Unmanned driving fault data processing method and system for well industry and mining

Abstract

The invention discloses a well mining unmanned driving fault data processing method and system, and relates to unmanned driving. The method comprises the following steps: obtaining multi-source heterogeneous data; performing space-time fusion processing on the multi-source heterogeneous data to obtain a fused data set; inputting the fused data set into a pre-trained LSTM-Attention prediction model, and outputting a fault probability prediction value of the unmanned vehicle; calculating a fault risk score according to the fault probability prediction value in combination with the fault influence level; according to the fault risk score, triggering a preset fault processing mechanism of a corresponding level; aiming at the problem that the influence of water mist spraying on equipment operation is not considered in traditional well industrial and mining unmanned driving fault prediction, the method introduces UWB positioning-based vehicle and spraying point distance calculation, adopts a Gaussian distance attenuation function to quantify a spraying influence weight, constructs a temperature compensation factor alpha fused with a spatial distance and a time accumulation effect, and performs fault prediction on equipment operation according to the temperature compensation factor alpha. And the compensation factor is used for dynamically correcting the temperature of the motor and the like, so that the interference of underground dust falling spraying on the temperature monitoring of the equipment is effectively eliminated.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned driving, and in particular to a mine unmanned driving fault data processing method and system. BACKGROUND

[0002] With the rapid development of unmanned driving technology, its application in mine scenes has made remarkable progress. At present, automatic driving technology has formed a mature solution in the vehicle scheduling and operation control of open-pit mines, realizing the automation and intelligentization of mine transportation operations. However, due to the special working environment of underground mines, there are still many challenges in key technical links such as real-time fault monitoring, intelligent diagnosis and predictive maintenance.

[0003] There are significant differences in working environment between underground mines and open-pit mines. The space of underground mine roadway is narrow, water mist spraying for dust reduction needs to be carried out all year round, and there are complex roadway stress distribution and other special working conditions. In particular, the water mist spraying operation of the underground dust reduction system has a significant impact on the running state of the vehicle equipment, such as reducing the motor temperature and changing the working environment of the sensor. However, the existing fault prediction technology of underground mine unmanned driving mainly follows the scheme of open-pit mine, without fully considering the influence of these environmental factors, resulting in low fault prediction accuracy and high false alarm rate. Specifically, there are the following problems:

[0004] (1) Inadequate data collection and fusion. The existing technology mainly relies on vehicle-mounted sensors to collect equipment operation data, without considering the working state of the water mist spraying system, the distribution of spraying points, the spatial position relationship between the vehicle and the spraying points, and other environmental data. Water mist spraying will directly affect the motor temperature reading. If the cooling effect of spraying is not considered, it is easy to misjudge the normal temperature drop as equipment failure, or ignore the abnormal temperature rise after spraying stops.

[0005] (2) Lack of dynamic compensation mechanism for environmental factors. Traditional fault diagnosis uses fixed threshold judgment and does not consider the spatio-temporal distribution characteristics of water mist spraying. The influence of spraying on vehicles is different at different positions, and the cooling effect of spraying has time accumulation. Simple threshold comparison cannot accurately reflect the real running state of the equipment, resulting in a false alarm rate of more than 40% in areas where spraying operations are frequent.

[0006] (3) Single fault handling mechanism and lagging response. The existing system only performs simple alarm after detecting an anomaly, and lacks adaptive processing capability according to the fault risk level. It is difficult to distinguish between slight deviations caused by environmental factors and real equipment failures. All anomalies need to be confirmed and disposed by manual on-site, which takes a long time to respond in the complex environment of underground mines, affecting production efficiency. SUMMARY

[0007] In view of the fact that the traditional unmanned fault prediction of underground mine does not consider the influence of water spray on equipment operation, the application provides an underground mine unmanned fault data processing method and system, which introduces the calculation of the distance between the vehicle and the spray point based on UWB positioning, quantifies the spray influence weight by using the Gaussian distance attenuation function, constructs the temperature compensation factor alpha which fuses the spatial distance and the time cumulative effect, and dynamically corrects the motor temperature by using the compensation factor, thereby effectively eliminating the interference of underground dust spray on equipment temperature monitoring.

[0008] One aspect of the application provides an underground mine unmanned fault data processing method, comprising: acquiring multi-source heterogeneous data, the multi-source heterogeneous data including equipment operation data, environmental monitoring data and historical fault data of the unmanned vehicle; performing spatio-temporal fusion processing on the multi-source heterogeneous data to obtain a fusion data set; inputting the fusion data set into a pre-trained LSTM-Attention prediction model to output a fault probability prediction value of the unmanned vehicle; calculating a fault risk score according to the fault probability prediction value and in combination with a fault influence level; and triggering a pre-set corresponding level fault processing mechanism according to the fault risk score.

