Intelligent health monitoring method and system for scraper conveyor chain by fusing multi-source sensing and deep learning

Abstract

This application discloses a method and system for intelligent health monitoring of scraper conveyor chains that integrates multi-source sensing and deep learning. The method includes: real-time acquisition and preprocessing of multi-source data during chain operation to form a multi-source time-series dataset; calculation of the chain's load rate, real-time operating speed, and cumulative wear based on the multi-source time-series dataset, followed by calculation of load correction coefficients, speed correction coefficients, wear correction coefficients, and temperature correction coefficients using temperature data, thereby determining a dynamic threshold; classification of the chain's state based on the dynamic threshold and real-time spring displacement, and location of chain breakage at the breakage level; simultaneously, real-time evaluation of the remaining chain levels and early warning of chain breakage risk using a machine learning model. This application achieves real-time assessment of chain health status, early warning of breakage risk, and precise location of chain breakage after it occurs.

Description

Technical Field

[0001] This invention relates to the field of mining equipment technology, specifically to a method and system for intelligent health monitoring of scraper conveyor chains that integrates multi-source sensing and deep learning. Background Technology

[0002] Scraper conveyors, as key continuous conveying equipment in heavy industries such as coal mines and metal mines, undertake the core task of transporting bulk materials over long distances and in large volumes. Their working principle involves a closed-loop arrangement of scraper chains circulating within a channel, continuously conveying materials in a predetermined direction. In the complex underground environment, scraper conveyors often operate continuously under harsh conditions of high load, strong vibration, and high dust levels. As the core component for power transmission and material carrying, the chain endures the combined effects of alternating stress, impact loads, and wear and corrosion, making it highly susceptible to fatigue damage and even sudden breakage. With the deepening and intensification of mining operations, the conveying distance and capacity of scraper conveyors are continuously increasing, placing higher demands on the reliability of the chains. Statistics show that chain breakage accounts for 35%-45% of all scraper conveyor failures, and these breaks often exhibit suddenness, concealment, and cascading damage characteristics.

[0003] Currently, scraper conveyor chain condition monitoring mainly relies on traditional methods such as regular manual inspections and local sensor monitoring. These methods have significant limitations. Manual inspections are highly subjective, lack comprehensive coverage, and are difficult to locate after a chain break, requiring an average of 2-4 hours of segment-by-segment inspection, severely impacting production efficiency and posing safety risks. While monitoring technologies based on eddy current, tension sensors, or vibration analysis can achieve some online monitoring, they generally suffer from limited monitoring range, susceptibility to signal interference, inability to accurately locate break points, and lack of early warning capabilities. First-generation wireless sensor node technology, although deploying sensors on the chain, has limited functionality, poor communication reliability, insufficient environmental adaptability, and can only alarm after a break occurs, failing to provide pre-emptive warnings and intelligent diagnosis. These technological bottlenecks have resulted in scraper conveyor chain monitoring remaining in a passive, reactive mode for a long time, failing to meet the development needs of modern mines for intelligent, less-manned, and more efficient operations.

[0004] Therefore, how to achieve early warning before chain breakage, how to achieve rapid and accurate positioning after chain breakage, and how to ensure the reliable operation of the monitoring system in harsh downhole environments are technical problems that urgently need to be solved in this field. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: In a first aspect, embodiments of this application provide a method for intelligent health monitoring of scraper conveyor chains that integrates multi-source sensing and deep learning, including: Multiple intelligent monitoring nodes are used to collect and preprocess multi-source data during the chain operation process in real time to form a multi-source time series dataset. The load rate, real-time operating speed, and cumulative wear of the chain are calculated based on the multi-source time-series dataset. The load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient are calculated based on the load rate, real-time operating speed, cumulative wear amount, and temperature data in the multi-source time series dataset. The dynamic threshold is determined based on the load correction factor, speed correction factor, wear correction factor, and temperature correction factor. The chain status is classified based on the dynamic threshold combined with real-time spring displacement, and the chain breakage location is located for chains at the breakage level. At the same time, the status of the remaining chains is evaluated in real time and a chain breakage risk warning is given through a machine learning model.

[0006] In one possible implementation, the step of collecting and preprocessing multi-source data during the chain operation process through multiple intelligent monitoring nodes to form a multi-source time-series dataset includes: Multiple intelligent monitoring nodes, installed at equal intervals in the grooves of the chain connecting rings of the scraper conveyor, collect pressure data, triaxial vibration data, temperature data, and node running time data in real time during the chain operation process. The collected pressure data, triaxial vibration data, temperature data, and node running time data are sliced ​​and normalized in a fixed time window to form a multi-source time series dataset.

