Coal Mine Equipment Fault Prediction System and Method Based on Big Data

Abstract

The present invention relates to the technical field of equipment fault prediction, and particularly to a coal mine equipment fault prediction system and method based on big data, which includes the following steps: collecting the thickness of the biofilm on the surface of the equipment, synchronously collecting the vibration acceleration of the equipment and the environmental hydrogen sulfide concentration, and constructing a corrosion multi-modal data set; performing electrochemical noise separation on the corrosion multi-modal data set to construct a corrosion activity fingerprint spectrum; decoupling the coupling effect of mechanical vibration and microbial corrosion to generate a corrosion-fatigue damage correlation map; constructing a dual-channel life prediction model based on the corrosion-fatigue damage correlation map and setting up an early warning mechanism. The present invention accurately extracts the characteristic current of microbial corrosion through wavelet packet decomposition and wavelet entropy analysis, and uses a graph attention network to establish a corrosion-vibration heterogeneous graph to realize the analysis of the corrosion-induced structural vibration enhancement effect. Compared with traditional single-signal-based prediction methods, it effectively overcomes problems such as the limited influence range of local corrosion and unstable vibration response.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault prediction, and particularly to a coal mine equipment fault prediction system and method based on big data. Background Art

[0002] During the operation of equipment, corrosion damage and fatigue cracks of metal structures are the two main factors affecting equipment safety and life. Especially for underground coal mine-related equipment, traditional corrosion prediction methods usually rely on electrochemical measurement means, which depend on laboratory tests or regular on-site monitoring, and it is difficult to achieve real-time tracking of the corrosion process. On the other hand, fatigue damage prediction mainly uses stress analysis and crack propagation models, such as the Paris formula. However, this method assumes that crack propagation is only affected by mechanical loads and does not consider the weakening effect of corrosion on the material microstructure, resulting in a decrease in prediction accuracy. In addition, corrosion and fatigue failures often interact with each other. Stress concentration caused by local corrosion will accelerate crack initiation, and structural vibration may also promote the diffusion of corrosive media, forming a vicious cycle. Therefore, it is difficult to accurately predict the comprehensive life of equipment only by relying on single failure mode analysis.

[0003] Existing crack propagation models do not consider the influence of corrosion, resulting in deviations in fatigue life prediction. Classic models such as the Paris crack propagation law and the Miner linear cumulative damage theory mainly calculate the crack growth rate based on mechanical stress and cyclic loading, and do not fully consider the accelerating effect of corrosion on crack propagation. Under actual working conditions, local corrosion will cause microscopic damage to the material and affect the stress distribution at the crack tip, making the fatigue crack propagation rate much higher than the theoretical prediction value. Summary of the Invention

[0004] The present invention provides a coal mine equipment fault prediction system and method based on big data.

[0005] The coal mine equipment fault prediction method based on big data includes the following steps:

[0006] S1: Deploy ultrasonic sensors on the metal surface of the equipment to collect the biofilm thickness on the equipment surface, synchronously collect the equipment vibration acceleration and the ambient hydrogen sulfide concentration, and construct a corrosion multi-modal data set;

[0007] S2: Perform electrochemical noise separation on the corrosion multi-modal data set, extract microbial corrosion characteristic current pulses through the wavelet entropy algorithm, and construct a corrosion activity fingerprint spectrum;

[0008] S3: Input the corrosion activity fingerprint spectrum into the graph attention network, decouple the coupling effect of mechanical vibration and microbial corrosion, and generate a corrosion-fatigue damage correlation map;

[0009] S4: Construct a dual-channel life prediction model based on the corrosion-fatigue damage correlation atlas, output a composite failure trajectory curve including the chemical corrosion rate and the mechanical crack propagation rate, and set up an early warning mechanism based on the composite failure trajectory curve.

[0010] Optionally, the S1 specifically includes:

[0011] S11: The ultrasonic sensor emits ultrasonic pulses, monitors the reflected echo, and calculates the acoustic wave propagation time difference: , where is the biofilm thickness, is the ultrasonic propagation speed in the biofilm, taking 1550 m / s, is the propagation time delay of the ultrasonic wave;

[0012] S12: Install a three-axis MEMS vibration sensor on the device surface, and set the vibration sampling frequency to 10 kHz;

[0013] S13: Arrange multiple hydrogen sulfide electrochemical sensor arrays circumferentially around the device to obtain ambient hydrogen sulfide concentration data;

[0014] S14: Unify the sampling time reference of each sensor through a hardware clock synchronizer to generate a raw data stream with a unified timestamp;

[0015] S15: Generate a corrosion multi-modal dataset including biofilm thickness, device vibration acceleration, and ambient hydrogen sulfide concentration.

[0016] Optionally, the S2 specifically includes:

[0017] S21: Based on the db4 wavelet basis function, perform multi-level decomposition on the electrochemical noise signals in the corrosion multi-modal dataset. In the wavelet decomposition, each layer of the signal is split into a low-frequency part and a high-frequency part to form multiple sub-bands. The decomposition process continues for eight levels, so that the signal is divided into multiple frequency ranges;

[0018] S22: Calculate the wavelet entropy value of each sub-band: After obtaining the target sub-band signal, analyze the energy distribution characteristics, use wavelet entropy analysis to quantify the complexity of the signal, judge whether the signal contains corrosion characteristics, count the proportion of different energy components in each sub-band to obtain the energy distribution characteristics, and calculate the wavelet entropy value based on the energy distribution characteristics to measure the randomness and structural information content of the signal. The larger the wavelet entropy value, the higher the signal complexity, and it may contain more corrosion characteristics. In all sub-band signals, set an entropy threshold. When the wavelet entropy value of a certain sub-band exceeds the entropy threshold, it is judged that the current sub-band includes effective microbial corrosion information and is retained for subsequent feature extraction;

[0019] S23: Adaptively calculate the signal strength threshold for each sub-band to adapt to signal variations in different environments. When the actual signal strength exceeds the signal strength threshold, it is determined to belong to a corrosion current pulse. Among all signal segments exceeding the signal strength threshold, extract the key pulse parameters related to the corrosion process, including:

[0020] Pulse amplitude, which is used to characterize the intensity of the corrosion reaction;

[0021] Pulse duration, which reflects the time length of the corrosion reaction;

[0022] Pulse rise time, which is used to measure the suddenness of the corrosion process;

[0023] S24: After extracting all key pulse parameters, construct a three-dimensional feature vector. The three-dimensional feature vector includes the average amplitude of the pulse, the total duration, and the variation characteristics of the rise time. Use the K-means clustering algorithm to classify the corrosion activity, and divide the corrosion signals into low activity level, medium activity level, and high activity level. The classification of the levels is based on the pulse amplitude and duration indicators. Map the clustering results to the time-frequency dimension to form a corrosion activity fingerprint spectrum. In the corrosion activity fingerprint spectrum, the abscissa represents time, the ordinate represents the characteristic frequency, and the change in color reflects different levels of corrosion activity.