[0009] Further, the equipment operation data includes vibration data collected by a vehicle vibration sensor and chassis state data collected by a CAN bus; the environmental monitoring data includes stress data collected by a roadway pressure sensor and cumulative duration data of water spray; and the historical fault data includes vehicle historical fault records and point inspection records.

[0010] Furthermore, the multi-source heterogeneous data are subjected to spatiotemporal fusion processing to obtain a fused data set, including: using the PTP time protocol to perform time synchronization processing on the multi-source heterogeneous data, unifying the timestamps of the vibration data, chassis status data and stress data, and obtaining a time-aligned data set; according to the time-aligned data set, the three-dimensional coordinates of the vehicle corresponding to the vibration data are obtained through the UWB positioning system, and spatially associated with the monitoring point coordinates of the tunnel pressure sensor to generate spatiotemporal correlation data; the chassis status data in the spatiotemporal correlation data are subjected to Z-score normalization processing to obtain standardized equipment operation data; wherein, the chassis status data includes motor temperature and hydraulic pressure; the inspection records in the historical fault data are converted into a fault probability sequence in the range of 0 to 1, wherein each fault type corresponds to a probability value ; Use unique hot encoding to encode the discrete fault type labels in the fault probability sequence, and generate structured historical fault characteristics based on the encoded fault type and the corresponding fault probability value; Use the Kriging interpolation method to grid the discrete stress point data collected by the tunnel pressure sensor to generate grid data, and use principal component analysis to reduce the dimension of the grid data to obtain environmental stress characteristics; Extract the cumulative duration of water mist spraying from the environmental monitoring data, and encode the cumulative duration as a temperature compensation factor; Use the temperature compensation factor to dynamically correct the motor temperature in the standardized equipment operation data to obtain the temperature-compensated equipment operation data; Integrate the spatiotemporal correlation data, structured historical fault characteristics, environmental stress characteristics and temperature-compensated equipment operation data to obtain a fused data set.

[0011] Furthermore, the equipment operation data after temperature compensation is obtained, including: extracting the working state sequence of each water mist spray point from the environmental monitoring data , where i is the spray point number and t is the timestamp; the three-dimensional coordinates of the vehicle obtained based on the UWB positioning system and preset spray point coordinates , calculate the real-time distance between the vehicle and each spray point ; According to the distance decay function Calculate the influence weight of each spray point on the vehicle, where σ is the effective spray radius; calculate the temperature compensation factor α of the spatiotemporal coupling: ,in, is the cumulative time that the vehicle is within the effective range of the i-th spray point; is the temperature response time constant; the temperature compensation factor α is used to dynamically correct the motor temperature in the standardized equipment operation data to obtain the compensated motor temperature value ; The motor temperature value after compensation Replace the original motor temperature standardization value to obtain the equipment operation data after temperature compensation.

[0012] Particularly, in the narrow roadway environment, the vehicle needs to frequently cross the densely distributed water mist spraying area. Although these spraying systems are essential for dust reduction and improvement of the working environment, the uneven cooling effect caused by them will seriously interfere with the accurate monitoring of the motor temperature. The traditional fixed threshold monitoring method cannot distinguish between spraying cooling and equipment abnormalities, resulting in a large number of false positives of the fault prediction system, which seriously affects the reliability of the unmanned system.

[0013] In the spatial dimension, the application adopts a local compensation strategy based on UWB high-precision positioning. By calculating the spatial distance between the vehicle and each spraying point in real time, and introducing a Gaussian distance decay function to quantify the spatial distribution of spraying influence, the compensation intensity changes continuously with the distance. Only the vehicles that are truly within the spraying influence range will receive the corresponding temperature compensation, fundamentally solving the problem of false compensation at a distance.

[0014] In the time dimension, the application introduces an exponential decay-based time response model The cooling effect does not reach the maximum value instantaneously, but gradually stabilizes with the increase of spraying time. By reasonably setting the time constant τ, the model can accurately depict the complete process from the vehicle entering the spraying area to the temperature reaching the steady state, avoiding the data mutation problem caused by simple on-off compensation.

[0015] In addition, in key areas such as the intersection of underground mine tunnels and loading points, there are often multiple spraying points operating simultaneously. The temperature compensation factor α accurately reflects the superimposed cooling effect of multi-point spraying by weighting and summing the influence of all effective spraying points.

