Belt conveyor fault intelligent inspection diagnosis method and system based on digital twinning

Abstract

The application discloses a kind of based on digital twinning's belt conveyor fault intelligent patrol inspection diagnosis method and system, the diagnosis method includes: the construction of belt conveyor digital twinning model, fusion three-dimensional geometric model and physical behavior model, based on digital twinning model constructs the fault simulation model of simulating the association of different fault types and physical parameters;Collect and preprocess multi-source data, load fault detection model identification image and fuse physical parameter data;Intelligent inference model carries out fault inference;Digital twinning model is based on fault simulation model and loads solution simulation to reproduce fault scene;Result is effective then issues instruction;The diagnosis method builds the digital twinning model associated with equipment behavior, combines image data and physical parameter data comprehensive diagnosis, realizes fault and behavior association, accurately locates fault;Fault simulation model cooperates digital twinning model to reproduce fault scene, loads inference result simulation, virtual-real joint mechanism control instruction, guarantees operation safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent conveying, monitoring and fault diagnosis technology, specifically to an intelligent inspection and diagnosis method and system for belt conveyors based on digital twins. Background Technology

[0002] Belt conveyors are core equipment in port bulk cargo transportation, mainly used to connect berths, storage yards, and loading and unloading equipment to achieve continuous transfer of bulk cargo such as coal, ore, and grain. Core components include a drive system, conveyor belt, idler rollers, tensioning device, braking system, and frame. Port belt conveyors must meet the demands of large-volume, long-distance, and highly continuous operations. Their operational stability directly affects the smoothness of the port's logistics chain. A malfunction can not only interrupt loading and unloading operations but also cause cargo delays and ship congestion, impacting port turnover efficiency. Furthermore, conveyor belt tears and idler roller jamming can lead to material leakage and pollution, equipment damage from impacts, and even threaten the safety of port personnel.

[0003] Currently, there are three main types of inspection and fault diagnosis methods for belt conveyors: First, manual inspection, where maintenance personnel carry specialized instruments to inspect along the conveyor line and work with basic protection devices to achieve fault early warning; second, sensor monitoring and robot inspection, which involves deploying special monitoring equipment and combining it with track-mounted inspection robots to collect data, with some systems introducing image detection models to identify visually identifiable faults; and third, expert system-based diagnostic methods, which typically build a rule base based on fault tree analysis and combine it with a case library to achieve fault diagnosis through rule-based reasoning and case-based reasoning algorithms.

[0004] However, existing technologies have key flaws: First, expert system reasoning lacks verification of equipment operating status, relying solely on preset rules or historical cases without combining real-time operating parameters and physical behavior verification results, which can easily lead to misdiagnosis due to rule conflicts or case matching deviations; Second, fault detection is disconnected from equipment behavior, with existing image detection technology relying only on conveyor belt surface images without integrating physical parameters, making it impossible to correlate faults with equipment dynamics and difficult to locate the root cause of the fault; Third, there is a lack of virtual-real linkage mechanism, with no virtual pre-verification step between fault diagnosis results and control commands, and direct issuance of commands can easily induce secondary accidents, failing to guarantee operational safety.

[0005] Therefore, a digital twin-based intelligent inspection and diagnosis method and system for belt conveyor faults is provided to solve the problems of lack of verification of expert system reasoning, disconnect between fault detection and equipment behavior, and lack of pre-verification of operation and maintenance instructions in the existing inspection and fault diagnosis technology of belt conveyors, which can easily induce secondary accidents. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for intelligent inspection and diagnosis of belt conveyor faults based on digital twins, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] A digital twin-based intelligent inspection and diagnosis method for belt conveyor faults includes the following steps and contents:

[0009] S1. Build a digital twin model of the belt conveyor, which integrates the three-dimensional geometric model and physical behavior model of the belt conveyor; construct a fault simulation model based on the digital twin model that can simulate the correlation between different fault types and physical parameters; establish a rule base for subsequent reasoning.

[0010] S2. Real-time acquisition and preprocessing of multi-source data, including image data and physical parameter data, loading of a trained fault detection model for visual fault recognition, fusion of physical parameter data, and output of detection results; calling the rule base and, based on the rule reasoning mechanism, the intelligent reasoning model performs fault reasoning on the detection results, and outputs the fault cause and its solution.

[0011] S3. Inject fault parameters into the digital twin model, call the fault simulation model to reproduce the fault scenario, so that the simulated fault data is consistent with the actual fault data, load the solution for online simulation, and verify the effectiveness of the reasoning results.

[0012] S4. If the solution effectively resolves the fault, the inference result is output, and a correction instruction is issued in conjunction with manual triggering. If the inference result fails to effectively resolve the fault, the rule base is updated based on the data generated by the verification, and the process returns to step S2 to re-perform the inference until the inference result is verified to be effective or the preset iteration threshold is reached.

[0013] According to one aspect of this disclosure, the intelligent reasoning model performs comprehensive reasoning based on preset rules and historical failure cases. Step S1 also includes building a case library. In step S2, the intelligent reasoning model combines the rule-based reasoning mechanism and the case-based reasoning mechanism to perform failure reasoning. The case-based reasoning mechanism retrieves similar cases by calling the case library and using the trained weighted KNN algorithm, calculates their similarity, and outputs similar sample case data.

[0014] The intelligent reasoning model performs weighted fusion of the confidence of the fault causes output by the rule-based reasoning mechanism and the similarity of the fault causes of similar cases output by the case-based reasoning mechanism, and outputs a comprehensive reasoning result.

[0015] According to one aspect of this disclosure, the training method for the weighted KNN algorithm is as follows:

[0016] S21. Construct a judgment matrix that satisfies the consistency test, and determine the initial weights of the index attributes of each case;

[0017] S22. Select data from several historical failure cases as training samples. Divide the training samples into a failure training set and a failure validation set using cross-validation. Calculate the sample similarity of the failure training set and determine the K value.

[0018] S23. Iteratively optimize the weights of the case index attributes, retaining the weight combination with the highest accuracy.

[0019] S24. Validate the weighted KNN algorithm and weight combination using a fault verification set to ensure an accuracy of ≥90%;

[0020] S25. Deploy the trained weighted KNN algorithm into the intelligent inference model.

[0021] According to one aspect of this disclosure, the fault detection model performs transfer learning based on a pre-trained detection model, collects fault images corresponding to various fault types, freezes the backbone network layers of the pre-trained detection model, retains its neck layer structure and head layer structure, and performs adaptive adjustments to it, thereby training and deploying the fault detection model:

[0022] M1. Perform image enhancement processing on the collected fault images to obtain a fault image set. Divide the fault image set into a training set, a validation set, and a test set, and label the fault areas and fault types in the fault images.

[0023] M2. Set training parameters, start the training script, train the fault detection model and save;

[0024] M3, the fault detection model, integrates image detection results and physical parameter data to output detection results.

