Belt conveyor operation monitoring method and system based on big data

Abstract

The invention discloses a belt conveyor operation monitoring method and system based on big data, and relates to the technical field of intelligent operation and maintenance of industrial equipment, and the method comprises the steps: collecting the operation data of a belt conveyor through deploying multiple sensors, extracting image features through employing a YOLOv7 model, and carrying out the multi-modal feature fusion based on the image features; based on the fused multi-modal features, obtaining a low-dimensional feature vector by using an auto-encoder, constructing a three-dimensional digital twinborn model, updating the three-dimensional digital twinborn model by using a Kalman filtering method, generating a final state vector, and obtaining a health index of the belt conveyor based on the final state vector; the wear depth is calculated, the belt health index, the wear depth and the final state vector serve as input of an LSTM model, the remaining use time of the belt is output, and the fault occurrence probability is obtained through a causal Bayesian network inference method. According to the method, the state sensing precision and the operation and maintenance decision reliability of the belt conveyor under the complex working condition are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for industrial equipment, and in particular to a method and system for monitoring the operation of belt conveyors based on big data. Background Technology

[0002] Large conveying equipment, such as belt conveyors, is increasingly automated and intelligent in critical infrastructures like mines, ports, and power plants. As core equipment for continuous material transport, belt conveyors operate under harsh conditions such as high loads, strong dust, and complex temperature and humidity, making them prone to failures like belt tearing, drum wear, and tension imbalance. This can lead to unplanned downtime or even safety accidents. Traditional monitoring methods mainly rely on single sensors (such as vibration or temperature) for threshold alarms, lacking the ability to collaboratively perceive and deeply analyze the multi-dimensional state of equipment, making it difficult to achieve early fault identification and remaining life prediction. The development of technologies such as multi-sensor fusion, digital twins, deep learning, and edge computing has provided new technical paths for equipment condition monitoring. Although existing technologies have made some progress in equipment monitoring, significant shortcomings remain. Most methods rely on single-modal information, neglecting the state correlation under multi-physics coupling, leading to… Fault characterization is incomplete; traditional feature fusion strategies often employ simple splicing or weighted averaging, lacking a modeling mechanism for dynamic adjustment of semantic associations and confidence levels between modalities, thus affecting fusion accuracy; existing digital twin models are mostly static geometric mappings, lacking the ability to dynamically update parameters and correct states based on real-time observation data, making it difficult to reflect the nonlinear evolution characteristics of equipment degradation processes; remaining life prediction is mostly based on empirical models or shallow networks, failing to fully integrate health status, wear depth, and multi-dimensional operating parameters, resulting in insufficient robustness of prediction results; at the level of fault attribution and decision support, existing methods mostly remain at the level of correlation analysis, lacking causal reasoning mechanisms to distinguish between interfering factors and root causes, and maintenance strategy generation largely relies on human experience, lacking intelligent decision support based on multi-objective optimization; monitoring data is susceptible to communication interference or malicious tampering, making data credibility difficult to guarantee, thus restricting the application of the system in scenarios with high safety requirements. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, this invention provides a belt conveyor operation monitoring method and system based on big data, which solves the problems of low early warning accuracy, poor diagnostic reliability, and delayed maintenance response of belt conveyors under complex working conditions.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a belt conveyor operation monitoring method based on big data, which includes: collecting belt conveyor operation data by deploying multiple sensors, extracting image features using the YOLOv7 model, and performing multimodal feature fusion based on the image features; Based on the fused multimodal features, a low-dimensional feature vector is obtained using an autoencoder, a three-dimensional digital twin model is constructed, the three-dimensional digital twin model is updated using the Kalman filter method, and a final state vector is generated. The health index of the belt conveyor is obtained based on the final state vector. The wear depth is calculated, and the belt health index, wear depth and final state vector are used as input to the LSTM model. The remaining service time of the belt is output. The probability of failure is obtained by using the causal Bayesian network inference method. The most probable cause of failure is obtained based on the probability of failure. The risk level of the belt conveyor is obtained by combining the most probable cause of failure with the health index. The risk level is then used to perform multi-source information fusion using a fuzzy logic controller to generate an early warning level. A cached dataset is generated based on the warning level. The integrity of the dataset is verified by blockchain. The root cause of the anomaly is identified by combining the remaining service time of the belt. Based on the root cause of the anomaly and the blockchain-stored evidence data, the best maintenance strategy is selected.

[0006] As a preferred embodiment of the belt conveyor operation monitoring method based on big data described in this invention, the steps of collecting belt conveyor operation data by deploying multiple sensors, extracting image features using the YOLOv7 model, and performing multimodal feature fusion based on the image features refer to deploying multiple source sensors and collecting data, and obtaining a multi-source data stream based on the collected data. The acquired laser-assisted imaging images are input into the YOLOv7 model for inference, outputting the category, bounding box, and confidence score for each detected defect. Based on the CNN backbone network, features are extracted layer by layer from the laser-assisted imaging images to generate deep feature maps. The model accuracy is calculated based on these features. and recall rate The harmonic mean of precision and recall was obtained. Calculate the credibility weight Before multimodal fusion, the image features are weighted to obtain weighted image features. Based on image features Perform a linear transformation to obtain the fusion features .