[0016] Further, the temperature compensation factor α is used to dynamically correct the motor temperature in the standardized equipment operation data, using the following formula: wherein, is the standardized motor temperature value, and β is the temperature compensation coefficient, with a value range of [0.1, 0.3];

[0017] Furthermore, the fused dataset is input into the pre-trained LSTM-Attention prediction model to output the predicted value of the failure probability of the unmanned vehicle, including: extracting and dividing three types of feature data from the fused dataset, where the first type is historical fault feature data, the second type is environmental parameter feature data, and the third type is operating indicator feature data; using a sliding window to perform time series sampling on the three types of feature data to construct an input sequence matrix; using Min Max Scaler to normalize the input sequence matrix to obtain a standardized input matrix X; inputting the standardized input matrix X into the bidirectional LSTM layer for time series feature extraction and outputting a hidden state sequence H; inputting the hidden state sequence H into the attention layer, using the scaled dot product attention mechanism to calculate the attention weight, and performing weighted aggregation on the hidden state sequence to obtain a context vector; inputting the context vector into a fully connected layer containing N output neurons, and outputting an N-dimensional fault probability vector through a Sigmoid activation function, where each dimension corresponds to the predicted fault probability of a fault type. ;

[0018] Furthermore, based on the predicted value of the fault probability and the fault impact level, the fault risk score is calculated, including: threshold filtering of the N-dimensional fault probability vector to obtain the predicted fault probability Fault types greater than the preset threshold are considered high-confidence fault types; the underground environment is complex, and sensors are susceptible to electromagnetic interference and dust, which can generate noise. Low-probability predictions may be environmental interference rather than real faults. Through threshold filtering, only high-confidence fault types are retained for risk assessment, avoiding the interference of false alarms on safety decisions. Specifically, for the braking system, the preset threshold is 0.3; for motor overheating, the preset threshold is 0.4; for the hydraulic system, the preset threshold is 0.5. For each high-confidence fault type, the corresponding fault impact coefficient is obtained according to the preset fault impact level matrix ; Based on the predicted failure probability and Failure Impact Factor , calculate the failure risk score R, ;

[0019] Furthermore, the brake system failure influence coefficient , hydraulic system failure influence coefficient , motor overheating fault influence coefficient ;

[0020] Further, according to the fault risk score, a preset corresponding level fault handling mechanism is triggered, including: when 0.7≤R<0.85, a switching instruction is sent to the vehicle-mounted control system to switch the main control system to the backup automatic driving system, and all operation data at the switching time is recorded; when 0.85≤R<0.95, the fault vehicle information, vehicle three-dimensional coordinates and fault type are uploaded to the cloud control platform, the cloud control platform re-computes and issues optimized operation path data according to the full mine vehicle position distribution and roadway traffic state; when R≥0.95, an isolation area data containing the position of the fault vehicle is generated, the area boundary coordinates and emergency escape roadway path data are pushed to all related vehicles, and an emergency plan triggering signal is sent to the mine management platform.

[0021] Another aspect of the present application also provides an unmanned mine fault data processing system, comprising: a data acquisition module for acquiring multi-source heterogeneous data, including equipment operation data, environmental monitoring data and historical fault data of unmanned vehicles; a space-time fusion module for performing space-time fusion processing on the multi-source heterogeneous data to obtain a fusion data set; a fault prediction module for inputting the fusion data set into a pre-trained LSTM-Attention prediction model to output a fault probability prediction value of the unmanned vehicle; a risk assessment module for calculating a fault risk score according to the fault probability prediction value and combining the fault influence level; and a fault handling module for triggering a corresponding level fault handling mechanism according to the fault risk score.

[0022] Compared with the prior art, the present application has the following advantages:

[0023] In view of the problems in the prior art that multi-source heterogeneous data of unmanned mine vehicles is difficult to effectively fuse in a complex roadway environment, traditional fault prediction methods do not fully consider the influence of water spray and roadway stress and other environmental factors on equipment operation, and there is a lack of adaptive hierarchical fault handling mechanism, the present application provides an unmanned mine fault data processing method, which realizes accurate space-time fusion of multi-source data by using PTP time protocol and UWB positioning system, introduces a temperature compensation factor based on distance attenuation and time accumulation effect to dynamically correct the motor temperature, constructs an LSTM-Attention deep learning model integrating historical fault, environmental parameter and operation index three types of features for fault probability prediction, and automatically triggers a three-level fault handling mechanism of equipment level, system level and platform level according to the fault risk score, which can realize accurate prediction and adaptive hierarchical response of unmanned vehicle faults in the special environment of a mine, and effectively reduce the false positive rate caused by environmental interference. BRIEF DESCRIPTION OF DRAWINGS

[0024] The present application will be further described in the manner of exemplary embodiments, which will be described in detail by means of the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same numbers represent the same structures, wherein:

[0025] Figure 1 is an exemplary flow chart of a mine unmanned fault data processing method according to some embodiments of the present application;

[0026] Figure 2 is an exemplary flow chart of generating a fusion data set according to some embodiments of the present application;

[0027] Figure 3 is an exemplary flow chart of calculating temperature-compensated equipment operation data according to some embodiments of the present application;

[0028] Figure 4 is an exemplary flow chart of outputting an N-dimensional fault probability vector method according to some embodiments of the present application. DETAILED DESCRIPTION

[0029] The method and system provided by the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings.