[0024] In calculating the wavelet entropy value of each sub-band in S22, for each sub-band , its wavelet entropy value is calculated as:

[0025] , where is the energy proportion of the th layer and the th node:

[0026] , where represents the energy of this node. When , it is determined that this layer contains effective corrosion features, is the entropy value threshold.

[0027] Optionally, the signal strength threshold is calculated as: , where is the standard deviation of the signal of the th layer, is the number of sampling points of this layer.

[0028] Optionally, S3 specifically includes:

[0029] S31: Construct a heterogeneous graph structure to describe the association between corrosion nodes and vibration nodes. The corrosion nodes are the pulse feature vectors extracted from the corrosion activity fingerprint spectrum, and the vibration nodes are the band energy entropy vectors extracted from the vibration time-frequency spectrum. After constructing the nodes, define the association relationship between corrosion and vibration, and adopt a weighted mechanism based on time difference to assign an edge weight to each corrosion-vibration connection edge. The value of the edge weight decays as the time difference between the corrosion node and the vibration node increases, ensuring that events closer in time have higher correlation.

[0030] S32: Based on the heterogeneous graph structure, construct a two-channel graph attention network, including a corrosion feature enhancement channel and a vibration-corrosion coupling channel, where:

[0031] The corrosion feature enhancement channel is used to analyze the mutual correlation between corrosion nodes. The multi-head attention mechanism is adopted to calculate the attention degree of each corrosion node to other corrosion nodes, adjust the feature expression of the corrosion nodes. Each corrosion node calculates its own correlation with other corrosion nodes, and uses the attention weighting method to update its own features to enhance the corrosion features.

[0032] The vibration-corrosion coupling channel is used to analyze the influence of vibration signals on the corrosion process. The gating mechanism is adopted to screen effective association edges. The gating mechanism calculates a dynamic weight according to the joint features of corrosion and vibration to correct the initial time association edge weight. The corrected edge weight is used to adjust the influence degree of vibration on corrosion nodes.

[0033] S33: Use the singular value decomposition method to perform dimensionality reduction on the embedded features of corrosion-vibration nodes and decouple the coupling effect, including calculating the mutual information between corrosion and vibration in the reduced feature space. The mutual information measures the statistical correlation between vibration signals and corrosion signals. The larger the value, the stronger the coupling degree between the two. When the mutual information exceeds a predetermined threshold, it indicates that the vibration signal and the corrosion signal have a strong dependence relationship, indicating that the corrosion process exacerbates the fatigue vibration of the equipment.

[0034] S43: After decoupling the corrosion-vibration relationship, construct a generated corrosion-fatigue damage association map to describe their coupling mode. The association map includes the horizontal axis and the vertical axis of the map, where:

[0035] The horizontal axis of the map represents the corrosion activity intensity, which is derived from the chromaticity information of the corrosion activity fingerprint spectrum.

[0036] The vertical axis of the map represents the vibration energy entropy, which is derived from the integration result of the time-frequency spectrum.

[0037] The edge weight is defined as: , where and represent the timestamps of the corrosion pulse event and the vibration event respectively. is the time - associated scale factor.

[0038] The corrosion feature enhancement channel uses a multi - head attention mechanism (4 heads) to calculate the spatial correlation degree between corrosion nodes: , where and are learnable parameter matrices, is the feature dimension, is the attention weight between corrosion nodes, representing the attention score of the -th corrosion node to the -th corrosion node, which is used to calculate the spatial correlation degree between nodes, represents the feature vector of corrosion node , represents the feature vector of corrosion node ;

[0039] The vibration - corrosion coupling channel filters out effective associated edges through a gated attention mechanism:

[0040] ;

[0041] ;

[0042] where is the Sigmoid function, is the gated weight matrix; represents the vector concatenation operation, is the gated attention value of the vibration - corrosion coupling channel, is the corrected edge weight, represents the pulse feature vector, represents the frequency - band energy entropy vector.

[0043] Optionally, the S4 specifically includes:

[0044] S41: Construct a chemical corrosion rate prediction channel; extract the average amplitude of the pulse current from the corrosion activity fingerprint spectrum, obtain the hydrogen sulfide concentration gradient and the growth rate of the biofilm thickness, use a temporal convolutional network to learn the kinetic characteristics of chemical corrosion, capture the corrosion change trends at different time scales, and establish a mapping relationship between the average amplitude, hydrogen sulfide concentration gradient and corrosion rate. Among them, the pulse current amplitude is used to measure the basic intensity of corrosion, the hydrogen sulfide concentration gradient affects the non - linear change of the corrosion reaction rate, and the growth rate of the biofilm thickness is used to reflect the contribution of microbial activities to corrosion;

[0045] S42: Construct a prediction channel for the mechanical crack growth rate; extract the vibration-corrosion mutual information from the corrosion-vibration correlation atlas, obtain the maximum principal stress and the amplitude of the stress intensity factor to characterize the direct effect of mechanical load on crack growth. In crack growth prediction, based on the traditional Paris law and considering the acceleration effect of corrosion on crack growth;

[0046] S43: Composite failure trajectory synthesis; establish a mapping relationship between the time scale of chemical corrosion and the number of mechanical cycles of mechanical fatigue, introduce time-axis alignment processing, correlate the evolution process of chemical corrosion with the crack growth process of fatigue cracks, so that multiple failure modes can be analyzed in the same time coordinate system. This mapping relationship adjusts the cumulative mode of fatigue cycles by considering the influence of corrosion on the operating frequency of the equipment. After time-axis alignment processing, calculate the composite failure trajectory curve. The composite failure trajectory curve describes the cumulative damage of corrosion and crack growth during service. Adopt the calculation method of equivalent service time, combine the chemical corrosion time and the number of mechanical cycles, and define the composite damage degree;

[0047] S44: Early warning mechanism; after obtaining the composite damage degree, set up an early warning mechanism according to different damage thresholds. When the damage degree exceeds the set damage threshold but still does not reach the critical value, trigger a first-level early warning and recommend planned maintenance to prevent further deterioration. If the damage degree continues to increase and its growth rate exceeds the critical value, trigger a second-level early warning and take emergency shutdown measures.

[0048] Optionally, in the early warning mechanism of S44:

[0049] When trigger a first-level early warning;

[0050] When the following two conditions are met, trigger a second-level early warning:

[0051] , where represents the composite damage degree, represents the change rate of the composite damage degree, is hour, is the equivalent service time.