[0007] As a preferred embodiment of the belt conveyor operation monitoring method based on big data described in this invention, the method involves: using fused multimodal features to obtain low-dimensional feature vectors via an autoencoder, constructing a three-dimensional digital twin model, updating the three-dimensional digital twin model using a Kalman filter, generating a final state vector, obtaining the belt conveyor health index based on the final state vector, and calculating the total loss function using the autoencoder. Based on the total loss function Once the trained autoencoder is obtained, the features will be fused. Input encoder, output low-dimensional feature vector Calculate tension A 3D digital twin model is constructed in Unity based on the geometric parameters of the belt conveyor, generating an initial state vector. ; The initial state vector As input to the Unity model, a 3D digital twin model is constructed, and the Kalman filter method is used to obtain the updated state vector. The updated state vector As input parameters to the 3D digital twin model, the updated state vector is mapped to the physical properties of the model through the 3D digital twin model. Physical quantities are extracted in real time from the simulation results as high-order features, and then fused with the updated state vector and the low-dimensional image feature vector z output by the autoencoder to generate the final state vector. Based on the state vector at the current moment The health index (PHI) of the belt conveyor was obtained.

[0008] As a preferred embodiment of the belt conveyor operation monitoring method based on big data described in this invention, the following steps are taken: The wear depth is calculated by using the health index, wear depth, and final state vector as inputs to an LSTM model, outputting the remaining service time of the belt, obtaining the probability of failure using a causal Bayesian network inference method, determining the most probable cause of failure based on the failure probability, fitting the wear model to obtain the current degradation rate using the defect detection results output by the YOLOv7 model, and finally obtaining the belt wear depth. , the state vector Health Index (PHI) and Wear Depth As input to the LSTM model, predict future states. Calculate the failure time The remaining service life is calculated based on the failure time. The probability of failure occurrence is calculated using the causal Bayesian network inference method. By using the chain rule, each path C is decomposed into the product of the conditional probabilities of each node. Calculate conditional probability ; For the fault node Given path C and observational evidence Under the given conditions, the probability of failure occurrence is calculated using a causal Bayesian network recommendation method. Identify the probability of failure occurrence based on observational evidence through causal intervention. .

[0009] As a preferred embodiment of the belt conveyor operation monitoring method based on big data described in this invention, the risk level of the belt conveyor is obtained by combining the most probable cause of failure with a health index. Using this risk level, a fuzzy logic controller is employed to generate an early warning level, dividing the risk into probability of occurrence and severity of consequences. The probability and severity are combined through a Cartesian product mapping, and a lookup table rule is set to output the risk level R, which is divided into four risk levels. The rate of health deterioration is then calculated. Based on risk level R and rate of health deterioration This is the input to the fuzzy logic controller, and the output is warning level B.

[0010] As a preferred embodiment of the belt conveyor operation monitoring method based on big data described in this invention, the step of generating a cached dataset based on the warning level, verifying the integrity of the dataset through blockchain, and identifying the root cause of the anomaly in conjunction with the remaining belt time refers to generating a cached dataset based on warning level B. Perform a hash operation on the cached dataset to obtain the hash value. , hash value Upon on-chain verification, it is confirmed that the hash has been uploaded and has not been tampered with. If the verification fails, a data anomaly alarm is triggered, subsequent processes are paused, and data integrity is checked by tracing back the sensor or communication link. If the verification passes, it is determined whether the high-risk intervention condition is met, i.e., when the remaining usage time RUL is less than a preset threshold. And when the warning level B is higher than or equal to high, proceed... If the problem is not identified, it is moved to the low-frequency monitoring queue. Causal intervention analysis is then employed to define the dominant causal path and analyze each candidate parent node within that path. Perform virtual intervention operations on nodes in the causal graph. Perform intervention actions and define intervention impact scores. .

[0011] As a preferred embodiment of the belt conveyor operation monitoring method based on big data described in this invention, the step of selecting the best maintenance strategy based on the root cause of the anomaly and blockchain-stored evidence data refers to normalizing all feasible strategies based on the confirmed root cause of the anomaly and the verification data stored on the blockchain, constructing a three-dimensional target space, solving the Pareto optimal front, selecting the non-dominated solution set, selecting the strategy with the smallest Euclidean distance to the weighted ideal point as the recommended decision, loading a three-dimensional digital twin scene, coloring the equipment according to risk level, and generating a health trend curve, an early warning heat map, and a maintenance path trajectory.

[0012] Secondly, the present invention provides a belt conveyor operation monitoring system based on big data, comprising, The multimodal feature fusion and digital twin modeling module is used to generate low-dimensional feature vectors and combine them with a three-dimensional digital twin model to dynamically update the state using Kalman filtering to obtain the belt conveyor health index. The life prediction and fault inference module is used to calculate the wear depth and combine it with the belt conveyor health index to predict the remaining service life using an LSTM network. The risk assessment and intelligent early warning module is used to construct a two-dimensional risk matrix to determine the risk level, and uses a fuzzy logic controller to process the risk level and the rate of health deterioration to output graded early warning signals. The blockchain evidence storage and anomaly tracing module is used to ensure data integrity by putting datasets on the blockchain. When under high-risk conditions, it identifies the root cause of failure through causal intervention analysis. The maintenance decision generation and visualization monitoring module is used to match the optimal maintenance strategy based on the confirmed root cause and perform Pareto optimal solution.

[0013] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the belt conveyor operation monitoring method based on big data as described in the first aspect of the present invention.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the belt conveyor operation monitoring method based on big data as described in the first aspect of the present invention.

[0015] The beneficial effects of this invention are as follows: By introducing the YOLOv7 model and combining it with laser-assisted imaging, high-precision image feature extraction is achieved. A confidence-weighted multimodal feature fusion mechanism is designed to overcome the problems of missing information in a single modality and noise interference. Furthermore, a quantum variational autoencoder is used to perform nonlinear dimensionality reduction on the fused features. Combined with Kalman filtering to dynamically update the three-dimensional digital twin model, closed-loop feedback and state correction between the physical entity and the virtual model are realized, solving the evolutionary lag problem caused by traditional static modeling of digital twins. An LSTM lifetime prediction model is constructed by fusing health index, wear depth, and dynamic state vectors. A causal Bayesian network is innovatively introduced for intervention analysis, leaping from statistical correlation to causal interpretability, accurately identifying the root cause of failure. Finally, through the collaborative mechanism of fuzzy logic and blockchain, risk warning, data storage, and Pareto optimal maintenance strategy generation are achieved, which not only enhances the credibility and security of the system, but also realizes a technological leap from passive response to active prediction and autonomous decision-making. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a belt conveyor operation monitoring method based on big data in Example 1.

