Mine equipment operation safety situation assessment method and system

Abstract

The invention provides a mining equipment operation safety situation assessment method and system, and belongs to the technical field of mining equipment, in the method, sensor data, management data and image data of the mining equipment are collected and preprocessed in real time, and weights of the sensor data, the management data and the image data of the mining equipment are constructed, so that the operation safety situation of the mining equipment is assessed. And finally, inputting sensor data, management data and image data of the mining equipment into the dynamic risk assessment system to obtain a safety situation score and a risk level of the mining equipment. According to the mining equipment operation safety situation assessment method, the mining equipment is comprehensively monitored and assessed in real time, potential problems can be found in the early stage, measures can be taken in time, accidents are effectively prevented, and mine production safety is guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of mining equipment, and in particular to a method and system for assessing the operating safety status of mining equipment. Background Art

[0002] The safe operation of mining equipment ensures production efficiency and protects personnel. However, due to the complexity of the mining environment and the harsh operating conditions of mining equipment, methods based on regular maintenance and manual inspection often fail to detect potential problems in a timely manner, leading to mining equipment failures or accidents. Summary of the invention

[0003] The embodiment of the present application provides a method and system for evaluating the safety situation of mining equipment operation. The present application adopts the following technical solutions: In a first aspect, a method for assessing the safety situation of mining equipment operation is provided, the method comprising: Collect sensor data, management data and image data of mining equipment in real time; Preprocessing the collected sensor data, management data and image data, including data cleaning, feature engineering and data fusion of the sensor data, management data and image data; wherein the data cleaning is used to remove outliers and noise, the feature engineering is used to extract key features of the sensor data, management data and image data, and the data fusion is used to fuse the key features of the sensor data, management data and image data into a comprehensive data set; Based on the comprehensive data set, a multi-dimensional evaluation index system including the status of mining equipment itself, mining equipment operating environment indicators, human operation factors and management mechanisms is constructed; A hierarchical analysis method and / or a fuzzy comprehensive evaluation method are used to determine the indicator weights, and a dynamic risk assessment system is constructed, wherein the dynamic risk assessment system includes an LSTM model and a random forest model, wherein the LSTM model is suitable for predicting the remaining useful life RUL of the mining equipment, wherein the LSTM model input includes historical fault data and real-time sensor data, and the random forest model is suitable for classifying risk levels, and the random forest model input includes the comprehensive data set; The sensor data, the management data and the image data collected in real time are input into the LSTM model and the random forest model, and the safety situation score and the risk level of the mining equipment are calculated by the LSTM model and the random forest model. When the safety situation score is lower than the preset safety situation value and / or the risk level is higher than the preset risk level value, an early warning is triggered.

[0004] In one embodiment of the present application, the real-time collection of sensor data of mining equipment includes: first sensor data, wherein the first sensor data includes a vibration frequency of the mining equipment, a temperature of the mining equipment, a pressure of the mining equipment, a wear degree of the mining equipment, and a lubrication state of the mining equipment; Second sensor data, the second sensor data includes methane concentration, carbon monoxide concentration, oxygen concentration, dust concentration, and temperature and humidity; The management data include mine equipment maintenance records, personnel training files and accident reports; The image data includes mining equipment status images, environmental risk images and personnel behavior images.

[0005] In one embodiment of the present application, the feature engineering is further suitable for performing time alignment processing, sliding window processing and normalization processing on the key features of the sensor data, the management data and the image data; wherein The time alignment process is suitable for synchronizing key features of the sensor data, the management data and the image data according to a unified time reference to facilitate subsequent joint analysis; The key features of the sensor data, the management data and the image data after the time alignment process are constructed into a data stream, and the sliding window process is suitable for dividing the data stream into time periods of fixed length, and the data in each time period is regarded as a sample; The normalization process is suitable for converting the key features of the sensor data, the management data and the image data to the same scale range to avoid the key features of the sensor data, the management data and / or the image data having a large numerical range and dominating the model training process.

[0006] In one embodiment of the present application, the LSTM model constructs a neural network structure including at least two layers of LSTM units, and trains the LSTM model to prevent overfitting of the training data. The mean absolute error is used as the loss function to train the LSTM model, and the prediction results are evaluated using the PHM08 Challenge standard.

[0007] In one embodiment of the present application, the random forest model includes: Encode static and dynamic features and handle class imbalance issues; Find the optimal parameter combination and use the F1 score as the main evaluation indicator; Analyze the importance of output features to improve the interpretability of the random forest model; The RUL value output by the LSTM model is used as one of the input features, and the risk level classification is further implemented through the random forest model to form a cascade risk assessment process; The static features include: basic information of the mining equipment, maintenance records of the mining equipment, and fixed parameters of the operating environment of the mining equipment; The dynamic features include: real-time collection of sensor data of mining equipment, short-term operating behavior, changes in environmental factors, and the remaining useful life RUL of the mining equipment predicted by the LSTM model.

[0008] In one embodiment of the present application, it also includes: Collecting multi-source image data, the multi-source image data includes a first image, a second image and a third image, the first image is the mine equipment status image, the second image is the environmental risk image, and the third image is the personnel behavior image, and performing image enhancement processing, spatial alignment processing and time synchronization processing on the multi-source image data; The image features in the multi-source image data are obtained, and the image features are input together with the sensor data of the real-time mining equipment into the LSTM model and the random forest model for comprehensive risk assessment.

[0009] In one embodiment of the present application, the LSTM model is compressed by TensorFlow Lite, and the LSTM model is deployed on a mining edge gateway device to quickly predict the remaining useful life RUL of the mining equipment, thereby supporting safe monitoring and rapid decision-making of the mining equipment.

