starts, it drives the main shaft and sprocket to rotate, which drives the ring chain to move in a cycle. The scraper fixed on the chain moves synchronously with the chain, continuously pushing the powder and debris accumulated at the bottom of the collection box to the discharge port at the end of the box, realizing fully automatic continuous cleaning and material recovery without manual intervention, completely replacing manual cleaning operations.

[0054] The roller wear visual monitoring unit includes a first industrial camera, a dustproof supplementary light source, a lens self-cleaning and blowing assembly, and a sealed protective housing. The camera is fixedly mounted on the side of the conveyor frame, with the lens aimed at the working surface of the return roller, capturing high-definition images of the roller surface throughout the entire process. This enables 24 / 7 non-contact online monitoring of the roller's condition, avoiding the drawbacks of traditional manual inspections that require machine shutdown. An independent adjustable supplementary light source adaptively corrects uneven lighting conditions, ensuring image quality is unaffected by changes in ambient light. The camera's entire structure uses a sealed dustproof housing to isolate it from dust and moisture, making it suitable for harsh industrial conditions. For example, the first industrial camera is a 2-megapixel camera with a resolution of 1920×1080 pixels, a frame rate of 30fps, and a lens focal length of 12mm. It is mounted at a 45° angle to the side of the roller, 800mm from the roller surface. The camera is fixed to the side of the conveyor frame via a pan-tilt mount with multi-degree-of-freedom adjustment, allowing for easy adjustment of the shooting angle and field of view to adapt to different installation spaces.

[0055] The visual monitoring unit for dust accumulation inside the collection box includes a second industrial camera, a sealed protective cover, a built-in supplementary lighting module, a lens cleaning mechanism, and a wear-resistant, non-stick, tempered glass viewing window. The camera is sealed and installed at a pre-set viewing window position on the upper part of the collection box, with the lens vertically downwards facing the bottom collection area of ​​the box, continuously acquiring images of the powder accumulation state inside the box for real-time quantification of dust volume. The viewing window uses wear-resistant, non-stick tempered glass with a nano-silica non-stick coating to reduce powder adhesion and accumulation, ensuring long-term image clarity. For example, the second industrial camera is a 2-megapixel industrial camera of the same model as the first industrial camera, installed at the center of the top of the collection box, with the lens vertically downwards, 1000mm from the bottom of the box.

[0056] The image closed-loop dustproof self-cleaning protection subsystem includes compressed air piping, an electromagnetic control valve, a fan-shaped air knife nozzle, and a pressure monitoring sensor. The nozzle is aimed at the camera lens and viewing window surface. The fan-shaped air knife nozzle is mounted via a miniature bracket with a ball joint, allowing manual adjustment of the blowing angle to ensure airflow covers the entire lens or viewing window surface. The ball joint adjustment structure allows the blowing range to flexibly adapt to different camera mounting angles. The camera housing is equipped with a heat-insulating layer and a drying chamber to resist high temperatures and moisture corrosion, preventing lens fogging and condensation. For example, the compressed air supply pressure is 0.6 MPa, and the electromagnetic control valve is a two-position three-way solenoid valve with a response time ≤20 ms, ensuring timely blowing response.

[0057] The original dual-vision collaborative three-dimensional linkage diagnostic algorithm of this invention specifically includes the following steps: Step 1: Extraction of roller wear parameters and calculation of uneven wear The first industrial camera continuously acquires a time-series image sequence of the surface of the belt return roller. The roller wear visual recognition algorithm performs the following processing: (1) The collected roller images are subjected to noise reduction, defogging, illumination correction and dynamic dust filtering to accurately eliminate image interference caused by suspended dust and light and shadow deviation on site, and improve image quality and analysis accuracy.

[0058] (2) By using image semantic segmentation technology, the cylindrical working surface area of ​​the roller is accurately extracted, the interference of frame, belt and background debris is eliminated, the effective detection area is locked, and the interference of non-target areas on wear analysis is avoided.

[0059] (3) Retrieve the standard reference profile model of the roller from the factory, compare and inspect the surface condition of the roller in real time, identify various wear defects such as uniform wall thickness reduction, local pits, cracks, pitting and perforation, and quantify the wear area. and equivalent wear depth ,in The circumferential angle of the roller (0°-360°). For time. By comparing with the factory baseline model, it is possible to accurately distinguish between normal use wear and abnormal wear, providing reliable data for life prediction.

[0060] (4) By using texture and grayscale features, the powder adhering to the roller surface can be accurately distinguished from the actual wear defects, thus avoiding misjudging the surface adhering material as wear damage and improving the accuracy of diagnosis.

[0061] (5) Calculate the roller wear coefficient based on the wear depth distribution at different circumferential angles of the roller. and the angle of the wear position : wear coefficient ,in This represents the maximum circumferential wear depth. This represents the minimum circumferential wear depth. This represents the original wall thickness of the roller. This coefficient reflects the unevenness of roller wear; a larger value indicates more severe uneven wear.

