A safety coal conveying belt intelligent inspection system

Abstract

The application discloses a kind of safety coal conveying with coal conveying belt intelligent inspection system, including data acquisition module from multiple data sources collection data;Relay data processing and analysis module is used to determine and handle the temperature abnormal rise data detected, data can be transmitted to the device where next sensor is located;Data preprocessing module, the data collected are standardized, and accurate timestamp and spatial location information are marked for each data record;Space-time alignment module, the data from different sources after being processed by data preprocessing module are accurately aligned in time dimension;Identify data with spatial attributes, and establish the spatial relationship between different data points;Improved Transform model, process the data after being processed by space-time alignment module;Fault prediction and result display module predicts fault, and intuitively displays fault location.The application breaks through traditional single sensor threshold detection, realizes composite fault accurate positioning.

Description

Technical Field

[0001] This invention relates to the field of safe coal conveying technology, specifically to an intelligent inspection system for safe coal conveying conveyor belts. Background Technology

[0002] In the coal transportation sector, traditional manual inspections suffer from low efficiency, limited coverage, and a high risk of missed inspections. Furthermore, the long distances and diverse operating conditions of coal conveyor belts increase the risk of fires, wire breakage, belt tears, and belt misalignment. Existing monitoring systems often rely on single-point sensors or single monitoring methods, making it difficult to meet the demands for multi-source heterogeneous data fusion and analysis in complex environments. Summary of the Invention

[0003] Therefore, the present invention provides an intelligent inspection system for safe coal conveying belts to solve the problems in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] An intelligent inspection system for safe coal conveying belts includes a data acquisition module, a data preprocessing module, a spatiotemporal alignment module, an improved Transformer model, a relay data processing and analysis module, and a fault prediction and result display module.

[0006] The data acquisition module collects data from multiple data sources;

[0007] The relay data processing and analysis module is installed on mobile inspection equipment, fixed monitoring equipment at bends, and lower conveyor belt monitoring equipment; it is used to identify and process detected abnormal temperature rise data and can transmit the data to the equipment where the next sensor is located.

[0008] The data preprocessing module cleans the collected data, removing noise, outliers, and missing values; it standardizes the source data, mapping it to a set numerical range, and labels each data record with an accurate timestamp and spatial location information to ensure that subsequent spatiotemporal alignment operations can be performed.

[0009] The spatiotemporal alignment module precisely aligns the data from different sources after processing by the data preprocessing module in the time dimension, ensuring that the timestamps of each data source are consistent; it also identifies data with spatial attributes and establishes spatial relationships between different data points.

[0010] The improved Transformer model processes data after it has been processed by the spatiotemporal alignment module. The improved Transformer model includes an input layer, a multi-head self-attention mechanism layer, a feedforward neural network layer, residual connections, and an output layer.

[0011] The fault prediction and result display module predicts the fault based on the output result of the improved Transformer model and intuitively displays the fault position.

[0012] Further, the data collected by the data collection module includes data of various sensors, device log data, and surrounding environment monitoring data; various sensors are integrated on a plurality of mobile inspection devices, turning fixed monitoring devices, and lower belt monitoring devices.

[0013] Further, the mobile inspection device is erected above the coal conveying belt. Since the conveying distance of the coal conveying belt is very long, each long slope section and horizontal conveying section is divided into a plurality of small sections, and a reciprocating mobile inspection device is arranged in each small section. A turning fixed monitoring device is erected at each turning of the coal conveying belt. A plurality of lower belt monitoring devices are arranged above the lower belt of the coal conveying belt to monitor the condition of the empty belt.

[0014] Further, the next sensor re-detects the belt point with abnormally high temperature detected for the first time. When the next sensor still detects abnormally high temperature after comparison, the device where the next sensor is located transmits the comparison data to the device where the next sensor is located. The next sensor re-detects the point, and when it is confirmed that the temperature is indeed abnormally high and exceeds the set threshold, the device where the next sensor is located directly determines that the point is burning with hidden fire or open fire, transmits the data directly to the fault prediction and result display module, and performs fire warning or alarm. The data is also transmitted to the data collection module.

[0015] When the data of the next comparison abnormal point initial sensor is compared, it is found that the temperature of the temperature abnormality point is decreasing or the abnormality of the point disappears. Then, the comparison data is sent to the subsequent two sensors, and when it is found that the temperature abnormality point indeed disappears, the data transmission is ended.