[0030] Embodiment 1

[0031] As shown in Figure 1 , multi-source heterogeneous data is obtained, and the multi-source heterogeneous data includes equipment operation data of an unmanned vehicle, environmental monitoring data, and historical fault data; the multi-source heterogeneous data is subjected to spatio-temporal fusion processing to obtain a fusion data set; the fusion data set is input into a pre-trained LSTM-Attention prediction model to output a fault probability prediction value of the unmanned vehicle; a fault risk score is calculated according to the fault probability prediction value in combination with a fault influence level; and a pre-set corresponding level fault handling mechanism is triggered according to the fault risk score.

[0032] As shown in Figure 2As shown, the PTP precise time protocol is used for time synchronization processing of multi-source heterogeneous data, the time stamps of vehicle vibration sensor, roadway pressure sensor and CAN bus collected data are unified, and a time-aligned data set is formed; based on the time-aligned data set, the three-dimensional coordinate information of the vehicle vibration data is obtained through the UWB positioning system, and is real-time spatially correlated with the coordinates of the roadway stress monitoring points to generate spatio-temporal correlation data; the multi-dimensional parameters including motor temperature, hydraulic pressure collected through the CAN bus in the spatio-temporal correlation data are subjected to dimensionless processing by using the Z-score standardization method to obtain standardized equipment operation parameters; the historical point inspection records are converted into a fault probability sequence in the interval [0, 1], and the discrete fault type labels are subjected to one-hot encoding processing to generate structured historical fault features; the discrete stress point data obtained by the roadway pressure sensor is subjected to gridding data generation by using the Kriging interpolation method to generate 50x50 gridding data, and is reduced to 10 dimensions by principal component analysis to obtain compressed environmental stress features; the cumulative length of the water mist spraying is encoded as a temperature compensation factor, which is associated with the temperature data in the standardized equipment operation parameters to dynamically correct the temperature threshold; the spatio-temporal correlation data, the standardized equipment operation parameters, the structured historical fault features, the compressed environmental stress features and the temperature compensated data after the above processing are integrated to form a fusion data set for inputting into a prediction model.

[0033] As shown in Figure 3 , the working state sequence of each water mist spraying point is extracted from the environmental monitoring data , wherein i is the spraying point number, t is the time stamp, indicates that the i-th spraying point is in working state at time t, indicates that it is in closed state;

[0034] The three-dimensional coordinates of the vehicle obtained based on the UWB positioning system and the preset spraying point coordinates , the real-time distance between the vehicle and the i-th spraying point is calculated according to the formula ;

[0035] The effective action radius of spraying σ is calculated according to the formula , wherein Q is the spraying flow, the unit is L / min, k is the roadway environment coefficient, the value range is [0.8, 1.2];

[0036] The influence weight of the i-th spraying point on the vehicle is calculated according to the distance attenuation function , when , set ; ;

[0037] The temperature compensation factor of spatio-temporal coupling is calculated: ​wherein, is the cumulative time length of the vehicle in the 3σ range of the i-th spray point 3 and the spray point is in the working state, and the unit is second; τ is the temperature response time constant, which is 300 seconds; and the summation range is all spray points;

[0038] Obtaining the motor temperature standardized value in the standardized equipment operation data ;

[0039] According to the formula Calculating the compensated motor temperature value wherein, is the standardized motor temperature value, β is the temperature compensation coefficient, and the value range is [0.1, 0.3]; and α is the spatio-temporal coupling temperature compensation factor, and the value range is [0, 1];

[0040] Replacing the original motor temperature standardized value in the standardized equipment operation data with the compensated motor temperature value , while keeping the hydraulic pressure and other chassis state data unchanged, to obtain the equipment operation data after temperature compensation.

[0041] The motor temperature in the standardized equipment operation data is dynamically corrected by using the temperature compensation factor α, and the following formula is adopted:

[0042] wherein, is the standardized motor temperature value, and β is the temperature compensation coefficient, and the value range is [0.1, 0.3];

[0043] As shown in FIG. 1, three types of feature data are extracted and divided from the fusion data set: Figure 4

[0044] The first type is historical fault feature data. 10-dimensional features are extracted from structured historical fault features, a sliding window with a length of 24 hours is used to count and statistically analyze the fault records, and the fault type label after one-hot encoding is retained. In particular, the mine vehicle fault has obvious cumulative effect and periodicity. The 10-dimensional features combined with the one-hot encoded fault type label enable the model to learn the time distribution pattern of a specific fault type, such as that the brake system fault occurs mostly in the night shift of heavy load downhill sections, while the hydraulic system fault is related to the frequent loading and unloading operations in the day shift.

