Deep learning-based belt conveyor multi-state fusion intelligent online control method

Abstract

The invention relates to the technical field of automatic control, in particular to a belt conveyor multi-state fusion intelligent online control method based on deep learning. The method comprises the following steps: acquiring multi-modal data of the belt conveyor, and preprocessing the multi-modal data; a cross-modal deep learning fusion network of a multi-source data co-processing platform based on Transform is constructed, and full-link monitoring of the belt conveyor is achieved; constructing a multi-parameter collaborative dynamic adjustment mechanism based on material characteristics and equipment parameters, and constructing a mapping relation between flow and speed; deep learning is utilized to drive fault traceability analysis, and a potential fault mode is identified; and on the basis of a robust closed-loop control architecture, predicting a running track of the belt, generating an adjustment strategy, and realizing multi-state control of the belt conveyor. According to the method, deep learning, multi-mode sensing, advanced control and predictive analysis are deeply fused, and the purpose of achieving safety, high efficiency, reliability and intelligence of belt conveyor operation is achieved.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and in particular to a deep learning-based multi-state fusion intelligent online control method for belt conveyors. Background Technology

[0002] In industries such as ports, metallurgy, and mining, belt conveyors are core equipment for continuous material transport. Although the technology is mature, there are still many pain points and problems in practical applications, mainly concentrated in the following aspects: belt misalignment, material flow overload, and misalignment, causing material spillage and dust pollution; idler roller failure; low efficiency of belt conveyor operation under no-load and low-load conditions, resulting in a large workload for daily maintenance and high operating costs.

[0003] Existing technical measures have the following limitations: Taking belt alignment as an example, the current method of using alignment rollers to solve the problem of belt misalignment is limited to belt misalignment switches and mechanical passive alignment rollers. The misalignment signals are relatively simple, with a first-level alarm (slight misalignment) and a second-level shutdown (severe misalignment). The alignment rollers cannot act immediately according to the instructions, resulting in a lag.

[0004] There is no systematic approach to solving the various problems that occur with belt conveyors, resulting in a piecemeal approach. For example, if a belt conveyor experiences uneven loading and material spillage during operation, it could be due to belt misalignment or excessive flow leading to overflow. In such cases, a comprehensive diagnostic method is lacking, making it impossible to take accurate and effective measures.

[0005] The lack of a systematic intelligent algorithm means that the safe and reliable operation of belt conveyors directly impacts port production efficiency. At best, it reduces transport efficiency; at worst, it leads to production stoppages, significant economic losses, and even threats to personnel and property safety. Currently, monitoring and managing belt conveyor systems typically relies on worker inspections, with repairs only performed after equipment malfunctions. This approach cannot guarantee the safety and reliability of belt conveyor operation.

[0006] Therefore, there is a need to provide a control method that can ensure the stable operation of belt conveyors. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a deep learning-based intelligent online control method for multi-state fusion of belt conveyors. The invention primarily strengthens the correlation of features within each mode, dynamically establishes inter-modal correlations, and constructs a cross-modal deep learning fusion architecture based on a Transformer-based multi-source data collaborative processing platform. It employs a dynamic adjustment mechanism based on the coordination of multiple parameters such as material properties and equipment parameters, and utilizes deep learning-driven fault tracing analysis to construct an intelligent control system with a robust closed-loop control architecture based on model prediction.

[0008] The technical means employed in this invention are as follows: A deep learning-based multi-state fusion intelligent online control method for belt conveyors includes: acquiring multi-modal data of the belt conveyor and preprocessing the multi-modal data; constructing a cross-modal deep learning fusion network based on a Transformer-based multi-source data collaborative processing platform to achieve full-link monitoring of the belt conveyor; constructing a multi-parameter collaborative dynamic adjustment mechanism based on material characteristics and equipment parameters, and establishing a mapping relationship between flow rate and speed; using deep learning to drive fault source analysis to identify potential fault modes; and predicting the belt running trajectory and generating adjustment strategies based on a robust closed-loop control architecture to achieve multi-state control of the belt conveyor.

[0009] Furthermore, the multimodal data includes: belt conveyor deviation amount, belt conveyor deviation direction, belt conveyor cross-section material type, flow rate, material spillage, idler roller abnormal noise location, material blockage information inside the head chute, and belt conveyor operating power; the preprocessing includes noise reduction, formatting, and standardization operations.

[0010] Furthermore, the cross-modal deep learning fusion network is used to fuse the collected multimodal data, build a Transformer-based cross-modal feature fusion platform, perform spatiotemporal correlation modeling on the multimodal data, generate equipment health index, and construct a full-link monitoring system for multi-dimensional states of equipment vibration, deviation, and material blockage.