[0062] Offset grinding position angle This refers to the circumferential angle position where the wear depth is greatest, used to locate the roller side with the most severe wear.

[0063] For example, with the roller wall thickness Taking a working condition of 10mm as an example, the algorithm identifies the circumferential angle. The wear depth at the 90° (right side of the roller) position is 3.5mm. If the wear depth at the 270° (left side of the roller) position is 1.2mm, then the wear coefficient is... =(3.5-1.2) / 10=0.23, the angle of the uneven wear position. =90°.

[0064] Step 2: Extraction of Ash Distribution and Calculation of Off-center Load The second industrial camera continuously acquires a sequence of images inside the collection bin. The visual recognition algorithm for dust accumulation in the collection bin performs the following processing: (1) Preprocessing the acquired dust accumulation images and accurately distinguishing between suspended dust and stationary accumulated powder through a multi-frame dynamic filtering algorithm. This algorithm is based on N frames (e.g., N=5) of continuously acquired images. Corner detection and optical flow tracing are performed on each frame to calculate the motion vector of each feature point. Spatiotemporal clustering is performed on the motion vectors. Feature points with high motion speed and random direction are marked as "suspended dust"; feature points with zero or extremely low motion speed are marked as "stationary accumulated powder". By statistically analyzing the pixel areas marked as "accumulated powder" within a certain time period, the powder accumulation boundary can be determined. The purpose of this algorithm is to eliminate the interference of suspended dust in the bin on the calculation of dust accumulation volume and accurately extract the effective accumulation area.

[0065] (2) Pre-store the empty reference image of the box, establish a three-dimensional spatial calibration model of the box, accurately lock the effective dust accumulation detection area, and provide a spatial reference for volume conversion.

[0066] (3) By using grayscale layer recognition and contour extraction technology, the powder accumulation boundary, coverage area and accumulation thickness are accurately identified, and the volume of ash accumulation on the left side of the box is calculated. and the volume of dust accumulation on the right side This step achieves precise quantification of the ash accumulation volume by extracting multi-dimensional features from the accumulation region.

[0067] (4) Calculate the ash accumulation eccentric load coefficient The direction of the off-center load is Pointing to the left, The value points to the right. This coefficient reflects the degree of asymmetry in ash accumulation inside the box, providing a basis for diagnosing eccentric loading.

[0068] For example, the algorithm identifies the volume of ash accumulated on the left side of the box. =1.2m 3 , Dust accumulation volume on the right side =0.3m 3 Therefore, the ash accumulation eccentric load factor D = (1.2 - 0.3) / (1.2 + 0.3) = 0.6, and the eccentric load direction is to the left.

[0069] Step 3: Spatial coordinate mapping and 3D linkage diagnosis The angle of the uneven wear position Mapped to the same spatial coordinate system as the off-center load direction (based on the belt running direction, roller circumferential angle) =0° corresponds to the top of the belt =90° corresponds to the right side of the belt =180° corresponds to the area below the belt. =270° corresponds to the left side of the belt). The off-center load direction is also mapped to this coordinate system (left-side off-center load corresponds to...). =270°, right-side off-center load corresponds to =90°). The purpose of establishing this unified spatial coordinate system is to enable quantitative comparison between the wear position and the load direction under the same reference system.

[0070] The innovation of this diagnostic method lies in its utilization of the fact that belt misalignment, as a cause, simultaneously produces two corresponding spatial effects: unilateral wear on the rollers and unilateral dust accumulation on the collection bin. By simultaneously monitoring these secondary phenomena generated by these two independent systems and verifying their spatial consistency, the occurrence of misalignment can be indirectly determined with high confidence without directly observing the belt edge. This method is particularly suitable for situations where the belt edge is obstructed by the guide chute or dust cover, making it impossible to install a direct misalignment detection camera, thus filling the diagnostic blind spot in such scenarios.

[0071] The edge controller simultaneously determines the following three conditions. When all three conditions are met, it determines that the belt misalignment is the root cause of uneven wear, dust accumulation, and uneven load, and outputs a misalignment diagnostic signal: Condition 1: Ash accumulation eccentric load factor The absolute value is greater than the first preset threshold. (For example =0.3), indicating that the dust accumulation shows significant off-center loading; Condition 2: Roller wear coefficient Greater than the second preset threshold (For example =0.2), indicating that the roller has significant uneven wear; Condition 3: Angle of uneven wear position The absolute value of the spatial angle between the angle indicated by the off-center load direction and the angle indicated by the off-center load direction is less than a preset angle threshold. (For example =30°), indicating that the location of the most severe wear is spatially highly consistent with the direction of the dust accumulation load.

[0072] The simultaneous fulfillment of these three conditions is the key to this diagnostic logic. Any single condition may be caused by other factors (such as uneven feeding on one side leading to uneven load, or uneven material of the roller itself leading to uneven wear), but if all three conditions are met at the same time, it is highly likely to point to the common root cause of belt misalignment.