[0045] The second type is environmental parameter feature data. The first three principal components of the environmental stress feature, the temperature compensation factor α, and the roadway humidity value are extracted. The specific processing includes: obtaining the real-time humidity value from the humidity sensor in the roadway, and the value range is [0, 100]%;

[0046] ​According to the humidity-stress coupling relationship, the humidity compensation coefficient is calculated: wherein, is a humidity influence coefficient, the value range is [0.1, 0.3], is a standard reference humidity, the value is 60%;

[0047] The first three principal components of the environmental stress characteristics Humidity compensation is performed: wherein, i = 1, 2, 3, the compensated stress value reflects the correction of the softening effect of humidity on the roadway rock;

[0048] The compensated principal components are subjected to noise suppression by Kalman filtering, the process noise covariance of the filter Q = 0.01, and the measurement noise covariance R = 0.1;

[0049] The first two principal components are selected from the filtered three principal components Combined with the temperature compensation factor a, a 3-dimensional environmental feature vector is formed The first two dimensions represent the stress distribution characteristics of the roadway, and the third dimension represents the cooling effect of the water mist spray on the vehicle.

[0050] In particular, the complexity of the mine roadway environment is concentrated in the coupling effect of humidity, stress, and temperature. Firstly, the humidity-stress coupling compensation mechanism is introduced. The high humidity environment in the mine (usually 80% to 90%) will significantly affect the mechanical properties of the surrounding rock of the roadway. For every 10% increase in humidity, the strength of the rock can decrease by 5% to 15%. The stress principal components are corrected by the humidity compensation coefficient γ, which accurately reflects the influence of surrounding rock softening on the stability of the roadway. This compensation enables the model to distinguish between the stable state of high stress + low humidity and the potentially dangerous state of medium stress + high humidity. Secondly, Kalman filtering is used to process the stress signal. Instantaneous stress disturbances may be generated by underground blasting operations and large equipment. Kalman filtering can effectively separate these short-term disturbances from the true stress trend changes, ensuring the stability and reliability of the environmental features. Finally, the stress and temperature compensation factors are combined. The first two stress principal components and the temperature compensation factor a form a 3-dimensional vector, realizing the trinity of surrounding rock-environment-equipment features. This design enables the model to learn the influence mechanism of environmental coupling effects on equipment failure - such as the fatigue acceleration of equipment support structure caused by high stress area superimposed with strong spray.

[0051] The third type is the running index feature data. The vibration data, compensated motor temperature, and hydraulic pressure are extracted from the temperature-compensated equipment running data. The vibration data is subjected to FFT transformation to extract 100 frequency domain features, combined with 20 time domain statistical features such as mean, variance, and peak value, to form a 120-dimensional feature vector;

[0052] In particular, the mine vehicle runs on the uneven roadway floor, and the vibration signal contains rich fault information. The design of extracting 100 FFT frequency domain features fully considers the complex spectral characteristics of underground vibration: the low frequency band (0-50 Hz) reflects the road surface excitation and suspension system state, the medium frequency band (50-500 Hz) reflects the transmission system fault, and the high frequency band (above 500 Hz) indicates the early damage of precision components such as bearings.

[0053] Combined with 20 time domain statistical features, a 120-dimensional high-dimensional feature space is formed, which can comprehensively describe the vibration mode of the equipment. In particular, the peak value, kurtosis and other high-order statistics are highly sensitive to common impact faults (such as gear pitting and bearing spalling) in the mine. At the same time, the temperature-compensated rather than the original temperature is used to ensure that the operation index is not affected by the environment and truly reflects the internal state of the equipment.

[0054] As shown in Figure 4 , a sliding window with a time step of 12 is used to time sequence sample three types of feature data, and each time step corresponds to 5 minutes of data to construct an input sequence matrix; specifically, the sliding window length is set to 12 time steps, corresponding to 60 minutes of historical data; at time t, 12 time steps of data are extracted; the 133-dimensional feature vector (10-dimensional historical fault + 3-dimensional environmental parameter + 120-dimensional operation index) of each time step is vertically spliced;

[0055] An input sequence matrix is generated: for each sliding window, a two-dimensional matrix of dimensions (12, 133) is constructed , where the i-th row corresponds to the 133-dimensional feature vector of the i-th time step; the sliding window slides forward by 1 time step (5 minutes), generating a continuous input sequence matrix sequence ; the structure of each input sequence matrix is: the first 10 columns are the historical fault feature time sequence, the 11th-13th columns are the environmental parameter feature time sequence, and the 14th-133rd columns are the operation index feature time sequence;

[0056] ​The input sequence matrix is normalized using Min Max Scaler to map all feature values to the interval [0, 1], obtaining a standardized input matrix X with dimensions (12, 133); the standardized input matrix X is input into the bidirectional LSTM layer for time series feature extraction, with 64 hidden units for both forward and backward LSTM, and the output dimension is (12, 128) hidden state sequence H; the hidden state sequence H is input into the Attention layer, the scaled dot-product attention mechanism is used to calculate the attention weight, and the hidden state is weighted and aggregated to obtain a 128-dimensional context vector; the context vector is input into a fully connected layer containing 8 output neurons, and a 8-dimensional fault probability vector is output through the Sigmoid activation function, each dimension corresponds to the prediction probability of a fault type.

