CN-121980444-A - Coal mine equipment abnormality monitoring method with depth features and manual features fused

CN121980444ACN 121980444 ACN121980444 ACN 121980444ACN-121980444-A

Abstract

The application relates to the technical field of equipment monitoring and fault diagnosis, and discloses a coal mine equipment anomaly monitoring method with depth characteristics and manual characteristics fused, which comprises the following steps of carrying out standardized processing on vibration acceleration signals to obtain standardized signals; the method comprises the steps of carrying out Hilbert-Huang transform decomposition and multi-scale entropy calculation on the standardized signals, inputting the standardized signals into a one-dimensional convolutional neural network and a two-way long-short-term memory network which are connected in series, carrying out weighting processing on the deep hidden characteristic sequence by using the attention weight, inputting the comprehensive characteristic vector into a multi-layer perceptron, and generating a monitoring result representing the abnormal state of coal mine equipment through the full-connection layer and the activation function processing of the multi-layer perceptron. According to the application, through the fusion of the attention mechanism guided by the local entropy value and the depth characteristic, the fault impact is automatically focused under strong noise, the complementation of the mechanism and the data advantage is realized, and the robustness and the accuracy of the abnormality monitoring of the coal mine equipment are improved.

Inventors

LI ZHIRONG
HUANG YANDE
WANG WEITAO
Kang shuai
ZHU XINLONG
Qiao Jinniu
XU YONGXIANG
LIU JUNFENG
LI JIAN

Assignees

陕煤集团神木柠条塔矿业有限公司
天地科技股份有限公司
中煤科工开采研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20251229

Claims (10)