[0007] In one possible implementation, the load rate, real-time operating speed, and cumulative wear of the chain are calculated based on the multi-source time-series dataset, including: The real-time load rate and real-time operating speed of the chain are calculated based on the pressure data and the signal time difference between adjacent nodes in the multi-source time-series dataset. The formula for calculating the load rate is as follows: in, For load rate, For the real-time load of the chain, This is the rated load of the chain; The vibration signal time series data is input into the wear quantization model, and the time-domain vibration signal is subjected to fast Fourier transform to obtain the power spectral density function, and then the high-frequency energy ratio, vibration energy entropy and impact pulse intensity are extracted. The instantaneous wear rate of the chain is calculated by using a preset linear weighted model to calculate the instantaneous wear rate in the current time window, and then the instantaneous wear rate is integrated on the time axis to obtain the cumulative wear amount of the chain.

[0008] In one possible implementation, the instantaneous wear rate of the current time window is calculated using a preset linear weighted model, and then the instantaneous wear rate is integrated over the time axis to obtain the formula for calculating the cumulative wear of the chain: in, This represents the cumulative wear and tear at the current moment. This represents the cumulative wear and tear at the previous moment. Instantaneous wear rate, This is the load-velocity weighting factor. The sampling interval is... For high-frequency energy ratio, The sensitivity coefficient for high-frequency energy ratio. The vibrational energy entropy, The sensitivity coefficient is the vibration energy entropy. The intensity of the impact pulse. The sensitivity coefficient is the value of the impact pulse intensity. Based on the wear rate, For real-time load, This is the rated load of the chain. For real-time running speed, This is the rated operating speed.

[0009] In one possible implementation, the formulas for calculating the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient based on the load rate, real-time operating speed, cumulative wear, and temperature data in the multi-source time-series dataset are as follows: in, This is the load correction factor. For speed correction factor, This is a temperature correction factor. This is the wear correction factor. For load rate, For real-time running speed, For the rated transport speed, The local temperature of the chain. This serves as a reference threshold for cumulative wear. This represents the cumulative wear and tear.

[0010] In one possible implementation, the formula for determining the dynamic threshold based on the load correction factor, speed correction factor, wear correction factor, and temperature correction factor is as follows: in, For dynamic thresholds, Based on the threshold, This is the load correction factor. For speed correction factor, This is a temperature correction factor. This is the wear correction factor.

[0011] In one possible implementation, the chain state is graded based on the dynamic threshold combined with real-time spring displacement, and the breakage location of chains at the breakage level is located. Simultaneously, a machine learning model is used to perform real-time assessment of the state of the remaining chains and provide breakage risk warnings, including: The real-time spring displacement of the sensing spring is calculated by inverting the pressure sensor signal. The calculation formula is as follows: in, For spring displacement, This is the real-time spring force measured by the pressure sensor. The spring force under initial compression. This is the stiffness coefficient of the sensing spring; Compare the real-time spring displacement with the dynamic threshold; When the real-time spring displacement is less than the first preset dynamic threshold, it is classified as normal. When the real-time spring displacement is greater than the first preset dynamic threshold but less than the dynamic threshold, it is classified as a level of concern; When the real-time spring displacement is greater than the dynamic threshold, it is classified as a high-risk level; When the real-time spring displacement is greater than the second preset dynamic threshold, it is classified as a fracture occurrence level; The system locates the breakage position of chains at different breakage levels, and uses machine learning models to conduct real-time assessments and provide early warnings of breakage risks for chains at normal, watch, and high-risk levels.

[0012] In one possible implementation, the location of the chain break at the breakage level includes: The last communication node is identified using the network topology, and the initial disconnection interval is determined based on the last communication node. High-speed sampling commands are sent to multiple nodes upstream and downstream of the initial chain break interval, and data from multiple nodes is collected and transmitted to the cloud. After receiving the multi-node data, the cloud performs feature extraction and fusion analysis on the multi-node data, and calculates the vibration energy ratio and temperature gradient of each node; The location of the chain break is accurately determined based on the vibration energy ratio and temperature gradient of each node.