[0018] Figure 2 This is a schematic diagram of a belt conveyor operation monitoring system based on big data, as shown in Example 1.

[0019] Figure 3 This is a diagram showing the data flow and module interaction in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figures 1 to 3 This is the first embodiment of the present invention, which provides a method for monitoring the operation of a belt conveyor based on big data, including the following steps: S1. Collect belt conveyor operation data by deploying multiple sensors, extract image features using the YOLOv7 model, and perform multimodal feature fusion based on the image features; Based on the fused multimodal features, a low-dimensional feature vector is obtained using an autoencoder, a three-dimensional digital twin model is constructed, the three-dimensional digital twin model is updated using the Kalman filter method, and a final state vector is generated. The health index of the belt conveyor is obtained based on the final state vector. Specifically, multi-sensor data acquisition of belt conveyor operation is performed, and image features are extracted using the YOLOv7 model. Multi-modal feature fusion is then conducted based on these image features. This involves deploying vibration sensors, temperature sensors, a Hikvision camera, velocity sensors, and acceleration sensors to collect vibration intensity v, temperature θ, rotational speed ω, and belt vibration acceleration data, respectively. ; The image resolution is designed based on the belt width (e.g., 900mm) and height (e.g., 600mm) (e.g., 2×900=1800 (horizontal pixels, based on belt width 900mm), 2×600=1200 (vertical pixels, based on belt height 600mm), resulting in a resolution of 1800×1200 (total 2,160,000 pixels)). The collected data is transmitted to the AWS platform via the MQTT protocol for preprocessing operations such as noise reduction, format standardization, and time synchronization, outputting a multi-source data stream. (m=1,2,3,4,5 correspond to vibration, temperature, image, rotational speed and vibration acceleration respectively). The Hikvision camera acquires laser-assisted imaging images of the belt surface, which are then cropped and scaled to obtain a 640×640×3 resolution image of the laser-assisted imaging. Laser-assisted imaging images are input into a YOLOv7 model deployed on a Jetson TX2 for inference. The model outputs the category (e.g., "tear", "scratch"), bounding box (center coordinates, width and height), and confidence score ([0,1]) for each detected defect. Overlapping boxes are removed using non-maximum suppression (NMS). Specifically, a single-stage object detection architecture based on the PyTorch framework is adopted, including a backbone network (e.g., EfficientNet or CSPDarknet), a Feature Pyramid Network (FPN) as the neck structure, and an anchor-free detection head. Classification and localization accuracy are jointly optimized using cross-entropy loss and GIoU loss. During training, a dataset of 5000 mine conveyor belt images is used, divided into training and validation sets at an 80%:20% ratio. Data augmentation techniques (e.g., random cropping, flipping, color dithering) are combined to improve generalization ability. Iterative training is performed using the Adam optimizer and a learning rate decay strategy to obtain a fine-tuned model. The laser-assisted imaging images are scaled to 640×640 and input into the model. Inference is performed on TX2, outputting the category (e.g., "tear", "scratch"), bounding box (center coordinates, width and height) and confidence score ([0,1]) for each detected defect, and removing overlapping boxes through non-maximum suppression (NMS); The bounding boxes output by the model are matched with the manually labeled ground truth labels using the Intersection over Union (IoU). If the IoU is greater than a set threshold (e.g., 0.5) and the categories are consistent, it is determined as a true positive (TP), which is a correctly detected defect. If the predicted box does not match any ground truth label, it is a false positive (FP), which is a falsely detected non-defect. A true defect that is not matched by any predicted box is a false negative (FN), which is a missed defect. Based on a CNN backbone network, features are extracted layer by layer from laser-assisted imaging images to generate deep feature maps. Specifically, through convolutional layers, pooling layers, and activation functions, feature maps of different scales and abstract levels are extracted layer by layer, capturing changes from low-level visual features (such as edges and colors) to high-level semantic information (such as the parts and overall structure of objects). Deep within the network, such as before global average pooling, one or more deep feature maps containing rich semantic information are obtained. Feature vectors are extracted through specific layers, flattened, and compressed into vectors of a specified dimension (e.g., 512-dimensional) through fully connected layers, denoted as image features. ); Accuracy of the calculation model Recall rate The harmonic mean of precision and recall The formula is: , in, To output the proportion of correctly detected defects out of all detection results, and to measure the degree of false alarms, To output the proportion of correctly detected defects out of the total actual defects, and to measure the degree of missed detection, The harmonic mean of precision and recall is used to comprehensively measure detection performance, and H is the coefficient of the harmonic mean. Calculate credibility weight The formula is: , Where A is the upper limit of the weight (i.e., the maximum is 1, without amplification). This is the baseline score of the model on the offline validation set, representing a normal performance level; Before multimodal fusion, the image features are weighted to obtain weighted image features. The formula is: , Based on image features Perform a linear transformation to obtain the fusion features The formula is: , in, Layer normalization, used to stabilize training and accelerate convergence, normalizes the feature vectors. The output of the cross-attention mechanism represents the correlation features between different modalities (such as images and vibrations).