[0016] When one of the subsequent two sensors detects abnormally high temperature again, the sensor that detects the abnormally high temperature again is taken as the abnormal point initial sensor, and the cycle is restarted to repeat the previous detection and comparison. At the same time, the data is recorded and transmitted to the fault prediction and result display module as suspicious fire warning.

[0017] Further, the space-time alignment module mainly realizes its function through time alignment and space alignment.

[0018] Time alignment, according to the timestamp information of the data processed by the data preprocessing module, all data are converted to a unified time reference; for time series data, resampling technology is used to unify the sampling rate of different data; for event-driven data, sorting alignment is performed according to the sequence of event occurrence; a time alignment matrix is constructed to represent the correspondence of different data sources at each time point;

[0019] Spatial alignment, for data with spatial position attributes, a spatial coordinate system is established to map the data of different devices or monitoring points to the coordinate system; spatial interpolation algorithm is used to encrypt sparse spatial data, making the data distribution in the whole monitoring area more uniform, facilitating spatial feature extraction and fusion; and the spatial correlation matrix between data is calculated to measure the similarity or correlation degree between different position data, providing weight basis for subsequent spatial feature fusion.

[0020] Further, the input layer takes the multi-source heterogeneous data after time and space alignment as the input of the improved Transformer model, and organizes the data into a suitable input tensor form according to the feature dimension of the data;

[0021] Multi-head self-attention mechanism layer, multi-head self-attention mechanism is introduced, which allows the improved Transformer model to simultaneously focus on different position information and feature information when processing input data; each head calculates the attention weight, then performs weighted summation to obtain the output, and finally splices the outputs of all heads;

[0022] Feedforward neural network layer, connected after the multi-head self-attention mechanism layer is the feedforward neural network FFN, which consists of two fully connected layers with ReLU activation function in between; the feedforward neural network FFN is used for further feature transformation and nonlinear mapping of the output of the self-attention mechanism;

[0023] Residual connection, a residual connection is added between the multi-head self-attention mechanism layer and the feedforward neural network layer; the residual connection directly adds the input to the output;

[0024] Output layer, through a fully connected layer, the fault type, confidence and position coordinates of the fault probability graph are output, and the number of neurons of the output layer is set according to the classification number of fault types, the quantization level of confidence and the accuracy requirement of position coordinates.

[0025] Further, the model training of the improved Transformer model mainly uses the following methods;

[0026] 1) Loss function design: design a comprehensive loss function, including cross-entropy loss for fault type prediction, mean square error loss for confidence prediction, and Euclidean loss for position coordinate prediction; by combining these loss terms, balance the learning goals of different tasks, guide the model to optimize fault diagnosis accuracy, confidence evaluation rationality and position positioning accuracy during training;

[0027] 2) Optimization algorithm selection: use Adam optimization algorithm for model parameter update, Adam optimization algorithm combines the advantages of momentum method and adaptive learning rate, which can automatically adjust the learning rate during training, accelerate the convergence speed and improve the stability of the model;

[0028] 3) Data division: divide the preprocessed data set into training set, validation set and test set according to certain proportion; the training set accounts for 70% - 80% of the total data set, which is used for model training; the validation set accounts for 10% - 15%, which is used to adjust the model hyperparameters and monitor whether the training process is overfitting; the test set accounts for 10% - 15%, which is used to evaluate the final performance of the model; in the training process, stratified sampling or random sampling method is used to ensure that each batch of data can fully represent the distribution characteristics of the whole data set.

[0029] Further: use the time correlation matrix and space correlation matrix calculated in the space-time alignment process to provide prior knowledge for the multi-head self-attention mechanism, guide the improved Transformer model to pay more attention to the space-time related data features when feature fusion;

[0030] In the output results of the multi-head self-attention mechanism, the feature vectors of different data sources are weighted and summed according to the attention weights to obtain the fused feature representation;

[0031] In the different levels or different modules of the improved Transformer model, different scale features are processed and fused respectively.

[0032] Further: according to the fault type probability distribution output by the improved Transformer model, the threshold judgment or Top-K selection method is used to determine the most possible fault type.

[0033] Further: when displaying the prediction results, for each predicted fault point, mark its position on the coal conveying belt map with different colors or symbols, and display the corresponding fault type and confidence information, and mark the fault point position in real time according to the belt conveying speed, and display it on the coal conveying belt.