[0010] In one embodiment of the present application, it also includes: The assessment results are displayed in a visual form, including the status of mining equipment, risk level and early warning information, and the assessment results are integrated into the mine management system to achieve data sharing and decision support; Regularly collect new data to update the dynamic risk assessment system, and conduct model drift detection and A / B testing to ensure the accuracy and stability of the dynamic risk assessment system; The model drift detection triggers model retraining by monitoring changes in feature distribution, and the A / B test compares the warning accuracy by running the new and old models in parallel to select the optimal model for deployment.

[0011] In one embodiment of the present application, the visualization form includes but is not limited to a cockpit interface, reports and charts, so that safety management personnel can intuitively understand the safety situation and risk level of the mining equipment and take timely intervention measures.

[0012] In the second aspect, based on the same inventive concept, a mine equipment operation safety situation assessment system is provided, the system is applied to the mine equipment operation safety situation assessment method described in any one of the above embodiments, the system includes: a data acquisition module, a data preprocessing module, a database support module, a safety situation assessment module and a visualization module; wherein The data acquisition module is used to obtain sensor data, management data and image data of the mining equipment; The data preprocessing module is suitable for performing data cleaning, feature engineering and data fusion on the sensor data, the management data and the image data; The database support module is used for structured storage of the mining equipment data; The safety situation module is adapted to input the sensor data, the management data and the image data collected in real time into the LSTM model and the random forest model, calculate the safety situation score and risk level of the mining equipment through the LSTM model and the random forest model, and trigger an early warning when the safety situation score is lower than a preset safety situation value and / or the risk level is higher than a preset risk level value; The visualization module is used to display the safety situation score of the mining equipment and the risk level of the mining equipment.

[0013] In summary, the above-mentioned mining equipment operation safety situation assessment method and system have the following technical effects: The embodiments of the present application can detect potential problems at an early stage and take timely measures to effectively prevent accidents and ensure the safety of mine production by conducting comprehensive and real-time monitoring and evaluation of mining equipment. Moreover, the present application can also provide intuitive safety situation scores and risk levels to help managers make more scientific and reasonable decisions. Of course, based on accurate risk assessment results, managers can reasonably arrange maintenance plans and resource allocation to reduce unnecessary downtime and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of the steps of a method for assessing the operating safety status of mining equipment provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0016] Reference Figure 1 The embodiment of the present application provides a method for evaluating the safety status of mining equipment operation, the method comprising the following steps: Collect sensor data, management data and image data of mining equipment in real time; It is understandable that by collecting sensor data, management data and image data of mining equipment in real time, all-round information about the status of mining equipment can be obtained in real time. Sensor data includes but is not limited to physical parameters such as vibration frequency, temperature, pressure, etc. of mining equipment, and sensor data directly reflects the working status of mining equipment. Management data includes but is not limited to maintenance records, training files and accident reports, etc., which provide information at the management and operation level. Image data includes but is not limited to video monitoring or image capture, which is used to monitor environmental risks and personnel behavior.

[0017] Preprocessing the collected sensor data, management data and image data, including data cleaning, feature engineering and data fusion; data cleaning is used to remove outliers and noise, feature engineering is used to extract key features of sensor data, management data and image data, and data fusion is used to fuse the key features of sensor data, management data and image data into a comprehensive data set; It is understandable that the sensor data, management data and image data collected in real time need to be preprocessed. The preprocessing includes data cleaning, feature engineering and data fusion of the above data. Here, data cleaning is used to remove outliers and noise to ensure the quality and reliability of the above data; feature engineering is used to extract key features of sensor data, management data and image data to facilitate subsequent analysis. For example, feature engineering can be to convert time series data into a form suitable for LSTM model input; data fusion is used to fuse the key features of sensor data, management data and image data into a comprehensive data set, so that data from different sources (sensor data, management data and image data) can be integrated to form a comprehensive data set, which is convenient for unified processing and analysis.

[0018] Based on the comprehensive data set, a multi-dimensional evaluation index system including the status of mining equipment itself, mining equipment operating environment indicators, human operation factors and management mechanisms is constructed; It is understandable that through the above settings, a multi-dimensional evaluation index system can be established, including the equipment status, operating environment indicators, human operation factors and management mechanisms, thus providing a comprehensive perspective to evaluate the safety status of mining equipment. It is worth mentioning that the above design takes into account the influence of multiple aspects such as technology, environment, personnel and management, so as to achieve a more comprehensive evaluation of the operating safety situation of mining equipment.

[0019] A hierarchical analysis method and / or a fuzzy comprehensive evaluation method are used to determine the indicator weights, and a dynamic risk assessment system is constructed, wherein the dynamic risk assessment system includes an LSTM model and a random forest model, wherein the LSTM model is suitable for predicting the remaining useful life RUL of the mining equipment, wherein the LSTM model input includes historical fault data and real-time sensor data, and the random forest model is suitable for classifying risk levels, and the random forest model input includes the comprehensive data set; In some embodiments, the analytic hierarchy process (AHP) is used to decompose the problem into multiple levels (such as goals, criteria, sub-criteria, and options) and determine the importance weight of each factor through a pairwise comparison matrix. In the safety assessment of mining equipment, the analytic hierarchy process can be used to evaluate the impact of different dimensions (such as equipment status, environmental indicators, human operating factors, and management mechanisms) on the overall safety situation and assign corresponding weights to each dimension.