[0024] By deploying multiple sensors (such as vibration, temperature, speed, and acceleration sensors) and a Hikvision laser-assisted imaging camera, comprehensive multi-source data on the operation of the belt conveyor is collected. High-resolution images are used to capture details of the belt surface, and the YOLOv7 model is used to extract deep features and detect targets in the images, achieving accurate identification of belt defects (such as tears and scratches). Non-maximum suppression technology is combined to remove overlapping bounding boxes, improving detection accuracy. By evaluating the model output with manually labeled results using IoU matching, precision, recall, and their harmonic mean are calculated, thereby quantifying model performance. A credibility weight adjustment mechanism is introduced to further ensure the effectiveness of features, and multimodal feature fusion is performed based on this, improving the robustness and reliability of the overall system. This provides strong technical support for real-time monitoring and maintenance of belt conveyors, significantly reducing the risk of missed detections and false alarms.

[0025] Furthermore, based on the fused multimodal features, a low-dimensional feature vector is obtained using an autoencoder to construct a three-dimensional digital twin model. The three-dimensional digital twin model is then updated using a Kalman filter method to generate a final state vector. Based on this final state vector, the belt conveyor health index is obtained, and the total loss function is calculated using a QVAE (Quick Visual Encoder). The formula is: , in, KL divergence measures the distribution of a quantum encoder. With prior distribution The difference between (standard normal distributions) To measure the reconstruction error, the L2 norm is used to measure the fused features of the original input. (Right now ) and decoder reconstruction features differences The encoder distribution generated by parameterized quantum circuits, The prior distribution (standard normal distribution) (0,1)), where U is the expected value, which is approximated by Monte Carlo sampling; The variational parameters (such as the rotating gate angle) in the quantum circuit are adjusted using gradient descent (e.g., the Adam optimizer), and the process is iterated repeatedly to minimize the variational parameters. Once the loss function converges, the encoder and decoder parameters are fixed, and the trained autoencoder model is obtained. Features are then fused. The input is an encoder, and the output is a compressed low-dimensional feature vector. ; Calculate tension (The tension of the belt during the operation of the belt conveyor), the formula is: , in, N represents the vibration acceleration collected in real time by the accelerometer, m represents the normal force data, and μ represents the load mass and the coefficient of friction. Based on the geometric parameters of the belt conveyor (width 900mm, length 100m), a 3D digital twin model (including components such as belt, rollers, and motor) was constructed in Unity (simulation software). This model was then used to construct the digital twin model using tension. The initial state vector is constructed by combining rotational speed ω, vibration intensity v, and temperature θ. ; The initial state vector As input to the Unity model, through the physics parameter interface in the Unity engine, each item is mapped to the corresponding physical component of the 3D digital twin model (such as the ADT model): This includes assigning tension M to the tension property of the belt, driving the corresponding deformation and stress distribution in the mechanical simulation; setting the rotational speed ω as the angular velocity parameter of the motor or drive roller to control the belt running rhythm; applying vibration intensity v as an external excitation input to the support structure or roller to trigger dynamic response and generate vibration waveform; inputting temperature θ to the material thermodynamics module to adjust the thermal expansion coefficient and elastic modulus of the belt and metal parts to simulate thermal deformation behavior; and injecting the low-dimensional feature vector z output by the autoencoder as a hidden state variable into the model to adjust the nonlinear degradation function (such as wear rate and fatigue accumulation) to reflect the damage evolution trend revealed by the image features. For complex rendering tasks with high dust, the Jetson TX2 at the edge only sends rendering instructions and key parameters to the Unity instance in the AWS cloud, and the cloud completes the high-load graphics calculation and sends the image stream back to the local machine to achieve lightweight operation and build a 3D digital twin model. The updated state vector is obtained using the Kalman filter method. The formula is: , in, The Kalman gain matrix at time t is calculated adaptively using covariance to balance the observation and prediction weights. The observation matrix maps the state space to the observation space. Update the state vector As input parameters for the 3D digital twin model, the model is driven in real time via the Unity engine's API interface. The updated state vector is mapped to the model's physical properties—for example, tension M is set as the belt's pulling force parameter, rotational speed ω is assigned to the angular velocity of the driving roller, vibration intensity v is applied as an external excitation to the supporting structure, and temperature θ adjusts the material's thermodynamic properties. The digital twin model runs a physics simulation engine based on these parameters, dynamically calculating the equivalent stress σ, contact wear rate w, and fatigue cumulative damage in each region of the belt. The physical quantities are extracted in real time from the simulation results as high-order features, and then fused with the updated state vector [M,ω,v,θ] and the low-dimensional image feature vector z output by QVAE to generate a final state vector containing both physical and data-driven information. Obtain the state vector at the current moment. Subtract their respective historical minimum values and divided by the historical maximum value Compared with historical minimum The difference between the values ​​is normalized and then multiplied by the weights calculated by PCA. The weighted terms are then summed to obtain the health index PHI of the belt conveyor, which is within the range of [0,1], thus realizing a fusion quantitative assessment of the equipment's health status.

[0026] By fusing multimodal sensor data and deep learning features, and using an autoencoder to reduce and compress high-dimensional fused features, low-dimensional latent features characterizing equipment status are extracted. An initial state vector is then constructed using physical parameters to drive a Unity-based 3D digital twin model for high-fidelity simulation. Kalman filtering is introduced to dynamically optimize the state, improving estimation accuracy. The updated state is fed back to the digital twin model, achieving not only visualization but also physical computability. A simulation engine generates high-order physical features such as stress, wear rate, and fatigue damage in real time. These simulation features are deeply integrated with original sensor data and image features to form a final state vector that comprehensively reflects the equipment's operating status. The health index PHI is then quantified using normalization and PCA weighting methods. This approach overcomes the limitations of traditional monitoring systems that rely solely on a single data source or static threshold judgment, achieving health assessment through multi-source heterogeneous data fusion, dynamic state estimation, and physical mechanism synergy. This improves the accuracy of conveyor belt operating status perception and early fault identification capabilities.