[0034] This invention has the following advantages: It breaks through the traditional single-sensor threshold detection and achieves accurate location of complex faults; through the spatiotemporal alignment module and the improved Transformer model, it effectively integrates multimodal data such as infrared thermal imaging and laser scanning, and improves the early identification capability of complex faults; the relay data processing module compares abnormal data step by step through multiple sensors, which significantly reduces the false alarm rate caused by environmental noise and reduces the amount of data processed by the data acquisition module.

[0035] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. Attached Figure Description

[0036] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0037] Fig. 1 This is a system block diagram of an intelligent inspection system for a safe coal conveying belt, provided as an embodiment of this application.

[0038] Fig. 2 This is a schematic diagram showing the position and structure of the inspection equipment on the coal conveyor belt in an intelligent inspection system for safe coal conveying according to the present invention.

[0039] Fig. 3 This is a cross-sectional structural diagram showing the installation positions of the mobile inspection equipment and the fixed monitoring equipment on the lower conveyor belt in the intelligent inspection system for coal conveyor belts of the present invention.

[0040] In the picture:

[0041] 1. Coal conveyor belt; 2. Mobile inspection equipment; 3. Fixed monitoring equipment at bends; 4. Fixed monitoring equipment for the lower conveyor belt; 5. Drive mechanism. Detailed Implementation

[0042] The following embodiments of the present application are illustrated by way of specific examples, and other advantages and effects of the present application will be readily understood by those skilled in the art from the disclosure of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. It should be understood that these embodiments are only for further illustrating the present application, and cannot be understood as limiting the scope of protection of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] Please refer to Figs. 1-3 A safety coal conveying belt intelligent inspection system for coal conveying, comprising a data acquisition module, a data preprocessing module, a space-time alignment module, an improved Transformer model, a relay data processing and analysis module, and a fault prediction and result display module.

[0044] The data acquisition module collects relevant data from multiple data sources, including but not limited to various sensor data, device log data, and surrounding environment monitoring data (such as humidity, electromagnetic interference, etc.).

[0045] A plurality of sensors are integrated in a plurality of mobile inspection devices 2, turning fixed monitoring devices 3, and lower belt fixed monitoring devices 4.

[0046] The mobile inspection device 2 is erected above the coal conveying belt 1, and a driving mechanism 5 is provided on both sides of the mobile inspection device 2. The driving mechanism 5 can move the mobile inspection device 2 on the coal conveying belt 1. Since the conveying distance of the coal conveying belt 1 is very long, each long slope section and horizontal conveying section is divided into several small sections, and a reciprocating mobile inspection device 2 is provided in each small section. A turning fixed monitoring device 3 is erected at each turning of the coal conveying belt 1. And a plurality of lower belt fixed monitoring devices 4 are provided above the lower belt of the coal conveying belt 1 to monitor the condition of the empty belt.

[0047] The sensors on the mobile inspection device 2, the turning fixed monitoring device 3, and the lower belt fixed monitoring device 4 mainly include:

[0048] A high-frame-rate infrared thermal imager (resolution 0.05℃) is used to detect temperature anomalies on the surface of the belt;

[0049] A linear laser scanner (accuracy ±0.1mm) is used to measure the belt deviation in real time;

[0050] An acoustic emission sensor array (frequency range 20kHz-1MHz) is used to capture internal steel wire rope fracture signals of the belt;

[0051] Multispectral camera (covering visible light + near-infrared band), identifying surface oil, foreign matter and micro-cracks.

[0052] Relay data processing and analysis module, arranged on the mobile inspection device 2, the turning fixed monitoring device 3 and the lower belt fixed monitoring device 4; mainly used for determining and processing the temperature abnormally high data (abnormal point initial sensor data) detected by the high frame rate infrared thermal imager, and transmitting the data to the device where the next sensor is located;

[0053] The second sensor detects the belt point detected by the abnormal point initial sensor again; after comparison, if the second sensor still detects temperature abnormally high, the device where the second sensor is located transmits the comparison data to the device where the third sensor is located; the third sensor detects the point again, and when it is confirmed that the temperature is indeed abnormally high and exceeds the set threshold, the device where the third sensor is located directly determines that the point is burning with hidden fire or open fire, transmits the data directly to the fault prediction and result display module, and gives a fire warning or alarm; at the same time, the data is transmitted to the data acquisition module.