[0052] A coal mine equipment fault prediction system based on big data, used to implement the above-mentioned coal mine equipment fault prediction method based on big data, includes the following modules:

[0053] Data acquisition module: Collect the thickness of the biofilm on the surface of the equipment through an ultrasonic sensor, and synchronously obtain the vibration acceleration of the equipment and the environmental hydrogen sulfide concentration to construct a corrosion multi-modal data set;

[0054] Signal processing module: It is used to separate the electrochemical noise from the corrosion multi-modal dataset, extract the characteristic current pulses of microbial corrosion by using the wavelet entropy algorithm, and generate the corrosion activity fingerprint spectrum;

[0055] Coupling analysis module: It is used to input the corrosion activity fingerprint spectrum into the graph attention network, decouple the coupling effect of mechanical vibration and microbial corrosion, and generate the corrosion-fatigue damage correlation map;

[0056] Life prediction and early warning module: It is used to construct a dual-channel life prediction model based on the corrosion-fatigue damage correlation map, calculate the composite failure trajectory curve of the chemical corrosion rate and the mechanical crack growth rate, and set a dynamic threshold early warning mechanism based on the composite failure trajectory curve to achieve the fault prediction and risk early warning of coal mine equipment.

[0057] Compared with the prior art, the beneficial effects of the present invention include:

[0058] Based on multi-modal data such as biofilm thickness, corrosion current pulse, vibration time-frequency characteristics, and hydrogen sulfide concentration gradient, the present invention constructs a corrosion-fatigue composite damage prediction model, accurately extracts the characteristic current of microbial corrosion through wavelet packet decomposition and wavelet entropy analysis, and uses the graph attention network to establish a corrosion-vibration heterogeneous graph to realize the analysis of the corrosion-induced structure vibration enhancement effect. Compared with the traditional single-signal-based prediction method, the present invention can jointly model in three dimensions of time, frequency, and space, effectively overcome problems such as the limited influence range of local corrosion and unstable vibration response, and improve the prediction accuracy of corrosion-fatigue damage.

[0059] Based on the Paris crack growth formula, the present invention introduces the vibration-corrosion mutual information quantity, extracts the corrosion-induced characteristics in the vibration signal through singular value decomposition, and calculates the dynamic correction of the corrosion influence factor on the crack growth rate. This method can accurately identify the local stress concentration area caused by microbial corrosion and predict the non-linear crack growth trend, thereby reducing the prediction error of crack initiation time and effectively extending the service life evaluation accuracy of the equipment.

[0060] The present invention proposes a time axis alignment algorithm. By establishing a non-linear mapping relationship between the chemical corrosion rate and the number of fatigue cycles, it realizes the synchronous calculation of the corrosion rate and the crack growth rate, ensures the consistency of failure prediction. At the same time, it uses a composite damage degree calculation model to quantify the cumulative damage degree of the equipment, and introduces a dynamic threshold early warning mechanism to trigger different levels of early warning when the damage degree exceeds the set value. Brief description of the drawings

[0061] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0062] Figure 1 Schematic flowchart of the prediction method according to an embodiment of the present invention;

[0063] Figure 2 Schematic diagram of the dual-channel attention network according to an embodiment of the present invention;

[0064] Figure 3 Schematic diagram of the functional modules of the prediction system according to an embodiment of the present invention. Detailed implementation manners

[0065] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0066] It should be noted that in the specification, when referring to "an embodiment", "embodiment", "exemplary embodiment", "some embodiments", etc., it indicates that the described embodiment may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. Additionally, when combining an embodiment to describe a specific feature, structure or characteristic, implementing such feature, structure or characteristic in combination with other embodiments (whether explicitly described or not) should be within the knowledge scope of those skilled in the relevant art.

[0067] Generally, terms can be understood at least in part from their use in the context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but rather, at least in part depending on the context, allowing for the existence of other factors that may not be explicitly described.

[0068] As Figure 1 - Figure 2 shown, the coal mine equipment fault prediction method based on big data includes the following steps:

[0069] S1: Deploy ultrasonic sensors on the metal surface of the device to collect the thickness of the biofilm on the device surface, synchronously collect the vibration acceleration of the device and the ambient hydrogen sulfide concentration, and construct a corrosion multi-modal dataset;

[0070] S2: Perform electrochemical noise separation on the corrosion multi-modal dataset, extract the characteristic current pulses of microbial corrosion through the wavelet entropy algorithm, and construct a corrosion activity fingerprint spectrum;

[0071] S3: Input the corrosion activity fingerprint spectrum into the graph attention network, decouple the coupling effect of mechanical vibration and microbial corrosion, and generate a corrosion-fatigue damage correlation map;

[0072] S4: Construct a dual-channel life prediction model based on the corrosion-fatigue damage correlation map, output a composite failure trajectory curve including the chemical corrosion rate and the mechanical crack growth rate, and set up an early warning mechanism based on the composite failure trajectory curve.

[0073] S1 specifically includes:

[0074] S11: The ultrasonic sensor emits ultrasonic pulses, monitors the reflected echo, and calculates the acoustic wave propagation time difference: , where is the biofilm thickness, is the ultrasonic propagation speed in the biofilm, taking 1550 m / s, is the propagation time delay of the ultrasonic wave;

[0075] S12: Conformally install a triaxial MEMS vibration sensor within 5 cm of the biofilm sensor on the device surface, and set the vibration sampling frequency to 10 kHz;

[0076] S13: Arrange multiple hydrogen sulfide electrochemical sensor arrays circumferentially around the device, and the electrochemical sensor arrays are densely arranged at a spacing of 30 cm to obtain ambient hydrogen sulfide concentration data;

[0077] S14: Use a hardware clock synchronizer to unify the sampling time reference of each sensor, control the clock deviation within ±100 μs, and generate an original data stream with a unified timestamp;

[0078] S15: Generate a corrosion multi-modal dataset including biofilm thickness, device vibration acceleration, and ambient hydrogen sulfide concentration.

[0079] S2 specifically includes:

[0080] S21: Based on the db4 wavelet basis function, perform multi-level decomposition on the electrochemical noise signals in the corrosion multi-modal dataset. In wavelet decomposition, each layer of the signal is split into a low-frequency part and a high-frequency part, forming multiple sub-bands. The decomposition process continues for eight levels, so that the signal is divided into multiple frequency ranges. Among all the sub-bands, it is necessary to focus on the characteristic signals related to microbial corrosion. Experimental studies have shown that the current pulses generated during microbial corrosion are mainly concentrated in a specific frequency range. Therefore, select the sub-band signals from the third layer to the fourth layer, and the corresponding frequency range is about 125 to 500 Hz. These sub-band signals contain the electrochemical perturbations caused by microbial activities and can better reflect the characteristics of the microbial corrosion process; Electrochemical noise is the current or potential fluctuation caused by the corrosion process. The electrochemical noise signal mainly comes from the electrochemical current fluctuation measured during the monitoring of the biofilm thickness, vibration acceleration, and environmental hydrogen sulfide concentration. They themselves do not belong to the electrochemical noise signal, but will indirectly affect the corrosion process. Mechanical vibration will change the ion distribution at the corrosion interface, resulting in current fluctuations. High-concentration hydrogen sulfide can accelerate the growth of the biofilm, thereby enhancing the electrochemical noise;