[0027] S2. Calculate the wear depth. Take the belt health index, wear depth and final state vector as input to the LSTM model, output the remaining service time of the belt, use the causal Bayesian network inference method to obtain the failure probability, and obtain the most probable failure cause based on the failure probability. The risk level of the belt conveyor is obtained based on the most probable cause of failure, and a fuzzy logic controller is used to generate an early warning level based on the belt conveyor health index and the risk level. Specifically, the wear depth is calculated by using the health index, wear depth, and final state vector as inputs to an LSTM model. The remaining service time of the belt is output. The probability of failure is obtained using a causal Bayesian network inference method. Based on the failure probability, the most probable cause of failure is determined. The defect detection results output by the YOLOv7 model are used to fit the wear model to obtain the current degradation rate, thus yielding the belt wear depth. The formula is: , in, This represents the initial wear depth. and The fitting coefficients (obtained from multi-source data streams) The analysis, based on multi-dimensional time-series data and actual degradation status, demonstrates the support of multi-source information for wear trend modeling. The state vector Health Index (PHI) and Wear Depth As input to the LSTM model, predict future states. By statistically analyzing historical fault data, a threshold A is set, and when the wear depth is predicted... When the threshold A is exceeded, it is determined that the device will fail at some point in the future. A linear interpolation method is then used to calculate the failure rate from the current time. Starting with, we work backwards along the predicted wear trend curve to... The time when A equals the failure time is the time of failure. The remaining service life is calculated based on the failure time. The formula is: , , in, Predict the state of the next moment (such as wear and temperature trends). For LSTM hidden states (memory cells) The current time; Using the Causal Bayesian Network (CBN) inference method, calculate the conditional probability of the current node (e.g., "belt breakage") occurring (the probability of a node occurring given a parent node (cause), i.e., the probability of failure occurring). The formula reflects the statistical dependency among various failure factors: , in, For the current node (e.g., "belt breakage"), for The parent node (e.g., "high vibration", "high temperature") Statistical frequency of historical events; Using the chain rule, each path C (e.g., "overload → increased tension → roller wear") is decomposed into the product of the conditional probabilities of each node. The formula is: , In intervention operations (This indicates an intervention (such as manually adjusting the speed or stopping the load), used in causal reasoning to distinguish between correlation and causation.) Given a sequence of observational evidence from time 1 to the current time t (such as sensor data (vibration, temperature), defect labels), the influence of parent nodes (referring to the antecedent variables in the causal graph that directly cause the current node, such as "excessive load" being the parent node of "increased tension") is corrected using a causal graph pruning method. The conditional probability of the path after the correction intervention is calculated. ; Based on historical maintenance data (such as maintenance records and downtime events) and equipment operation logs, and combined with Failure Mode and Effects Analysis (FMEA) methods, typical fault types of belt conveyors are identified, such as "belt tear," "drum wear," and "motor overheating," as candidate fault nodes. These candidate nodes are then labeled and verified using field fault cases and real-time alarm information (such as vibration exceeding limits and temperature alarms) to confirm their validity under the current operating conditions. The labeled and verified fault events are then defined as fault nodes. ; For the fault node Given path C and observational evidence Under conditions such as vibration or excessive temperature, the probability of a fault occurring is calculated using a causal Bayesian network recommendation method. The formula is: , in, This is a normalization factor (evidence probability) to ensure the total probability is 1. The parent node, which directly affects the fault in the cause-effect graph. Preceding factors (such as "high vibration" or "high temperature"); Identify the most likely cause of the failure through causal intervention (do operator) and quantify the probability of occurrence. The formula is as follows: , in, The probability of a failure occurring under observed evidence (such as vibration, temperature).

[0028] By integrating YOLOv7 detection results with multi-source time-series data, a dynamic wear model is established to accurately quantify the wear depth of the belt. The health index, wear depth, and final state vector containing both physical and data-driven information are jointly input into an LSTM network to effectively capture equipment degradation trends. Based on the prediction curve and failure threshold, the remaining service life is inferred, improving the accuracy and timeliness of RUL prediction. A causal Bayesian network (CBN) is constructed, and fault nodes are defined by combining historical operation and maintenance data with FMEA analysis. Using Bayesian inference and causal intervention methods, correlation and causality are distinguished under complex operating conditions. The most probable cause of failure is identified from multiple possible paths, achieving a leap from "phenomenon warning" to "root cause localization". This overcomes the shortcomings of traditional diagnosis, which relies on experience thresholds, is difficult to predict service life, and cannot trace the root cause. It significantly enhances the intelligence and interpretability of the belt conveyor fault prediction and health management system.

[0029] Furthermore, the risk level of the conveyor belt is obtained based on the most probable cause of failure. A fuzzy logic controller is then used to generate an early warning level based on the conveyor belt's health index and risk level. This level divides the risk into probability of occurrence and severity of consequences, with the probability of occurrence based on the probability of failure. Mapped to qualitative levels: for example, when A value >0.8 is considered "extremely high," and 0.6 to 0.8 is considered "high." Severity is based on the PHI (Health Index) and the real-time status of key equipment components (such as rollers and motors). This is achieved by monitoring key operating parameters of critical components (such as temperature, vibration, and wear) and comparing them with the normal operating range. If a value exceeds the normal range, it indicates a potential problem that will affect the severity of the consequences of a failure. The probability and severity are combined using a Cartesian product mapping, and a lookup table rule is set to output the risk level R, which is divided into four risk levels (low, medium, high, and extremely high). A fuzzy logic controller (FLC) is used for multi-source information fusion. Specifically, the health deterioration rate is calculated. The formula is: , Based on risk level R and rate of health deterioration The input variables reflect the current risk level and the trend of state degradation. The input variables are divided into three fuzzy sets: "low", "medium", and "high" using a triangular membership function. Each set is represented by a triangle, with the vertex corresponding to the typical value with a membership degree of 1 and the two bottom corners corresponding to the boundary values ​​with a membership degree of 0. The triangles of adjacent sets partially overlap to achieve a smooth transition. The "low risk" triangle covers the lower interval of the input range, the "medium risk" triangle is in the middle, and the "high risk" triangle covers the higher interval. A sigmoid-type output membership function is used to map the result to the [0,1] interval. Combined with the "if...then..." fuzzy rule (e.g., high risk and rapid deterioration correspond to a red warning), the centroid method is used to defuzzify the data and output the warning level B.