[0013] In one possible implementation, a machine learning model is used to perform real-time assessments of the status of normal, monitored, and high-risk chains and to provide early warnings of chain breakage risks, including: When the chain status is classified into normal level, attention level or high risk level, the multi-source time series dataset is input in real time into the long short-term memory network in the deep learning model that integrates the pre-trained long short-term memory network and random forest. Long Short-Term Memory (LSTM) networks are used to extract deep dynamic features from multi-source time-series data, capture the temporal evolution of fault symptoms, and output multi-dimensional feature vectors. The multidimensional feature vector is concatenated with manually designed statistical features to form a fused feature vector; Random forest is used to classify and regress the fused feature vectors, and the output is a multi-dimensional evaluation result including the real-time health score of the chain, the probability of breakage risk, and the prediction of wear trend. When the health score falls below a preset threshold or the risk probability exceeds a critical value, a tiered warning is issued to enable proactive intelligent operation and maintenance decision-making before the chain breaks.

[0014] Secondly, embodiments of this application provide an intelligent health monitoring system for scraper conveyor chains that integrates multi-source sensing and deep learning, comprising: The multi-source data acquisition module is used to collect and preprocess multi-source data in real time during the chain operation through multiple intelligent monitoring nodes to form a multi-source time series dataset. The first calculation module is used to calculate the load rate, real-time operating speed and cumulative wear of the chain based on the multi-source time-series dataset. The second calculation module calculates the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient based on the load rate, real-time operating speed, cumulative wear amount, and temperature data in the multi-source time-series dataset. The determination module is used to determine the dynamic threshold based on the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient; The health monitoring module classifies the chain's state based on the dynamic threshold and real-time spring displacement, locates the breakage position of chains at the breakage level, and uses a machine learning model to evaluate the state of the remaining chains in real time and provide early warning of chain breakage risk.

[0015] Compared with the prior art, the beneficial effects of this application are as follows: This application constructs an integrated intelligent monitoring node that deeply couples three functional modules: multi-source sensor data acquisition, adaptive threshold decision-making, and two-way self-organizing network communication, forming a closed-loop system of perception-decision-execution. This enables real-time assessment of the chain's health status, early warning of breakage risks, and second-level precise positioning after a chain breakage occurs, fundamentally improving the operational safety and maintenance efficiency of scraper conveyors.

[0016] This application addresses the shortcomings of existing technologies, which can only locate faults after the fact, lack early warning capabilities, and have poor anti-interference capabilities. It integrates multi-source sensors at chain nodes, achieves redundant data transmission through bidirectional self-organizing network communication, and introduces an adaptive threshold adjustment mechanism to dynamically adjust chain breakage judgment conditions based on real-time operating conditions. Through machine learning models, the health status of the chain can be assessed in real time, and breakage risk warnings can be issued. Furthermore, it links with a cloud monitoring platform to achieve remote diagnosis and intelligent operation and maintenance. This application represents a technological leap from passive location to proactive early warning and intelligent diagnosis, significantly improving the operational safety and maintenance efficiency of scraper conveyors. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an intelligent health monitoring method for scraper conveyor chains that integrates multi-source sensing and deep learning, provided as an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the intelligent monitoring node provided in the embodiments of this application; Figure 3 This is a schematic diagram of the mechanical installation of nodes provided in an embodiment of this application; Figure 4 This is a bidirectional self-organizing network topology diagram provided in the embodiments of this application; Figure 5 The adaptive threshold adjustment path diagram provided in the embodiments of this application; Figure 6 A path diagram for the chain break location method provided in the embodiments of this application; Figure 7 A machine learning model architecture diagram provided for embodiments of this application; Figure 8 This is a schematic diagram of a health monitoring interface provided in an embodiment of this application. Detailed Implementation

[0018] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0019] Figure 1 A flowchart illustrating an intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning, provided in an embodiment of this application, is shown below. Figure 1This embodiment of the intelligent health monitoring method for scraper conveyor chains, which integrates multi-source sensing and deep learning, includes: S101 collects and preprocesses multi-source data in real time during the chain operation through multiple intelligent monitoring nodes to form a multi-source time-series dataset.

[0020] In this embodiment, multiple intelligent monitoring nodes installed at equal intervals in the grooves of the chain connecting rings of the scraper conveyor collect pressure data, triaxial vibration data, temperature data, and node running time data in real time during the chain operation. The collected pressure data, triaxial vibration data, temperature data, and node running time data are sliced ​​and normalized with a fixed time window to form a multi-source time series dataset.