[0081] S22: Calculate the wavelet entropy values of each sub-band. After obtaining the target sub-band signals, analyze the energy distribution characteristics, use wavelet entropy analysis to quantify the complexity of the signal, and judge whether the signal includes corrosion characteristics. Statistically analyze the proportion of different energy components in each sub-band to obtain the energy distribution characteristics. Calculate the wavelet entropy values based on the energy distribution characteristics to measure the randomness and structural information content of the signal. The larger the wavelet entropy value, the higher the signal complexity and the more corrosion characteristics it may contain. Among all the sub-band signals, set an entropy threshold. When the wavelet entropy value of a certain sub-band exceeds the entropy threshold, it is judged that the current sub-band includes effective microbial corrosion information and is retained for subsequent feature extraction;

[0082] S23: For the selected sub-band signals, extract the microbial corrosion current pulse signals. Since the electrochemical noise signal has large fluctuations, it is difficult to accurately extract weak corrosion pulses using the traditional fixed threshold method. Therefore, adopt the dynamic threshold method for adaptive determination. Adaptively calculate the signal intensity threshold of each sub-band to make it adapt to the signal changes in different environments. When the actual signal intensity exceeds the signal intensity threshold, it is judged that it belongs to the corrosion current pulse. Among all the signal segments that exceed the signal intensity threshold, extract the key pulse parameters related to the corrosion process, including:

[0083] Pulse amplitude, used to characterize the intensity of the corrosion reaction;

[0084] Pulse duration, reflecting the time length of the corrosion reaction;

[0085] Pulse rise time, used to measure the suddenness of the corrosion process;

[0086] These parameters can comprehensively describe the characteristics of transient current pulses during microbial corrosion, providing basic data for subsequent analysis;

[0087] S24: After extracting all key pulse parameters, construct a three-dimensional feature vector. The three-dimensional feature vector includes the average amplitude of the pulse, the total duration, and the variation characteristics of the rise time. Use the K-means clustering algorithm to classify the corrosion activity, divide the corrosion signal into low-activity level, medium-activity level, and high-activity level. The classification is based on the pulse amplitude and duration indicators to ensure that the classification results can effectively correspond to different intensities of microbial corrosion behavior. Map the clustering results to the time-frequency dimension to form a corrosion activity fingerprint spectrum. In the corrosion activity fingerprint spectrum, the abscissa represents time, the ordinate represents the characteristic frequency, and the change in color reflects different levels of corrosion activity. Through this fingerprint spectrum, the intensity change of microbial corrosion at different time points can be clearly observed, providing a reliable basis for equipment fault prediction and maintenance.

[0088] Perform wavelet packet decomposition on the electrochemical noise signal in the corrosion multimodal dataset. Select the db4 wavelet basis function for 8-layer decomposition, and extract the microbial corrosion characteristic subbands in the 3rd - 4th layers (corresponding to the frequency range of 125 - 500 Hz); The db4 wavelet has 4 vanishing moments and can better extract mutation features (corrosion current pulses). For the input signal, wavelet packet decomposition uses a pair of filters, including a low-pass filter (approximate component) and a high-pass filter (detail component). Each decomposition generates two sub-signals, including a high-frequency component and a low-frequency component. Use the db4 wavelet basis to decompose the signal into 8 layers. Each layer of decomposition divides the signal into a low-frequency (approximate component) and a high-frequency (detail component). After 8-layer decomposition, there are 256 subbands in total. The frequency range of each layer is recursively divided by the dichotomy method. Extract the microbial corrosion characteristic subbands in the 3rd - 4th layers (corresponding to the frequency range of 125 - 500 Hz), and select the 125 - 500 Hz frequency band as the microbial corrosion signal characteristic interval.

[0089] In calculating the wavelet entropy value of each subband in S22, for each subband , its wavelet entropy value is calculated as:

[0090] , where is the energy proportion of the th layer and the rd node:

[0091] , where represents the energy of this node. When , it is determined that this layer contains effective corrosion features, is the entropy value threshold.

[0092] Signal intensity threshold It is calculated as: , where is the standard deviation of the signal of the th layer, and

[0093] is the number of sampling points of this layer; For all pulses exceeding the signal intensity threshold

[0094] the following key parameters are extracted: : ;

[0095] Pulse duration : ;

[0096] Rise time : ;

[0097] where and are the start and end times of the pulse respectively, is the time point at which the maximum pulse amplitude appears, is the current signal, representing the instantaneous current value at time for detecting corrosion pulses:

[0098] When constructing the corrosion activity fingerprint spectrum, all pulse parameters need to be statistically analyzed to generate a three-dimensional feature vector :

[0099] , where is the average amplitude of all pulses, is the logarithm of the total duration of all pulses, is the standard deviation of the rise time;

[0100] K-means clustering is used to classify corrosion activity into low / medium / high levels:

[0101] Low activity: ;

[0102] Medium activity: ;

[0103] High activity: ;

[0104] The clustering results are mapped to the time-frequency dimension to generate a fingerprint spectrum, including:

[0105] Abscissa: Time ;

[0106] Ordinate: Characteristic frequency ;

[0107] The chromaticity represents the corrosion activity intensity (from blue (low) to red (high)).

[0108] S3 specifically includes:

[0109] S31: Construct a heterogeneous graph structure to describe the association between corrosion nodes and vibration nodes. The corrosion nodes are pulse feature vectors extracted from the corrosion activity fingerprint spectrum, and the vibration nodes are band energy entropy vectors extracted from the vibration time-frequency spectrum. After constructing the nodes, define the association relationship between corrosion and vibration. Since there may be a certain time offset between corrosion and vibration events, a weighted mechanism based on time difference is adopted to assign an edge weight to each corrosion-vibration connection edge. The value of the edge weight decays as the time difference between the corrosion node and the vibration node increases, ensuring that events closer in time have higher correlation;

[0110] S32: Based on the heterogeneous graph structure, construct a dual-channel graph attention network, including a corrosion feature enhancement channel and a vibration-corrosion coupling channel, where:

[0111] The corrosion feature enhancement channel is used to analyze the mutual correlation between corrosion nodes. The multi-head attention mechanism is adopted to calculate the attention of each corrosion node to other corrosion nodes, adjust the feature expression of the corrosion nodes. Each corrosion node calculates its own correlation with other corrosion nodes, and uses the attention weighting method to update its own features, enhancing the corrosion features to make them more in line with the actual corrosion damage pattern;

[0112] The vibration-corrosion coupling channel is used to analyze the influence of vibration signals on the corrosion process. Since vibration signals may contain a large amount of irrelevant or noisy information, when calculating the corrosion-vibration association, a gating mechanism is first used to screen effective association edges. The gating mechanism calculates a dynamic weight based on the joint features of corrosion and vibration to correct the initial time association edge weight, making the final corrosion-vibration coupling relationship more accurate. Finally, the corrected edge weight is used to adjust the influence degree of vibration on corrosion nodes;