[0057] According to the fault probability prediction value, combined with the fault influence level, the fault risk score is calculated, including: threshold filtering is performed on the N-dimensional fault probability vector to obtain the predicted fault probability greater than the preset threshold, as a high confidence fault type; for each high confidence fault type, the corresponding fault influence coefficient is obtained according to the preset fault influence level matrix ; according to the predicted fault probability and the fault influence coefficient , the fault risk score R is calculated, ; the brake system fault influence coefficient , the hydraulic system fault influence coefficient , the motor overheating fault influence coefficient ;

[0058] According to the numerical interval of the fault risk score R, the corresponding level of fault handling mechanism is triggered:

[0059] When 0.7≤R<0.85, the device-level self-healing mechanism is triggered, the switching instruction is sent to the vehicle-mounted control system, the main control system is switched to the standby automatic driving system, and all running data at the switching time is recorded; among them, this level corresponds to medium risk faults, such as sensor deviation, communication delay, etc. Recoverable problems; the underground electromagnetic environment is complex, and the main control system may be subject to transient interference. By switching to the standby automatic driving system, normal operation can be maintained without stopping, avoiding congestion of subsequent vehicles due to sudden stopping in narrow lanes. The standby system usually uses different sensor combinations or control algorithms, which can avoid systematic failures of the main system.

[0060] When 0.85≤R<0.95, trigger the system-level self-healing mechanism, upload the fault vehicle information, vehicle three-dimensional coordinates and fault type to the cloud control platform, and the cloud control platform recalculates and issues the optimized operation path data according to the mine vehicle position distribution and roadway traffic state; wherein this level corresponds to high-risk faults, such as brake performance decline, steering delay and other conditions that affect operation but have not yet lost control. The intervention of the cloud control platform realizes the upgrade from single vehicle risk avoidance to global optimization. By obtaining the accurate three-dimensional coordinates of the fault vehicle and the distribution of all mine vehicles, the system can intelligently plan: selecting a slower slope and larger width alternative path for the fault vehicle; planning an avoidance route for other vehicles to prevent meeting in narrow roadways; prioritizing the use of maintenance roadways for fault vehicles to reduce the impact on main transportation channels. This dynamic path optimization based on global information makes full use of the redundancy of underground mine roadways, ensuring the safety of fault vehicles while minimizing the impact on overall production.

[0061] When R≥0.95, trigger the platform-level self-healing mechanism, automatically generate isolation area data containing the location of the fault vehicle, push the area boundary coordinates and emergency escape roadway path data to all related vehicles, and send an emergency plan trigger signal to the mine management platform. Among them, this level corresponds to extremely high-risk faults, such as complete brake failure, hydraulic system collapse and other conditions that may lead to catastrophic consequences.

[0062] Embodiment 2

[0063] Device layer: collect vibration data through vibration sensors installed on autonomous driving mine trucks. Collect vehicle chassis state data through the autonomous driving system, merge autonomous driving state node state data upload. Obtain the point inspection results, point inspection pictures and fault prediction data of the autonomous driving mine truck before operation. Vehicle historical fault reporting data

[0064] Environment layer: collect mine point cloud data through the laser radar of the autonomous driving mine truck, and construct a three-dimensional point cloud model of the roadway. At the same time, the vehicle-mounted laser radar of the autonomous driving mine truck will collect the front roadway surface data at a frequency of 10Hz during operation. Embed fiber Bragg grating pressure sensors in the roadway roof / two sides, use E-shaped concentric resistance structure to enhance signal stability, and monitor stress data in the roadway in real time. Real-time transmission of stress data to edge computing nodes, and upload to the cloud control platform after processing. The cloud control platform interacts with the mine management system to obtain the duration of water mist spraying in the mine, the concentration of main gases such as gas, etc.

[0065] Communication layer: real-time reporting of device state and environmental data through the deployment of 5G+ optical fiber dual-channel redundant transmission.

[0066] The vehicle vibration data (latitude + longitude + elevation) is associated with the roadway stress monitoring point coordinates in real time through the UWB positioning system, and the PTP precise time protocol is used to unify the time stamp of the mine truck vibration sensor and the roadway pressure sensor. Through the CAN bus, 20-dimensional parameters such as motor temperature and hydraulic pressure are collected, and Z-score standardization is used to eliminate dimensional differences. At the same time, the PTP precise time protocol is used to align with the vibration data timestamp. The above point inspection history point inspection record is converted into a fault probability sequence (0-1), and a one-hot encoding process is used to handle discrete labels. The roadway stress discrete point data obtained by the pressure sensor is used to generate a 50X50 grid using ordinary Kriging interpolation, and PCA is used to reduce the dimension to 10. The cumulative duration of the water mist spray is encoded as a temperature compensation factor to dynamically adjust the temperature threshold.