1. The coal mine equipment abnormality monitoring method based on depth feature and manual feature fusion is characterized by comprising the following steps of: S1, acquiring vibration acceleration signals of coal mine equipment, and performing standardized processing on the vibration acceleration signals to obtain standardized signals; S2, performing Hilbert-Huang transform decomposition and multi-scale entropy calculation on the standardized signal to obtain a local entropy sequence and a global entropy vector; S3, inputting the standardized signals into a one-dimensional convolutional neural network and a bidirectional long-short-time memory network which are connected in series to obtain a deep hidden characteristic sequence containing time step information; S4, calculating attention weights based on the local entropy value sequence and the depth hidden feature sequence, and carrying out weighting processing on the depth hidden feature sequence by using the attention weights to obtain weighted depth feature vectors; And S5, splicing the global entropy value vector and the weighted depth feature vector to obtain a comprehensive feature vector, inputting the comprehensive feature vector into a multi-layer perceptron, and generating a monitoring result representing the abnormal state of the coal mine equipment through the full-connection layer and the activation function processing of the multi-layer perceptron.
2. The method for monitoring anomalies in coal mine equipment with fusion of depth features and manual features according to claim 1, wherein in step S1, the vibration acceleration signal comprises: a non-stationary time series signal acquired from a sensor of a shearer cutting section or a drag conveyor reduction box; the time series signal comprises equipment vibration data when the coal mine underground working condition operates.
3. The method for monitoring abnormality of coal mine equipment by combining depth features and manual features according to claim 1, wherein in step S1, the normalizing the vibration acceleration signal specifically comprises: Calculating the mean value and standard deviation of the acquired vibration acceleration signals; Subtracting the mean value from the vibration acceleration signal, and dividing the mean value by the standard deviation to obtain data after Z-score standardization; dividing the standardized data into a plurality of sample fragments according to a preset time length, and taking the sample fragments as the standardized signal.
4. The method for monitoring abnormality of coal mine equipment by fusion of depth features and manual features according to claim 1, wherein in step S2, the performing hilbert-yellow transform decomposition and multi-scale entropy calculation comprises: Decomposing the standardized signal into a plurality of inherent mode function components and monotonic trend term residual errors by adopting an empirical mode decomposition method; Respectively carrying out Hilbert transformation on the natural mode function components obtained by decomposition to obtain instantaneous frequency distribution and instantaneous amplitude distribution of each component; Constructing a reconstruction vector for the normalized signal or the natural mode function component based on a phase space reconstruction technique; Calculating Chebyshev distances among the reconstruction vectors, and introducing an exponential form fuzzy membership function to calculate the similarity among the vectors, wherein independent variables of the fuzzy membership function comprise the Chebyshev distances, boundary gradient parameters and similarity tolerance; and calculating to obtain the fuzzy entropy value according to the logarithmic difference value of the average value of the vector similarity under different dimensions.
5. The method for monitoring abnormality of coal mine equipment by fusion of depth features and manual features according to claim 1, wherein in step S2, the local entropy value sequence and the global entropy value vector specifically include: Setting sliding windows moving along a time axis, calculating a fuzzy entropy value in each sliding window according to the standardized signal, and arranging the fuzzy entropy values calculated by all the sliding windows according to time sequence to form a local entropy value sequence corresponding to the time step of the depth hidden characteristic sequence; And calculating an arrangement entropy mean value and a fuzzy entropy mean value of the whole section of the standardized signal, extracting energy entropy values of a plurality of inherent mode function components in the standardized signal, and combining the arrangement entropy mean value, the fuzzy entropy mean value and the energy entropy values to form the global entropy vector.
6. The method for monitoring abnormality of coal mine equipment by combining depth features and manual features according to claim 1, wherein in step S3, the inputting the standardized signals into the serial one-dimensional convolutional neural network and the bidirectional long-short-term memory network specifically comprises: Carrying out convolution operation on the standardized signal by utilizing a convolution layer of a one-dimensional convolution neural network, extracting local spatial features and generating a feature map; inputting the characteristic diagram into a bidirectional long-short-time memory network, processing the characteristic diagram by a forward long-short-time memory unit to obtain forward hidden states of all time steps, and processing the characteristic diagram by a backward long-short-time memory unit to obtain backward hidden states of all time steps; And vector splicing is carried out on the forward hidden state and the backward hidden state under the same time step to obtain a depth hidden feature vector corresponding to the time step, and the depth hidden feature vectors of all the time steps jointly form the depth hidden feature sequence.
7. The method for monitoring anomalies in coal mine equipment with fusion of depth features and manual features according to claim 1, wherein in step S4, the calculating the attention weight includes: extracting depth hidden feature vectors corresponding to time steps in the depth hidden feature sequence and local entropy values corresponding to the same time steps in the local entropy value sequence; Performing matrix multiplication operation on the depth hidden feature vector and the depth feature weight matrix to obtain a first feature component; multiplying the local entropy value and the physical entropy value mapping weight, and mapping the operation result to the same dimension as the first characteristic component to obtain a second characteristic component; summing the first characteristic component, the second characteristic component and the bias term, and performing activation processing on a summation result by using a hyperbolic tangent function to obtain attention energy scores of time steps; and calculating the attention energy scores of all the time steps by using a normalization exponential function to obtain normalization values corresponding to the time steps, wherein the normalization values are used as the attention weights.
8. The method for monitoring abnormality of coal mine equipment by fusion of depth features and manual features according to claim 1, wherein in step S4, the weighting process for the depth hidden feature sequence specifically comprises: performing scalar multiplication operation on the depth hidden feature vector corresponding to each time step in the depth hidden feature sequence and the attention weight corresponding to the time step respectively to obtain a temporary weighting vector corresponding to the time step; And carrying out vector summation operation on the temporary weighted vectors obtained by calculation of all time steps, and taking the vectors obtained by summation as the weighted depth characteristic vectors.
9. The method for monitoring abnormal conditions of coal mine equipment by fusion of depth features and manual features according to claim 1, wherein in step S5, the monitoring result for characterizing abnormal conditions of coal mine equipment specifically comprises: The first full-connection layer of the multi-layer perceptron is utilized to carry out linear mapping on the comprehensive feature vector, and the mapped vector is processed through correcting a linear unit activation function to obtain a dimension-reduced intermediate feature vector; Inputting the intermediate feature vector after the dimension reduction into an output layer of the multi-layer perceptron, and calculating an output value by using a Sigmoid activation function, wherein the output value represents the probability that the current equipment belongs to an abnormal state; Comparing the magnitude relation between the output value and a preset threshold, judging that the monitoring result is in an abnormal state when the output value is larger than the preset threshold, and judging that the monitoring result is in a normal state when the output value is smaller than or equal to the preset threshold.
10. A computing device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements a method of coal mine equipment anomaly monitoring for fusion of depth features and hand features as claimed in any one of claims 1 to 9.