[0113] S33: Use the singular value decomposition method to perform dimensionality reduction on the embedded features of corrosion-vibration nodes and decouple the coupling effect, including calculating the mutual information between corrosion and vibration in the reduced feature space. The mutual information measures the statistical correlation between vibration signals and corrosion signals. The larger the value, the stronger the coupling degree between the two. When the mutual information exceeds a predetermined threshold, it indicates that the vibration signal and the corrosion signal have a strong dependence relationship, indicating that the corrosion process exacerbates the fatigue vibration of the equipment;

[0114] S43: After decoupling the corrosion-vibration relationship, construct a generated corrosion-fatigue damage association map to describe their coupling mode. The association map includes the horizontal axis and the vertical axis of the map, where:

[0115] The horizontal axis of the spectrum represents the corrosion activity intensity, which is derived from the chromaticity information of the corrosion activity fingerprint spectrum;

[0116] The vertical axis of the spectrum represents the vibration energy entropy, which is derived from the integration result of the time-frequency spectrum;

[0117] The correlation strength coding is visualized through colors:

[0118] In the strong coupling region, that is, the region with a higher corrosion-vibration mutual information, it is marked as a red contour line.

[0119] In the independent action region, that is, the region where there is no obvious correlation between corrosion and vibration, it is marked as a blue background.

[0120] To further identify the critical points that may cause equipment damage, the critical value of fatigue crack initiation is marked on this spectrum. When the weighted combination of corrosion activity and vibration energy entropy in a specific frequency band exceeds the highest point, it is considered that the equipment is in a high-risk state of fatigue crack initiation and maintenance or protection measures need to be taken.

[0121] When constructing the heterogeneous graph structure, two types of nodes are defined:

[0122] Corrosion nodes: Pulse feature vectors extracted from the corrosion activity fingerprint spectrum:

[0123] ;

[0124] Vibration nodes: Band energy entropy vectors extracted from the vibration time-frequency spectrum:

[0125] , separately extract the energy entropy values of the 3 kHz, 5 kHz, and 8 kHz frequency bands in the vibration time-frequency spectrum to measure the complexity of the signal energy distribution in this frequency band;

[0126] The edge weight is defined as: ;

[0127] Among them, and respectively represent the timestamps of the corrosion pulse event and the vibration event, is the time correlation scale factor.

[0128] The corrosion feature enhancement channel uses a multi-head attention mechanism (4 heads) to calculate the spatial correlation degree between corrosion nodes: , among which, and are learnable parameter matrices, is the feature dimension, , is the attention weight between corrosion nodes, indicating the th corrosion node's pair with the The attention score of a corrosion node, which is used to calculate the spatial correlation degree between nodes, represents the corrosion node 's eigenvector, represents the corrosion node 's eigenvector;

[0129] The vibration-corrosion coupling channel filters out effective associated edges through a gated attention mechanism:

[0130] ;

[0131] ;

[0132] Among them, is the Sigmoid function, is the gated weight matrix; represents the vector concatenation operation, is the gated attention value of the vibration-corrosion coupling channel, is the corrected edge weight, represents the pulse eigenvector, represents the band energy entropy vector.

[0133] In decoupling the coupling effect: perform singular value decomposition (SVD) on the node embedding: , where, is the output feature matrix of the -th layer network. Only the first 3 singular vectors are retained to construct the decoupling subspace, represents the result of singular value decomposition (SVD): is the left singular vector matrix, representing the main feature directions of the corrosion and vibration nodes, is the singular value matrix, representing the importance of each principal component, is the right singular vector matrix, representing the projection relationship of features between different nodes. After singular value decomposition, the corrosion-vibration features are dimensionally reduced and decoupled. The original embedded feature may contain the coupling information of corrosion and vibration, but it may also have redundant or irrelevant parts. After decomposition, only the first 3 singular vectors are retained to construct the decoupling subspace, that is, the main associated features are retained while reducing noise interference. The feature subspace after SVD dimensional reduction can accurately represent the core relationship between corrosion and vibration without being affected by data redundancy.

[0134] Calculate the vibration-corrosion mutual information within the decoupling subspace :

[0135] ;

[0136] When: , is a predetermined threshold value, and it is determined that there is a strong coupling effect between vibration and corrosion. represents the joint probability distribution and represents the characteristics of vibration nodes and the characteristics of corrosion nodes The probability of co-occurrence, represents the marginal probability distribution of the characteristics of vibration nodes, The marginal probability distribution of the characteristics of corrosion nodes.

[0137] In the corrosion-fatigue damage correlation map:

[0138] 1. Horizontal axis: Corrosion activity intensity (fingerprint spectrum chromaticity value);

[0139] 2. Vertical axis: Vibration energy entropy (integral value of time-frequency spectrum);

[0140] 3. Correlation strength coding:

[0141] Red contour line: When it is marked as a strong coupling area;

[0142] Blue area: When it is marked as an independent action area;

[0143] 4. Mark the critical point of fatigue crack initiation: , where As the fatigue crack initiation index under the combined action of corrosion and vibration, when this value exceeds 0.75, it is determined that the equipment enters the high-risk fatigue damage stage. The 5 kHz frequency band is usually related to the vibration characteristics of material fatigue crack propagation, and the energy change of the high-frequency vibration signal in this frequency band can reflect the crack initiation and propagation process.

[0144] The frequency band energy entropy vector extracted from the vibration time-frequency spectrum is specifically as follows:

[0145] 1. Generation of time-frequency spectrum: Perform short-time Fourier transform on the original vibration signal :

[0146] ;

[0147] where is the Nuttall window, the window length is 50 ms, and the frequency resolution is 0.5 Hz (meeting the detection requirements of the high-frequency vibration characteristics of coal mine equipment), represents the time-frequency spectrum, which is the representation of the vibration signal in the time-frequency domain obtained by short-time Fourier transform, representing the energy distribution at time t and frequency f.

[0148] 2. Divide the key frequency bands based on the equipment failure mechanism:

[0149] 3 kHz: 2.8 - 3.2 kHz: corresponding to the defect characteristic frequency of the bearing outer ring;

[0150] 5 kHz: 4.8 - 5.2 kHz: corresponding to the gear meshing sub - harmonic resonance frequency;

[0151] 8 kHz: 7.6 - 8.4 kHz: corresponding to the high - frequency resonance frequency of micro - cracks.

[0152] 3. Band energy Calculation:

[0153] Perform energy integration for each frequency band: , where represents the integration duration, which refers to the time - window length during the calculation of band energy, , covering the fluctuation range of characteristic frequencies, represents the center frequency, is the spectrum value after Fourier transform, Take 3 kHz, 5 kHz, 8 kHz, represents the start time of the window, respectively represent the integration micro - elements of the time variable and the frequency variable;

[0154] 4. Solving the energy entropy value:

[0155] Normalization: , is the normalized band energy;

[0156] Calculate the band energy entropy: ;

[0157] The final band energy entropy vector: .