[0030] By integrating the probability of failure with health indices and the status of key components, a two-dimensional risk matrix is ​​constructed to achieve multi-dimensional quantitative assessment of belt conveyor risks. An innovative fuzzy logic controller (FLC) is introduced, using risk level and health deterioration rate as inputs. Through fuzzification, rule-based reasoning, and defuzzification mechanisms, static risks and dynamic degradation trends are nonlinearly fused, effectively overcoming the shortcomings of traditional early warning systems that rely on fixed thresholds, have delayed responses, and struggle to adapt to complex operating conditions. Through interpretable fuzzy rules such as "if high risk and rapid deterioration occur, a high-level early warning will be triggered," the system upgrades from "simple alarms" to "intelligent grading," significantly enhancing the sensitivity, adaptability, and decision support capabilities of early warnings. This ensures timely and accurate early warnings in situations of rapid equipment degradation or high-risk scenarios, providing maintenance personnel with reliable and tiered response guidelines.

[0031] S3. Generate a cached dataset based on the warning level, verify the integrity of the dataset through blockchain, and identify the root cause of the anomaly by combining the remaining service time of the belt. Based on the root cause of the anomaly and the blockchain evidence data, select the best maintenance strategy. Specifically, a cached dataset is generated based on the warning level, the integrity of the dataset is verified through blockchain, and the root cause of anomalies is identified by combining the remaining time of the conveyor belt. This is based on the cached dataset generated at warning level B. Perform hash operations on the cached dataset to ensure data immutability and calculate the hash value. The formula is: , hash value The data is uploaded to the blockchain via Hyperledger Fabric (using the PBFT consensus mechanism) and written to the distributed ledger, enabling trusted storage and audit trails across multiple nodes. A query is initiated to the Hyperledger Fabric blockchain network to verify that the hash has been uploaded and has not been tampered with. If the verification fails, a data anomaly alarm is triggered, subsequent processes are paused, and data integrity is checked by tracing back to sensors or communication links. If the verification passes, it is determined whether high-risk intervention conditions are met, and the threshold is adjusted based on the lifespan curve provided by the equipment manufacturer and actual operating conditions. That is, when the remaining usage time RUL < preset threshold (e.g., 24 hours) and when the warning level B ≥ high (i.e., the risk level is "high" or "extremely high"), the causal intervention analysis process is automatically triggered, and the abnormal root cause identification stage is entered. Otherwise, it is transferred to the low-frequency monitoring queue, and the source tracing is temporarily suspended to avoid wasting resources. When validation is successful and high-risk intervention conditions are met, causal intervention analysis is used to verify which node is the key driver leading to the current failure state, and the dominant causal path is defined as follows: → →⋯→ (e.g., "load fluctuation → tension increase → roller wear"), for each candidate parent node in the path (e.g., "load fluctuations"), execute virtual intervention operations. For nodes in a causal graph Perform intervention to forcibly restore the system to a normal state (i.e., manually eliminate its abnormal effects). This refers to a baseline state where the variable is in normal, fault-free condition; Define the impact score of the intervention The formula is: , in, For current observation data (Including historical and real-time sensor information, early warning status, etc.) under the condition of fault The probability of occurrence; like The larger the negative value, the greater the decrease in the probability of failure after intervention, indicating a stronger driving effect of the node on the failure. The smallest (i.e. the most significant suppression effect) node is used to confirm the root cause of the anomaly.

[0032] By hashing a cached dataset consisting of warning levels, health indices, and remaining service life using SHA-256 and uploading it to the Hyperledger Fabric blockchain, the PBFT consensus mechanism ensures the immutability and auditability of critical operational data, effectively solving the problems of easily modified data and difficult traceability in traditional systems. In high-risk scenarios, causal intervention analysis is triggered by the remaining service life threshold. Virtual intervention (do operator) quantifies the impact of each candidate cause on the probability of failure, and the Intervention Impact Score (IES) identifies the root driving factors causing anomalies, avoiding reliance on experience and misjudgment, and achieving a leap from "passive response" to "proactive attribution." This not only improves the objectivity and credibility of fault tracing but also prevents resource waste through a conditional triggering mechanism, constructing a safe, intelligent, and traceable closed-loop diagnostic system, significantly enhancing the robustness and decision reliability of the belt conveyor health management system.

[0033] Furthermore, based on the root cause of the anomaly and the blockchain-stored evidence data, the optimal maintenance strategy is selected. This involves generating and visualizing the decision based on the confirmed root cause of the anomaly (such as "load fluctuation") and the blockchain-stored verification data (including PHI, RUL, warning level, causal path, and timestamp). The specific operations are as follows: Match the corresponding set of maintenance actions according to the type of root cause of the anomaly. For example, the optional strategies for "load fluctuation" are: {adjust the feed rate, calibrate the weighing sensor, add buffer roller support, and plan a shutdown for maintenance}. Each strategy is associated with three evaluation indicators, including cost: including manpower, spare parts, and downtime losses, obtained through the maintenance knowledge base; risk reduction: by replaying the ADT model, simulating the expected increase in PHI after implementing the strategy, and calculating the risk level reduction value; and execution time: obtaining the historical average time or expert estimate from the work order system. Normalize all feasible strategies, construct a three-dimensional target space, solve the Pareto optimal front, filter out the non-dominated solution set, and select the strategy with the smallest Euclidean distance to the weighted ideal point as the recommended decision (e.g., "adjust the feed rate and complete it within 24 hours"). After the decision is generated, the instruction is encapsulated in JSON format and sent to the edge execution unit (such as PLC or maintenance terminal) via MQTT protocol. The 3D digital twin scene based on CesiumJS is loaded in the WebGIS platform, and the equipment is colored according to the risk level (gradient from green to red). The health trend curve, early warning heat map and maintenance path trajectory generated by ECharts are overlaid to realize spatial dynamic display. Operators can view the fault location, the evidence chain of the root cause of the abnormality and the handling suggestions in real time through the browser.