[0011] Furthermore, the dynamic adjustment mechanism constructs a mapping relationship between flow rate and speed through a lookup table method and piecewise linear interpolation to ensure the interpretability of adjustment parameters when the material flow is abnormal; it adopts a control structure that combines feedforward and feedback, and when the lidar detects a sudden change in material flow, it triggers the belt conveyor to slow down and the diversion device to work together through a preset logic chain to achieve a rapid response.

[0012] Furthermore, the fault tracing analysis is driven by deep learning, deploying a convolutional autoencoder to learn from historical deviation data and material flow fluctuation signals to construct a normal state feature space for equipment operation; when the control strategy fails to adjust after execution, the deep learning model is automatically invoked to reconstruct the features of the sensor data stream, and potential fault modes are identified through anomaly score detection, providing data-driven improvement suggestions for control parameter optimization, forming a closed-loop iterative mechanism.

[0013] Furthermore, the robust closed-loop control architecture is a predictive control framework based on a physical model. It combines real-time collected belt misalignment data, predicts the belt running trajectory through a state-space model, and generates the optimal adjustment parameter sequence using a rolling optimization algorithm. The control strategy integrates a fuzzy logic rule base and matches predefined reverse adjustment strategies for different loads and belt speeds to achieve industrial-grade control stability.

[0014] Compared with the prior art, the present invention has the following advantages: This invention provides a deep learning-based multi-state fusion intelligent online control method for belt conveyors, establishing a model capable of predicting the future dynamic behavior of the system. Utilizing a Transformer fusion model or a specially trained deep learning model, based on the current state (speed, material flow rate) and control inputs (belt offset, material flow command), the method predicts the system state over a future period, ensuring that the predicted system behavior closely approximates the setpoint trajectory while satisfying various constraints and optimizing the objective function. Only the first optimized control input is actually sent to the actuator. The process is repeated in the next cycle, i.e., "rolling" optimization. This invention also compares the actually measured system state with the predicted state, using online error correction to adjust the prediction model or optimization process, enhancing robustness.

[0015] In summary, the method of this invention deeply integrates deep learning, multimodal perception, advanced control, and predictive analytics, aiming to achieve safety, efficiency, reliability, and intelligence in the operation of belt conveyors. Furthermore, this invention enables comprehensive, accurate, and real-time perception of the belt conveyor's operating status, and based on this, performs predictive maintenance and dynamic optimization control, ultimately improving conveying efficiency, reducing energy consumption, minimizing downtime, and ensuring the safety of personnel and equipment.

[0016] Based on the above reasons, this invention can be widely applied in fields such as automatic control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the deep learning-based multi-state fusion intelligent online control method for belt conveyors in this invention.

[0019] Figure 2 This is a structural diagram of a deep learning-based multi-state fusion intelligent online control system for belt conveyors in an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. 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.

[0022] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0023] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0024] like Figure 1 As shown, this invention provides a deep learning-based multi-state fusion intelligent online control method for belt conveyors, comprising: The multimodal data of the belt conveyor is acquired and preprocessed. In a preferred embodiment of the present invention, the multimodal data includes: belt conveyor deviation amount, belt conveyor deviation direction, belt conveyor cross-section material type, flow rate, material spillage, idler roller abnormal noise location, material blockage information inside the head chute, and belt conveyor operating power. The preprocessing includes noise reduction, formatting, and standardization.

[0025] During implementation, multimodal data is acquired collaboratively through multiple sensors. Belt conveyor misalignment data is collected using laser or ultrasonic rangefinder sensors. A pair of sensors is deployed at intervals along both sides of the belt conveyor, with the installation angle perpendicular to the belt edge cross-section; this accurately monitors the online misalignment amount and direction. Belt conveyor material flow status monitoring data uses 3D lidar sensors, installed near the outlet of the conveyor's guide chute, positioned at an upper distance. The installation principle allows the sensor to comprehensively scan the belt conveyor cross-section; this monitors real-time information such as material type, flow rate, and material spillage. The belt conveyor fiber optic diagnostic system module collects data on idler noise and location information. Material status data within the head chute of the belt conveyor is collected using a radar level gauge in conjunction with a mechanical level gauge, monitoring for material blockage information in the head chute in real-time. Belt conveyor drive unit motor current and voltage data are collected to monitor the belt conveyor's operating power information in real-time.