[0158] S4 specifically includes:

[0159] S41: Construct a chemical corrosion rate prediction channel; extract the average amplitude of the pulsed current from the corrosion activity fingerprint spectrum, obtain the hydrogen sulfide concentration gradient and the growth rate of the biofilm thickness, use a time - convolutional network to learn the kinetic characteristics of chemical corrosion, capture the corrosion change trends at different time scales, and establish a mapping relationship between the average amplitude, hydrogen sulfide concentration gradient and corrosion rate. Among them, the pulsed current amplitude is used to measure the basic intensity of corrosion, the hydrogen sulfide concentration gradient affects the non - linear change of the corrosion reaction rate, and the growth rate of the biofilm thickness is used to reflect the contribution of microbial activity to corrosion;

[0160] S42: Construct a prediction channel for the mechanical crack growth rate; extract the vibration-corrosion mutual information from the corrosion-vibration correlation map, obtain the maximum principal stress and the amplitude of the stress intensity factor to characterize the direct effect of mechanical loads on crack growth. In crack growth prediction, based on the traditional Paris law and considering the acceleration effect of corrosion on crack growth, the crack growth rate depends not only on the amplitude of the stress intensity factor but also on the vibration-corrosion mutual information. When the coupling degree of corrosion and vibration is high, the crack growth rate will increase significantly, thus accelerating material failure;

[0161] S43: Composite failure trajectory synthesis; establish a mapping relationship between the time scale of chemical corrosion and the number of mechanical cycles of mechanical fatigue, introduce time-axis alignment processing, correlate the evolution process of chemical corrosion with the crack growth process of fatigue cracks, and analyze multiple failure modes in the same time coordinate system. This mapping relationship adjusts the cumulative mode of fatigue cycles by considering the influence of corrosion on the operating frequency of the equipment. After time-axis alignment processing, calculate the composite failure trajectory curve. The composite failure trajectory curve describes the cumulative damage of corrosion and crack growth during service. Using the calculation method of equivalent service time, combine the chemical corrosion time and the number of mechanical cycles to define the composite damage degree. The composite damage degree takes into account the influence of corrosion rate and crack length. The calculation method of the damage degree adopts a non-linear combination method to avoid the irrationality that may be caused by simple linear superposition;

[0162] S44: Early warning mechanism; after obtaining the composite damage degree, set the early warning mechanism according to different damage thresholds. When the damage degree exceeds the set damage threshold but still does not reach the critical value, trigger a first-level warning and recommend planned maintenance to prevent further deterioration. If the damage degree continues to increase and its growth rate exceeds the critical value, trigger a second-level warning and take emergency shutdown measures to prevent sudden failure of the equipment.

[0163] The threshold setting of the early warning mechanism refers to the international material durability standard and the failure assessment standard of pipeline engineering to ensure the reliability and applicability of the early warning system. The threshold of the first-level warning corresponds to the safe range of the equipment durability limit, while the threshold of the second-level warning corresponds to the failure critical point in engineering applications. By dynamically adjusting the early warning threshold, false alarms and missed alarms can be effectively reduced, and the prediction accuracy can be improved.

[0164] The input features when constructing the chemical corrosion rate prediction channel include:

[0165] : The pulse average amplitude in the corrosion activity fingerprint spectrum;

[0166] : The hydrogen sulfide concentration gradient, and the hydrogen sulfide concentration at different positions is measured by a hydrogen sulfide electrochemical sensor array , and then calculate the rate of change of concentration along the space: , where is the difference in hydrogen sulfide concentration measured by adjacent sensors, is the sensor spacing;

[0167] : growth rate of biofilm thickness, is the unit change in biofilm thickness, is the time increment of the unit change in biofilm thickness;

[0168] The calculation of the chemical corrosion rate is based on the Temporal Convolutional Network (TCN):

[0169] ,

[0170] where is the proportionality coefficient (empirical parameter) in the corrosion rate calculation, used to adjust the influence of the pulse average amplitude on the chemical corrosion rate, is the normalization parameter of the hydrogen sulfide concentration gradient, used to normalize the influence of the hydrogen sulfide concentration gradient on the corrosion rate, is the influence coefficient of the growth rate of biofilm thickness, indicating the contribution of the change in biofilm thickness to the chemical corrosion rate, , , , all of which are material-related coefficients.

[0171] The input features when constructing the mechanical crack growth rate prediction channel include:

[0172] : vibration-corrosion mutual information;

[0173] : maximum principal stress, obtained by measuring with a strain gauge sensor, based on the material stress condition: , where is the external load, is the stress cross-sectional area;

[0174] : stress intensity factor amplitude, based on crack size and load changes, according to the classical formula: , where is the maximum principal stress, is the crack length;

[0175] Modified Paris law model (coupling corrosion and crack growth effects):

[0176] , where , , (representing the coupling effect coefficient) is the change in crack length, representing the growth of the crack in one loading cycle. represents a small increment in the number of loading cycles, indicating the fatigue load cycle step experienced during crack propagation.

[0177] In the synthesis of composite failure trajectories:

[0178] Time-axis alignment processing: Establish the chemical corrosion time scale through the following formula and the number of mechanical cycles to establish the mapping relationship between them: , where is the fundamental frequency of the equipment , represents the corrosion-frequency coupling coefficient;

[0179] Parameterized representation of the failure trajectory curve:

[0180] Abscissa: Equivalent service time : , where (is the weighting factor);

[0181] Ordinate: Composite damage degree : , where (represents the critical corrosion rate), (represents the crack threshold length).

[0182] In the early warning mechanism:

[0183] When , trigger a first-level early warning;

[0184] When the following two conditions are met, trigger a second-level early warning:

[0185] , where represents the composite damage degree, represents the change rate of the composite damage degree, is hours, is the equivalent service time.

[0186] As Figure 3 shown, a coal mine equipment fault prediction system based on big data is used to implement the above prediction method, including the following modules:

[0187] Data acquisition module: Collect the thickness of the biofilm on the equipment surface through an ultrasonic sensor, and synchronously obtain the equipment vibration acceleration and environmental hydrogen sulfide concentration to construct a corrosion multi-modal dataset;

[0188] Signal processing module: used to separate electrochemical noise from the corrosion multi-modal dataset, extract microbial corrosion characteristic current pulses using the wavelet entropy algorithm, and generate a corrosion activity fingerprint spectrum;

[0189] Coupling analysis module: used to input the corrosion activity fingerprint spectrum into the graph attention network, decouple the coupling effect of mechanical vibration and microbial corrosion, and generate a corrosion-fatigue damage correlation map;

[0190] Life prediction and early warning module: used to construct a dual-channel life prediction model based on the corrosion-fatigue damage correlation map, calculate the composite failure trajectory curve of the chemical corrosion rate and the mechanical crack growth rate, and set a dynamic threshold early warning mechanism based on the composite failure trajectory curve to achieve fault prediction and risk early warning of coal mine equipment.