[0054] When the second sensor compares the data of the abnormal point initial sensor and finds that the temperature of the temperature abnormally high point is decreasing or the abnormal high data of the point disappears, the comparison data is transmitted to the subsequent two sensors for three times of comparison, and when it is found that the temperature abnormally high point indeed disappears, the data transmission is ended.

[0055] When one of the subsequent two sensors detects temperature abnormally high again, the sensor that detects temperature abnormally high again is taken as the abnormal point initial sensor, and the above process is repeated; at the same time, the data is recorded and transmitted to the fault prediction and result display module as a suspicious fire warning.

[0056] For example, the sensor of the mobile inspection device 2 at point A detects that the point X of the belt is abnormal, and the data of point A is transmitted to the device at point B (mobile inspection device 2 / turning fixed monitoring device 3 / lower belt fixed monitoring device 4), the device at point B detects point X again, and when it is detected that the temperature is abnormally high again, the data of points A and B is transmitted to the device at point C, and the device at point C detects point X again; when it is confirmed that the temperature is abnormal, the device at point C directly transmits the data to the fault prediction and result display module; when point B does not detect temperature abnormality, it is confirmed again by points C and D, and when the temperature is normal, the data transmission is ended and the data is recorded; when point B detects that the temperature decreases or the abnormal temperature disappears, and point C detects that the temperature increases again, point C is reset as the sensor point where the abnormal temperature is first detected, and then the previous operation is repeated, the data of point C is uploaded to the fault prediction and result display module, and a suspicious fire warning is given.

[0057] The data preprocessing module cleans the collected data, removes noise, outliers and missing values, and uses appropriate interpolation methods or filling strategies based on statistical rules to handle missing data.

[0058] Due to the large difference in data dimension and value range of different data sources, standardization processing is performed on each source data, such as Z-score standardization or Min-Max standardization, to map the data to a set value interval, so as to improve the stability and convergence speed of model training.

[0059] In addition, accurate timestamps and spatial location information are labeled for each data record to ensure subsequent spatio-temporal alignment operations. For data that cannot directly obtain spatial location, the relative position can be determined through association with device installation location or other indirect ways; for example, the spatial position of the sensor data can be determined according to the conveying speed of the belt, combined with the section where the mobile inspection device 2 is located, and the moving speed of the mobile inspection device 2 in the section.

[0060] The spatio-temporal alignment module accurately aligns data from different sources (such as device operation data, sensor monitoring data, etc.) in the time dimension, ensuring that the timestamps of each data source are consistent, so that subsequent analysis can be based on a unified time reference; for data with spatial attributes (such as sensor data), the spatio-temporal alignment module can identify and establish spatial relationships between different data points.

[0061] The spatio-temporal alignment module mainly realizes its functions through time alignment and space alignment.

[0062] Time alignment: according to the timestamp information of the data processed by the data preprocessing module, all data is converted to a unified time reference; for time series data, resampling technology is used to unify the sampling rate of different data; for event-driven data, sorting alignment is performed according to the sequence of event occurrence.

[0063] Construct a time alignment matrix to represent the correspondence of different data sources at each time point, so that the same time data can be accurately combined during subsequent fusion operations.

[0064] Spatial alignment: for data with spatial location attributes, a spatial coordinate system (such as a two-dimensional plane coordinate system or a three-dimensional space coordinate system) is established, and the data of different devices or monitoring points is mapped into the coordinate system.

[0065] The sparse spatial data is encrypted by using a spatial interpolation algorithm, so that the data distribution in the entire monitoring area is more uniform, facilitating spatial feature extraction and fusion; and a spatial correlation matrix between the data is calculated to measure the similarity or correlation degree between data at different positions, providing a weight basis for subsequent spatial feature fusion.

[0066] The improved Transformer model processes the data processed by the spatio-temporal alignment module, and the improved Transformer model includes an input layer, a multi-head self-attention mechanism layer, a feedforward neural network layer, a residual connection, and an output layer.

[0067] The input layer: the multi-source heterogeneous data after spatio-temporal alignment is taken as the input of the improved Transformer model, and the data is organized into a suitable input tensor form according to the feature dimension of the data. For example, for each combination of time points and spatial positions, the corresponding multi-source data features are spliced into a vector as an element of the input.