[0034] Based on the confirmed root cause of the anomaly and the complete evidence chain stored on the blockchain, a multi-objective optimization decision-making mechanism for maintenance strategies is constructed. The Pareto optimal solution is screened through three dimensions: cost, risk reduction, and execution time. The weighted ideal point method is combined to recommend the maintenance scheme with the best overall benefits, overcoming the shortcomings of traditional operation and maintenance that rely on human experience, have strong decision-making subjectivity, and lack quantitative basis. The decision results are distributed to the edge execution unit in real time via the MQTT protocol to achieve closed-loop control. At the same time, digital twin models, health trends, early warning heat maps, and maintenance paths are integrated into the WebGIS 3D scene based on CesiumJS, supporting the visualization of fault location, causal tracing, and handling suggestions. This significantly improves the scientific nature, transparency, and execution efficiency of maintenance decisions, realizing the full-process automation and visualization management from "fault alarm - root cause analysis - trusted evidence storage - intelligent decision-making - remote monitoring".

[0035] This embodiment also provides a belt conveyor operation monitoring system based on big data, including: The multimodal feature fusion and digital twin modeling module is used to generate low-dimensional feature vectors and combine them with a three-dimensional digital twin model to dynamically update the state using Kalman filtering to obtain the belt conveyor health index. The life prediction and fault inference module is used to calculate the wear depth and combine it with the belt conveyor health index to predict the remaining service life using an LSTM network. The risk assessment and intelligent early warning module is used to construct a two-dimensional risk matrix to determine the risk level, and uses a fuzzy logic controller to process the risk level and the rate of health deterioration to output graded early warning signals. The blockchain evidence storage and anomaly tracing module is used to ensure data integrity by putting datasets on the blockchain. When under high-risk conditions, it identifies the root cause of failure through causal intervention analysis. The maintenance decision generation and visualization monitoring module is used to match the optimal maintenance strategy based on the confirmed root cause and perform Pareto optimal solution.

[0036] This embodiment also provides a computer device applicable to a belt conveyor operation monitoring method based on big data, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the belt conveyor operation monitoring method based on big data as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the belt conveyor operation monitoring method and system based on big data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] In summary, this invention achieves high-precision image feature extraction by introducing the YOLOv7 model combined with laser-assisted imaging, and designs a confidence-weighted multimodal feature fusion mechanism to overcome the problems of missing information and noise interference in single modalities. Furthermore, it employs a quantum variational autoencoder to perform nonlinear dimensionality reduction on the fused features, and combines this with Kalman filtering to dynamically update the 3D digital twin model, realizing closed-loop feedback and state correction between the physical entity and the virtual model, thus solving the evolutionary lag problem caused by traditional static modeling of digital twins. It integrates health index, wear depth, and dynamic state vectors to construct an LSTM lifetime prediction model, and innovatively introduces a causal Bayesian network for intervention analysis, leaping from statistical correlation to causal interpretability, accurately identifying the root cause of failures. Finally, through a fuzzy logic and blockchain collaborative mechanism, it achieves risk warning, data storage, and Pareto optimal maintenance strategy generation, not only enhancing the system's credibility and security but also realizing a technological leap from passive response to proactive prediction and autonomous decision-making.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring the operation of a belt conveyor based on big data, characterized in that: include, By deploying multiple sensors to collect belt conveyor operation data, using the YOLOv7 model to extract image features, and performing multimodal feature fusion based on the image features; Based on the fused multimodal features, a low-dimensional feature vector is obtained using an autoencoder, a three-dimensional digital twin model is constructed, the three-dimensional digital twin model is updated using the Kalman filter method, and a final state vector is generated. The health index of the belt conveyor is obtained based on the final state vector. The wear depth is calculated, and the belt health index, wear depth and final state vector are used as input to the LSTM model. The remaining service time of the belt is output. The probability of failure is obtained by using the causal Bayesian network inference method. The most probable cause of failure is obtained based on the probability of failure. The risk level of the belt conveyor is obtained based on the most probable cause of failure, and a fuzzy logic controller is used to generate an early warning level based on the belt conveyor health index and the risk level. A cached dataset is generated based on the warning level. The integrity of the dataset is verified by blockchain. The root cause of the anomaly is identified by combining the remaining service time of the belt. Based on the root cause of the anomaly and the blockchain-stored evidence data, the best maintenance strategy is selected.

2. The belt conveyor operation monitoring method based on big data as described in claim 1, characterized in that: The process involves deploying multiple sensors to collect belt conveyor operation data, extracting image features using the YOLOv7 model, and performing multimodal feature fusion based on these image features. This process involves deploying multiple source sensors and collecting data, and obtaining a multi-source data stream based on the collected data. The acquired laser-assisted imaging images are input into the YOLOv7 model for inference, outputting the category, bounding box, and confidence score for each detected defect. Based on the CNN backbone network, features are extracted layer by layer from the laser-assisted imaging images to generate deep feature maps. The model accuracy is calculated based on these features. and recall rate The harmonic mean of precision and recall was obtained. Calculate the credibility weight Before multimodal fusion, the image features are weighted to obtain weighted image features. Based on image features Perform a linear transformation to obtain the fusion features .