[0067] The above collected and fused data is mainly divided into three categories as the input layer of the LSTM-Attention model. The first category is historical faults, with a feature dimension of 10, and the processing method is sliding window counting + one-hot encoding. The second category is environmental parameters, with a feature dimension of 3, and the processing method is Kalman filtering + humidity compensation. The third category is operating indicators, with a feature dimension of 120, and the processing method is FFT spectrum + time domain features. A sliding window with a time step of 12 is used to process the data, and finally the data is normalized to the [0, 1] interval by MinMax Scaler.

[0068] The fault probability output by the LSTM-Attention dynamic prediction model is filtered by a preset threshold (≥0.85), and only high-confidence predictions trigger the self-healing process to avoid low-probability false actions.

[0069] According to the fault probability output by the model, combined with the current device state and the impact level of the fault, a comprehensive risk score of the fault is obtained. According to the risk score, different levels of self-healing mechanisms are triggered, and the alarm mechanism of the cloud control platform is also triggered synchronously.

[0070] Device-level self-healing: When the risk score is between 0.7 and 0.85, the fault is generally a fault that occurs in the autonomous driving truck itself, triggering a first-level self-healing to switch to a backup autonomous driving system.

[0071] System-level self-healing: When the risk score is between 0.85 and 0.95, it is generally a fault in the environment, path planning, electronic fence, etc., triggering a second-level self-healing. The cloud control platform re-plans the operating route of all autonomous driving trucks based on their positions and the conditions in the mine roadway, and adjusts the formation of the faulty truck if necessary to prevent affecting the operation progress of other healthy trucks.

[0072] Platform level self-recovery: when the risk score is higher than 0.95, generally, the failure occurs or the environment of the mine blasting area changes, etc., at this time, the third level self-recovery is triggered, the regional isolation is automatically created, the vehicle planning enters the emergency escape roadway, and the emergency plan is pushed to the mine management platform system.

[0073] The above has described the application creation and its implementation mode schematically, which is not limited, and the application can be realized in other specific forms without departing from the spirit or essential characteristics of the application. The embodiment shown in the drawings is only one of the application creations, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired thereby, without departing from the purpose of the creation, similar structure modes and embodiments can be designed without creativity, which should belong to the protection scope of the application. In addition, the word "comprising" does not exclude other elements or steps, and the word "one" before the element does not exclude "multiple" elements. The words "first", "second" and the like are used to indicate names, and do not mean any specific order.

Claims

1. A method for processing fault data of unmanned driving in underground mines, characterized in that: include: Acquire multi-source heterogeneous data, including unmanned vehicle equipment operation data, environmental monitoring data, and historical fault data; Perform spatiotemporal fusion processing on multi-source heterogeneous data to obtain a fused data set; The fused dataset is input into the pre-trained LSTM-Attention prediction model to output the failure probability prediction value of the autonomous vehicle; Calculate the fault risk score based on the predicted fault probability value and the fault impact level; According to the fault risk score, the preset fault handling mechanism of the corresponding level is triggered.

2. The method for processing fault data of unmanned driving in underground mines according to claim 1, characterized in that: Equipment operation data includes: vibration data collected by vehicle vibration sensors and chassis status data collected through the CAN bus; Environmental monitoring data includes: stress data collected by tunnel pressure sensors and accumulated duration data of water mist spraying; Historical fault data includes: vehicle historical fault records and inspection records.

3. The method for processing fault data of unmanned driving in underground mines according to claim 2, characterized in that: Perform spatiotemporal fusion processing on multi-source heterogeneous data to obtain a fused dataset, including: The PTP time protocol is used to synchronize multi-source heterogeneous data, unify the timestamps of vibration data, chassis status data, and stress data, and obtain a time-aligned data set. Based on the time-aligned data set, the UWB positioning system is used to obtain the three-dimensional coordinates of the vehicle corresponding to the vibration data, and spatially correlated with the monitoring point coordinates of the roadway pressure sensor to generate spatiotemporal correlation data; Perform Z-score normalization on the chassis status data in the spatiotemporal correlation data to obtain standardized equipment operation data; the chassis status data includes motor temperature and hydraulic pressure; Convert the inspection records in the historical fault data into a fault probability sequence between 0 and 1, where each fault type corresponds to a probability value; One-hot encoding is used to encode the discrete fault type labels in the fault probability sequence. Based on the encoded fault types and corresponding fault probability values, structured historical fault features are generated. The Kriging interpolation method is used to grid the discrete stress point data collected by the tunnel pressure sensor to generate grid data. The grid data is then reduced in dimension using principal component analysis to obtain the environmental stress characteristics. Extract the cumulative duration of water mist spraying from environmental monitoring data and encode the cumulative duration as a temperature compensation factor; Use the temperature compensation factor to dynamically correct the motor temperature in the standardized equipment operation data to obtain the temperature-compensated equipment operation data; The spatiotemporal correlation data, structured historical fault characteristics, environmental stress characteristics and temperature-compensated equipment operation data are integrated to obtain a fused data set.