[0021] See Figure 2 and Figure 3 In this embodiment, each intelligent monitoring node is integrated into the mounting box and fixed to a dedicated groove in the middle of the chain connecting ring by a snap-fit. After installation, the node does not protrude from the chain outline and does not affect the normal meshing of the chain and sprocket. The mounting box uses a high-strength aluminum alloy shell with an IP67 protection rating to adapt to the harsh environment of dampness and dust in underground mines. An internal rubber damping layer is provided to isolate the impact vibration generated by chain operation. The box integrates multiple sensors, including a miniature pressure sensor, vibration sensor, temperature sensor, microprocessor unit, dual-mode wireless communication module, and battery module. Sensing springs on both sides extend outside the box through spring outlet slots, with wear-resistant ceramic contacts at their ends for tight contact with the preceding and following chain links. The multiple sensors are connected to the springs to sense chain tension, vibration, and temperature changes in real time. All modules are connected via a flexible circuit board and are powered by a unified intelligent power management circuit.

[0022] This embodiment does not simply stack multiple sensors within the same node, but rather constructs a collaborative sensing mechanism with dual verification of pressure and vibration. One end of the sensing spring contacts the chain, and the other end is connected to a pressure sensor to sense changes in chain tension in real time, while the vibration sensor simultaneously collects high-frequency vibration signals. The microprocessor embeds a logic AND gate algorithm to determine whether an alarm is triggered. Only when the pressure sensor signal indicates that the spring tension exceeds a dynamic threshold and the vibration sensor detects an impact pulse exceeding 10 times the normal value within a very short time (e.g., 0.5 seconds) is a genuine chain breakage event confirmed. This mechanism utilizes vibration characteristics to verify the effectiveness of pressure surges, effectively eliminating false alarms caused by non-breakage factors such as normal chain vibration or foreign object obstruction, reducing the false alarm rate from over 15% in existing technologies to below 3%.

[0023] Existing wireless monitoring solutions mostly employ unidirectional serial communication. A single point of disconnection leads to the loss of connection for all subsequent nodes, making it impossible to distinguish between node damage and chain breakage. Positioning accuracy can only be roughly estimated based on the signal interruption point. This embodiment constructs a bidirectional self-organizing network communication architecture, such as... Figure 4 As shown, after power-on, the nodes automatically perform network discovery and networking, establishing connections with neighboring nodes and edge computing gateways deployed at the head and tail of the conveyor via a handshake protocol, forming a wireless sensor network with multi-path transmission and self-healing capabilities. The intelligent monitoring nodes arranged along the scraper conveyor chain, together with the edge computing gateways deployed at the head and tail, constitute a Mesh self-organizing network. Nodes not only communicate with adjacent nodes but also establish connections with non-adjacent nodes through multi-hop routing, forming multiple redundant transmission paths. The network uses a TDMA (Time Division Multiple Access) mechanism to coordinate transmission time slots and combines frequency hopping technology to resist downhole electromagnetic interference. When a node or link fails, the network can automatically reconstruct routes, achieving self-healing and ensuring the reliability of data transmission.

[0024] S102, calculates the load rate, real-time operating speed and cumulative wear of the chain based on the multi-source time series dataset.

[0025] In this embodiment, the real-time load rate and real-time operating speed of the chain are calculated based on the pressure data and the signal time difference between adjacent nodes in the multi-source time-series dataset. The formula for calculating the load rate is as follows: in, For load rate, For the real-time load of the chain, Assuming the chain's rated load, the vibration signal time-series data is input into the wear quantization model. A fast Fourier transform is performed on the time-domain vibration signal to obtain the power spectral density function, thereby extracting the high-frequency energy ratio, vibration energy entropy, and impact pulse intensity. In this embodiment, the wear quantization model is an Arcard wear model used to construct a dynamic wear model for the sprocket. The core idea of ​​this model is that there is a definite mapping relationship between the degree of chain wear and the vibration characteristics generated during operation. As chain wear intensifies, the gap between chain links increases, and the meshing impact intensifies. These physical changes are reflected in the spectral structure and time-domain characteristics of the vibration signal in a quantifiable way. The high-frequency energy ratio is defined as the ratio of the energy in the 500Hz to 2000Hz frequency band to the total energy in the 10Hz to 2000Hz frequency band, reflecting the proportion of high-frequency impact components caused by wear. The vibration energy entropy is calculated by dividing the analysis frequency band into 16 sub-bands and then calculating the information entropy of the energy distribution in each sub-band, reflecting the degree of spectral structure complexity caused by wear. The impact pulse intensity is defined as the time integral of the ratio of the impact amplitude exceeding three times the effective value to the effective value in the time-domain signal, reflecting the severity of the transient impact generated by the worn chain links.