[0026] A cross-modal deep learning fusion network based on a Transformer-based multi-source data collaborative processing platform is constructed to achieve full-link monitoring of belt conveyors. In a preferred embodiment, the cross-modal deep learning fusion network is used to fuse collected multimodal data, building a Transformer-based cross-modal feature fusion platform. This platform performs spatiotemporal correlation modeling of the multimodal data to generate an equipment health index, achieving a paradigm shift from "fault detection" to "predictive maintenance." The system integrates a fiber optic diagnostic system through a standardized interface, constructing a full-link monitoring system for multi-dimensional states such as equipment vibration, misalignment, and material blockage, providing comprehensive data support for preventative maintenance.

[0027] During implementation, a cross-modal sensing fusion technology based on deep learning is constructed, namely a multi-source heterogeneous sensing network including laser displacement, radar scanning, and fiber optic acoustic signals, breaking through the limitations of traditional single-sensor detection. Through a fusion model of convolutional neural networks and long short-term memory networks, spatiotemporal features of conveyor belt deviation, material flow status, and chute blockage characteristics are extracted. The system automatically learns material dynamics (such as the correlation characteristics of angle of repose and friction coefficient), enabling the detection system to adaptively optimize in harsh environments such as dust and vibration. A cross-modal deep learning fusion architecture based on a Transformer-based multi-source data collaborative processing platform is constructed. This architecture encodes the spatiotemporal features of geometric position data from the laser displacement sensor, material point cloud data from the lidar, and acoustic vibration signals from fiber optic auscultation. An attention mechanism is used to automatically extract the correlation features of abnormal states such as deviation and blockage. The model employs a transfer learning strategy, rapidly adapting to different working conditions through pre-trained models in scenarios with limited data in industrial settings.

[0028] Based on material characteristics and equipment parameters, a multi-parameter collaborative dynamic adjustment mechanism is constructed to establish a mapping relationship between flow rate and speed. In specific implementation, as a preferred embodiment of the present invention, the dynamic adjustment mechanism constructs the mapping relationship between flow rate and speed through table lookup and piecewise linear interpolation to ensure the interpretability of adjustment parameters when the material flow is abnormal. A combined feedforward and feedback control structure is adopted. When the lidar detects a sudden change in material flow, the belt conveyor is slowed down and the diversion device is linked through a preset logic chain, which, in conjunction with the edge computing unit, can achieve a fast response of 20ms.

[0029] Deep learning is used to drive fault source analysis and identify potential fault modes. In a preferred embodiment of this invention, fault source analysis is driven by deep learning, deploying a convolutional autoencoder to learn from historical deviation data and material flow fluctuation signals to construct a feature space of normal equipment operation. When the control strategy fails to adjust, the deep learning model is automatically invoked to reconstruct the features of the sensor data stream. Potential fault modes, such as roller jamming and material adhesion, are identified through anomaly score detection, providing data-driven improvement suggestions for control parameter optimization and forming a closed-loop iterative mechanism.

[0030] Based on a robust closed-loop control architecture, the belt conveyor's trajectory is predicted, and adjustment strategies are generated to achieve multi-state control of the belt conveyor. In a preferred embodiment of this invention, the robust closed-loop control architecture is a predictive control framework based on a physical model, combined with real-time data on belt misalignment, achieving an accuracy of [insert accuracy here]. The system predicts the belt running trajectory through a state-space model and generates the optimal adjustment parameter sequence using a rolling optimization algorithm. The control strategy integrates a fuzzy logic rule base and matches predefined reverse adjustment strategies for different loads and belt speeds to achieve industrial-grade control stability.

[0031] Example like Figure 2 As shown, the present invention also provides a multi-state fusion intelligent online control system for belt conveyors based on deep learning, which first arranges the following multi-source heterogeneous sensing devices at the corresponding measurement positions of the belt conveyor.

[0032] Laser displacement sensors are used to accurately monitor the online deviation and direction of high-precision belt conveyors. LiDAR scanning provides wide-range, non-contact measurement of material flow volume, profile, and velocity, and real-time monitoring of belt conveyor cross-sectional material type, flow rate, and material spillage. Fiber optic acoustic sensors utilize optical fibers laid along the belt conveyor as "nerves" to continuously sense vibrations (idler jamming, bearing failure, material impact, belt joint abnormalities), acoustics (tearing, scratching, abnormal friction), and even temperature signals along the entire belt (distributed, long-distance, high-sensitivity), collecting information such as idler noise and positioning.