4. The method for processing fault data of unmanned driving in underground mines according to claim 3, characterized in that: The device operation data after temperature compensation is obtained, including: Extracting the working state sequence of each water mist spray point from environmental monitoring data , where i is the sprinkler point number and t is the timestamp; Vehicle three-dimensional coordinates obtained based on UWB positioning system and preset spray point coordinates , calculate the real-time distance between the vehicle and each spray point ; According to the distance decay function Calculate the influence weight of each spray point on the vehicle, where σ is the effective spray radius; Calculate the temperature compensation factor α for spatiotemporal coupling: ,in, is the cumulative time that the vehicle is within the effective range of the i-th spray point; τ is the temperature response time constant; Use the temperature compensation factor α to dynamically correct the motor temperature in the standardized equipment operation data to obtain the compensated motor temperature value ; The motor temperature value after compensation Replace the original motor temperature standardization value to obtain the equipment operation data after temperature compensation.

5. The method for processing fault data of unmanned driving in underground mines according to claim 4, characterized in that: The temperature compensation factor α is used to dynamically correct the motor temperature in the standardized equipment operation data using the following formula: ,in, is the normalized motor temperature value, and β is the temperature compensation coefficient.

6. The method for processing fault data of unmanned driving in underground mines according to claim 4, characterized in that: Input the fusion dataset into the pre-trained LSTM-Attention prediction model, including: Extract and divide three types of feature data from the fused data set: the first type is historical fault feature data, the second type is environmental parameter feature data, and the third type is operating indicator feature data; A sliding window is used to perform time series sampling on the three types of feature data to construct an input sequence matrix; Use Min Max Scaler to normalize the input sequence matrix to obtain the standardized input matrix X; The standardized input matrix X is input into the bidirectional LSTM layer for time series feature extraction, and the hidden state sequence H is output; The hidden state sequence H is input into the attention layer, the scaled dot product attention mechanism is used to calculate the attention weight, and the hidden state sequence is weightedly aggregated to obtain the context vector; The context vector is input into a fully connected layer containing N output neurons, and an N-dimensional fault probability vector is output through the Sigmoid activation function. Each dimension corresponds to the predicted fault probability of a fault type. .

7. The method for processing fault data of unmanned driving in underground mines according to claim 6, characterized in that: Calculates a failure risk score, including: Perform threshold filtering on the N-dimensional fault probability vector to obtain the predicted fault probability Fault types that are greater than the preset threshold are considered high-confidence fault types; For each high-confidence fault type, the corresponding fault impact coefficient is obtained according to the preset fault impact level matrix ; According to the predicted failure probability and Failure Impact Factor , calculate the failure risk score R, .

8. The method for processing fault data of unmanned driving in underground mines according to claim 7, characterized in that: Braking system failure influence coefficient , hydraulic system failure influence coefficient , motor overheating fault influence coefficient .

9. The method for processing fault data of unmanned driving in underground mines according to claim 7, characterized in that: Based on the fault risk score, the preset fault handling mechanism of the corresponding level is triggered, including: When 0.7≤R<0.85, a switching command is sent to the vehicle control system to switch the main control system to the backup automatic driving system, and all operating data at the time of switching are recorded; When 0.85≤R<0.95, the fault vehicle information, vehicle 3D coordinates, and fault type are uploaded to the cloud control platform. The cloud control platform recalculates and issues optimized operation path data based on the vehicle location distribution and roadway traffic status throughout the mine. When R≥0.95, the isolation area data containing the location of the faulty vehicle is generated, the area boundary coordinates and emergency avoidance tunnel path data are pushed to all relevant vehicles, and an emergency plan trigger signal is sent to the mine area management platform.

10. A fault data processing system for unmanned driving in underground mines, characterized in that: include: The data acquisition module acquires multi-source heterogeneous data, including equipment operation data of unmanned vehicles, environmental monitoring data, and historical fault data; The spatiotemporal fusion module performs spatiotemporal fusion processing on multi-source heterogeneous data to obtain a fused data set; The fault prediction module inputs the fused dataset into the pre-trained LSTM-Attention prediction model and outputs the predicted value of the autonomous vehicle's fault probability; The risk assessment module is used to calculate the fault risk score based on the predicted fault probability value and the fault impact level; The fault handling module is used to trigger the corresponding level of fault handling mechanism according to the fault risk score.

Images