[0020] The fuzzy comprehensive evaluation method is suitable for dealing with problems with high uncertainty, ambiguity and subjectivity. The fuzzy comprehensive evaluation method uses fuzzy mathematics theory to conduct comprehensive evaluation by establishing a fuzzy relationship matrix and determining a weight vector. In the evaluation of mining equipment, the fuzzy comprehensive evaluation method can be used to deal with factors that are difficult to accurately quantify but are very important (such as geological risk level), thereby providing a more comprehensive risk assessment. Therefore, the weight of each evaluation indicator is determined by AHP or fuzzy comprehensive evaluation method, ensuring that information of different dimensions is reasonably valued and avoiding a single dimension dominating the evaluation results.

[0021] The LSTM model (Long Short-Term Memory Network) is suitable for processing time series data. The LSTM model can capture long-term dependencies to predict the remaining service life (RUL) of equipment. The input of the LSTM model includes historical fault data (equipment downtime records, maintenance logs, etc.) and real-time sensor data (vibration frequency, temperature, pressure, etc. of mining equipment). The LSTM model can predict the remaining service life (RUL) of mining equipment through the input of historical fault data and real-time sensor data. By predicting the future health status of mining equipment, possible faults can be identified in advance, thereby helping to formulate maintenance plans and reduce unplanned downtime.

[0022] The random forest model improves prediction accuracy and controls overfitting by constructing multiple decision trees and aggregating the results of multiple decision trees. The input of the random forest model includes a comprehensive data set, that is, the input of the random forest model includes key features extracted from sensor data, management data, and image data, as well as the RUL value predicted by the LSTM model. The random forest model outputs the risk level classification result through the input comprehensive data set. By analyzing the comprehensive data, the random forest model can evaluate the risk level of mining equipment based on the current status information, thereby guiding the adoption of corresponding preventive measures.

[0023] Therefore, by combining the time series prediction ability of LSTM and the classification ability of random forest, the accuracy of the prediction of the remaining life and risk level of mining equipment can be significantly improved, thereby providing intuitive safety situation scores and risk levels, helping managers make more scientific and reasonable decisions, optimizing maintenance strategies and resource allocation, and helping to reduce the accident rate and ensure mine production safety.

[0024] The sensor data, the management data and the image data collected in real time are input into the LSTM model and the random forest model, and the safety situation score and the risk level of the mining equipment are calculated by the LSTM model and the random forest model. When the safety situation score is lower than the preset safety situation value and / or the risk level is higher than the preset risk level value, an early warning is triggered.

[0025] In some embodiments, sensor data, management data, and image data of mining equipment collected in real time are input into LSTM and random forest models to calculate the safety situation score and risk level of the mining equipment. When the calculated safety situation score is lower than a preset safety situation value or the risk level is higher than a preset risk level value, the system automatically triggers an early warning, prompting to take corresponding measures to prevent potential safety accidents.

[0026] According to the mine equipment operation safety situation assessment method of the present application, through comprehensive and real-time monitoring and assessment of mine equipment, potential problems can be discovered at an early stage, and timely measures can be taken to effectively prevent accidents and ensure mine production safety. Moreover, the present application can also provide intuitive safety situation scores and risk levels to help managers make more scientific and reasonable decisions. Of course, based on accurate risk assessment results, managers can reasonably arrange maintenance plans and resource allocation to reduce unnecessary downtime and maintenance costs.

[0027] It is worth mentioning that the mining equipment operation safety situation assessment method of the present application can be applied to mining equipment such as coal mining machines or conveyor belts.

[0028] According to some embodiments of the present invention, real-time collection of sensor data of mining equipment includes: first sensor data and second sensor data, the first sensor data includes mining equipment vibration frequency, mining equipment temperature, mining equipment pressure, mining equipment wear and lubrication status of mining equipment, and the second sensor data includes methane concentration, carbon monoxide concentration, oxygen concentration, dust concentration, and temperature and humidity; wherein management data includes mining equipment maintenance records, personnel training files and accident reports; image data includes mining equipment status images, environmental risk images and personnel behavior images.

[0029] In some embodiments, the vibration frequency data of mining equipment is used to monitor the working status of the mechanical parts of the mining equipment (abnormal vibration may be an early signal of mining equipment failure); the temperature data of mining equipment is suitable for monitoring the temperature changes during the operation of mining equipment (excessive temperature may indicate problems such as overload or poor lubrication); for hydraulic systems or other mining equipment that rely on pressure to operate, the pressure data of mining equipment can reflect the health status of the mining equipment; the wear data of mining equipment can be used to evaluate the degree of wear of key components through sensors or regular inspection records to help predict maintenance needs; the lubrication status data of mining equipment is suitable for monitoring whether the mining equipment has good lubrication, and the status of the lubricating oil (such as the degree of contamination, viscosity, etc.) directly affects the service life of the mining equipment.

[0030] Methane concentration data, carbon monoxide concentration data, and oxygen concentration data are mainly used to monitor the air quality in underground mining environments and prevent explosions and suffocation accidents; for dust concentration, high dust concentration not only affects the health of workers, but may also cause dust explosions; ambient temperature and humidity can affect the performance of mining equipment and the health of workers (especially in special environments such as underground mines).

[0031] Mining equipment maintenance record data is suitable for recording the maintenance history of mining equipment. Mining equipment maintenance record data includes information such as maintenance time, replacement parts, and failure causes, which is helpful for analyzing the historical performance of mining equipment and optimizing future maintenance plans; personnel training file data includes records of employees' safety training to ensure that all operators have the necessary safety knowledge and skills (specifically, personnel training files are scored through expert group decision-making evaluation to derive the degree of influence of personnel training files in the mining equipment operation safety situation assessment method); accident report data includes records of past accidents and their handling processes to provide a reference for preventing similar accidents in the future.