[0073] For example, the angle of the off-center wear position =90° (severe wear on the right side of the roller), the calculation results of the ash accumulation off-center load show... That is, the off-center loading direction is to the right (corresponding to an angle of 90°), and the spatial angle is 0° < 30°, and =0.6>0.3, =0.23>0.2, all three conditions are met simultaneously, and the edge controller outputs a diagnostic signal of "belt deviating to the right".

[0074] To quantify the severity of the deviation for subsequent active correction control, this algorithm further calculates a deviation index. This index combines the degree of dust accumulation and uneven load with the degree of roller wear, for example... The edge controller will As a quantification value of the deviation degree, and based on the mapping relationship in the table below, a correction instruction is generated: This mapping relationship can be modified based on on-site commissioning to ensure the stability of the closed-loop correction and adapt to the dynamic response characteristics of different belt conveyors.

[0075] Step 4: Predicting the Remaining Life of the Roller Based on the wear parameters accumulated over time, especially the maximum wear depth Using time-series data, fit the roller wear rate curve: Wear rate The data is obtained by performing linear regression on historical wear depth data and is used to reflect wear trends.

[0076] Simultaneously, the edge controller collects real-time current data of the drive motor of the chain scraper automatic cleaning mechanism and calculates the current change rate. Current data can indirectly reflect changes in material characteristics such as hardness and moisture content, thereby assisting in correcting wear predictions. For example, when the real-time current... Relative to no-load reference current When the rate of change exceeds the preset value, the coefficient calibrated in the experiment is used. Correct the wear rate: .generally The value ranges from 0.5 to 1.5, and the specific value is determined experimentally based on the material characteristics. The remaining service life is predicted using the corrected wear rate. ,in The maximum allowable wear depth for the roller (typically 60% of the wall thickness). This represents the current maximum wear depth. Incorporating current data into the wear rate correction is intended to compensate for the impact of changes in material properties on the wear rate, thereby improving the accuracy and adaptability of predictions.

[0077] For example, the current maximum wear depth =3.5mm, maximum allowable wear depth =6mm, linear wear rate =0.5mm / month, after correction =0.55mm / month, then the remaining service life =(6-3.5) / 0.55≈4.5 months.

[0078] Step 5: Comprehensive Health Assessment Based on the roller wear coefficient The absolute value of the ash accumulation eccentric load factor Based on the deviation diagnosis results, calculate the overall health score of the equipment: in: A preset warning threshold for roller wear coefficient (e.g., set to 0.5) is used to measure the severity of wear. Penalty factor for deviation: when deviation diagnosis signal is present. When there is no deviation diagnosis signal The purpose of introducing this penalty factor is to significantly reduce the score of equipment with deviation faults, so as to remind maintenance personnel to deal with them in a timely manner. , , These are the weighting coefficients. , ∈[0.3,0.5], ∈[0.2,0.4], ∈[0.1,0.3]. The purpose of setting the weighting coefficients is to differentiate the weighting based on the degree of influence of each indicator on the safe operation of the equipment.

[0079] For example, take =0.4, =0.3, =0.3, =0.23, =0.5, =0.6, indicating a deviation diagnostic signal. ,but =0.4×(1-0.23 / 0.5)+0.3×(1-0.6)=0.216+0.12=0.336. According to the scoring grading: ≥0.8 is excellent, 0.6≤ <0.8 is good, 0.4≤ <0.6 is considered normal. A value less than 0.4 indicates the need for maintenance. The purpose of this scoring system is to integrate multi-dimensional monitoring data into a single, intuitive quantitative indicator, enabling maintenance personnel to quickly grasp the overall health status of the equipment and formulate maintenance plans accordingly.

[0080] It is particularly important to emphasize that the active belt misalignment correction function triggered by 3D linkage diagnostics has the highest priority. Once a belt misalignment diagnostic signal is output, regardless of the current system alarm level, the edge controller will prioritize executing the active belt misalignment correction control procedure. This procedure is as follows: the edge controller immediately reads the misalignment index. As a quantification value of the degree of deviation, based on the preset... - The belt alignment mapping relationship generates control commands containing the alignment direction and amount, driving the belt alignment device to adjust the belt position in real time. After the alignment action is completed, the vision system continues to monitor. If the misalignment signal diagnosed in the next cycle has not disappeared, the above alignment process is repeated until the misalignment diagnostic signal disappears, forming a closed-loop alignment. Setting the highest priority is to ensure that belt misalignment, a fault that may cause serious equipment damage, can be handled first, avoiding delays in alignment due to waiting for the routine alarm handling process.