[0025] According to one aspect of this disclosure, the fault detection model integrates a dark channel prior algorithm, which is used to dehaze the fault images used for training and the test images to be detected in real time.

[0026] According to one aspect of this disclosure, several samples of physical parameter data of a belt conveyor under normal operating conditions are collected for field testing, and the data are divided into a physical test set and a physical verification set. The method for constructing a physical behavior model is as follows:

[0027] S11. Import the 3D geometric model and set the material parameters;

[0028] S12, Add constraints and driving torque;

[0029] S13. The physical behavior model is based on the multibody dynamics equations:

[0030] ;

[0031] in, The inertia matrix, For generalized coordinates, For generalized speed, For generalized acceleration, Let the Coriolis force and centrifugal force be vectors. The vector of the gravity term. For driving torque;

[0032] S14. Identify the parameters to be identified in the physical test set and determine the identification parameters;

[0033] S15. Integrate the identification parameters to build a physical behavior model, input physical verification to verify the accuracy of the physical behavior model, adjust the physical behavior model and save the final model.

[0034] According to one aspect of this disclosure, the digital twin model is dynamically updated based on a hierarchical update strategy, so that it is synchronized with the actual operating status of the belt conveyor in real time; the update of the three-dimensional geometric model is triggered by real-time scanning to obtain entity point cloud data at a preset first frequency, and the geometric deviation between the current three-dimensional geometric model is calculated. If the deviation exceeds a preset value, the three-dimensional geometric model is locally updated based on an incremental algorithm; the update of physical parameters is triggered by real-time acquisition of physical parameter data at a preset second frequency, and the corresponding physical quantities in the physical behavior model are adjusted based on a parameter correction algorithm.

[0035] According to one aspect of this disclosure, the preprocessing in step S2 is based on a multi-source data fusion model. The multi-source data fusion model performs noise reduction processing and fault state estimation of physical parameter data based on Kalman filtering. At the same time, based on DS inference theory, it performs preliminary fault symptom inference and fusion on physical parameter data, and outputs the processed physical parameter information and its fusion result.

[0036] According to one aspect of this disclosure, a method for constructing a fault simulation model is as follows:

[0037] Step 1: Organize the parameter changes corresponding to typical fault phenomena of belt conveyors and establish a fault tree to establish the correlation between fault phenomena, fault types and fault parameters.

[0038] Step 2: Inject fault tree data into the physical behavior model to construct fault simulation models for each fault type;

[0039] Step 3: Associate the simulation data of the fault simulation model with the data and patterns of the fault parameters, and add fault labels to the fault simulation model.

[0040] A digital twin-based intelligent fault inspection and diagnosis system for belt conveyors includes a physical layer, a data acquisition layer, a digital twin modeling layer, an intelligent reasoning layer, a database, and a human-computer interaction layer; among which,

[0041] The physical layer includes belt conveyors, inspection robots, and sensor arrays, which include vibration sensors, temperature sensors, and image sensors.

[0042] The data acquisition layer is based on edge computing nodes, on which a multi-source data fusion model is deployed to initially fuse physical parameter data collected by the physical layer.

[0043] The digital twin modeling layer is used to build a digital twin model of the belt conveyor, including a three-dimensional geometric model, a physical behavior model, and a dynamic update module. The dynamic update module updates the three-dimensional geometric model and the physical behavior model at a fixed frequency based on the physical parameter data of the physical layer.

[0044] The database includes a rule base and a case base. The rule base consists of production rules built based on intermediate nodes of the fault tree, while the case base stores historical fault cases that have been standardized according to case attributes.

[0045] The intelligent inference layer includes a detection module, an inference module, and a virtual verification module. The detection module deploys a fault detection model that processes and detects image data, receives physical parameter data uploaded from edge computing nodes, and further fuses the image data and physical parameter data to output the detection result. The inference module deploys an intelligent inference model that performs comprehensive inference based on a rule base and a case base, and outputs the comprehensive inference result to the virtual verification module. The virtual verification module deploys a fault simulation model that calls the corresponding fault simulation model based on the inference result, injects it into the digital twin model for fault simulation, reproduces actual faults, dynamically updates the solution loaded with the inference result, and verifies the effectiveness of the inference result.

[0046] The human-machine interaction layer includes a virtual-real linkage display module and an instruction control module. The virtual-real linkage display module is used to monitor the operating status of the belt conveyor and display the fault verification report of the intelligent inference layer. The instruction control module provides manual instruction input and sends the solution verified by simulation to the control system of the belt conveyor based on the triggering of the manual instruction.

[0047] According to one aspect of this disclosure, the modeling of the three-dimensional geometric model is based on real-time scanning by a laser scanner and supplemented by key geometric parameters from equipment drawings. The physical behavior model is established by using physical parameter data of the belt conveyor under normal operating conditions from on-site testing.

[0048] Compared with the prior art, the present invention provides a digital twin-based intelligent inspection and diagnosis method and system for belt conveyor faults, which has the following advantages:

[0049] This method constructs a digital twin model associated with the physical behavior of the equipment, integrates conveyor belt surface image data and physical parameter data for comprehensive fault diagnosis, and overcomes the diagnostic limitations of existing methods that rely solely on surface images. It achieves the correlation between faults and equipment behavior, can capture latent fault signs, and accurately locate the root cause of faults, providing core support for subsequent accurate monitoring and maintenance. The fault simulation model injects fault parameters into the digital twin model to reproduce the fault scenario, loads the inference results of the intelligent inference model for online simulation, monitors whether the equipment operating status has returned to normal, and sends control commands after the simulation passes, avoiding the direct issuance of manual commands that may induce secondary accidents, and ensuring the reliability and safety of maintenance operations.

[0050] In addition, the fault detection module first uses the dark channel prior algorithm to remove fog, and then fuses the image data with the physical parameters of the physical layer of the entity to form a structured fusion diagnostic dataset; the intelligent reasoning model performs comprehensive reasoning based on the rule base and case base to improve the accuracy of fault diagnosis and avoid misdiagnosis caused by rule conflicts or case bias.

[0051] The system's data acquisition layer collects multi-source data from all dimensions of the physical layer. The detection module performs fault detection on image data and integrates physical parameter data. The intelligent inference layer infers fault causes and solutions based on an improved RBR+CBR inference mechanism. Through collaboration between the virtual verification module and the digital twin modeling layer, the results are extracted and the solutions are simulated online in the virtual-physical mapping model of the belt conveyor in the digital twin modeling layer. Control commands are issued through collaboration between the human-machine interaction layer and the digital twin modeling layer. Through seamless collaboration at all levels, the entire system achieves fully automated closed-loop management of the entire process, from automatic fault detection, intelligent inference, online solution verification, to virtual-physical linkage control. This significantly reduces the cost of manual intervention, improves the efficiency and accuracy of fault handling, and provides full-stack technical support for the unmanned and intelligent operation and maintenance of belt conveyors. Attached Figure Description

[0052] Figure 1 This is a flowchart of the intelligent inspection and diagnosis method for belt conveyor faults based on digital twins disclosed in this invention.