[0033] Design or employ appropriate deep neural network feature alignment and correlation for each modality of data to address the differences in spatiotemporal scale, data format, and sampling rate among different sensors. Utilize attention mechanisms, graph neural networks, or specially designed fusion networks to learn deep correlations and complementarities between features of different modalities.

[0034] The precise material thickness detected by laser is cross-checked and complemented by the overall material volume detected by radar. Abnormal vibration acoustic characteristics sensed by fiber optics at specific locations are correlated with structural deformations (such as idler roll sagging) detected by laser / radar at those locations, jointly indicating faults. Visually detected belt misalignment is corroborated by the belt edge position measured by laser and the lateral friction force changes sensed by fiber optics. The goal is to learn a fused, robust, and more informative joint feature representation that more accurately reflects the true state of the belt conveyor (e.g., "normal conveying," "slight misalignment," "early stage of idler roll jamming," "large foreign object," "minor longitudinal tear in the belt," etc.).

[0035] Cross-modal sensing fusion technology transforms raw data or preliminary features (from feature extraction networks) from various modalities into a unified sequence of embedded vectors, including location encoding (spatial location, timestamp). For example, radar volume data at a given time point can be used to "query" the vibration intensity of an optical fiber at the same moment to determine whether it is normal material flow or foreign object impact. This significantly improves the comprehensiveness, accuracy (reducing false alarms and missed alarms), robustness (resistance to single-modal failures or interference), and ability to detect weak or early-stage faults.

[0036] Employing the Transformer architecture, it can effectively capture long-distance dependencies and complex relationships within sequences, making it suitable for handling systems with spatiotemporal continuity, such as belt conveyors. For example, an anomalous signal (such as vibration) at a location may originate from changes in upstream drive or downstream load, and it allows the model to simultaneously focus on different "aspects" of information from different modes, time steps, and spatial locations, improving the efficiency of large-scale data processing and making it suitable for processing time-series sensor data (fiber optic acoustic vibration, velocity signals, etc.).

[0037] After multi-layer Transformer encoding, a perception feature is obtained that integrates spatiotemporal information from all modes. It can be used for fault diagnosis (bearing damage level, tearing, etc.); for state quantity estimation (accurate material flow rate, belt tension prediction, etc.); and for predicting future states (material accumulation risk, fault development).

[0038] Because the ideal operating conditions of belt conveyors (such as optimal speed and tension) are highly dependent on the characteristics of the conveyed material (density, particle size, humidity, viscosity) and the equipment condition (belt wear, idler resistance changes, ambient temperature / humidity), traditional fixed-parameter control or simple feedback-based control struggles to adapt to these complex changes, leading to inefficiency or accelerated wear. A multi-parameter collaborative dynamic adjustment mechanism achieves "on-demand customization" and "refined management" of belt conveyor operation, significantly improving energy efficiency (avoiding idling or overloading), reducing equipment wear (optimizing tension and speed), and enhancing system adaptability and throughput efficiency.

[0039] By employing multi-parameter sensing and leveraging online monitoring and perception capabilities, real-time acquisition of material characteristics (volume – from laser / radar fusion; particle size – can be combined with visual or vibration analysis; humidity – optional dedicated sensors) and equipment operating status parameters (belt stiffness / wear estimation – from tension, misalignment, and acoustic vibration fusion analysis; idler roller rotation resistance – from motor current and acoustic vibration; ambient temperature and humidity) is achieved through deep learning-driven relational modeling. A deep neural network is trained to establish a complex nonlinear mapping relationship between the aforementioned multi-parameters (material characteristics, equipment status, environment) and the optimal control objective.

[0040] The system generates dynamically optimized setpoints, inputting multi-parameter data collected online into a trained deep learning model in real time. The model then outputs suggested optimal setpoints for the current operating conditions. These optimal setpoints are then used as setpoints for the control unit. The system requires coordination among multiple actuators, including the drive motor (speed regulation), the alignment mechanism (alignment), the tensioning device (tension adjustment), and the feeding equipment (flow rate adjustment), to achieve smooth, rapid, and accurate dynamic adjustment.

[0041] Simply detecting a "fault" or "part failure" is insufficient. It's crucial to quickly and accurately pinpoint the exact location of the fault (e.g., which idler roller? Which drum? Which belt section?) and analyze its root cause (installation issues? material impact? poor lubrication? fatigue aging?) to guide efficient and precise maintenance, preventing the fault from escalating or recurring. Employing a deep learning-driven fault tracing analysis module significantly shortens troubleshooting time, enables precise maintenance (avoiding blind replacement), and provides a deep understanding of the fault mechanism. This data support optimizes maintenance strategies, improves equipment design, and adjusts operating procedures, upgrading predictive maintenance to knowledge-driven precision maintenance.