[0081] Meanwhile, this invention also incorporates anomalies such as roller wear, dust accumulation, and mechanical failures into a unified three-level classification and handling logic: Level 1 Warning (roller wear depth < 20% of wall thickness, or dust accumulation volume in the chamber reaches the preset upper limit threshold of 60%, or slight fluctuation in drive motor current of 5%): Only local pop-up notifications and log recordings are displayed; production operation is not interfered with. The purpose of this level is to provide a notification and record minor anomalies for maintenance personnel to check as needed.

[0082] Level 2 Alarm (roller wear depth reaches 30% to 60% of wall thickness, or dust accumulation in the bin continuously exceeds the upper limit threshold without decreasing, or motor current remains abnormal): Local audible and visual alarms, mobile push notifications for maintenance, and automatic activation of the cleaning mechanism. This level is designed to issue clear alerts for moderate anomalies, reminding maintenance personnel to arrange for prompt handling.

[0083] Level 3 Emergency Interlock (Risk of Roller Perforation, Risk of Dust Overflow, or Motor Jam and Overload): Continuous audible and visual alarms, push emergency fault signals, and can, according to preset modes, delay and then trigger the belt conveyor to decelerate or stop. The purpose of this level is to take emergency protection measures for serious faults to prevent further equipment damage or safety accidents.

[0084] The above strategy supports dual-mode switching between production priority mode and equipment protection mode: in production priority mode, only early warnings and fault records are provided, and shutdown is recommended at an opportune time to ensure continuous completion of production tasks; in equipment protection mode, the belt conveyor is decelerated or stopped after a preset delay period is reached. This dual-mode switching capability allows the system to flexibly adapt to the management needs and priority strategies of different factories.

[0085] This invention employs a three-layer hierarchical control architecture that integrates dual algorithms with a single controller: Perception layer: Includes dual vision acquisition units, motor current sensor, temperature sensor, air pressure sensor and belt conveyor operation signal acquisition interface, responsible for comprehensively collecting equipment operation status data.

[0086] Edge Control Layer: Employing a single edge controller as its core, this layer runs two sets of vision algorithms locally in parallel, independently performing image analysis, logical judgment, device control, and alarm output. It supports autonomous operation even when the network is down; all monitoring and control functions function normally, and the entire dataset is cached locally for more than 7 days, automatically re-uploading the cached data upon network recovery. The purpose of this layer is to achieve real-time intelligent control at the field level, independent of network connectivity.

[0087] Cloud-based operations and maintenance layer: This layer receives all data uploaded from edge devices, enabling centralized monitoring of multiple devices, wear curve tracing, automatic generation of consumable ledgers, energy consumption statistics, and fault record archiving. Its role is to provide remote centralized management and long-term data storage and analysis capabilities.

[0088] The overall collaborative workflow of this invention is as follows: 1. When the belt conveyor starts running, all the fine ash and debris spilled during the conveying process fall into the bottom sealed collection box, achieving fully enclosed collection.

[0089] 2. Two sets of vision monitoring units acquire images synchronously and continuously, and the edge controller runs dual vision algorithms in parallel to simultaneously complete the quantitative monitoring of roller wear and the accurate monitoring of dust accumulation.

[0090] 3. The edge controller executes a three-dimensional linkage diagnostic algorithm. Once belt misalignment is indirectly diagnosed through spatial consistency verification, it immediately outputs a misalignment diagnostic signal and a quantified misalignment index. .

[0091] 4. In response to the diagnostic signal and Upon receiving the value, the edge controller immediately generates a correction control command containing the correction direction and correction amount with the highest priority, driving the belt correction device to perform active correction, thus achieving precise closed-loop control of "diagnosis equals correction".

[0092] 5. The dust accumulation algorithm determines the amount of dust accumulation in real time based on the volume of dust in the bin, and automatically starts and stops the cleaning mechanism as needed to achieve unattended intelligent cleaning.

[0093] 6. The roller wear algorithm continuously tracks the operating status of consumables, predicts their service life, and issues warnings based on a three-level standard when abnormalities occur.

[0094] 7. The self-cleaning protection system automatically performs lens dust removal and re-inspection in dual modes according to the timed cycle and real-time image clarity status, forming a complete self-closed loop.

[0095] 8. The system synchronously collects motor current and temperature data, and intelligently identifies mechanical faults such as chain jamming and scraper jamming.

[0096] 9. The system automatically wakes up and monitors the system in response to the start and stop status of the conveyor belt, and adjusts the computing power output as needed to achieve energy-saving operation.

[0097] 10. All operation, monitoring, and alarm data are cached locally and uploaded to the cloud platform simultaneously to achieve full-process digital operation and maintenance management.

[0098] Example 1 This embodiment is applied to a construction waste crushing and recycling aggregate production line, equipped with a 1000mm wide belt conveyor for continuous conveying of crushed concrete solids. The on-site working conditions are characterized by high dust, high wear, and large fluctuations in ash production, which perfectly matches the characteristics of the equipment of this invention.