[0053] Figure 2 This is a schematic diagram of the intelligent inspection and diagnosis system for belt conveyors based on digital twins disclosed in this invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely the best embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The term "embodiment" as used herein means that a particular method, step, or content described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0056] This embodiment provides an intelligent fault inspection and diagnosis system for belt conveyors based on digital twins, such as... Figure 2 As shown, it includes a physical layer, a data acquisition layer, a digital twin modeling layer, an intelligent reasoning layer, a database, and a human-computer interaction layer, with each layer communicating with each other in sequence; among them,

[0057] The physical layer of this implementation includes the belt conveyor itself, the inspection robot, and a sensor array, which includes vibration sensors, temperature sensors, and image sensors. The following details the specific deployment steps, installation parameters, and debugging methods for each piece of hardware in the physical layer to ensure that the hardware system meets the requirements for comprehensive monitoring of the port's belt conveyor, providing reliable physical support for subsequent digital twin modeling and intelligent reasoning diagnosis.

[0058] (1) Select the belt conveyor commonly used in port bulk cargo transportation as the core equipment of the physical layer and confirm its key parameters, including the width of the conveyor belt, the conveying speed, the rated speed and rated power of the drive motor, the diameter of the idler roller, the spacing between the idler rollers, the tensioning method and the adjustable range of the tension force.

[0059] A three-dimensional coordinate system is established with the center of the drive drum at the head of the conveyor as the origin. The X-axis is along the conveying direction, the Y-axis is perpendicular to the ground, and the Z-axis is horizontal and perpendicular to the conveying direction. The key installation coordinates of the motor, reducer, hydraulic coupler, conveyor belt, and idler group are recorded.

[0060] (2) To comprehensively monitor the operating status of the port belt conveyor, the sensor group also includes a methane sensor and a smoke sensor; the sensor group includes a methane sensor, a smoke sensor, a vibration sensor, a temperature sensor, and a high-definition camera; the cables of the sensor group are all introduced into the explosion-proof junction box in the port operation area through explosion-proof cable holes, and the installation steps are as follows:

[0061] Step 1: Install a temperature sensor at each critical rotation node: including the front and rear bearing housings of the drive motor, the input and output shaft end caps of the reducer, and the drive roller bearing housing, ensuring that the temperature sensor is in close contact with the metal surface of the component being measured.

[0062] Step 2: Install vibration sensors at key transmission nodes: The vibration sensors are fixed to the drive motor housing near the bearing, the reducer housing at the intermediate shaft position, and above the idler roller bearing seat via magnetic bases (select 1 set for installation for every 10 idler roller sets). The vibration sensor sampling frequency is set to 1000Hz, the range is ±5g, and the output signal is 4-20mA.

[0063] Step 3: Install the methane sensor and smoke sensor: Both are installed on the top of the working area at the head, middle, and tail of the conveyor using expansion bolts. The methane sensor has a range of 0-100% LEL and an accuracy of ±3%, while the smoke sensor has a range of 0-2000 ppm and an accuracy of ±7%.

[0064] Step 4: Install image sensors, such as high-definition cameras, at key nodes of the conveyor belt. The high-definition cameras are installed at a height of 2.5m above the surface of the conveyor belt using angle steel brackets. The high-definition cameras are arranged along both sides of the conveyor, respectively targeting the contact point between the drive roller and the head of the conveyor belt, the contact point between the middle conveyor belt surface and the idler roller (arranged at intervals of 50m / camera in the middle), and the redirecting roller. The high-definition cameras use 4 megapixels, 25fps frame rate, and the lens angle adjustment range is not less than 120°. They are connected to the port operation area wireless base station via industrial Ethernet.

[0065] (3) The inspection robot can perform autonomous inspections along the port's belt conveyor. The inspection robot is equipped with a main control chip, a DC motor for walking, and a lithium battery, and is equipped with a matching track and charging device. Its deployment steps are as follows:

[0066] Step 1: Install the inspection track: The inspection track is arranged diagonally above the belt conveyor, with the center line of the track parallel to the center line of the conveyor, and the installation height is 0.8m from the surface of the conveyor belt;

[0067] Step 2: Install the inspection robot body: Hoist the inspection robot, including the walking mechanism, main control box, and sensor module, onto the inspection track. Adjust the contact gap between the walking wheels and the track to ensure that the robot can hover stably on a 40° slope. Install a vision extension device at the bottom of the inspection robot and load a camera at the end. The camera should be aimed at the rollers on the edge of the conveyor belt. A laser scanner should be placed at the bottom of the inspection robot.

[0068] Step 3: Install autonomous charging piles: Install autonomous charging piles at the starting point of the track, i.e., the head of the conveyor. The contact charging method is adopted. When the battery level of the inspection robot is ≤20%, it will automatically return to charge. The full charge time is ≤2 hours.

[0069] In this embodiment, the data acquisition layer is based on the constructed edge computing nodes. It preprocesses and stores multi-source physical parameter data acquired by the physical layer. According to the requirements of data acquisition, transmission, and preliminary processing of port belt conveyors, the selection and specific deployment and installation methods of the relevant hardware of the edge computing nodes are as follows:

[0070] Step 1: Install the industrial control computer: Install the industrial control computer in the explosion-proof control cabinet at the head of the conveyor, connect it to the port operation area switch via network cable, and configure a static IP address and subnet mask;

[0071] Step 2: Deploy the software system: Install the Windows 10 embedded system on the industrial control computer to deploy the data acquisition and preprocessing software, and install a database to store real-time sensor data. The sampling periods for various types of data are as follows: temperature data 1s, gas data 1s, vibration data 100ms, and image data 25fps.

[0072] Step 3: Connect sensor signals: Connect the sensor group signals to the edge computing node according to their type. RS485 interface sensors, including methane, smoke, and temperature sensors, are connected to the industrial computer through a converter. Vibration sensors are connected through a data acquisition card. Image data is transmitted directly through industrial Ethernet.

[0073] In addition, check the compatibility of the existing protection devices and sensor groups of the belt conveyor, including coal pile protection, anti-deviation device, and overheating sprinkler device. Connect the contact signal of the coal pile protection to the PLC input module, and connect the offset signal of the anti-deviation device with a range of 0-150mm to the edge computing node in sync with the vibration sensor signal to ensure that the physical layer data collection is complete and meets the data monitoring requirements of the continuous operation of the port belt conveyor.

[0074] Step 4: Develop a fusion algorithm: In the data processing software, develop a multi-source data fusion model program based on Kalman filtering and DS inference theory. Kalman filtering is used to suppress noise in vibration and temperature data, with a process noise variance of 0.01 and a measurement noise variance of 0.05. Performance status estimation is then performed. Next, based on preset fault rules, DS inference theory is used to infer and fuse fault symptoms in the noise-processed multi-sensor data, such as calculating the confidence level of "temperature overheating + abnormal vibration" as a drive motor fault.