[0032] Images of the appearance and internal structure of mining equipment are obtained through cameras or other imaging devices to obtain mining equipment status image data. Mining equipment status image data is used to detect physical damage such as cracks and deformation on the surface of mining equipment; environmental risk image data includes monitoring the mining operating environment to identify potential dangerous areas, such as landslides, water accumulation, etc.; personnel behavior image data includes recording and analyzing the behavior of staff to ensure that staff comply with safety operating procedures and promptly discover and correct violations.

[0033] It is understandable that by collecting different types of sensor data, management data and image data, we can fully understand the operating status of mining equipment and its surrounding environment from multiple perspectives, which will help to more accurately assess the safety situation of mining equipment. Using sensors for real-time data collection can detect abnormal situations in a timely manner and trigger early warning mechanisms, so that measures can be taken quickly to avoid accidents.

[0034] It should be noted that sensor data provides direct feedback on the operating status of mining equipment. By analyzing sensor data, it is possible to predict problems that may occur in mining equipment and arrange maintenance work in advance. Management data provides information about the maintenance history of mining equipment, employee training, and past accidents to help formulate more scientific and reasonable management and maintenance strategies; image data can not only intuitively display the status of equipment and environment, but also automatically identify potential risk factors through image processing technology to improve safety management efficiency.

[0035] According to some embodiments of the present invention, feature engineering is also suitable for time alignment processing, sliding window processing and normalization processing of key features of sensor data, management data and image data; wherein the time alignment processing is suitable for synchronizing the key features of sensor data, management data and image data according to a unified time reference to facilitate subsequent joint analysis; the key features of sensor data, management data and image data that have undergone time alignment processing are constructed as a data stream, and the sliding window processing is suitable for dividing the data stream into time periods of fixed length, with the data in each time period being regarded as a sample; the normalization processing is suitable for converting the key features of sensor data, management data and image data to the same scale range to avoid the key features of sensor data, management data and image data having an excessively large numerical range and dominating the model training process.

[0036] In some embodiments, since different types of devices or systems may have different sampling frequencies or recording times, directly merging these data may result in time misalignment. Therefore, the time alignment processing of the present application can synchronize the key features of sensor data, management data, and image data according to a unified time base. Through the time alignment processing, it can be ensured that all data points correspond to the same timestamp for joint analysis. For example, the vibration data at a certain moment is synchronized with the methane concentration, maintenance records, and video surveillance images at that moment to form a complete time series data set. Therefore, through the time alignment processing, it can be ensured that all data are analyzed under the same time base to avoid information loss or errors caused by time misalignment.

[0037] Sliding window processing is suitable for dividing the data stream after time alignment into time periods of fixed length. The data in each time period is regarded as a sample, and each sample is input into the LSTM model and the random forest model respectively, so as to capture the temporal patterns in sensor data, management data and image data. For example, with a window length of 10 minutes and sliding every 5 minutes, a series of samples containing all sensor readings, management records and image features within 10 minutes are generated to help identify short-term trends and changes.

[0038] Normalization is suitable for converting data from different sources to the same scale range, so that the LSTM model and the random forest model can treat each feature more fairly and avoid certain features dominating the model training process due to large differences in numerical ranges. For example, the range of vibration frequency may be 0 to 1000 Hz, while the range of temperature may be -20 to 50 degrees Celsius. Therefore, normalization can prevent the LSTM model and the random forest model from giving priority to features with larger values ​​(such as vibration frequency) and ignoring other equally important features (such as temperature).

[0039] Therefore, through time alignment, sliding window and normalization processing, the quality and consistency of sensor data, management data and image data are significantly improved, which helps to improve the prediction accuracy and stability of machine learning models (such as LSTM models and random forest models), and thus better capture the status changes and potential risks of mining equipment. Of course, precise data processing reduces unnecessary waste of computing resources and improves the efficiency of the system.

[0040] According to some embodiments of the present invention, the LSTM model constructs a neural network structure including at least two layers of LSTM units, and trains the LSTM model to prevent overfitting of the training data. The mean absolute error is used as the loss function to train the LSTM model, and the prediction results are evaluated using the PHM08 Challenge standard.

[0041] In some embodiments, the LSTM model structure includes a neural network structure of at least two layers of LSTM units to capture more complex temporal dependencies. In the assessment of the operating safety status of mining equipment, the use of at least two layers of LSTM units can better capture the temporal dynamic characteristics of equipment state changes to improve the understanding and prediction capabilities of the health status of mining equipment.

[0042] As an example of this application, the LSTM model is trained using the Early Stopping method, which can monitor the performance indicators (such as the loss function value) on the validation set during the training process. It is worth mentioning that if the performance on the validation set does not improve after several consecutive iterations, the training is terminated early. That is, through the Early Stopping method, the training can be stopped before the LSTM model starts to overfit, thereby maintaining the generalization ability of the LSTM model on unseen data.

[0043] Methods that dynamically adjust the learning rate (such as ReduceLROnPlateau) can automatically reduce the learning rate according to the performance of the validation set during the training process to help the model converge to a better solution. The method of dynamically adjusting the learning rate can avoid the problem of too fast or too slow learning speed that may be caused by a fixed learning rate, so that the LSTM model can gradually find the optimal parameters during the training process; the mean absolute error (MAE) is used as a loss function, and its calculation formula is the average of the absolute values ​​of the prediction errors of all samples. Compared with the mean square error (MSE), MAE is less sensitive to outliers and is more suitable for regression problems. When predicting the remaining useful life RUL of mining equipment, using MAE as a loss function can help the LSTM model learn the trend of mining equipment status changes more robustly, reduce the impact of extreme errors, reduce the impact of outliers on the model training process, and improve the stability and reliability of the LSTM model.