[0099] Configuration: A 1.5-meter standard modular collection box is assembled along its entire length, with each section secured by flange bolts. Oil-resistant rubber gaskets are sandwiched between the flange faces. A dedicated 280mm maintenance space is reserved between the top of the box and the bottom of the return roller, allowing for roller replacement without disassembly. Roller replacement time is reduced from the traditional 2.5 hours to 25 minutes, improving maintenance efficiency by 83%. Extended inclined guide plates are installed on both sides of the box at a 45° angle to the horizontal plane. The upper edge of the guide plates extends to the outer edges of the belt, allowing material spilled from the belt edges to smoothly slide into the box, ensuring complete collection. An internal chain scraper circulating cleaning mechanism, compatible with a 1.1kW drive motor, meets the cleaning needs under conditions of large fluctuations in ash production.

[0100] Monitoring Deployment: A 2-megapixel roller wear monitoring camera with a resolution of 1920×1080 and a frame rate of 30fps is fixedly mounted on the side of the rack via a pan-tilt bracket. A dust accumulation monitoring camera is sealed on the top of the enclosure, with its lens pointing vertically downwards towards the bottom of the enclosure through a wear-resistant, non-stick, tempered glass viewing window. A fully automatic compressed air dual-mode purging system is included, with a timed purging cycle set at 1.5 hours and an image clarity threshold set at 70% of the standard value. If the image clarity falls below the threshold, an emergency purging and re-inspection are automatically triggered. The camera housing is equipped with a heat-insulating and drying structure to adapt to the high temperatures and high humidity conditions encountered in the field during summer.

[0101] Intelligent control: A single edge controller locally runs dual vision algorithms. Cleaning automatically starts when dust accumulation reaches the upper limit of 60% of the cabinet volume and automatically stops when it drops to the lower limit of 10%, achieving energy-saving cleaning on demand. Real-time quantitative monitoring of roller wear status is implemented, and cleaning is initiated when the wear coefficient simultaneously meets the specified requirements. >0.2、 When the value is greater than 0.3 and the spatial angle is less than 30°, the system automatically outputs a belt misalignment diagnostic signal and a misalignment index. and immediately according to The system sends a correction command to the matching belt alignment device. A level two warning is triggered when the wear depth reaches 30% of the wall thickness, and a level three emergency interlock is triggered when it reaches 60%. The system defaults to equipment protection mode, but can be switched to production priority mode with a single click based on the production plan.

[0102] Operation and maintenance results: On-site ground dust was completely eliminated, spilled material was thoroughly swept away without residue, and the material recovery rate was 100%. Roller wear and tear belt failures decreased by 92%. The cleaning mechanism no longer requires 24-hour continuous operation, saving 38% of electricity, and the scraper chain lifespan was extended by 55%. The lens cleaning and maintenance cycle was extended from the traditional 5 days to 35 days. Complete elimination of reliance on manual inspection and cleaning, achieving fully unattended intelligent operation and maintenance.

[0103] Example 2 This embodiment is applied to an iron ore dry beneficiation workshop, using a 1200mm wide belt conveyor. The material is a mixture of 0-15mm iron ore and waste rock. The working conditions are characterized by extremely high material hardness and abrasiveness, and the production line has extremely high requirements for continuous production, which poses a severe test to the reliability of the equipment and the ability of intelligent operation and maintenance.

[0104] Implementation configuration: A 1.5-meter standard modular enclosure is used, with a 250mm maintenance space reserved. The first industrial camera is equipped with a splash guard to withstand the impact of splashed high-hardness materials. The scraper is made of wear-resistant ceramic composite material to cope with long-term wear from highly abrasive materials. The rest is the same as in Example 1.

[0105] Operational Results: After the device was put into operation, unplanned downtime due to roller wear-through no longer occurred, and equipment availability significantly improved. Roller wear coefficient W W When the dust concentration first exceeds 0.2, the system actively corrects its course, significantly reducing the risk of belt edge tearing. The image self-cleaning system successfully resists the adhesion of high concentrations of metal dust, maintaining a maintenance cycle of approximately 30 days. The on-demand cleaning mode extends the scraper chain's lifespan by 60% compared to the 24-hour continuous operation mode.

[0106] Example 3 This embodiment is applied to a building materials factory with a belt width of 800mm. The conveyed material is a mixture of clay and lime with a moisture content of 15%, which is very easy to stick to the belt and rollers and may caking inside the box. It is a typical working condition of wet and sticky materials.

[0107] Implementation Configuration: To address the sticky, wet characteristics, the scraper was replaced with a flexible polyurethane scraper, and the gap between the scraper and the bottom of the casing was increased to prevent material from sticking and jamming. A stainless steel liner was added to the bottom of the casing to reduce the coefficient of friction and aid material sliding. The dust accumulation volume threshold in the algorithm was adjusted to be more sensitive to changes in dust accumulation and to trigger cleaning promptly. In the diagnostic logic, because wet material is more likely to cause increased roller wear, the wear coefficient warning threshold was increased. The value was lowered from 0.5 to 0.35 to make the early warning more sensitive.

