Abstract: In coal-fired power plants, coal conveyor belts often experience misalignment, small cracks, or, in severe cases, belt tears during long-term use. If belt misalignment or cracking is not detected promptly, it can cause continued belt damage and, in serious cases, damage supporting structures, gearboxes, motors, or other equipment. Applying AI image recognition technology to coal conveyor monitoring and protection systems can improve detection coverage and increase the reliability of conveyor protection.
Preface
Belt conveyor systems are widely used for bulk-material handling in the power sector due to their safety, reliability, and adaptability. Conveyor belts represent a significant portion of operating cost for coal conveying systems, accounting for roughly 30% of total system expense. Effective monitoring and protection can extend belt life and support a more complete belt life management process.
1. Implementation Background
A power plant has dual-route belt conveyor systems arranged on its coal handling trestles, with redundant A and B lines on the same trestle. Systems A, B, and C include 38 belt conveyors in total. System A comprises 9 sections with 18 belt conveyors; System B comprises 10 sections with 20 belt conveyors. The A and B systems use DT75 fixed-type belt conveyors produced by a local heavy-lift machinery manufacturer. Those belts have canvas carcasses with rubber cover and ply-bonding material, four plies in total, with top cover rubber thickness of 6 mm and bottom cover 3 mm. The C system uses DTII fixed-type rubber conveyor belts produced by another local manufacturer; those are steel-cord belts. Existing belt misalignment protection for the A and B systems includes belt deviation switches and guiding idlers.
Coal conveyor belts are essential to plant operation. During use, installation errors during belt replacement, improper splice location, misaligned discharge points, foreign objects between the drum and belt, damaged or missing idlers, sticking or damage from debris, and other factors can cause belt misalignment, scratches, and fatigue cracks. If these faults are not detected and addressed promptly, they will lead to continued belt deterioration. Delayed belt shutdown can lead to damage to frames, gearboxes, motors, and other equipment, and may result in economic loss or personal injury.
Currently, belt misalignment detection mainly uses fixed-position deviation switches, which lack flexibility and have limited detection accuracy and reliability. There is no widely effective automated method for detecting belt cracking; inspections are often manual and performed at belt inspection speed, resulting in low efficiency, high labor intensity, poor accuracy, and safety risks. Therefore, research and application of detection methods for belt misalignment and cracks have significant operational value.
An intelligent belt detection system based on image recognition and AI algorithms covers all coal conveyor belts, performing intelligent detection, analysis, and alarm functions for belt misalignment, cracks, and tears. This can prevent belt rupture and reduce abnormal wear caused by frequent misalignment, improving system safety and production efficiency while lowering belt replacement costs.
2. System Functions
The plant installs real-time video monitoring devices at the head and tail of 38 belt conveyors on the coal A/B lines, using network cameras of 4 megapixels or higher. Explosion-proof network cameras are used where flammable or explosive hazards exist. Video storage supports at least 45 days of retention and a stream bitrate of no less than 2 Mbps. To prevent video loss when front-end cameras lose their connection to the central video storage device, the system supports resume-on-connect functionality. The video monitoring system includes transmission equipment, storage devices, video monitoring hardware, and the video monitoring software platform.
The intelligent belt detection system based on image recognition and AI algorithms was developed through research into image preprocessing, feature extraction, classification, and image matching algorithms, comparing the advantages and disadvantages of various approaches. Leveraging computer vision techniques, the system implements automatic detection. When a serious condition is detected, the system can automatically stop the conveyor according to authorization, preventing economic loss from belt damage.
- Equipment monitoring: Detect operational conditions of primary conveyor equipment, such as belt misalignment, tearing, or spillage.
- Safety monitoring: Detect foreign objects on the belt and issue alarms to prompt staff action and prevent accidents. Detect personnel entry into restricted areas and issue immediate warnings.
3. Technical Approach
The system uses machine vision combined with auxiliary structured-light contouring to identify and evaluate surface damage on rubber conveyor belts in real time, issuing alarms or triggering shutdowns based on the assessment. The principle uses a specialized optoelectronic camera to image the belt surface. Structured light projects a contour line that conforms to the belt surface; real-time image algorithms analyze changes in the contour line to detect anomalies.
3.1 Major Equipment Components
- Laser emitter
Presents the belt surface contour completely. A wide-angle light-source design ensures accurate contour imaging regardless of belt shape. - Explosion-proof camera
(1) A 2048-pixel industrial line-scan camera ensures that within a 40 cm lateral field of view the minimum resolvable element is 0.4 mm, meeting imaging precision requirements.