[0044] It is worth mentioning that using the PHM08 Challenge standard to evaluate the prediction results of the LSTM model can comprehensively measure the performance of the model in actual application scenarios, encouraging the LSTM model to not only accurately predict the failure time, but also issue early warning signals as early as possible, so as to help take timely measures to avoid mining equipment failures and production interruptions.

[0045] According to some embodiments of the present invention, the random forest model includes: encoding static features and dynamic features, and dealing with category imbalance problems; finding the optimal parameter combination, and using the F1 score as the main evaluation indicator; analyzing the importance of output features to improve the interpretability of the random forest model; using the RUL value output by the LSTM model as one of the input features, and further implementing risk level classification through the random forest model to form a cascade risk assessment process; wherein the static features include: basic information of mining equipment, maintenance records of mining equipment, and fixed parameters of the operating environment of mining equipment; dynamic features include: real-time collection of sensor data of mining equipment, short-term operating behavior, changes in environmental factors, and the remaining service life RUL of mining equipment predicted by the LSTM model.

[0046] In some embodiments, static features and dynamic features are encoded and converted to ensure that they can be correctly understood and processed by the random forest model, for example, converting categorical variables into numerical variables. By combining static features and dynamic features, the random forest model can fully capture various factors that affect the safety of mining equipment, from the state of the mining equipment itself to changes in the external environment.

[0047] As an embodiment of the present application, the present application uses SMOTE technology to solve the problem of class imbalance, and balances the proportion between different classes by generating synthetic instances of minority class samples, thereby improving the recognition ability of the random forest model on the minority class. In the risk level classification, if the occurrence rate of a certain category of faults or high-risk events is low, the use of SMOTE can help the random forest model better learn the characteristics of such samples and avoid the random forest model from being biased towards majority class samples. Of course, the present application uses grid search (GridSearchCV) to find the optimal hyperparameter setting. The grid search will automatically perform cross-validation, evaluate the performance of each set of parameter combinations, and select the best performing set of parameters, with the F1 score as the main evaluation indicator (the F1 score is the harmonic mean of precision and recall). At the same time, the present application analyzes the importance of output features through SHAP values ​​(SHapleyAdditive exPlanations) to quantify the contribution of each feature to the prediction results of the random forest model. By performing SHAP value analysis on the random forest model, it is possible to clearly understand which features are most critical to the final risk level classification decision, thereby improving the transparency and interpretability of the random forest model.

[0048] It is worth mentioning that the equipment remaining service life RUL predicted by the LSTM model is used as one of the input features, and is input into the random forest model together with static features and dynamic features for risk level classification. This not only utilizes the historical operation data of mining equipment (obtained through the LSTM model), but also combines the current operating environment and behavioral characteristics, thereby achieving a more comprehensive risk assessment. That is, the LSTM model first predicts the RUL, and then the RUL is input into the random forest model together with static features and dynamic features for further risk level classification, forming a two-level assessment process, which improves the accuracy and reliability of the overall assessment.

[0049] In some embodiments, the LSTM model is: Python from tensorflow.keras.models import Sequential from tensorflow.keras.layers import LSTM, Dense, Dropout model = Sequential() model.add(LSTM(64, input_shape=(window_length, feature_dim), return_sequences=True))# Input dimension: time step × number of features model.add(Dropout(0.2)) model.add(LSTM(32)) model.add(Dense(16, activation='relu')) model.add(Dense(1, activation='linear'))# Output RUL value model.compile(loss='mae', optimizer='adam')# Loss function: mean absolute error (MAE).

[0050] The random forest model is: Python from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import GridSearchCV # Define parameter grid param_grid = { 'n_estimators': [100, 200], 'max_depth': [10, 20, None], 'class_weight': ['balanced', {0:1, 1:5, 2:10, 3:20}] } # Grid Search Optimization model = GridSearchCV( RandomForestClassifier(), param_grid, cv=5, scoring='f1_macro' ) model.fit(X_train, y_train).

[0051] Moreover, the cascade risk assessment is: sensor data → LSTM predicts RUL → RUL as feature input → random forest classification risk level.

[0052] It is worth mentioning that in the hierarchical analysis method, the weight of mining equipment itself accounts for 40%, the environmental weight accounts for 30%, the human weight accounts for 20%, and the management weight accounts for 10%.

[0053] In some embodiments, the warning levels that can be output by the random forest model include four levels, namely level one (0~25%), level two (25%~50%), level three (50%~75%) and level four (75%~100%). When the warning level is level one, the safety situation of the mine equipment operation is safe (marked with green here), when the warning level is level two, the safety situation of the mine equipment operation is low risk (marked with blue here), when the warning level is level three, the safety situation of the mine equipment operation is medium risk (marked with yellow here), and when the warning level is level four, the safety situation of the mine equipment operation is high risk (marked with red here). It is worth mentioning that when the warning level is level three, the staff can shut down the mine equipment for maintenance, and when the warning level is level four, the staff needs to evacuate the mine in time.

[0054] It should be noted that when the RUL predicted value is less than the threshold, the risk level will be forced to increase (such as from level 2 to level 3); when the methane concentration, carbon monoxide concentration and dust concentration exceed the standard, additional weighting will also be triggered.