[0108] Operational Results: The device successfully addressed the challenge of material caking, with no scraper jamming occurring. Through real-time monitoring and early warning of wear rates, maintenance personnel can replace the rollers in a planned manner, preventing accidents caused by rapid deterioration of wear due to material adhesion. This embodiment illustrates that the device and method of the present invention can be adapted to different material characteristics through simple configuration adjustments, exhibiting good adaptability and portability to different operating conditions.

[0109] Example Effect Comparison Table: The modular sunken aggregate box of this invention has a dedicated, unobstructed vertical space of 200-300mm reserved between its top surface and the bottom surface of the return roller. Referring to Embodiment 1, the roller replacement operation time is reduced from 2.5 hours in the traditional solution to 25 minutes, improving maintenance efficiency by 83%. This structural design allows the device to be installed independently under existing conveyor belts without requiring modifications to the corridor, significantly lowering the modification threshold and construction period.

[0110] This invention creatively proposes an indirect diagnostic method based on spatial consistency verification of secondary phenomena. The edge controller integrates the dust accumulation off-center loading coefficient. and roller wear coefficient When the direction of dust accumulation aligns spatially with the location of uneven wear, belt misalignment is indirectly diagnosed. More importantly, the active correction function triggered by this diagnostic signal has the highest priority, forming a closed-loop error correction system that does not rely on direct observation of the belt edge. In Example 2 (iron ore condition), this system successfully addressed the issue at the uneven wear coefficient... When the first deviation exceeds the standard, proactive correction is performed, which significantly reduces the risk of belt edge tearing, with a correction success rate of over 98%.

[0111] The image closed-loop dustproof self-cleaning subsystem of this invention uses the image sharpness parameter output by the vision algorithm itself as the sole basis for triggering and verification. When the algorithm detects that the sharpness is below a threshold, it automatically blows the image and re-checks it until it meets the standard. If it fails to meet the standard three times consecutively, an alarm is triggered. This closed loop achieves a precise response to the actual dust accumulation state of the lens. According to data from Examples 1 to 3, the cleaning and maintenance cycle of the camera lens is generally extended from about 5 days in the traditional solution to 28-35 days, ensuring the long-term stable operation of the dual vision system in harsh industrial environments.

[0112] This invention utilizes real-time quantitative data of the dust accumulation volume inside the bin from a second visual monitoring unit to enable on-demand start and stop of the cleaning mechanism (e.g., starting when dust reaches 60% and stopping when it drops to 10%). Referring to Embodiment 1, this mode eliminates the need for the cleaning mechanism to operate continuously for 24 hours, resulting in a comprehensive power saving of 38% and extending the service life of easily worn parts such as the scraper chain by 55%.

[0113] This invention's three-tiered handling strategy differentiates between different levels of anomalies and creatively provides a dual "production priority / equipment protection" mode. This design makes equipment operation and maintenance no longer a black-and-white choice of shutdown or non-shutdown, but can flexibly adapt to the management needs of different factories.

[0114] In this invention, the ring chain drive can be replaced by a synchronous belt drive or a wire rope traction drive; the scraper can be a steel wear-resistant scraper, a flexible rubber scraper, or a polyurethane ceramic composite scraper, adaptable to the cleaning needs of different materials, whether dry or wet, soft or hard. The collection box can be arranged horizontally or slightly inclined to assist in the sliding discharge of materials; a bottom wear-resistant liner and an end discharge gate can be added to improve wear resistance and controllability. The inclined guide plate can be replaced by an arc-shaped guide cover or a flexible baffle curtain, adaptable to different material spreading trajectory scenarios.

[0115] In this invention, monocular vision can be replaced with binocular vision, a line scan camera, or laser contour vision to further improve the detection accuracy of wear depth and dust accumulation thickness. Ultrasonic level sensors and infrared temperature sensors can be superimposed and cross-checked with the visual algorithm to form a multi-dimensional fusion monitoring system. Compressed air purging can be replaced with electric wiper-type cleaning, ultrasonic dust removal, etc., to adapt to application scenarios without an air source.

[0116] In this invention, edge local computing can be replaced by a PLC's built-in vision module or centralized cloud computing, adapting to different field equipment configurations. It can be equipped with an incremental learning model to continuously adapt to changes in field conditions and automatically iterate and optimize the recognition model. Early warning outputs can be connected to MES systems, WeChat, SMS, or industrial walkie-talkies to achieve multi-channel, hierarchical alarm push notifications.

[0117] This invention replaces the fixed start-stop mode, employing variable frequency speed control logic for high-speed cleaning when dust accumulation is high and low-speed intermittent cleaning when dust accumulation is low, achieving precise adaptive control. The scraper mechanism can alternately operate in forward and reverse directions to prevent powder caking and jamming, adapting to cleaning scenarios involving damp and easily caking materials. It can be linked with a belt correction device and a spray dust suppression system to achieve a fully automated closed-loop process from monitoring to execution.