[0075] The fusion result output cycle is 500ms, and the data format is JSON, which includes timestamp, sensor ID, preliminary fusion inference result and confidence level.

[0076] To ensure that each hardware device in the physical layer and data acquisition layer operates normally independently, the sensors, inspection robots, and edge computing nodes need to be debugged separately. The specific debugging parameters and procedures are as follows:

[0077] Step 1: Sensor debugging: Connect the corresponding voltage to each sensor, read the data of each sensor through the host computer software, adjust the threshold of each sensor, simulate the fault to confirm that the sensor can trigger the alarm and upload data. For example, use a hot air gun to heat the motor to 85°C and test whether the temperature sensor is triggered.

[0078] Step 2: Debug the inspection robot: Remotely control the inspection robot to run along the track, test the adjustable range of walking speed, turning flexibility and the extension and positioning error of the vision extension device robotic arm, and ensure the success rate of reading the NFC module on the inspection robot.

[0079] Step 3: Debug edge computing nodes: Use the Ping command to test the network connectivity between the edge computing nodes and the sensor group, and between the belt conveyor PLC and the wireless base station, to confirm the integrity of data storage, i.e., no packet loss and continuous timestamps.

[0080] Step 4: Start the belt conveyor under no-load operation. The inspection robot patrols according to the preset inspection route. The edge computing node collects and merges multi-source physical parameter data in real time. The merged data is then uploaded to the digital twin modeling layer with a delay of ≤100ms. This confirms that the physical layer and the data acquisition layer are working together normally, with no data interruption or false alarms, thus meeting the monitoring requirements for the continuous and stable operation of the port belt conveyor.

[0081] The digital twin modeling layer in this embodiment is used to construct a digital twin model of a belt conveyor. The digital twin model includes a three-dimensional geometric model and a physical behavior model. The dynamic update module periodically updates the three-dimensional geometric model and the physical behavior model based on data from the physical layer. Combining the structural characteristics of the belt conveyor with port operating parameters, the following details the construction steps, parameter settings, and verification methods for the three-dimensional geometric model, physical behavior model, and dynamic update module in the digital twin modeling layer. This achieves accurate mapping between the physical layer equipment and the virtual twin model, providing a virtual verification environment for subsequent intelligent reasoning and diagnosis.

[0082] (1) Collect modeling data

[0083] A laser scanner was used to scan the actual belt conveyor with a scanning accuracy of 0.1 mm, and the three-dimensional point cloud data of the drive system, conveyor belt, idler group and frame were obtained and exported as STL format.

[0084] Based on the supplementary drawings of the coordinates of each key node of the belt conveyor and the key geometric parameters, the three-dimensional point cloud data was adjusted and supplemented.

[0085] Key physical parameters of the belt conveyor were determined through field testing and literature review, including the elastic modulus, Poisson's ratio, and density of the conveyor belt; the friction coefficient of the idler rollers under normal and jamming conditions; the rotational inertia and rated torque of the drive motor; the transmission efficiency of the reducer; and the belt tension, idler roller speed, and motor current of the belt conveyor under the rated transportation conditions of ore in the port.

[0086] (2) Constructing a three-dimensional geometric model

[0087] To accurately reproduce the physical structure of the port belt conveyor, professional 3D software was used to construct 3D models of each component based on the collected modeling data and complete the assembly of the whole machine. This ensured that the assembly gap between each component after assembly was ≤1mm, the moving parts could rotate freely, and the overall assembly model file was saved. The model accuracy was controlled within 0.1mm.

[0088] (3) Construct a physical behavior model

[0089] To simulate the actual operating state of a port belt conveyor, the three-dimensional geometric model needs to be imported into dynamics software and a physical behavior model needs to be constructed. The specific construction steps are as follows:

[0090] S11. Import the final assembly model and set the material parameters;

[0091] S12. Add constraints and driving torque: Add rotary joints, translational joints and fixed joints according to the motion nodes of the belt conveyor, and set the constraint friction coefficient of the kinematic joints; control the addition of rotary driving torque according to the rated characteristics of the motor.

[0092] S13. Establish the multibody dynamics equations:

[0093] ;

[0094] in, The inertia matrix, For generalized coordinates, For generalized speed, For generalized acceleration, Let the Coriolis force and centrifugal force be vectors. The vector of the gravity term. The vector of rolling resistance / friction terms related to speed and temperature t. For driving torque;

[0095] S14. Parameter Identification: Several data points of normal operation of the belt conveyor are collected through the physical layer, including motor speed, drum speed, conveyor belt speed, and idler temperature. These are imported into the dynamics software, and the least squares method is used to identify the parameters to be identified for constructing the multibody dynamics equations. The parameters to be identified include at least the motor's moment of inertia parameter and the rolling resistance / friction parameters related to the idlers. The multibody dynamics equations are updated based on the identified parameters. and For example: the rated speed of the motor is 1480 r / min, adjust... The rotational inertia parameter of the motor rotor is adjusted to ensure that the error between the simulated speed and the actual speed is ≤5%. The rolling resistance / friction parameter is adjusted to ensure that the error between the simulated conveyor belt speed and the actual speed is ≤3%. Simulation conditions are set including no-load, rated load, and 120% overload. The simulation data is compared with the actual data to ensure that the model error is ≤5%. Among them, the idler roller temperature is used to characterize the friction state of the idler roller bearing and the change of rolling resistance. The rolling resistance / friction parameter is corrected or abnormal operating conditions are verified with the model. It is not directly used to identify the inertia matrix, Coriolis term, and centrifugal force term.

[0096] S15. Integrate various identification parameters, build a physical behavior model based on ADAMS, input an independent validation set to verify the error of the physical behavior model, adjust the physical behavior model, control the model error within 5%, and save the final physical behavior model.

[0097] (4) Deployment of the dynamic update module

[0098] To ensure that the digital twin model is synchronized with the actual operating status of the port belt conveyor in real time, the dynamic update module uses Socket communication to achieve data interaction with the edge computing node and receives physical parameters collected in real time by the data acquisition layer, including temperature, vibration and conveyor belt offset.

[0099] The incremental update program of the dynamic update module adopts a layered update strategy. The 3D geometric model is triggered to update every 1 hour. The data acquisition layer receives the latest point cloud data from the laser scanner. The edge computing node compares the geometric deviation between the current model and the point cloud data. If the deviation is >0.5mm, the local structure of the 3D geometric model is updated using an incremental algorithm. The physical behavior model is triggered to update every 100ms. Real-time acquired physical parameter data is input into the physical behavior model. The physical parameters in the physical behavior model are adjusted through a "parameter correction algorithm". For example, when the actual value of the motor temperature is 5℃ higher than the simulation value, the motor heat dissipation coefficient in the physical behavior model is automatically reduced by 10% to ensure that the output of the physical behavior model is consistent with the actual operating state. The program feeds back the updated physical behavior model status to the edge computing node for logging and anomaly alarms.