[0068] The multi-head self-attention mechanism layer: the multi-head self-attention mechanism is introduced, which allows the improved Transformer model to simultaneously focus on different position information and feature information when processing input data; each head calculates the attention weight, then performs weighted summation to obtain the output, and finally splices the outputs of all heads; through the multi-head self-attention mechanism, the improved Transformer model can automatically learn the complex dependency between data, including spatio-temporal correlation and feature interaction.

[0069] The feedforward neural network layer: a feedforward neural network (FFN) is connected after the multi-head self-attention mechanism layer, which consists of two fully connected layers with a ReLU activation function in between; FFN is used to further transform and nonlinearly map the results output by the self-attention mechanism, enhancing the expression ability of the model.

[0070] Residual connection: in order to alleviate the problem of gradient disappearance and accelerate the model training convergence of the improved Transformer model, a residual connection is added between the multi-head self-attention mechanism layer and the feedforward neural network layer; the residual connection directly adds the input to the output, making it easier for the improved Transformer model to optimize the gradient flow in the deep network.

[0071] The output layer: after the above structure is stacked for multiple layers, the three-dimensional information of the fault probability map, i.e. fault type, confidence and position coordinates, is finally output through a fully connected layer. The number of neurons in the output layer is set according to the classification number of fault types, the quantization level of confidence, and the accuracy requirement of position coordinates, and the Softmax function is used to predict the probability distribution of fault types and normalize the confidence.

[0072] The model training of the improved Transformer model mainly adopts the following methods;

[0073] 1) Loss function design: design a comprehensive loss function, including cross-entropy loss for fault type prediction, mean square error loss for confidence prediction, and Euclidean loss for position coordinate prediction. By combining these loss terms by weighting, the learning goals of different tasks are balanced, guiding the model to optimize fault diagnosis accuracy, confidence evaluation rationality and position positioning accuracy during training.

[0074] 2) Optimization algorithm selection: use Adam optimization algorithm for model parameter update. Adam optimization algorithm combines the advantages of momentum method and adaptive learning rate, which can automatically adjust the learning rate during training, accelerate the convergence speed and improve the stability of the model. Set appropriate initial learning rate, learning rate decay strategy and training round number and other hyperparameters, and adjust and optimize according to the performance on the validation set.

[0075] 3) Data division: divide the preprocessed data set into training set, validation set and test set according to certain proportion. Usually, the training set accounts for 70%-80% of the total data set, which is used for model training; the validation set accounts for 10%-15%, which is used to adjust model hyperparameters and monitor whether the training process is overfitting; the test set accounts for 10%-15%, which is used to evaluate the final performance of the model. In the training process, stratified sampling or random sampling method is used to ensure that each batch of data can fully represent the distribution characteristics of the whole data set.

[0076] In addition, the time correlation matrix and space correlation matrix calculated in the space-time alignment process are used to provide prior knowledge for the multi-head self-attention mechanism, guiding the improved Transformer model to pay more attention to the space-time related data features during feature fusion. For example, when calculating the attention weight, the attention score can be weighted and adjusted according to the time interval and spatial distance, so that the data features of adjacent time points and similar spatial positions obtain higher weights;

[0077] In the output results of the multi-head self-attention mechanism, the feature vectors of different data sources are weighted and summed according to the attention weight to obtain the fused feature representation; this feature fusion method based on attention mechanism can dynamically adapt to the internal structure and correlation of the data, highlighting important feature information and suppressing irrelevant noise.

[0078] Considering that multi-source heterogeneous data may have different feature scales (such as time scale, space scale and feature value scale), a multi-scale feature fusion strategy is adopted; different scales of features are processed and fused in different levels or modules of the improved Transformer model.

[0079] For example, short-term feature extraction and fusion can be performed on time series data at the bottom layer, mid-scale fusion can be performed by combining spatial features at the middle layer, and large-scale feature fusion can be performed by considering global spatio-temporal information at the high layer to generate the final composite fault probability map.

[0080] The fault prediction and result display module predicts faults based on the output results of the improved Transformer model, and after confidence evaluation and calibration, visualizes the location coordinates on a two-dimensional or three-dimensional spatial map, and can use geographic information system (GIS) software or professional visualization tool libraries (such as Matplotlib, Plotly, etc.) to realize intuitive display of fault locations.