[0118] The present invention has been described in detail above with reference to the embodiments. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A dual-vision intelligent monitoring and cleaning device for conveyor belt spillage, characterized in that, include: Belt conveyor frame; The modular sealed collection box is arranged horizontally below the return section of the belt conveyor. The collection box is spliced ​​together from multiple standard length units. A vertical maintenance space is reserved between the top surface of the box and the bottom surface of the return roller. The upper part of both sides of the collection box is provided with outwardly expanding inclined guide plates. The upper edge of the guide plate extends to the outer edge of both sides of the belt, and the lower edge connects to the side wall of the box. An automatic cleaning mechanism with a chain scraper is assembled inside the collection box; The first visual monitoring unit includes a first industrial camera, which is installed on the side of the belt conveyor frame with the lens aimed at the working surface of the return roller. The second visual monitoring unit includes a second industrial camera, which is installed on the upper part of the collection box, with the lens vertically downwards and aimed at the material collection area at the bottom of the box. The image closed-loop dustproof self-cleaning protection subsystem includes a compressed air pipeline, an electromagnetic control valve, and an air nozzle, with the air nozzle aimed at the lens surfaces of the first visual monitoring unit and the second visual monitoring unit. The edge controller is connected to the first visual monitoring unit, the second visual monitoring unit, the drive module of the chain scraper automatic cleaning mechanism, the electromagnetic control valve of the image closed-loop dustproof self-cleaning protection subsystem, and the drive mechanism of a belt correction device. The edge controller is configured as follows: The comprehensive sharpness evaluation parameter of the current frame image from the first or second visual monitoring unit is calculated in real time. When the comprehensive sharpness evaluation parameter is lower than a preset threshold, the electromagnetic control valve is controlled to trigger a purge to restore sharpness. The first wear recognition algorithm is run to process the images captured by the first industrial camera to quantify the wear parameters at different circumferential angles of the roller, and to calculate the roller wear coefficient. and the angle of the wear position ; The second ash accumulation recognition algorithm is run to process the images captured by the second industrial camera to calculate the ash accumulation volume on the left and right sides of the box, and to calculate the ash accumulation off-center loading coefficient. and the direction of off-center loading; A collaborative diagnostic approach is performed that leverages the spatially consistent causal relationship between belt misalignment causing uneven roller wear and uneven dust accumulation in the collection bin, by verifying the angle of the uneven wear position. Whether the belt misalignment is spatially aligned with the off-center loading direction is used to indirectly determine belt misalignment. When the following conditions are met simultaneously, a belt misalignment diagnostic signal and a quantified value of the misalignment degree are generated and output: a) The ash accumulation eccentric load coefficient The absolute value is greater than the first threshold. ; b) The roller wear coefficient Greater than the second threshold ; c) The angle of the eccentric wear position The absolute value of the spatial angle difference between the load direction and the off-center load direction is less than the third threshold. ; In response to the output belt misalignment diagnostic signal, a belt misalignment control command is generated and output to the drive mechanism of the belt misalignment device based on the quantized value of the misalignment degree and the preset mapping relationship between the quantized value and the correction command. The command includes the correction direction and correction amount to actively correct the belt misalignment.

2. The dual-vision intelligent monitoring belt conveyor material spillage and cleaning device according to claim 1, characterized in that, The automatic cleaning mechanism of the chain scraper includes a drive motor, a reduction transmission assembly, a main shaft, a ring transmission chain, wear-resistant scrapers, and an adjustable tension bracket; sprockets are fixedly installed at both ends of the main shaft, and the ring chain surrounds the sprockets to form a closed-loop transmission structure; multiple sets of wear-resistant scrapers are evenly fixed on the outside of the chain; the adjustable tension bracket is used to adjust the installation position of the main shaft to fine-tune the chain tension.

3. The dual-vision intelligent monitoring belt conveyor material spillage and cleaning device according to claim 1, characterized in that, The image closed-loop dustproof self-cleaning protection subsystem also includes an air pressure monitoring sensor; the edge controller is also configured to: periodically control the electromagnetic control valve to perform purging; immediately re-check the clarity after a single purging, and if it is not restored, purge and re-check again; when it still does not meet the standard after 3 consecutive purgings, trigger a hardware fault alarm.

4. The dual-vision intelligent monitoring belt conveyor material spillage and cleaning device according to claim 1, characterized in that, The collaborative diagnostics performed by the edge controller is an indirect determination that does not rely on directly capturing and identifying images of the belt edge.