[0055] As an example of the present application, when the RUL prediction value is 12 hours, the SHAP value of the mining equipment life is 0.41, when the number of recent illegal operations is 3 times, the mining equipment operation SHAP value is 0.28, and when the methane concentration is 1.2%, the environmental SHAP value is 0.35. Finally, the calculated comprehensive risk score = 12*0.41+3*0.28+12*0.35=6.18, where the comprehensive risk score range is [0, 10]. Therefore, the comprehensive risk probability at this time = 6.18 / 10*100%=61.8%, and because the methane concentration exceeds the standard, the risk probability is increased by 20%, that is, the final comprehensive risk probability = 61.8%+20%=81.8%. The comprehensive risk probability of 81.8% has reached the fourth warning level, and the system can immediately trigger an emergency shutdown.

[0056] According to some embodiments of the present invention, the method for assessing the operating safety status of mining equipment also includes: collecting multi-source image data and obtaining image features in the multi-source image data, the multi-source image data including a first image, a second image and a third image, the first image being a mining equipment status image, the second image being an environmental risk image, and the third image being a personnel behavior image, and performing image enhancement processing, spatial alignment processing and time synchronization processing on the multi-source image data, and inputting the image features together with the sensor data of the mining equipment collected in real time into the LSTM model and the random forest model for comprehensive risk assessment.

[0057] In some embodiments, the first image is a mining equipment status image, which may be an image directly taken of the appearance or internal structure of the mining equipment, so as to detect physical damage such as cracks and deformation on the surface of the mining equipment; the second image is an environmental risk image, which includes photos or videos of monitoring the mining environment, such as monitoring landslides, water accumulation, etc., to help identify potential environmental risks; the third image is a personnel behavior image, which is used to record the behavior of the staff, ensure that the staff comply with safety operating procedures, and promptly discover and correct violations.

[0058] By performing enhancement processing (such as contrast adjustment and denoising) on ​​multi-source image data, the image quality can be improved, making subsequent feature extraction more accurate. For example, images taken under low-light conditions can be enhanced to improve clarity, making it easier to more accurately identify details in the equipment status or environment. Spatial alignment processing can ensure that images acquired from different perspectives or at different time points are correctly aligned to ensure the consistency and accuracy of feature extraction. For example, images from different angles of multiple cameras can be aligned to form a complete 3D view, which helps to comprehensively analyze the status of mining equipment. Time synchronization processing can keep multi-source image data consistent with other types of data (such as sensor data) in time for joint analysis, ensuring that the vibration data at a certain moment is synchronized with the image of the mining equipment status at that moment, thereby forming a complete time series data set, making risk assessment more accurate.

[0059] Use computer vision techniques (such as target detection, semantic segmentation, behavior recognition, etc.) to extract features from multi-source image data. It is worth mentioning that target detection is used to identify specific objects in multi-source image data, such as cracks on mining equipment or obstacles in the environment; semantic segmentation is used to assign each pixel in multi-source image data to a certain category, such as distinguishing different parts of mining equipment or identifying different environmental areas; behavior recognition is used to analyze personnel behavior images to determine whether there are any violations of safety regulations. The extracted image features are combined with the real-time mining equipment sensor data as the input of the LSTM model and the random forest model. The LSTM model then uses time series data (including sensor data and image features) to predict the remaining service life RUL of mining equipment. Finally, the random forest model combines static features, dynamic features, and image features to classify risk levels.

[0060] Therefore, by introducing multi-source image data and combining it with sensor data, the fusion of multimodal data is achieved, so as to fully understand the status of mining equipment and environment from more dimensions, providing a rich information basis for risk assessment. Moreover, through image enhancement, spatial alignment and time synchronization processing, the quality and consistency of multi-source image data are ensured, and the accuracy of feature extraction is improved. Of course, combining multi-source image data with sensor data and inputting them into LSTM and random forest models can form a more comprehensive and accurate risk assessment process.

[0061] According to some embodiments of the present invention, the LSTM model is compressed through TensorFlow Lite, and the LSTM model is deployed on a mining edge gateway device to achieve rapid prediction of the remaining useful life RUL of mining equipment, thereby supporting safe monitoring and rapid decision-making of mining equipment.

[0062] In some embodiments, the LSTM model is compressed (quantization technology) through TensorFlow Lite to reduce the size and computational complexity of the LSTM model, so that the LSTM model that originally required a large amount of computing resources can be run on edge devices, which not only reduces the storage requirements of the LSTM model, but also speeds up the reasoning speed. For example, by performing 8-bit integer quantization on the weights, the model size can be significantly reduced and the computational efficiency can be improved while maintaining a high prediction accuracy.

[0063] The mining edge gateway device can be Huawei Atlas 500 smart station, etc. The mining edge gateway device can be directly deployed on the mine site to process local data and reduce latency. Therefore, the LSTM model compressed by TensorFlow Lite is deployed on these edge gateway devices, so that the LSTM model can instantly analyze the data stream from the sensor to quickly predict the remaining service life of the mining equipment, avoiding the delay problem caused by data transmission to the cloud, so that the LSTM model can process sensor data in real time and perform RUL prediction.

[0064] Therefore, by using TensorFlow Lite to compress the LSTM model, the originally complex LSTM model can run efficiently on edge devices with limited resources, which not only ensures the performance of the LSTM model but also reduces the hardware cost. In addition, by deploying the LSTM model on the edge gateway device, local data processing is achieved, reducing the delay caused by network transmission and improving the response speed of the system.

[0065] According to some embodiments of the present invention, the method for assessing the operating safety situation of mining equipment also includes: displaying the assessment results in a visual form, the assessment results including the mining equipment status, risk level and early warning information, and integrating the assessment results into the mining management system to achieve data sharing and decision support; regularly collecting new data to update the dynamic risk assessment system, and performing model drift detection and A / B testing to ensure the accuracy and stability of the dynamic risk assessment system; wherein, model drift detection triggers model retraining by monitoring changes in feature distribution, and A / B testing compares the early warning accuracy by running the new and old models in parallel, and selects the optimal model for deployment.