[0191] This invention covers any alternatives, modifications, equivalent methods, and solutions made within the essence and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc., are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0192] The above are only the preferred embodiments of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. A method for predicting coal mine equipment failures based on big data, characterized in that, It includes the following steps: S1: Deploy ultrasonic sensors on the metal surface of the device to collect the thickness of the biofilm on the device surface, synchronously collect the vibration acceleration of the device and the ambient hydrogen sulfide concentration, and construct a corrosion multi-modal dataset. The corrosion multi-modal dataset includes electrochemical noise signals caused by current potential fluctuations during the device corrosion process; S2: Separate the electrochemical noise signals in the corrosion multi-modal dataset, extract the microbial corrosion characteristic current pulses through the wavelet entropy algorithm, and obtain the key pulse parameters including pulse amplitude, pulse duration, and pulse rise time; count all the key pulse parameters, construct a three-dimensional feature vector including average amplitude, total duration, and rise time fluctuation, use K-means clustering for activity grading, and map the clustering results to the time-frequency dimension to form a corrosion activity fingerprint spectrum. The abscissa of the corrosion activity fingerprint spectrum is time, the ordinate is frequency, and the chromaticity represents the corrosion activity level; S3: Input the corrosion activity fingerprint spectrum into the graph attention network to decouple the coupling effect of mechanical vibration and microbial corrosion, and generate a corrosion-fatigue damage association map, specifically including: based on the corrosion activity fingerprint spectrum and the vibration time-frequency spectrum, construct a heterogeneous graph structure including corrosion nodes and vibration nodes, establish association edges through a weighted mechanism based on time difference, assign an edge weight to each association edge, construct a two-channel graph attention network to extract corrosion features and vibration-corrosion coupling relationships, use singular value decomposition to decouple the node embedding features and calculate the vibration-corrosion mutual information, perform association strength encoding based on the value of the mutual information, and visualize the association strength encoding through colors to distinguish the strong coupling region and the independent action region, and finally generate a corrosion-fatigue damage association map including vibration-corrosion mutual information. The horizontal axis of the corrosion-fatigue damage association map is the corrosion activity intensity, and the vertical axis is the vibration energy entropy, which is used to characterize the damage mode caused by the coupling of the two; S4: Construct a two-channel life prediction model based on the corrosion-fatigue damage association map, output a composite failure trajectory curve including the chemical corrosion rate and the mechanical crack growth rate, set an early warning mechanism based on the composite failure trajectory curve. The two-channel life prediction model includes a chemical corrosion rate prediction channel and a mechanical crack growth rate prediction channel. Among them, the chemical corrosion rate prediction channel is based on the average amplitude of the pulse current, the hydrogen sulfide concentration gradient, and the biofilm thickness growth rate, and uses a temporal convolutional network to predict the corrosion rate; the mechanical crack growth rate prediction channel is based on the vibration-corrosion mutual information, the maximum principal stress, and the stress intensity factor amplitude, and modifies the Paris formula to predict the crack growth rate; introduce time axis alignment processing to associate the evolution process of chemical corrosion with the growth process of fatigue cracks, so that multiple failure modes can be analyzed in the same time coordinate system, calculate the composite failure trajectory and the equivalent service time, construct a composite damage degree based on the joint evaluation of the corrosion rate and the crack length, and set a multi-level early warning mechanism based on the composite damage degree to judge the equipment failure risk level.

2. The method for predicting coal mine equipment failures based on big data according to claim 1, wherein The specific content of S1 includes: S11: The ultrasonic sensor emits ultrasonic pulses, monitors the reflected echoes, and calculates the time difference of acoustic wave propagation: , where is the biofilm thickness, is the ultrasonic propagation speed in the biofilm, taking 1550 m / s, is the propagation time delay of the ultrasonic wave; S12: Install a three-axis MEMS vibration sensor on the device surface, and set the vibration sampling frequency to 10 kHz; S13: Arrange multiple hydrogen sulfide electrochemical sensor arrays circumferentially around the device to obtain ambient hydrogen sulfide concentration data; S14: Use a hardware clock synchronizer to unify the sampling time reference of each sensor and generate a raw data stream with a unified timestamp; S15: Generate a corrosion multi-modal dataset including biofilm thickness, device vibration acceleration, and ambient hydrogen sulfide concentration.

3. The method for predicting faults of coal mine equipment based on big data according to claim 1, wherein, The specific steps of S2 are as follows: S21: Based on the db4 wavelet basis function, perform multi-level decomposition on the electrochemical noise signals in the corrosion multi-modal dataset. During wavelet decomposition, each layer of the signal is split into a low-frequency part and a high-frequency part to form multiple sub-bands. The decomposition process continues for eight levels, dividing the signal into multiple frequency ranges; S22: Calculate the wavelet entropy value of each sub-band. After obtaining the target sub-band signal, analyze the energy distribution characteristics, use wavelet entropy analysis to quantify the complexity of the signal, determine whether the signal contains corrosion characteristics, count the proportion of different energy components in each sub-band to obtain the energy distribution characteristics, and calculate the wavelet entropy value based on the energy distribution characteristics to measure the randomness and structural information content of the signal. The larger the wavelet entropy value, the higher the signal complexity. Among all sub-band signals, set an entropy value threshold. When the wavelet entropy value of a certain sub-band exceeds the entropy value threshold, it is determined that the current sub-band contains effective microbial corrosion information and is retained for subsequent feature extraction; S23: Adaptively calculate the signal intensity threshold for each sub-band to adapt to signal changes in different environments. When the actual signal intensity exceeds the signal intensity threshold, it is determined that it belongs to a corrosion current pulse. Among all signal segments that exceed the signal intensity threshold, extract key pulse parameters related to the corrosion process, including: Pulse amplitude, which is used to characterize the intensity of the corrosion reaction; Pulse duration, which reflects the time length of the corrosion reaction; Pulse rise time, which is used to measure the suddenness of the corrosion process; S24: After extracting all key pulse parameters, construct a three-dimensional feature vector. The three-dimensional feature vector includes the average amplitude of the pulse, the total duration, and the change characteristics of the rise time. Use the K-means clustering algorithm to classify the corrosion activity, divide the corrosion signals into low activity level, medium activity level, and high activity level. The classification of the levels is based on the pulse amplitude and duration indicators. Map the clustering results to the time-frequency dimension to form a corrosion activity fingerprint spectrum. In the corrosion activity fingerprint spectrum, the abscissa represents time, the ordinate represents the characteristic frequency, and the change in color reflects different levels of corrosion activity.

4. The method for predicting faults of coal mine equipment based on big data according to claim 3, wherein, In calculating the wavelet entropy values of each subband in S22, for each subband , its wavelet entropy value is calculated as: , where is the energy proportion of the th node in the th layer: , where represents the energy of the th node in the th layer. When , it is determined that this layer contains effective corrosion features, is the entropy value threshold.

5. The method for predicting faults of coal mine equipment based on big data according to claim 4, wherein, The signal strength threshold is calculated as: , where is the standard deviation of the signals of the th layer, and is the number of sampling points of this layer.