5. A method for intelligent operation and maintenance and active deviation correction control of a belt conveyor system based on the device described in any one of claims 1 to 4, characterized in that, Includes the following steps: S1. Continuously acquire time-series image sequences of the surface of the return roller of the belt using a first industrial camera, extract the working surface area of ​​the roller, identify and quantify the wear parameters at different circumferential angles of the roller, including the wear area. and equivalent wear depth ,in For circumferential angle, The time is used to calculate the roller wear coefficient. and the angle of the wear position ; in, This represents the maximum circumferential wear depth. This represents the minimum circumferential wear depth. This represents the original wall thickness of the roller. S2. Continuously acquire image sequences of the inside of the collection bin using a second industrial camera. Use multi-frame dynamic filtering to distinguish between suspended dust and stationary powder, identify the powder accumulation boundary inside the bin, and calculate the dust accumulation volume on the left side. and the volume of dust accumulation on the right side Based on this, the ash accumulation eccentric load factor is calculated. and the direction of off-center loading; S3. Perform a collaborative diagnosis that utilizes the causal relationship between belt misalignment causing uneven roller wear and uneven ash accumulation in the collection bin, which are spatially consistent, and determines the position and angle of the uneven wear. Mapped to the same spatial coordinate system as the off-center load direction, when the following three conditions are met simultaneously, belt misalignment is indirectly determined to be the root cause, and a belt misalignment diagnostic signal and a quantified value of the misalignment degree are output: a) The ash accumulation eccentric load coefficient The absolute value is greater than the first preset threshold. ; b) The roller wear coefficient Greater than the second threshold ; c) The angle of the eccentric wear position The absolute value of the spatial angle difference between the load direction and the off-center load direction is less than the third threshold. ; S4. In response to the output belt misalignment diagnostic signal, based on the quantization value of the misalignment degree and the preset mapping relationship between the quantization value and the correction command, generate and output a correction control command to the drive mechanism of a belt correction device. The command includes the correction direction and correction amount to actively correct belt misalignment. S5. The maximum wear depth accumulated over time. Fitting the roller wear rate curve The system also collects real-time motor current data driving the chain scraper automatic cleaning mechanism and uses the current change rate to correct the wear rate curve. To predict the remaining service life of the roller ; S6. Based on the roller wear coefficient The absolute value of the ash accumulation eccentric load coefficient Based on the aforementioned deviation diagnosis results, calculate the overall health score of the equipment. ,in The preset wear coefficient warning threshold, As a penalty factor for deviation, when a deviation diagnosis signal exists, When there is no deviation diagnosis signal , Based on the comprehensive health score, a graded maintenance recommendation is output.

6. The control method according to claim 5, characterized in that, The quantification value of the deviation degree mentioned in step S3 is a deviation index. The index is based on the aforementioned wear coefficient. And ash accumulation eccentric load factor The absolute value weighted calculation is obtained. ,in and The preset weighting coefficients; the mapping relationship between the preset quantization value and the correction command mentioned in step S4 includes the deviation index. The corresponding correction direction and correction level are determined based on the threshold range in which the error occurs.

7. The method according to claim 5, characterized in that, The determination in step S3 utilizes the causal relationship and consistency between the dust accumulation and uneven load caused by belt misalignment and the uneven wear of the rollers in the spatial distribution, and is an indirect diagnosis that does not rely on direct imaging of the belt edge.

8. The method according to claim 5, characterized in that, The method for distinguishing between suspended dust and stationary accumulated powder by multi-frame dynamic filtering in step S2 is as follows: suspended dust exhibits random motion characteristics in continuous frames, while accumulated powder exhibits stationary or slow growth characteristics. Motion feature vectors are extracted through optical flow field analysis, and the categories of dust and materials are determined based on the temporal consistency of motion features.

9. The method according to claim 5, characterized in that, It also includes a closed-loop dustproof and self-cleaning protection mechanism for images: a preset timed blowing cycle, which periodically sprays compressed air onto the camera lens and viewing window; The system calculates the comprehensive image sharpness evaluation parameters in real time, integrating multi-dimensional indicators such as grayscale variance, edge gradient, and texture contrast. When the sharpness is lower than the preset threshold, the system automatically triggers a purge. After a single purge, a new image is acquired for sharpness re-check. If the sharpness recovers to above the threshold, the loop ends. If it still does not meet the standard, the purge is performed again for re-check. If it still does not meet the standard after three consecutive purges, a hardware fault alarm is triggered.

10. The method according to claim 5, characterized in that, The graded maintenance recommendations are based on a three-tiered handling strategy: Level 1 warnings are triggered when there is slight wear, dust accumulation is close to the upper limit, or there is a slight current fluctuation, with only local pop-up notifications and log recordings; Level 2 alarms are triggered when there is moderate wear, dust accumulation reaches the upper limit, or operating parameters are abnormal, with local audible and visual alarms and automatic activation of the cleaning mechanism. The three-level emergency interlock provides continuous audible and visual alarms and pushes emergency fault signals when there is a risk of roller perforation, dust overflow, or motor jamming and overload. The three-level graded handling strategy supports dual-mode switching between production priority mode and equipment protection mode. The active correction function has the highest priority and its execution is not constrained by the three-level graded handling strategy.

Images