[0026] After calculating the instantaneous wear rate of the current time window using a preset linear weighted model, the cumulative wear of the chain is obtained by integrating the instantaneous wear rate on the time axis. The calculation formula is as follows: in, This represents the cumulative wear and tear at the current moment. This represents the cumulative wear and tear at the previous moment. Instantaneous wear rate, This is a load-velocity weighted factor, reflecting the influence of operating conditions on the wear rate. The sampling interval is... For high-frequency energy ratio, The sensitivity coefficient for high-frequency energy ratio. The vibrational energy entropy, The sensitivity coefficient is the vibration energy entropy. The intensity of the impact pulse. The sensitivity coefficient is the value of the impact pulse intensity. Based on the wear rate, For real-time load, This is the rated load of the chain. For real-time running speed, This is the rated operating speed.

[0027] S103 calculates the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient based on the load rate, real-time operating speed, cumulative wear, and temperature data in the multi-source time-series dataset.

[0028] In this embodiment, the calculation formulas for the load correction factor, speed correction factor, wear correction factor, and temperature correction factor are as follows: in, This is the load correction factor. For speed correction factor, This is a temperature correction factor. This is the wear correction factor. For load rate, For real-time running speed, The rated transport speed is the standard chain speed designed for normal operating conditions of the scraper conveyor. The local temperature of the chain. This serves as a reference threshold for cumulative wear. This represents the cumulative wear and tear.

[0029] S104, the dynamic threshold is determined based on the load correction factor, speed correction factor, wear correction factor and temperature correction factor.

[0030] In this embodiment, the threshold for chain breakage detection is not a fixed value, but a variable dynamically adjusted based on real-time operating conditions. The dynamic threshold is obtained by multiplying a base threshold by a series of correction coefficients reflecting current load, operating speed, ambient temperature, and cumulative wear. The calculation formula is as follows: in, For dynamic thresholds, Based on the threshold, This is the load correction factor. For speed correction factor, This is a temperature correction factor. This is the wear correction factor.

[0031] Based on real-time sensor data and historical operating data, these correction coefficients are periodically calculated and dynamic thresholds are updated. This adaptive mechanism effectively reduces false alarms caused by normal fluctuations in operating conditions, such as start-up, shutdown, and load changes, while ensuring monitoring sensitivity under abnormal conditions.

[0032] S105 classifies the chain status based on dynamic thresholds and real-time spring displacement, locates the chain breakage position of the chain at the breakage level, and uses a machine learning model to evaluate the status of the remaining chain levels in real time and provide chain breakage risk warnings.

[0033] See Figure 5 In this embodiment, the real-time spring displacement of the sensing spring is calculated by inverting the pressure sensor signal. The calculation formula is as follows: in, For spring displacement, This is the real-time spring force measured by the pressure sensor. The spring force under initial compression. The stiffness coefficient of the sensing spring is used to compare the real-time spring displacement with the dynamic threshold. When the real-time spring displacement is less than 0.8 times the dynamic threshold, it is classified as normal. When the real-time spring displacement is greater than 0.8 times the dynamic threshold but less than the dynamic threshold, it is classified as a concern level. When the real-time spring displacement is greater than the dynamic threshold, it is classified as a high-risk level. When the real-time spring displacement is greater than 1.2 times the dynamic threshold, it is classified as a breakage level. The sampling frequency and monitoring strategy are adjusted accordingly. The chain breakage location is located for the chain in the breakage level. At the same time, the status of the normal, concern, and high-risk level chains is evaluated in real time and a chain breakage risk warning is given through a machine learning model.

[0034] When the state level reaches the point of fracture occurrence, a pre-judgment of the fracture event is triggered, and a subsequent dual verification process is initiated. The verification is initiated through the logic AND gate judgment algorithm built into the microprocessor. First, it is confirmed that the pressure sensor signal shows that the spring relaxation exceeds the calculated dynamic threshold. Second, within a preset very short time, it is confirmed that the vibration sensor detects an impact pulse that is more than ten times the normal value. Only when the above two conditions are met simultaneously within the time window is it confirmed as a real chain breakage event, and an emergency alarm is immediately issued through a high-priority channel. Otherwise, the event is judged as a false trigger caused by non-fracture factors.