[0042] High-precision fault location is achieved using distributed fiber optic sensing. Essentially, it's a linear sensor array that precisely locates the distance (along the fiber optic length) to the occurrence of abnormal vibration / acoustic events, pinpointing the event to a specific idler group, roller, or belt segment. It integrates other positioning information, such as localized deformation locations detected by laser / radar and image positions captured by cameras. Cross-verification of multi-source positioning information improves positioning accuracy and reliability.

[0043] Deep learning-driven fault diagnosis and tracing are based on feature extraction and fusion (especially multimodal features such as sound, vibration, and deformation near the fault location). The diagnostic model uses a deep classification network to identify specific fault types (such as: bearing inner ring damage, idler jamming, belt longitudinal tearing, joint delamination, and roller coating detachment).

[0044] The causal analysis and source tracing model utilizes graph neural networks to model the conveyor belt system, attaching perceived fault signals to nodes at corresponding locations. Through message passing, it learns how faults propagate and interact within the component network, thereby inferring root cause nodes and possible propagation paths. Combined with time-series models, it analyzes the temporal evolution of fault characteristic signals, distinguishing between instantaneous impacts and gradual damage, aiding in determining the root cause. The fault location, type, possible root causes, and recommendations are presented to maintenance personnel in an intuitive manner.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based multi-state fusion intelligent online control method for belt conveyors, characterized in that, include: Acquire multimodal data of the belt conveyor and preprocess the multimodal data; Construct a cross-modal deep learning fusion network based on a Transformer-based multi-source data collaborative processing platform to achieve full-link monitoring of belt conveyors; A multi-parameter collaborative dynamic adjustment mechanism is constructed based on material properties and equipment parameters to establish a mapping relationship between flow rate and speed; Utilize deep learning to drive fault tracing analysis and identify potential fault modes; Based on a robust closed-loop control architecture, the belt running trajectory is predicted and adjustment strategies are generated to achieve multi-state control of the belt conveyor.

2. The deep learning-based multi-state fusion intelligent online control method for belt conveyors according to claim 1, characterized in that, The multimodal data includes: belt conveyor deviation amount, belt conveyor deviation direction, belt conveyor cross-section material type, flow rate, material spillage, idler roller abnormal noise location, material blockage information inside the head chute, and belt conveyor operating power; the preprocessing includes noise reduction, formatting, and standardization operations.

3. The deep learning-based multi-state fusion intelligent online control method for belt conveyors according to claim 1, characterized in that, The cross-modal deep learning fusion network is used to fuse and process the collected multimodal data, build a cross-modal feature fusion platform based on Transformer, perform spatiotemporal correlation modeling on the multimodal data, generate equipment health index, and construct a full-link monitoring system for multi-dimensional states of equipment vibration, deviation, and material blockage.

4. The deep learning-based multi-state fusion intelligent online control method for belt conveyors according to claim 1, characterized in that, The dynamic adjustment mechanism constructs a mapping relationship between flow rate and speed through a lookup table method and piecewise linear interpolation, ensuring the interpretability of adjustment parameters when the material flow is abnormal. It adopts a control structure that combines feedforward and feedback. When the lidar detects a sudden change in the material flow, it triggers the belt conveyor to slow down and the diversion device to work together through a preset logic chain to achieve a rapid response.

5. The deep learning-based multi-state fusion intelligent online control method for belt conveyors according to claim 1, characterized in that, The fault tracing analysis is driven by deep learning. It deploys a convolutional autoencoder to learn from historical deviation data and material flow fluctuation signals to construct a feature space of normal operating conditions of the equipment. When the adjustment fails after the control strategy is executed, the deep learning model is automatically called to reconstruct the features of the sensor data stream. Potential fault modes are identified through anomaly score detection, providing data-driven improvement suggestions for control parameter optimization and forming a closed-loop iterative mechanism.

6. The deep learning-based multi-state fusion intelligent online control method for belt conveyors according to claim 1, characterized in that, The robust closed-loop control architecture is a predictive control framework based on a physical model. It combines real-time data on belt misalignment and predicts the belt's running trajectory using a state-space model. It then uses a rolling optimization algorithm to generate the optimal misalignment parameter sequence. The control strategy integrates a fuzzy logic rule base and matches predefined reverse adjustment strategies for different loads and belt speeds to achieve industrial-grade control stability.

Images