[0066] In some embodiments, the assessment results (such as mining equipment status, risk level, and early warning information) are displayed in an intuitive form to facilitate managers to quickly understand the current safety situation. A variety of visualization tools such as cockpit interfaces, reports, and charts can be used here to provide a multi-dimensional data view; the assessment results are integrated into the existing mine management system to achieve data sharing and cross-departmental collaboration. As a result, all relevant departments (such as maintenance teams, production management, safety management, etc.) can access the latest assessment results and improve information transparency, so that comprehensive data analysis results can be achieved to help management make more scientific and reasonable decisions and optimize resource allocation and maintenance plans.

[0067] Over time, the status and environmental conditions of mining equipment may change, causing the original LSTM model and random forest model to no longer be applicable. Therefore, this application continuously monitors the operating status of mining equipment, and regularly collects new sensor data, management data, and image data, and monitors changes in feature distribution to determine whether the current LSTM model and random forest model are still applicable. If a significant change in feature distribution is found, the LSTM model and random forest model retraining process is triggered to ensure the accuracy of the LSTM model and random forest model.

[0068] At the same time, this application uses A / B testing to identify the changing trend of feature distribution by comparing and analyzing historical data and new data. Once the performance of the LSTM model and / or random forest model is detected to have declined, the LSTM model and / or random forest model retraining process is started. By running the old and new models in parallel and comparing their performance in actual application scenarios (such as early warning accuracy), the model with better performance is selected for deployment to ensure that the LSTM model and / or random forest model finally deployed has the highest prediction accuracy and reliability.

[0069] According to some embodiments of the present invention, visualization forms include but are not limited to cockpit interfaces, reports and charts, so that safety managers can intuitively understand the safety situation and risk level of mining equipment and take timely intervention measures.

[0070] It is understandable that complex assessment results are presented to safety managers in an intuitive manner through cockpit interfaces, reports, charts and other forms, allowing safety managers to quickly understand and grasp the status and potential risks of mining equipment, thereby quickly formulating response strategies and improving decision-making efficiency.

[0071] Based on the same inventive concept, the present application proposes a mining equipment operation safety situation assessment system, which is applied to the mining equipment operation safety situation assessment method described in any one of the above embodiments, and the mining equipment operation safety situation assessment system includes: a data acquisition module, a data preprocessing module, a database support module, a safety situation assessment module and a visualization module; wherein the data acquisition module is used to obtain sensor data, management data and image data of the mining equipment; the data preprocessing module is suitable for data cleaning, feature engineering and data fusion of sensor data, management data and image data; the database support module is used for structured storage of mining equipment data; the safety situation module is suitable for inputting real-time collected sensor data, management data and image data into the LSTM model and the random forest model, and calculating the safety situation score and risk level of the mining equipment through the LSTM model and the random forest model, and triggering an early warning when the safety situation score is lower than the preset safety situation value and / or the risk level is higher than the preset risk level value; the visualization module is used to display the safety situation score of the mining equipment and the risk level of the mining equipment.

[0072] It should be noted that the overall technical route of the mining equipment operation safety situation assessment system of this embodiment is based on the big data architecture, is developed in mainstream languages, uses relational databases, non-relational databases, and index databases for database design and development, and designs each function as a loosely coupled module, providing an extensible interface to provide users with an on-demand application and development design system suitable for IoT digital, text data such as hidden dangers, risks and accidents, image and video data, etc.

[0073] Optionally, the mine equipment operation safety situation assessment system of this embodiment can adopt the B / S (Browser / Server) architecture and the front-end and back-end separation development mode, realize the front-end and back-end data docking through the data interface, separate the application layer display and the back-end business logic processing layer, maintain the big data visualization, back-end business logic processing and big data analysis model separation mode, and also ensure the independence of database logic processing. The overall system supports horizontal expansion, the back-end provides a standard RESTFUL interface, and the front-end can be displayed or customized according to the actual needs of the user.

[0074] It should be noted that the specific implementation of a mining equipment operation safety situation assessment system in an embodiment of the present application refers to the specific implementation of a mining equipment operation safety situation assessment method proposed in the first aspect of the aforementioned embodiment of the present application, and will not be repeated here.

[0075] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When a computer instruction or computer program is loaded or executed on a computer, a process or function according to an embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer program or instruction can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program or instruction can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0076] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0077] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0078] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0079] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0080] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0081] In the several embodiments provided in the present application, it should be understood that the disclosed systems, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the system or unit can be electrical, mechanical or other forms.

[0082] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0083] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit.

[0084] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0085] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for assessing the safety situation of mining equipment operation, characterized in that: The method comprises: Collect sensor data, management data and image data of mining equipment in real time; Preprocessing the collected sensor data, management data and image data, including data cleaning, feature engineering and data fusion of the sensor data, management data and image data; wherein the data cleaning is used to remove outliers and noise, the feature engineering is used to extract key features of the sensor data, management data and image data, and the data fusion is used to fuse the key features of the sensor data, management data and image data into a comprehensive data set; Based on the comprehensive data set, a multi-dimensional evaluation index system including the status of mining equipment itself, mining equipment operating environment indicators, human operation factors and management mechanisms is constructed; A hierarchical analysis method and / or a fuzzy comprehensive evaluation method are used to determine the indicator weights, and a dynamic risk assessment system is constructed, wherein the dynamic risk assessment system includes an LSTM model and a random forest model, wherein the LSTM model is suitable for predicting the remaining useful life RUL of the mining equipment, wherein the LSTM model input includes historical fault data and real-time sensor data, and the random forest model is suitable for classifying risk levels, and the random forest model input includes the comprehensive data set; The sensor data, the management data and the image data collected in real time are input into the LSTM model and the random forest model, and the safety situation score and the risk level of the mining equipment are calculated by the LSTM model and the random forest model. When the safety situation score is lower than the preset safety situation value and / or the risk level is higher than the preset risk level value, an early warning is triggered.