[0035] See Figure 6 In this embodiment, the chain breakage location for the chain at the breakage level includes: identifying the last communication node using network topology, determining the initial chain breakage interval based on the last communication node, sending high-speed sampling commands to three nodes upstream and downstream of the initial chain breakage interval, collecting multi-node data and transmitting it to the cloud, and performing feature extraction and fusion analysis on the multi-node data received by the cloud, calculating the vibration energy ratio and temperature gradient of each node; and finding nodes that meet preset conditions, which are: The nodal vibration energy ratio is greater than 10 and the temperature gradient is greater than 5 degrees Celsius. The vibration energy ratio of each node is less than 2. Based on the vibration energy ratio and temperature gradient of each node, the chain break location is accurately determined, a location report is generated, and pushed to the terminal. The entire positioning process can be completed within 50 seconds, with an accuracy of ±0.5-1 meter. The break location is determined by weighted fusion of the following five positioning information methods: Principle 1: Signal interruption positioning. Chain breakage causes nodes behind the break point to lose power or communication links. Identifying the last communication node allows for a preliminary determination of the chain break range. Principle 2: Spring tension detection positioning. Breakage increases the distance between adjacent chain links, causing the sensing springs close to the break area to tension, resulting in a decrease in the pressure sensor signal. Principle 3: Vibration shock wave positioning. The shock wave generated at the moment of chain breakage propagates along the chain to both ends. Analyzing the time difference of the vibration wave reaching each node allows for precise calculation of the wave source location. Principle 4: Received signal strength indication gradient positioning. Chain breakage increases the physical distance between adjacent nodes, causing a step decrease in the received signal strength indication value of the wireless signal. Principle 5: Temperature gradient positioning. The intense friction at the moment of chain breakage generates localized high temperatures, and analyzing the temperature gradient of adjacent nodes can help verify the location of the break. By utilizing the physical characteristics of chain breakage and the physical aspects of wireless communication (signal strength attenuation, timestamps) for multimodal fusion, a technological leap has been achieved from communication interruption alarm to precise physical breakage location.

[0036] See Figure 7 The machine learning model architecture diagram provided in this embodiment divides the multi-source time-series dataset into 60-second time windows when the chain status is classified into normal, attention, or high-risk levels. Data within each window is normalized and input into the Long Short-Term Memory (LSTM) network of a pre-trained deep learning model integrating a LSM network and a random forest. The LSM network extracts deep dynamic features from the multi-source time-series data, capturing the temporal evolution of fault symptoms and outputting a 128-dimensional feature vector. This 128-dimensional feature vector is then concatenated with manually designed statistical features to form a fused feature vector. These manually designed statistical features include mean and variance of pressure, effective vibration value and kurtosis, and temperature change rate. This fused feature vector is input into a random forest composed of 100 decision trees. A majority voting mechanism is used for binary classification as normal or abnormal, and the arithmetic mean of the prediction results from each decision tree is used to output a continuous health score and a probability of breakage risk. The node splitting of the random forest uses a mean squared error minimization criterion, and the importance ranking of each feature is output to enhance model interpretability. The hybrid architecture is deployed on a cloud platform, while a lightweight version is deployed on the edge gateway to ensure real-time responsiveness.

[0037] See Figure 8In this embodiment, the health monitoring interface provides real-time status display, early warning push notifications, and maintenance guidance. The interface uses green, yellow, and red to distinguish normal, warning, and fault nodes, with broken links highlighted by flashing. The panel displays real-time health scores, a list of warning events, and current operating parameters. The bottom area automatically generates maintenance suggestions based on the location results, including the specific location, repair path, and required spare parts. This interface can be deployed on a large screen in the wellhead monitoring center or on a mobile terminal for maintenance personnel, enabling real-time information visualization and decision support.

[0038] Corresponding to the above embodiment of the intelligent health monitoring method for scraper conveyor chains that integrates multi-source sensing and deep learning, this application also provides an embodiment of an intelligent health monitoring system for scraper conveyor chains that integrates multi-source sensing and deep learning.

[0039] This application provides an intelligent health monitoring system for scraper conveyor chains that integrates multi-source sensing and deep learning, including: The multi-source data acquisition module is used to collect and preprocess multi-source data in real time during the chain operation through multiple intelligent monitoring nodes to form a multi-source time-series dataset.