(2) Maximum capture speed of 81 frames per second and a 30 cm longitudinal field of view ensure complete imaging at object speeds up to 3 m/s.
(3) Exposure time range of 2 μs to 10 ms prevents motion blur when imaging high-speed objects. - Recognition host
An intelligent processing module performs image preprocessing. A photoelectric module handles remote transmission of preprocessed images, and a power cabinet supplies the recognition host. - High-performance compute servers
Algorithm processing time has been optimized to the millisecond range; single-frame processing is accelerated to under 30 ms. A cluster of high-performance servers provides real-time processing and response for each image unit, enabling centralized control actions such as shutdowns or alarms within short timeframes. - Core technology: AI intelligent algorithms
For contour-line characteristics—small local fluctuations that cross the field of view and generally lack large vertical movement—the line-segment extraction algorithm uses gradient search and preset search boundaries to reduce extraction time and ensure real-time performance.
3.2 Implemented Functions
System functions based on image recognition and AI algorithms include:
Belt tear detection
Belt tearing is a highly destructive failure that disrupts coal transport and can cause major economic loss if not handled promptly. Tearing is detected when a running belt shows gaps or overlaps. Cameras are installed at locations prone to tearing and provide real-time video via network protocols such as RTSP. Belt images are transmitted to the backend for analysis and recognition; when tearing is detected, the backend issues an automatic alarm.
Belt misalignment detection
Belt misalignment is a common fault caused by improper installation, material buildup on idlers, belt slack, uneven coal distribution, excessive vibration, and other issues. Misalignment can damage drum or idler bearings, cause belt tearing, and lead to spillage and dust. Two alignment reference lines are marked on the head and tail drums. Fixed overhead monitoring devices observe these lines. The backend analyzes live images to determine whether the reference lines are occluded. If either reference line is blocked, the system determines the belt is misaligned and issues an alarm.
Spillage detection
The system analyzes camera video to detect coal spillage along the belt edges and issues timely alarms when spillage is found.
Foreign object detection
The system analyzes video frames for foreign objects. When objects such as wooden sticks, steel pipes, or large stones are detected, alerts are generated.
Smoke and fire detection
Smoke and fire detection analyzes video to identify abnormal smoke or early fire indicators within monitored areas, issuing the fastest and most effective alarms and providing useful situational information. Real-time images can be reviewed for direct incident command and dispatch.
Intrusion detection
Non-safe zones can be defined within the monitoring area. When unauthorized entry is detected, the system generates an immediate alarm.
Unsafe behavior detection
The system detects unsafe personnel behaviors in video, such as not wearing safety helmets or dust masks.
Face recognition (personnel attendance)
Within predefined time windows, the system detects and recognizes faces appearing in video, automatically recording staff attendance and presence.
3.3 Alarm Management
- When the system detects an anomaly, it issues audible and visual alarms according to configured settings and continues the alarm until manually confirmed and cleared.
- Today's alarms can be presented in charts showing alarm type, severity, time, location, and handling status. Filters can be applied by alarm level, type, and source.
- Today's alarms can be categorized as pending, handled, or false positives. Historical alarm records can be displayed with details including alarm type, severity, time, area, and handling status.
- Filtering by alarm level, type, and source is supported across current and historical records.
3.4 Mobile App
A mobile app can push alarm notifications, allowing real-time reception of system alarms, alarm handling, and access to statistical reports.
4. Economic Benefit Analysis
The system automatically detects belt tearing via camera-based real-time monitoring, replacing traditional manual inspections. It can also infer tearing and severity by monitoring motor voltage and current variations driving the belt. Applying this system reduces replacement frequency of belts, drums, and other wear parts, lowering costs while improving operational safety.
5. Conclusion
Research and application of an intelligent belt detection system based on image recognition and AI algorithms address current problems in plant inspection such as low automation, difficult inspections, and untimely fault detection. Information-based methods can perform high-difficulty and high-intensity inspections, intelligent monitoring, and automated alarms that manual labor cannot achieve. By completing complex conveyor inspections with fewer personnel, the system improves efficiency, maintains work quality and personal safety, reduces belt replacement frequency and wear-part expenses, and supports performance and efficiency improvements for coal-fired power plants.