[0100] The incremental update program is compiled into an executable file and deployed to the industrial control computer. It starts automatically on boot and the task scheduler is set to trigger the update periodically according to the hierarchical update strategy.

[0101] (5) Validate the digital twin model

[0102] Based on the port's ore transportation conditions, the belt conveyor was started to operate at full load, and the digital twin model was run synchronously. The virtual and actual data were compared to verify the accuracy of the digital twin model.

[0103] In terms of static verification, if the deviation between the key dimensions of the 3D geometric model and the actual dimensions of the physical object is within the error range, then the static verification is satisfied.

[0104] In terms of dynamic verification, if the deviation between the simulated motor speed and the actual speed, and the deviation between the simulated conveyor belt speed and the actual conveying speed are within the error range during normal operation, then the dynamic verification is satisfied.

[0105] The database in this embodiment includes a rule base and a case base. The rule base consists of production rules built based on intermediate nodes of a fault tree. The case base stores historical fault cases with standardized case attributes. Case attributes include index attributes and association attributes. The index attributes are used for case retrieval by the intelligent inference layer. The following describes in detail the construction method of the rule base and case base in the intelligent inference diagnostic layer, using a digital twin model and multi-source physical parameters:

[0106] (1) Construction of the rule base

[0107] To achieve fault diagnosis based on preset rules, the rules in the correlation between fault phenomena and causes of belt conveyors are extracted and stored. The specific construction steps are as follows:

[0108] Step 1: Extract Production Rules: Referring to common fault types and phenomena of belt conveyors, extract several production rules with the fault phenomenon as the trigger condition and the fault cause and confidence level as the output. The rule format is "If fault phenomenon, then fault cause"; for example:

[0109] If the drive motor temperature is >80℃, then the fault causes are heat dissipation obstruction / overload operation / stator winding grounding, with confidence levels of 0.6, 0.3, and 0.1 for the three fault causes, respectively.

[0110] If the conveyor belt offset is greater than 70mm, then the causes of the fault are improper roller installation / idler jamming / uneven coal distribution, with confidence levels of 0.5, 0.3, and 0.2, respectively.

[0111] If the roller vibration value is greater than 0.5g and the temperature is greater than 70℃, then the cause of the failure is insufficient bearing lubrication / debris entering the bearing, with confidence levels of 0.7 and 0.3 respectively.

[0112] If the material accumulation height under the conveyor belt is >100mm, then the cause of the fault is coal accumulation, with a confidence level of 1.0.

[0113] Step 2: Store rules in the rule base: Create a "rule table" with fields including "rule ID", "fault phenomenon with parameter threshold", "fault cause", "confidence level" and "fault associated components". Import the extracted rules into the rule base according to the fields and assign them rule IDs.

[0114] (2) Construction of the case library

[0115] To achieve fault diagnosis based on historical experience, it is necessary to collect historical fault cases of port belt conveyors and store them in a standardized manner. The specific construction steps are as follows:

[0116] Step 1: Collect historical failure cases: Collect a number of historical failure cases through port operation and maintenance records, equipment manufacturer technical documents, simulation experiments, etc.

[0117] Step 2: Standardize the 8 attributes of each case: Standardize the fields of each case, including "Case ID", "Fault Area", "Fault Phenomenon", "Fault Parameters" (such as temperature, vibration, offset, stacking height, etc.), "Fault Cause", "Solution", "Loss Situation", and "Time of Occurrence".

[0118] For example, in the port operation and maintenance record case of the idler roller jamming, the fault area was the idler roller, the fault symptoms were high temperature and large vibration, the fault parameters were idler roller temperature 90℃ and vibration 0.8g, the fault cause was that ore debris entered the bearing and caused jamming, the solution was to clean the ore debris and replace the bearing, the loss was a 2-hour transportation delay, and the time of the incident was 2022-10-22.

[0119] For example, in the case of an equipment manufacturer, the fault was in the drive motor. The fault symptoms were a current of 100A and a temperature of 85℃. The fault parameters were a load of 120%. The cause of the fault was overload. The solution was to reduce the feed rate and check the heat dissipation. The loss was that the ship was delayed in port for 1 hour. The incident occurred on February 4, 2024.

[0120] The simulation experiment case is a coal pile failure case. The failure area is under the conveyor belt. The failure phenomenon is that the material pile height is 120mm and the conveyor belt speed is 2.0m / s. The failure parameters are that the material density is 2600kg / m³. The failure cause is that too much coal is piled up under the conveyor belt. The solution is to start the cleaning device and adjust the unloading speed. There is no loss. The occurrence time is 2023-06-15.

[0121] Step 3: Store the standardized fields in the case library and create a "case table". The fields correspond to the case attributes. Among them, "fault area", "fault phenomenon" and "fault parameters" are index case attributes, which support case retrieval based on the three fields. The rest are related attributes. The case data format is JSON, which is convenient for the intelligent inference layer to call.

[0122] The intelligent inference layer in this embodiment includes a detection module, an inference module, and a virtual verification module. The detection module deploys a fault detection model, which processes and detects image data and further integrates physical parameters and image data. The inference module deploys an intelligent inference model. Based on the detection results of the fault detection model, the intelligent inference model performs comprehensive inference based on a rule base and a case base, and outputs the inference results to the virtual verification module. The virtual verification module deploys a fault simulation model, which calls the corresponding fault simulation model based on the fault cause of the inference result and injects it into the digital twin model to simulate the fault, reproduce the actual fault, dynamically update the solution loaded by the module based on the inference result, and verify the effectiveness of the inference result.

[0123] (1) This embodiment takes four common types of faults as examples, including conveyor belt misalignment, conveyor belt tearing, idler jamming, and coal piling up; in order to achieve real-time fault detection based on images, the detection model is trained using an image dataset, and the dark channel prior algorithm is loaded to perform dehazing processing on the training images. The specific deployment steps of the fault detection model are as follows:

[0124] Step 1: Data Acquisition and Processing: Acquire original images of four types of faults, including several conveyor belt misalignment, several conveyor belt tears, several idler roller jams, and several coal pile-ups. Enhance the original images by rotating them 90° / 180° / 270° to generate 3 images each, flipping them horizontally / vertically to generate 2 images each, and adjusting the brightness by ±20% / contrast by ±15% to generate 2 images each. The enhanced images and the original images are combined to form the fault image set, which is then divided into a training set, a validation set, and a test set in a 6:2:2 ratio.

[0125] Use annotation software to annotate the fault areas of all the above images, and also annotate the fault category. The annotation format is VOC, and an annotation file corresponding to each image file is generated.

[0126] Step 2: Training the fault detection model: The fault detection model adopts transfer learning. Configure the training server, install the operating system, deep learning framework and image processing library, download the YOLOv5s pre-trained detection model, map the model output layer to the four types of faults, freeze the backbone network layer of the YOLOv5s pre-trained detection model, and retain its neck layer structure and head layer structure for fine-tuning adaptive learning.