[0040] The first calculation module is used to calculate the chain's load rate, real-time operating speed, and cumulative wear based on multi-source time-series datasets.

[0041] The second calculation module calculates the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient based on the load rate, real-time operating speed, cumulative wear amount, and temperature data in the multi-source time-series dataset.

[0042] The determination module is used to determine the dynamic threshold based on the load correction factor, speed correction factor, wear correction factor, and temperature correction factor.

[0043] The health monitoring module classifies the chain's condition based on dynamic thresholds and real-time spring displacement, locates the breakage position of chains at the breakage level, and uses a machine learning model to evaluate the condition of the remaining chains in real time and provide early warnings of chain breakage risk.

[0044] The entire system is designed to integrate with existing programmable logic controllers, monitoring and data acquisition systems, and production execution systems in the mine. By providing standardized data interfaces and communication protocols, this monitoring system can seamlessly integrate chain health status, early warning information, and maintenance suggestions into the mine's overall intelligent management platform. This provides crucial data support for predictive maintenance and intelligent production decision-making, ultimately achieving the core objectives of ensuring safe production, improving operational efficiency, and reducing maintenance costs.

[0045] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0046] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0047] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for intelligent health monitoring of a scraper conveyor chain by fusing multi-source sensing and deep learning, characterized in that, include: Multiple intelligent monitoring nodes are used to collect and preprocess multi-source data during the chain operation process in real time to form a multi-source time series dataset. The load rate, real-time operating speed, and cumulative wear of the chain are calculated based on the multi-source time-series dataset. The load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient are calculated based on the load rate, real-time operating speed, cumulative wear amount, and temperature data in the multi-source time series dataset. The dynamic threshold is determined based on the load correction factor, speed correction factor, wear correction factor, and temperature correction factor. The chain status is classified based on the dynamic threshold combined with real-time spring displacement, and the chain breakage location is located for chains at the breakage level. At the same time, the status of the remaining chains is evaluated in real time and a chain breakage risk warning is given through a machine learning model.

2. The method of claim 1, wherein, The process involves real-time collection and preprocessing of multi-source data during the chain operation through multiple intelligent monitoring nodes to form a multi-source time-series dataset, including: Multiple intelligent monitoring nodes, installed at equal intervals in the grooves of the chain connecting rings of the scraper conveyor, collect pressure data, triaxial vibration data, temperature data, and node running time data in real time during the chain operation process. The collected pressure data, triaxial vibration data, temperature data, and node running time data are sliced ​​and normalized in a fixed time window to form a multi-source time series dataset.

3. The method of claim 1, wherein, The load rate, real-time operating speed, and cumulative wear of the chain are calculated based on the multi-source time-series dataset, including: The real-time load rate and real-time operating speed of the chain are calculated based on the pressure data and the signal time difference between adjacent nodes in the multi-source time-series dataset. The formula for calculating the load rate is as follows: in, For load rate, For the real-time load of the chain, This is the rated load of the chain; The vibration signal time series data is input into the wear quantization model, and the time-domain vibration signal is subjected to fast Fourier transform to obtain the power spectral density function, and then the high-frequency energy ratio, vibration energy entropy and impact pulse intensity are extracted. The instantaneous wear rate of the chain is calculated by using a preset linear weighted model to calculate the instantaneous wear rate in the current time window, and then the instantaneous wear rate is integrated on the time axis to obtain the cumulative wear amount of the chain.

4. The intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning according to claim 3, characterized in that, After calculating the instantaneous wear rate of the current time window using a preset linear weighted model, the instantaneous wear rate is integrated over the time axis to obtain the formula for calculating the cumulative wear of the chain: in, This represents the cumulative wear and tear at the current moment. This represents the cumulative wear and tear at the previous moment. Instantaneous wear rate, This is the load-velocity weighting factor. The sampling interval is... For high-frequency energy ratio, The sensitivity coefficient for high-frequency energy ratio. For vibrational energy entropy, The sensitivity coefficient is the vibration energy entropy. The intensity of the impact pulse. The sensitivity coefficient is the value of the impact pulse intensity. Based on the wear rate, For real-time load, This is the rated load of the chain. For real-time running speed, This is the rated operating speed.