Description

Coal mine equipment abnormality monitoring method with depth features and manual features fused Technical Field The invention relates to the technical field of equipment monitoring and fault diagnosis, in particular to a coal mine equipment abnormality monitoring method with depth characteristics and manual characteristics fused. Background Comprehensive mechanized coal mining equipment for coal mines, such as coal mining machines and scraper conveyors, are core components of coal production systems. The equipment continuously operates under severe working conditions of high dust, high humidity and strong vibration in the pit for a long time, and key mechanical parts of the equipment are extremely easy to generate faults such as abrasion, fracture or spalling. Once the equipment is abnormal and can not be found in time, the production is interrupted, huge economic loss is caused, and serious safety accidents are easily caused. In the existing coal mine equipment state monitoring technology, vibration signal analysis is one of the most common means. Early monitoring methods rely mainly on traditional signal processing techniques such as fast fourier transforms, wavelet transforms, etc., in combination with manually extracted time-domain or frequency-domain statistical features for diagnosis. However, the underground working conditions of the coal mine are complex and changeable, the equipment load has high non-stationarity and non-linear characteristics, and the equipment load is often accompanied by strong background noise interference. On the one hand, the traditional signal processing method severely depends on expert experience to perform feature selection, and is difficult to adapt to the automatic processing requirement of massive monitoring data, and on the other hand, when facing to non-stable impact signals, the fixed basis function transformation is difficult to effectively extract weak fault features submerged in strong noise. With the development of deep learning technology, convolutional neural networks, long-short-term memory networks and other models are widely applied to fault diagnosis of industrial equipment. The data-driven method can automatically learn nonlinear characteristics from the original data, and solves the problem of difficult extraction of manual characteristics to a certain extent. However, pure data driven deep learning models often lack physical interpretability and require high quality input data. In a strong noise environment in a coal mine well, the neural network is easily misled by environmental noise, and non-fault-related false features are learned. In order to improve the robustness of monitoring, prior art attempts have been made to fuse manually extracted physical features with deep learning features. The current fusion mode mostly adopts a simple vector splicing strategy, namely, two types of features are directly connected in series at the input end or the output end of a network. This simple fusion approach ignores the inherent correlation of physical features with depth features in the time dimension. Existing attention mechanisms, while capable of weighting features, typically calculate their weights based only on similarities within the data, lacking a priori guidance on the physical mechanism level. When the signal is polluted by broadband noise, the attention weight calculated by data driving can not be accurately focused on the key time step containing fault impact, so that the false alarm rate of the monitoring model is higher, and the requirement of the coal mine site on high reliability of abnormal monitoring of equipment is difficult to meet. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a coal mine equipment anomaly monitoring method with fused depth features and manual features, which solves the problems of inaccurate feature extraction caused by lack of physical mechanism guidance of a depth learning model and poor equipment anomaly monitoring robustness and high false alarm rate caused by the fact that the traditional feature fusion mode ignores the inherent correlation between signal complexity and time sequence features in the prior art under the conditions of strong underground noise and non-steady underground coal mine. In order to achieve the above purpose, the invention is realized by the following technical scheme: The invention provides a coal mine equipment anomaly monitoring method with depth characteristics and manual characteristics fused, which comprises the following steps: S1, acquiring vibration acceleration signals of coal mine equipment, and performing standardized processing on the vibration acceleration signals to obtain standardized signals; S2, performing Hilbert-Huang transform decomposition and multi-scale entropy calculation on the standardized signal to obtain a local entropy sequence and a global entropy vector; S3, inputting the standardized signals into a one-dimensional convolu