3. The belt conveyor operation monitoring method based on big data as described in claim 2, characterized in that: The method involves using fused multimodal features to obtain low-dimensional feature vectors via an autoencoder, constructing a 3D digital twin model, updating the 3D digital twin model using Kalman filtering, and generating a final state vector. Based on the final state vector, the belt conveyor health index is obtained, and the total loss function is calculated using the autoencoder. Based on the total loss function Once the trained autoencoder is obtained, the features will be fused. Input encoder, output low-dimensional feature vector Calculate tension A 3D digital twin model is constructed in Unity based on the geometric parameters of the belt conveyor, generating an initial state vector. ; The initial state vector As input to the Unity model, a 3D digital twin model is constructed, and the Kalman filter method is used to obtain the updated state vector. The updated state vector As input parameters to the 3D digital twin model, the updated state vector is mapped to the physical properties of the model through the 3D digital twin model. Physical quantities are extracted in real time from the simulation results as high-order features, and then fused with the updated state vector and the low-dimensional image feature vector z output by the autoencoder to generate the final state vector. Based on the state vector at the current moment The health index (PHI) of the belt conveyor was obtained.

4. The belt conveyor operation monitoring method based on big data as described in claim 3, characterized in that: The calculation of wear depth involves using the health index, wear depth, and final state vector as inputs to an LSTM model. The output is the remaining service time of the belt. A causal Bayesian network inference method is used to obtain the probability of failure. Based on this probability, the most probable cause of failure is determined. The defect detection results from the YOLOv7 model are then used to fit the wear model to obtain the current degradation rate and thus the belt wear depth. , the state vector Health Index (PHI) and Wear Depth As input to the LSTM model, predict future states. Calculate the failure time The remaining service life is calculated based on the failure time. The probability of failure occurrence is calculated using the causal Bayesian network inference method. By using the chain rule, each path C is decomposed into the product of the conditional probabilities of each node. Calculate conditional probability For the faulty node Given path C and observational evidence Under the given conditions, calculate the probability of a fault occurring using Bayesian inference. Identify the probability of failure occurrence based on observational evidence through causal intervention. .

5. The belt conveyor operation monitoring method based on big data as described in claim 4, characterized in that: The risk level of the conveyor belt is obtained based on the most probable cause of failure. A fuzzy logic controller is used to generate an early warning level based on the conveyor belt's health index and risk level. This involves dividing the risk into probability of occurrence and severity of consequences, combining probability and severity through a Cartesian product mapping, and setting lookup table rules to output the risk level R, which is divided into four risk levels. The rate of health deterioration is then calculated. Based on risk level R and health condition rate of change This is the input to the fuzzy logic controller, and the output is warning level B.

6. The belt conveyor operation monitoring method based on big data as described in claim 5, characterized in that: The process of generating a cached dataset based on the warning level, verifying the dataset's integrity through blockchain, and identifying the root cause of anomalies by combining the remaining conveyor belt time refers to generating a cached dataset based on warning level B. Perform a hash operation on the cached dataset to obtain the hash value. , hash value Upon on-chain verification, it is confirmed that the hash has been uploaded and has not been tampered with. If the verification fails, a data anomaly alarm is triggered, subsequent processes are paused, and data integrity is checked by tracing back the sensor or communication link. If the verification passes, it is determined whether the high-risk intervention condition is met, i.e., when the remaining usage time RUL is less than a preset threshold. Furthermore, when the warning level B is higher than or equal to high, it enters the abnormal root state. In the cause identification phase, otherwise it is transferred to the low-frequency monitoring queue, where causal intervention analysis is adopted to set the dominant causal path and analyze each candidate parent node in the path. Perform virtual intervention operations on nodes in the causal graph. Perform intervention actions and define intervention impact scores. .

7. The belt conveyor operation monitoring method based on big data as described in claim 6, characterized in that: The process of selecting the best maintenance strategy based on the root cause of the anomaly and blockchain-stored evidence data involves normalizing all feasible strategies based on the confirmed root cause of the anomaly and the verified data stored on the blockchain, constructing a three-dimensional target space, solving the Pareto optimal front, filtering out the non-dominated solution set, selecting the strategy with the smallest Euclidean distance to the weighted ideal point as the recommended decision, loading a three-dimensional digital twin scene, coloring the equipment according to risk level, and generating a health trend curve, an early warning heat map, and a maintenance path trajectory.

8. A belt conveyor operation monitoring system based on big data, based on the belt conveyor operation monitoring method based on big data as described in any one of claims 1 to 7, characterized in that: include, The multimodal feature fusion and digital twin modeling module is used to generate low-dimensional feature vectors and combine them with a three-dimensional digital twin model to dynamically update the state using Kalman filtering to obtain the belt conveyor health index. The life prediction and fault inference module is used to calculate the wear depth and combine it with the belt conveyor health index to predict the remaining service life using an LSTM network. The risk assessment and intelligent early warning module is used to construct a two-dimensional risk matrix to determine the risk level, and uses a fuzzy logic controller to process the risk level and the rate of health deterioration to output graded early warning signals. The blockchain evidence storage and anomaly tracing module is used to ensure data integrity by putting datasets on the blockchain. When under high-risk conditions, it identifies the root cause of failure through causal intervention analysis. The maintenance decision generation and visualization monitoring module is used to match the optimal maintenance strategy based on the confirmed root cause and perform Pareto optimal solution.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the belt conveyor operation monitoring method based on big data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the belt conveyor operation monitoring method based on big data as described in any one of claims 1 to 7.

Images