5. The intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning according to claim 1, characterized in that, The formulas for calculating the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient based on the load rate, real-time operating speed, cumulative wear, and temperature data in the multi-source time-series dataset are as follows: in, This is the load correction factor. For speed correction factor, This is the temperature correction factor. This is the wear correction factor. For load rate, For real-time running speed, For the rated transport speed, The local temperature of the chain. This serves as a reference threshold for cumulative wear. This represents the cumulative wear and tear.

6. The intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning according to claim 1, characterized in that, The calculation formula for determining the dynamic threshold based on the load correction factor, speed correction factor, wear correction factor, and temperature correction factor is as follows: in, For dynamic thresholds, Based on the threshold, This is the load correction factor. For speed correction factor, This is the temperature correction factor. This is the wear correction factor.

7. The intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning according to claim 1, characterized in that, The chain state is classified based on the dynamic threshold combined with real-time spring displacement, and the breakage location of chains at the breakage level is located. Simultaneously, a machine learning model is used to perform real-time assessment of the state of the remaining chains and provide breakage risk warnings, including: The real-time spring displacement of the sensing spring is calculated by inverting the pressure sensor signal. The calculation formula is as follows: in, For spring displacement, This is the real-time spring force measured by the pressure sensor. The spring force under initial compression. This is the stiffness coefficient of the sensing spring; Compare the real-time spring displacement with the dynamic threshold; When the real-time spring displacement is less than the first preset dynamic threshold, it is classified as normal. When the real-time spring displacement is greater than the first preset dynamic threshold but less than the dynamic threshold, it is classified as a level of concern; When the real-time spring displacement is greater than the dynamic threshold, it is classified as a high-risk level; When the real-time spring displacement is greater than the second preset dynamic threshold, it is classified as a fracture occurrence level; The system locates the breakage position of chains at different breakage levels, and uses a machine learning model to conduct real-time assessments and provide early warnings of breakage risks for chains at normal, watch, and high-risk levels.

8. The intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning according to claim 7, characterized in that, Locating the breakage location of chains of varying breakage levels includes: The last communication node is identified using the network topology, and the initial disconnection interval is determined based on the last communication node. High-speed sampling commands are sent to multiple nodes upstream and downstream of the initial chain break interval, and data from multiple nodes is collected and transmitted to the cloud. After receiving the multi-node data, the cloud performs feature extraction and fusion analysis on the multi-node data, and calculates the vibration energy ratio and temperature gradient of each node; The location of the chain break is accurately determined based on the vibration energy ratio and temperature gradient of each node.

9. The intelligent health monitoring method for scraper conveyor chains integrating multi-source sensing and deep learning according to claim 7, characterized in that, The machine learning model is used to perform real-time assessments of the status of normal, monitored, and high-risk chains and to provide early warnings of chain breakage risks, including: When the chain status is classified into normal level, attention level or high risk level, the multi-source time series dataset is input in real time into the long short-term memory network in the deep learning model that integrates the pre-trained long short-term memory network and random forest. Long Short-Term Memory (LSTM) networks are used to extract deep dynamic features from multi-source time-series data, capture the temporal evolution of fault symptoms, and output multi-dimensional feature vectors. The multidimensional feature vector is concatenated with manually designed statistical features to form a fused feature vector; Random forest is used to classify and regress the fused feature vectors, and the output is a multi-dimensional evaluation result including the real-time health score of the chain, the probability of breakage risk, and the prediction of wear trend. When the health score falls below a preset threshold or the risk probability exceeds a critical value, a tiered warning is issued to enable proactive intelligent operation and maintenance decision-making before the chain breaks.

10. A smart health monitoring system for scraper conveyor chains integrating multi-source sensing and deep learning, characterized in that, include: The multi-source data acquisition module is used to collect and preprocess multi-source data in real time during the chain operation through multiple intelligent monitoring nodes to form a multi-source time series dataset. The first calculation module is used to calculate the load rate, real-time operating speed and cumulative wear of the chain based on the multi-source time-series dataset. The second calculation module calculates the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient based on the load rate, real-time operating speed, cumulative wear amount, and temperature data in the multi-source time-series dataset. The determination module is used to determine the dynamic threshold based on the load correction coefficient, speed correction coefficient, wear correction coefficient, and temperature correction coefficient; The health monitoring module classifies the chain's state based on the dynamic threshold and real-time spring displacement, locates the breakage position of chains at the breakage level, and uses a machine learning model to evaluate the state of the remaining chains in real time and provide early warning of chain breakage risk.

Images