[0127] M1. Set training parameters, including training batch, number of iterations and initial learning rate. Use cosine annealing learning scheduling strategy and CIoU Loss as the loss function with IoU threshold = 0.5.

[0128] M2. Set the training objective, start the training script, monitor the loss function curve in real time during training, control the training loss and validation loss, and save the weight file of the fault detection model after training. The training objective is: test set accuracy of not less than 96%, mAP=99.5% and detection speed FPS=57f / s.

[0129] M3, based on the physical parameters corresponding to the fault type and the abnormal physical parameters determined by the fused image detection, outputs fault information, including fault type, fault location, confidence level, and physical parameters, etc.

[0130] (2) As the best implementation of this embodiment, the intelligent reasoning model of this embodiment is based on the improved RBR+CBR reasoning mechanism, calls the corresponding rules of the rule base, and retrieves similar cases in the case base, and integrates rule reasoning and case reasoning to output a comprehensive reasoning result.

[0131] The intelligent inference model connects to the rule base through an interface, receives fault phenomena uploaded by edge computing nodes or detected and determined by the fault detection model, maps them to the corresponding rules in the rule base through the RBR inference mechanism, and outputs the fault cause and confidence level.

[0132] The intelligent inference model determines the inference algorithm and develops the inference program based on the CBR inference mechanism. The specific deployment steps are as follows:

[0133] S21. Determine the initial weights of index attributes: Assign initial weights to the attributes of each index case using the analytic hierarchy process (AHP), and construct a judgment matrix A sequentially using the fault area, fault phenomenon, and fault parameters as rows / columns respectively.

[0134] ;

[0135] Calculate the largest eigenvalue =3.003, Consistency Index CI=0.0015, Random Consistency Index RI=0.58, Consistency Ratio CR=CI / RI=0.0026≤0.1, the initial weights assigned to the above judgment matrix A are considered to have passed the consistency test. The initial weight values ​​for each index case attribute are then calculated: Fault Area =0.4931, Fault Phenomenon =0.1863, fault parameters =0.3206;

[0136] S22. Training the weighted KNN algorithm based on historical case data and determining the K value: Select 500 historical case data as training samples, and use cross-validation to divide the training and validation sets in an 8:2 ratio. Calculate the similarity between samples and determine the K value. The similarity calculation formula is as follows:

[0137] ;

[0138] Where X represents a historical case. Let Y be the value of the i-th attribute from the historical cases, and Y be the target case. For the i-th attribute value of the target case; For single-attribute similarity, a value of 1 is assigned when fault regions are the same and 0 when they are different. When fault parameters are ordered enumerations, the similarity is calculated as follows: calculate, The maximum parameter value for the current attribute. The minimum parameter value for the current attribute, such as the temperature parameter. , Target case temperature Temperature in a certain historical case Then the similarity of the fault parameters between the target case and the historical case is: ;

[0139] S23. The weighted KNN algorithm is trained with training samples, and the inference accuracy under different K values ​​is tested. It is determined that the algorithm has the highest diagnostic accuracy when K=10. The attribute weights are iteratively optimized. The weights are adjusted by 0.01 after 100 iterations, and the attribute weight combination with the highest accuracy is retained.

[0140] S24. After training, the algorithm is tested on the validation set to ensure a diagnostic accuracy of ≥90%, thus completing the training of the algorithm that meets the diagnostic requirements.

[0141] S25. Deploy the trained weighted KNN algorithm into the intelligent inference model.

[0142] (3) To ensure the accuracy of the fault reasoning results and the feasibility of the solutions, a fault simulation model is established, and the reasoning results are virtually verified using an online digital twin model. The deployment steps of the fault simulation model are as follows:

[0143] Step 1: Organize the parameter changes corresponding to typical fault phenomena of common fault types of belt conveyors to provide data support for the construction of fault simulation models; for example, when the idler is stuck, the idler speed drops to 0 and the bearing temperature rises to 90℃; when the conveyor belt runs off-center and the offset is ≥70mm, the idler is subjected to uneven force and the force on one side increases by 30%; when there is a coal pile-up fault, the material accumulation height under the conveyor belt is ≥100mm and the conveyor belt speed drops to below 2m / s.

[0144] Step 2: Based on the physical behavior model, inject fault tree data to construct a fault simulation model. Based on the correlation data between fault phenomena, fault types, and fault parameters, construct fault simulation models for four different fault types:

[0145] A model of idler roller jamming was constructed. A group of idler rollers was selected, and the rotational freedom of its rotating pair was locked at 0 r / min. At the same time, the friction coefficient of the idler roller bearing was increased from 0.02 to 0.1 to simulate the increased friction caused by idler roller jamming, which corresponds to the scenario of idler rollers being jammed by ore debris in port operations.

[0146] A conveyor belt misalignment fault model was constructed. The installation angle of the head redirecting roller was adjusted so that the angle between the roller centerline and the conveyor belt centerline increased from 0° to 2°. This simulated the conveyor belt shifting to one side. The location of the offset monitoring point was set to correspond to the misalignment of the port conveyor belt caused by uneven ore distribution.

[0147] An overload fault model of the drive motor was constructed. The driving torque was increased from 800 N·m to 1000 N·m to correspond to a 120% load. At the same time, the heat dissipation efficiency of the motor was reduced from 10 W / (m²·K) to 5 W / (m²·K) to simulate the overload temperature rise of the motor, corresponding to the working condition of increased ore transportation volume during the peak period of the port.

[0148] A coal pile failure model was constructed by adding material model density and pile height below the conveyor belt and setting the friction coefficient between the material and the conveyor belt to simulate the increased resistance of the conveyor belt due to material accumulation, corresponding to the scenario of material spillage during port unloading.

[0149] Step 3: Associate the physical parameter data or patterns of the fault. Associate the model simulation data after fault injection with the fault type and fault phenomenon, such as temperature change curves, offset curves when the conveyor belt runs off-track, etc., and add fault labels to the fault simulation model.

[0150] Step 4: Save the fault simulation model and deploy it to the virtual verification module.

[0151] The human-machine interaction layer includes a virtual-real linkage display module and an instruction control module. The virtual-real linkage display module can monitor the operating status through a digital twin model that is linked with the behavior of the belt conveyor, and receive and display the fault verification report from the intelligent inference layer. The instruction control module provides manual instruction input. The sending of manual instructions is based on whether the virtual verification result of the fault simulation model passes. If it passes, the manual instruction is sent to the PLC of the belt conveyor. The PLC controls the actuator to perform actions and eliminates the fault of the belt conveyor.

[0152] Based on the above intelligent inspection and diagnosis system, this embodiment provides a digital twin-based intelligent inspection and diagnosis method for belt conveyor faults, including the following steps and contents:

[0153] S1. Referring to the above-mentioned diagnostic system composition and deployment method, build a digital twin model of the current port belt conveyor, including a three-dimensional geometric model and a physical behavior model. Based on the digital twin model, construct a fault simulation model that can simulate the correlation between different fault types and physical parameters; build a rule base that can be mapped and a searchable case base for the intelligent reasoning model of the belt conveyor; train a fault detection model based on fault images and fuse physical parameters.