[0081] Fault prediction mainly determines the most likely fault type by threshold judgment or Top-K selection based on the probability distribution of fault types output by the improved Transformer model.

[0082] For example, set a probability threshold (such as 0.5), when the probability of a certain fault type exceeds the threshold, determine that the fault type occurs; or select the top K fault types with the highest probability as the candidate fault set, and further analyze and judge in combination with domain expert knowledge or actual maintenance experience.

[0083] Confidence evaluation and calibration post-process the confidence results of model prediction, and use the Platt Scaling calibration method to convert them into confidence estimates that are more consistent with actual situations.

[0084] Platt Scaling fits the relationship between model prediction probabilities and true labels on the validation set to obtain a calibration function, which is used to adjust the confidence prediction results on the test set to make them more reliably reflect the uncertainty of the model.

[0085] When displaying the prediction results, for each predicted fault point, mark its location on the map with different colors or symbols, and display the corresponding fault type and confidence information, so that users can quickly understand the location and possibility of fault occurrence.

[0086] In actual application scenarios, the model needs to be continuously monitored for performance changes after deployment; as equipment ages, operating environments change, and new fault patterns emerge, the performance of the model may gradually decline; therefore, it is necessary to regularly collect new data and retrain and update the model; by continuously iterating and optimizing the model, it can adapt to changing actual needs and maintain high fault diagnosis accuracy and reliability.

[0087] The application effectively integrates multi-modal data such as infrared thermal imaging and laser scanning through the space-time alignment module and the improved Transformer model, and improves the early identification capability of complex faults (such as dark fire and steel wire fracture).

[0088] The above merely describes preferred embodiments of the application and is not intended to limit the application. Any modification, equivalent replacement and improvement made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A coal conveying belt intelligent inspection system for safe coal conveying, characterized in that, The application relates to a coal belt monitoring system. The system comprises: a data acquisition module for acquiring data from multiple data sources, including data from various sensors, device log data and surrounding environment monitoring data; various sensors are integrated on a plurality of mobile inspection devices, turning fixed monitoring devices and lower belt monitoring devices; a relay data processing and analysis module is arranged on the mobile inspection devices, the turning fixed monitoring devices and the lower belt monitoring devices, and is used for determining and processing detected temperature abnormally high data, and can transmit the data to a device where the next sensor is located; the relay data processing and analysis module can also transmit the data to the data acquisition module or a fault prediction and result display module according to the data; a data preprocessing module is used for cleaning, removing noise, abnormal values and missing values of the acquired data, and standardizing each source data to map the data to a set value interval, and labeling accurate time stamps and spatial position information for each data record; a time-space alignment module is used for accurately aligning data from different sources in the time dimension after the data preprocessing module processes the data, and ensuring that the time stamps of the data sources are consistent; the time-space alignment module is also used for identifying data with spatial attributes, and establishing spatial relationships between different data points; an improved Transformer model is used for processing the data after the time-space alignment module processes the data, and the improved Transformer model comprises an input layer, a multi-head self-attention mechanism layer, a feedforward neural network layer, a residual connection and an output layer; a fault prediction and result display module is used for predicting faults and intuitively displaying fault positions based on the output results of the improved Transformer model; The time-space alignment module realizes its functions through time alignment and space alignment. Time alignment is used for converting all data to a unified time reference according to the time stamp information of the data after the data preprocessing module processes the data; for time series data, a resampling technology is used to unify the sampling rate of the data with different sampling frequencies; for event-driven data, the data is sorted and aligned according to the sequence of event occurrence; a time alignment matrix is constructed to represent the corresponding relationship of different data sources at each time point; Space alignment is used for establishing a spatial coordinate system to map data of different devices or monitoring points to the coordinate system for data with spatial position attributes; 2. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, A spatial interpolation algorithm is used to encrypt sparse spatial data, so that the data distribution in the whole monitoring area is more uniform, and spatial feature extraction and fusion are facilitated; and a spatial correlation matrix of the data is calculated to measure the similarity or correlation degree between data at different positions, and to provide a weight basis for subsequent spatial feature fusion. The mobile inspection device is arranged above a coal belt, and since the conveying distance of the coal belt is very long, each long slope section and horizontal conveying section is divided into a plurality of small sections, and a mobile inspection device reciprocally moves in each small section; a turning fixed monitoring device is separately arranged at each turning of the coal belt; and a plurality of lower belt monitoring devices are arranged above lower belts of the coal belt.

3. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, The next sensor detects the belt point position where the temperature abnormally rises again; after comparison, if the next sensor still detects the temperature abnormally rising, the device where the next sensor is located transmits the comparison data to the device where the next sensor is located; the next sensor detects the belt point position where the temperature abnormally rises again, and when it is confirmed that the temperature abnormally rises and exceeds the set threshold, the device where the next sensor is located directly determines that the point position is burning with hidden fire or open fire, transmits the data directly to the fault prediction and result display module, and performs fire warning or alarm; meanwhile, the data is transmitted to the data acquisition module; When the data of the next comparison abnormal point initial sensor is compared, it is found that the temperature of the temperature abnormally rising point is decreasing or the abnormal rising data of the point position disappears, then the comparison data is continuously transmitted to the subsequent two sensors, and after three comparisons, when it is confirmed that the temperature abnormally rising point disappears, the data transmission is ended; When one of the subsequent two sensors detects the temperature abnormally rising again, the sensor that detects the temperature rising again is taken as the abnormal point initial sensor, and the cycle is restarted to repeat the previous detection and comparison; meanwhile, the data is recorded and transmitted to the fault prediction and result display module as suspicious fire warning.

4. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, The input layer takes the multi-source heterogeneous data after spatio-temporal alignment as the input of the improved Transformer model, and organizes the data into a suitable input tensor form according to the feature dimension of the data; The multi-head self-attention mechanism layer introduces a multi-head self-attention mechanism, allowing the improved Transformer model to simultaneously focus on different position information and feature information when processing input data; each head calculates the attention weight, then performs weighted summation to obtain the output, and finally splices the outputs of all heads; The feedforward neural network layer is connected after the multi-head self-attention mechanism layer, and the feedforward neural network FFN consists of two fully connected layers with ReLU activation function in between; the feedforward neural network FFN is used for further feature transformation and nonlinear mapping of the self-attention mechanism output; The residual connection is added between the multi-head self-attention mechanism layer and the feedforward neural network layer; the residual connection directly adds the input to the output; The output layer outputs the fault type, confidence and position coordinates of the fault probability graph through a fully connected layer; the number of neurons in the output layer is set according to the classification number of fault types, the confidence quantization level and the accuracy requirement of position coordinates.

5. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, The model training of the improved Transformer model adopts the following methods: 1) Loss function design: a comprehensive loss function is designed, including cross-entropy loss of fault type prediction, mean square error loss of confidence prediction, and Euclidean loss of position coordinate prediction; by combining these loss terms, the learning goals of different tasks are balanced, and the model is guided to optimize fault diagnosis accuracy, confidence evaluation rationality and position positioning accuracy during the training process; 2) Optimization algorithm selection: Adam optimization algorithm is used for model parameter update. Adam optimization algorithm combines the advantages of momentum method and adaptive learning rate, which can automatically adjust the learning rate during training, accelerate the convergence speed and improve the stability of the model; 3) Data division: the preprocessed data set is divided into training set, validation set and test set according to certain proportion; The training set accounts for 70%-80% of the total data set, which is used for model training; The validation set accounts for 10%-15%, which is used to adjust the model hyperparameters and monitor whether the training process is overfitting; The test set accounts for 10%-15%, which is used to evaluate the final performance of the model; In the training process, stratified sampling or random sampling method is used to ensure that each batch of data can fully represent the distribution characteristics of the whole data set.

6. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, The time correlation matrix and space correlation matrix calculated in the space-time alignment process are used to provide prior knowledge for the multi-head self-attention mechanism, guiding the improved Transformer model to pay more attention to the space-time related data features when feature fusion; In the output results of the multi-head self-attention mechanism, the feature vectors of different data sources are weighted and summed according to the attention weights to obtain the fused feature representation; In the improved Transformer model, different scales of features are processed and fused in different levels or modules.

7. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, Fault prediction determines the most likely fault type according to the fault type probability distribution output by the improved Transformer model, using threshold judgment or Top-K selection method.

8. The intelligent inspection system for the coal conveying belt for the safe coal conveying according to claim 1, characterized in that, When displaying the prediction results, for each predicted fault point, mark its position on the coal conveying belt map with different colors or symbols, display the corresponding fault type and confidence information, and mark the fault point position on the coal conveying belt in real time according to the belt conveying speed, and display it.

Images