6. The method for predicting coal mine equipment failures based on big data according to claim 1, characterized in that The specific steps of S3 are as follows: S31: Construct a heterogeneous graph structure to describe the association between corrosion nodes and vibration nodes. Corrosion nodes are pulse feature vectors extracted from the corrosion activity fingerprint spectrum; vibration nodes are band energy entropy vectors extracted from the vibration time-frequency map. After constructing the nodes, define the association relationship between corrosion and vibration, and use a weighted mechanism based on time difference to assign an edge weight to each corrosion-vibration connection edge. The value of the edge weight decays as the time difference between the corrosion node and the vibration node increases, ensuring that events closer in time have higher correlation; S32: Construct a dual-channel graph attention network based on the heterogeneous graph structure, including an erosion feature enhancement channel and a vibration-erosion coupling channel, where: The erosion feature enhancement channel is used to analyze the mutual correlation between erosion nodes. The multi-head attention mechanism is adopted to calculate the attention of each erosion node to other erosion nodes, adjust the feature expression of the erosion nodes. Each erosion node calculates its own correlation with other erosion nodes, and uses the attention weighting method to update its own features to enhance the erosion features. The vibration-erosion coupling channel is used to analyze the influence of vibration signals on the erosion process. The gating mechanism is adopted to screen out effective associated edges. The gating mechanism calculates a dynamic weight according to the joint features of erosion and vibration to correct the initial time-associated edge weight. The corrected edge weight is used to adjust the influence degree of vibration on erosion nodes. S33: Use the singular value decomposition method to perform dimensionality reduction on the embedded features of erosion-vibration nodes and decouple the coupling effect, including calculating the mutual information between erosion and vibration in the reduced feature space. The mutual information measures the statistical correlation between vibration signals and erosion signals. The larger the value, the stronger the coupling degree between the two. When the mutual information exceeds a predetermined threshold, it indicates that the vibration signal and the erosion signal have a strong dependence relationship, indicating that the erosion process exacerbates the fatigue vibration of the equipment. After decoupling the relationship between erosion and vibration, construct a generated erosion-fatigue damage correlation map to describe their coupling mode. The correlation map includes the horizontal axis and the vertical axis of the map, where: The horizontal axis of the map represents the erosion activity intensity, which is derived from the chromaticity information of the erosion activity fingerprint spectrum. The vertical axis of the map represents the vibration energy entropy, which is derived from the integration result of the time-frequency map.

7. The method for predicting faults of coal mine equipment based on big data according to claim 6, characterized in that, The edge weight is defined as: , where and represent the timestamps of the corrosion pulse event and the vibration event respectively, is the time correlation scale factor; The corrosion feature enhancement channel uses a multi-head attention mechanism to calculate the spatial correlation degree between corrosion nodes: , where and are learnable parameter matrices, is the feature dimension, is the attention weight between corrosion nodes, representing the attention score of the th corrosion node to the th corrosion node, which is used to calculate the spatial correlation degree between nodes, represents the feature vector of the corrosion node , represents the feature vector of the corrosion node ; The vibration-erosion coupling channel screens out effective associated edges through the gating attention mechanism: ； ； Among them, is the Sigmoid function, is the gating weight matrix; represents the vector concatenation operation, is the gating attention value of the vibration-corrosion coupling channel, is the corrected edge weight, represents the pulse feature vector, represents the band energy entropy vector.

8. The method for predicting coal mine equipment faults based on big data according to claim 1, characterized in that, The specific steps of S4 include: S41: Construct a chemical corrosion rate prediction channel; extract the average amplitude of the pulsed current from the erosion activity fingerprint spectrum, obtain the hydrogen sulfide concentration gradient and the growth rate of the biofilm thickness. Use the time convolutional network to learn the kinetic characteristics of chemical corrosion, capture the corrosion change trends at different time scales, and establish the mapping relationship between the average amplitude, hydrogen sulfide concentration gradient and corrosion rate. Among them, the pulsed current amplitude is used to measure the basic intensity of corrosion, the hydrogen sulfide concentration gradient affects the non-linear change of the corrosion reaction rate, and the growth rate of the biofilm thickness is used to reflect the contribution of microbial activity to corrosion. S42: Construct a mechanical crack growth rate prediction channel; extract the vibration-erosion mutual information from the erosion-vibration correlation map, obtain the maximum principal stress and the amplitude of the stress intensity factor to characterize the direct influence of mechanical load on crack growth. In crack growth prediction, based on the traditional Paris law and considering the acceleration effect of erosion on crack growth. S43: Composite failure trajectory synthesis; establish a mapping relationship between the time scale of chemical corrosion and the number of mechanical cycles of mechanical fatigue, introduce time-axis alignment processing, correlate the evolution process of chemical corrosion with the propagation process of fatigue cracks, analyze multiple failure modes in the same time coordinate system. After time-axis alignment processing, calculate the composite failure trajectory curve. The composite failure trajectory curve describes the cumulative damage of corrosion and crack propagation during service. Using the calculation method of equivalent service time, combine the chemical corrosion time and the number of mechanical cycles to define the composite damage degree; S44: Early warning mechanism; after obtaining the composite damage degree, set the early warning mechanism according to different damage thresholds. When the damage degree exceeds the set damage threshold but still does not reach the critical value, trigger a first-level warning and recommend planned maintenance. If the damage degree continues to increase and its growth rate exceeds the critical value, trigger a second-level warning and take emergency shutdown measures.

9. The method for predicting faults of coal mine equipment based on big data according to claim 8, characterized in that, In the early warning mechanism of S44: When it triggers a first-level warning; When the following two conditions are met, trigger a second-level warning: , where represents the composite damage degree, represents the change rate of the composite damage degree, is in hours, is the equivalent service time.

10. A coal mine equipment fault prediction system based on big data, which is used to implement the coal mine equipment fault prediction method based on big data according to any one of claims 1-9, characterized in that, Including the following modules: Data acquisition module: Collect the biofilm thickness on the surface of the equipment through an ultrasonic sensor, and simultaneously obtain the equipment vibration acceleration and the ambient hydrogen sulfide concentration to construct a corrosion multi-modal data set; Signal processing module: Used to separate the electrochemical noise from the corrosion multi-modal data set, and extract the characteristic current pulse of microbial corrosion using the wavelet entropy algorithm to generate a corrosion activity fingerprint spectrum; Coupling analysis module: Used to input the corrosion activity fingerprint spectrum into the graph attention network, decouple the coupling effect of mechanical vibration and microbial corrosion, and generate a corrosion-fatigue damage correlation map; Life prediction and early warning module: Used to construct a two-channel life prediction model based on the corrosion-fatigue damage correlation map, calculate the composite failure trajectory curve of the chemical corrosion rate and the mechanical crack propagation rate, and set a dynamic threshold early warning mechanism based on the composite failure trajectory curve to achieve fault prediction and risk warning of coal mine equipment.

Images