[0154] S2. Real-time acquisition of multi-source data from sensors, cameras, scanners, and equipment bodies in the physical layer, including key node images, inspection images, and physical parameter data of the belt conveyor. Edge computing nodes process and analyze the physical parameter data in real time. A preliminary fusion model deployed on the edge computing nodes analyzes and fuses the physical parameter data. The multi-source data fusion model performs noise reduction based on Kalman filtering and estimates the fault state. Based on DS inference theory, preliminary fault symptom inference and fusion of multi-source physical data are performed. Finally, the processed physical parameter information and its fusion results are output.

[0155] The intelligent inference layer receives image data via the RTSP protocol, with a real-time video stream frame rate of 25fps. The dark channel prior algorithm integrated into the fault detection model performs dehazing on each frame of the original image, and then inputs it into the trained fault detection model for visual fault recognition, identifying the fault type, fault location, and fault parameters. Finally, it fuses the physical parameters uploaded by the edge nodes to output the detection results, including fault type confidence, fault location, fault timestamp, fault parameters, and physical parameters.

[0156] The dehazing process performed on the image to be tested includes the following steps and methods:

[0157] P1. Calculate the dark channel value of the image to be tested using the following formula:

[0158] ;

[0159] in, It is based on pixels The local image neighborhood centered on the image For fog-free images Dark channel pixel values, For fog-free images The c-th color channel pixel value, Let r be any pixel in the neighborhood of the local image, and g and b be the three color channels, respectively.

[0160] P2. Estimate the atmospheric light intensity B, and select the pixel values ​​in the original fog map corresponding to the first 0.1% of bright pixels in the dark channel.

[0161] P3. Estimating transmittance t(x):

[0162] ;

[0163] in, This means retaining an adjustment factor for a small amount of fog; Original fog map The pixel value of the c-th color channel; Let be the atmospheric light intensity of the c-th color channel;

[0164] P4, Recover the dehazed image J(x):

[0165] ;

[0166] in, I(x) represents the lower limit of transmittance, and I(x) represents the original haze map.

[0167] The intelligent inference layer performs inference and fuses the inference results to output the results, which includes the following steps:

[0168] Step 1: Call the rule base. Based on the detection results of the fault detection model, match the corresponding rules according to the RBR inference mechanism, and output the fault cause and confidence level: For example, if the fault type of the detection result is deviation, the confidence level is 0.92, the idler vibration data is 0.6g, and the idler temperature data is 35℃; the fault causes and confidence levels are: improper roller installation 0.5, idler jamming 0.3, and uneven coal distribution 0.2.

[0169] Step 2: Retrieve the case library, using the detection result as the target case. Input its attributes: fault area = conveyor belt, fault phenomenon = belt misalignment, and fault parameters = offset 80mm, vibration 0.6g, temperature 35℃. Perform fault reasoning based on the CBR inference mechanism. Use the trained weighted KNN algorithm to retrieve K similar cases and calculate the average similarity of the target case. If the training determines K=10, then the trained weighted KNN algorithm will retrieve 10 similar fault causes, among which 6 are "improper roller installation". =0.9, the average similarity of the four "idler jamming" issues is =0.8;

[0170] Step 3: Integrate the results of rule-based reasoning and case-based reasoning to avoid misdiagnosis caused by rule conflicts or case biases. After integration, output the cause of the fault and the solution. If the results of rule-based reasoning and case-based reasoning are integrated, and both rule-based reasoning and case-based reasoning are assigned a weight of 0.5, then the overall confidence level of "improper roller installation" is 0.5×0.5+0.9×0.5=0.7, and the overall confidence level of "idler jamming" is 0.3×0.5+0.8×0.5=0.55. Output the cause of the fault "improper roller installation" and the solution "adjust the angle of the redirecting roller to 0°".

[0171] S3. Use a digital twin model to reproduce the fault phenomenon and its fault parameters (such as the deviation of 80mm in the deviation fault), call the fault simulation model to reproduce the physical behavior of the fault scenario, so that the simulated fault data is consistent with the actual fault data, load the solution into the digital twin model with fault operation for online simulation, monitor the simulation operation parameters of the belt conveyor, and if the parameters return to normal, it indicates that the reasoning result can effectively solve the fault, and output the reasoning result.

[0172] If the inference results fail to effectively resolve the fault, the rule base is updated based on the data generated from the verification. The data generated from the verification includes injected fault parameters, simulation running data, actual fault data, and verification result labels. The structured verification data is matched and classified with the rule base. When the verification data largely conforms to the existing rules, the rule confidence is updated. If there are differences between the verification data and the existing rules, rule correction is triggered. If the verification data differs significantly from the existing rules or there is no match, a new rule is generated. The cause of the fault is determined by manual fault diagnosis using the new rule, and the fault case is stored in the case base.

[0173] After updating the rule base and case base, return to step S2 to re-perform reasoning until the reasoning result is verified to be valid, or until the preset iteration threshold is reached to complete the reasoning. The iteration threshold includes the number threshold and the convergence threshold. When the adjacent reasoning result reaches the convergence threshold or the number of iterations reaches the number threshold, and the fault is still not effectively resolved, it is considered that the reasoning has fallen into a local optimum, the iteration is stopped, manual intervention is triggered, and the failed cases are recorded.

[0174] S4. After the reasoning results are validated, combined with the triggering instructions from the human-machine interface layer, a correction instruction is sent to the PLC according to the solution to avoid the triggering instructions from the human-machine interface layer directly causing secondary accidents and to ensure the safety of operation and maintenance.

[0175] Furthermore, the software product of this intelligent fault inspection and diagnosis method is stored in a storage medium, including several instructions to cause a computer device, such as including but not limited to a personal computer, server, or network device, to store a computer program. When the computer program is executed by a processor, it can execute the intelligent fault inspection and diagnosis method for belt conveyors based on digital twins according to any embodiment of this application.

[0176] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent inspection and diagnosis of faults in a belt conveyor based on digital twins, characterized in that, Includes the following steps and content: S1. Construct a digital twin model of the belt conveyor, wherein the digital twin model integrates the three-dimensional geometric model and physical behavior model of the belt conveyor; construct a fault simulation model based on the digital twin model that can simulate the correlation between different fault types and physical parameters; establish a rule base for subsequent reasoning. S2. Real-time acquisition and preprocessing of multi-source data, including image data and physical parameter data, loading of a trained fault detection model for visual fault recognition, fusing the physical parameter data, and outputting the detection results; The rule base is invoked, and based on the rule reasoning mechanism, the intelligent reasoning model performs fault reasoning on the detection results, and outputs the fault cause and its solution; S3. Inject fault parameters into the digital twin model, call the fault simulation model to reproduce the fault scenario, so that the simulated fault data is consistent with the actual fault data, load the solution for online simulation, and verify the effectiveness of the reasoning results. S4. If the solution effectively resolves the fault, the reasoning result is output, and a correction command is issued in conjunction with manual triggering. If the reasoning result fails to effectively resolve the fault, the rule base is updated based on the data generated by the verification, and the process returns to step S2 to perform reasoning again until the reasoning result is verified to be valid or a preset iteration threshold is reached.

2. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 1, characterized in that: The intelligent reasoning model performs comprehensive reasoning based on preset rules and historical fault cases. Step S1 also includes building a case library. In step S2, the intelligent reasoning model combines the rule reasoning mechanism and the case reasoning mechanism to perform fault reasoning. The case reasoning mechanism retrieves similar cases by calling the case library and using the trained weighted KNN algorithm, calculates their similarity, and outputs similar sample case data. The intelligent reasoning model performs a weighted fusion of the confidence level of the fault cause output by the rule-based reasoning mechanism and the similarity level of the fault cause of similar cases output by the case-based reasoning mechanism, and outputs a comprehensive reasoning result.

3. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 2, characterized in that: The training method for the weighted KNN algorithm is as follows: S21. Construct a judgment matrix that satisfies the consistency test, and determine the initial weights of the index attributes of each case; S22. Select data from several historical failure cases as training samples, divide the training samples into a failure training set and a failure verification set using cross-validation, calculate the sample similarity of the failure training set and determine the K value. S23. Iteratively optimize the weights of the case index attributes, retaining the weight combination with the highest accuracy. S24. Verify the weighted KNN algorithm and the weight combination using the fault verification set to ensure an accuracy of ≥90%; S25. Deploy the trained weighted KNN algorithm into the intelligent inference model.

4. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 2, characterized in that: The fault detection model is based on a pre-trained detection model and undergoes transfer learning. Fault images corresponding to various fault types are collected, the backbone network layers of the pre-trained detection model are frozen, retaining its neck and head layer structures, and adaptive adjustments are made. The fault detection model is then trained and deployed. M1. Perform image enhancement processing on the acquired fault images to obtain a fault image set, divide the fault image set into a training set, a validation set, and a test set, and label the fault regions and fault types in the fault images; M2. Set the training parameters, start the training script, train the fault detection model and save it; M3. The fault detection model integrates the image detection results and the physical parameter data to output the detection results.

5. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 4, characterized in that: The fault detection model integrates a dark channel prior algorithm, which is used to dehaze the fault image used for training and the test image to be detected in real time.

6. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 2, characterized in that: Several physical parameter data samples of the belt conveyor under normal operating conditions were collected during on-site testing, and divided into a physical test set and a physical verification set. The method for constructing the physical behavior model is as follows: S11. Import the three-dimensional geometric model and set the material parameters; S12, Add constraints and driving torque; S13. The physical behavior model is based on the following multibody dynamics equations: ； in, The inertia matrix, For generalized coordinates, For generalized speed, For generalized acceleration, Let the Coriolis force and centrifugal force be vectors. The gravity term vector, For driving torque; S14. Identify the parameters to be identified in the physical test set and determine the identification parameters; S15. Integrate the identification parameters to build the physical behavior model, input the physical verification set to verify the accuracy of the physical behavior model, adjust the physical behavior model and save the final model.

7. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 6, characterized in that: The digital twin model is dynamically updated based on a hierarchical update strategy, ensuring real-time synchronization with the actual operating status of the belt conveyor. The update of the three-dimensional geometric model is triggered by real-time scanning at a preset first frequency to obtain entity point cloud data, calculate the geometric deviation between the point cloud data and the current three-dimensional geometric model, and if the deviation exceeds a preset value, a local update of the three-dimensional geometric model is performed based on an incremental algorithm. The update of the physical parameters is triggered by real-time acquisition of the physical parameter data at a preset second frequency, inputting it into the physical behavior model, and adjusting the corresponding physical quantities in the physical behavior model based on a parameter correction algorithm.

8. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 2, characterized in that: The preprocessing in step S2 is based on a multi-source data fusion model. The multi-source data fusion model performs noise reduction and fault state estimation of the physical parameter data based on Kalman filtering, and performs preliminary fault symptom inference and fusion of the physical parameter data based on DS inference theory, outputting the processed physical parameter information and its fusion result.

9. The intelligent inspection and diagnosis method for belt conveyor faults based on digital twins according to claim 1, characterized in that: The method for constructing the fault simulation model: Step 1: Organize the parameter changes corresponding to the typical fault phenomena of the belt conveyor, and establish a fault tree to establish the correlation between the fault phenomena, the fault types and the fault parameters. Step 2: Inject the fault tree data into the physical behavior model to construct the fault simulation model for each of the fault types; Step 3: Associate the simulation data of the fault simulation model with the data and patterns of the fault parameters, and add fault labels to the fault simulation model.

10. A digital twin-based intelligent inspection and diagnosis system for belt conveyor faults, used to implement the digital twin-based intelligent inspection and diagnosis method for belt conveyor faults as described in claim 8, characterized in that: It includes a physical layer, a data acquisition layer, a digital twin modeling layer, an intelligent reasoning layer, a database, and a human-computer interaction layer; among which, The physical layer includes the belt conveyor, the inspection robot, and the sensor group, which includes vibration sensors, temperature sensors, and image sensors. The data acquisition layer is based on edge computing nodes, and the multi-source data fusion model is deployed on the edge computing nodes for preliminary fusion of the physical parameter data acquired by the physical layer. The digital twin modeling layer is used to construct the digital twin model of the belt conveyor, including the three-dimensional geometric model, the physical behavior model, and a dynamic update module. The dynamic update module updates the three-dimensional geometric model and the physical behavior model at a fixed frequency according to the physical parameter data of the physical layer. The database includes the rule base and the case base. The rule base consists of production rules built based on intermediate nodes of the fault tree, and the case base stores the historical fault cases that have been standardized according to case attributes. The intelligent inference layer includes a detection module, an inference module, and a virtual verification module. The detection module deploys the fault detection model, which processes and detects image data, receives physical parameter data uploaded by the edge computing node, and further fuses the image data and physical parameter data to output the detection result. The inference module deploys the intelligent inference model, which performs comprehensive inference based on the rule base and the case base, and outputs the comprehensive inference result to the virtual verification module. The virtual verification module deploys the fault simulation model, which calls the corresponding fault simulation model based on the comprehensive inference result and injects it into the digital twin model for fault simulation to reproduce actual faults. The dynamic update module loads the solution from the comprehensive inference result and verifies its effectiveness. The human-machine interaction layer includes a virtual-real linkage display module and an instruction control module. The virtual-real linkage display module is used to monitor the operating status of the belt conveyor and display the fault verification report of the intelligent inference layer. The instruction control module provides human instruction input and, based on the triggering of the human instruction, sends the solution verified through simulation to the control system of the belt conveyor.

Images