2. The method for assessing the safety situation of mining equipment operation according to claim 1, characterized in that: The real-time collection of sensor data of mining equipment includes: first sensor data, wherein the first sensor data includes a vibration frequency of the mining equipment, a temperature of the mining equipment, a pressure of the mining equipment, a wear degree of the mining equipment, and a lubrication state of the mining equipment; Second sensor data, the second sensor data includes methane concentration, carbon monoxide concentration, oxygen concentration, dust concentration, and temperature and humidity; The management data include mine equipment maintenance records, personnel training files and accident reports; The image data includes mining equipment status images, environmental risk images and personnel behavior images.

3. The method for assessing the safety situation of mining equipment operation according to claim 2, characterized in that: The feature engineering is also suitable for performing time alignment processing, sliding window processing and normalization processing on key features of the sensor data, the management data and the image data; in The time alignment process is suitable for synchronizing key features of the sensor data, the management data and the image data according to a unified time reference to facilitate subsequent joint analysis; The key features of the sensor data, the management data and the image data after the time alignment process are constructed into a data stream, and the sliding window process is suitable for dividing the data stream into time periods of fixed length, and the data in each time period is regarded as a sample; The normalization process is suitable for converting the key features of the sensor data, the management data and the image data to the same scale range to avoid the key features of the sensor data, the management data and / or the image data having a large numerical range and dominating the model training process.

4. The method for assessing the safety situation of mining equipment operation according to claim 3, characterized in that: The LSTM model constructs a neural network structure including at least two layers of LSTM units, trains the LSTM model to prevent overfitting of the training data, and uses the mean absolute error as a loss function to train the LSTM model, and evaluates the prediction results using the PHM08 Challenge standard.

5. The method for assessing the safety situation of mining equipment operation according to claim 4, characterized in that: The random forest model includes: Encode static and dynamic features and handle class imbalance issues; Find the optimal parameter combination and use the F1 score as the main evaluation indicator; Analyze the importance of output features to improve the interpretability of the random forest model; The RUL value output by the LSTM model is used as one of the input features, and the risk level classification is further implemented through the random forest model to form a cascade risk assessment process; The static features include: basic information of the mining equipment, maintenance records of the mining equipment, and fixed parameters of the operating environment of the mining equipment; The dynamic features include: real-time collection of sensor data of mining equipment, short-term operating behavior, changes in environmental factors, and the remaining useful life RUL of the mining equipment predicted by the LSTM model.

6. The method for assessing the safety situation of mining equipment operation according to claim 5, characterized in that: Also includes: Collecting multi-source image data, the multi-source image data includes a first image, a second image and a third image, the first image is the mine equipment status image, the second image is the environmental risk image, and the third image is the personnel behavior image, and performing image enhancement processing, spatial alignment processing and time synchronization processing on the multi-source image data; The image features in the multi-source image data are obtained, and the image features are input together with the sensor data of the real-time mining equipment into the LSTM model and the random forest model for comprehensive risk assessment.

7. The method for assessing the safety situation of mining equipment operation according to claim 1, characterized in that: The LSTM model is compressed through TensorFlow Lite, and the LSTM model is deployed on a mining edge gateway device to quickly predict the remaining useful life RUL of the mining equipment, thereby supporting safe monitoring and rapid decision-making of the mining equipment.

8. The method for assessing the safety situation of mining equipment operation according to claim 1, characterized in that: Also includes: The assessment results are displayed in a visual form, including the status of mining equipment, risk level and early warning information, and the assessment results are integrated into the mine management system to achieve data sharing and decision support; Regularly collect new data to update the dynamic risk assessment system, and conduct model drift detection and A / B testing to ensure the accuracy and stability of the dynamic risk assessment system; The model drift detection triggers model retraining by monitoring changes in feature distribution, and the A / B test compares the warning accuracy by running the new and old models in parallel to select the optimal model for deployment.

9. The method for assessing the safety situation of mining equipment operation according to claim 8, characterized in that: The visualization forms include but are not limited to cockpit interfaces, reports and charts, so that safety management personnel can intuitively understand the safety situation and risk level of the mining equipment and take timely intervention measures.

10. A system for assessing the safety situation of mining equipment operation, the system being applied to the method for assessing the safety situation of mining equipment operation according to any one of claims 1 to 9, characterized in that: The system includes a data acquisition module, a data preprocessing module, a database support module, a security situation assessment module and a visualization module; in The data acquisition module is used to obtain sensor data, management data and image data of the mining equipment; The data preprocessing module is suitable for performing data cleaning, feature engineering and data fusion on the sensor data, the management data and the image data; The database support module is used for structured storage of the mining equipment data; The safety situation module is adapted to input the sensor data, the management data and the image data collected in real time into the LSTM model and the random forest model, calculate the safety situation score and risk level of the mining equipment through the LSTM model and the random forest model, and trigger an early warning when the safety situation score is lower than a preset safety situation value and / or the risk level is higher than a preset risk level value; The visualization module is used to display the safety situation score of the mining equipment and the risk level of the mining equipment.

Images