Belt conveyor fault prediction method based on deep learning
Through deep learning methods combined with TCN and iTransformer modules, the problem of unsatisfactory computing complexity and modeling effect in belt conveyor failure prediction is solved, and efficient and accurate fault prediction is achieved.
Patent Information
- Application Number
- CN202510525795.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-25
- Publication Date
- 2025-08-08
AI Technical Summary
The existing belt conveyor fault prediction methods have problems such as high computational complexity, difficulty in processing nonlinear relationships in sequence data, and poor long-distance dependency modeling effects.
Using a deep learning-based method, local features are captured through the TCN module and the iTransformer module to capture global features, and feature fusion is used to generate fault prediction values.
It improves the accuracy and robustness of fault prediction, reduces the computational complexity, and enhances the modeling ability of nonlinear relationships and long-distance dependencies.
Smart Images
Figure CN120448706A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of belt conveyor fault prediction, and in particular to a belt conveyor fault prediction method based on deep learning. Background Art
[0002] Belt conveyors, as a continuous material transport device, are widely used in coal mining due to their efficient and continuous transport capabilities. However, due to the long-term operation of belt conveyors in harsh environments such as mines, they often fail frequently, seriously affecting coal mine production and worker safety.
[0003] Currently, existing belt conveyor fault prediction methods are primarily based on machine learning, with a smaller number of methods based on deep learning. These methods all have limitations. For example, machine learning-based methods suffer from high computational complexity, difficulty processing sequential data, and difficulty modeling nonlinear relationships. Deep learning-based methods, on the other hand, rely solely on conventional neural networks, making them less effective at modeling long-range dependencies. Summary of the Invention
[0004] The purpose of this paper is to provide a deep learning-based method for predicting belt conveyor faults. This method captures both local and global features in the time series data of belt conveyor operating parameters and fuses these features to generate a prediction value. By considering both local and global features, it is possible to accurately predict possible future faults.
[0005] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is as follows: The present invention provides a belt conveyor fault prediction method based on deep learning, comprising the following steps:
[0006] S1: Preprocessing the time series data of belt conveyor operating parameters collected by sensors, including timestamp alignment, missing value filling and standardization;
[0007] S2: Use the TCN module to extract local features from time series data. The TCN module expands the receptive field by dilated causal convolution to capture multi-scale local time series features, and uses residual connections and ReLU activation functions to build a deep network;
[0008] S3: Extract global features from time series data using the iTransformer module. The iTransformer module encodes the time series of each parameter into tokens independently through the embedding layer and captures global correlations based on a multivariate attention mechanism and a feedforward neural network.
[0009] S4: fusing the local features and the global features through a gated feature fusion module, generating a fused feature representation through a projection operation and gated weight calculation, and outputting a predicted value of the belt conveyor operating parameter using a residual connection;
[0010] S5: Generate fault prediction information based on the comparison result between the predicted value and the preset threshold value, and visualize it through the interactive interface.
[0011] Preferably, the dilated causal convolution operation in step S2 is defined as:
[0012]
[0013] Among them, y t represents the convolution result corresponding to time step t, w i Represents the i-th weight parameter of the convolution kernel, l represents the size of the convolution kernel, x represents the element in the input sequence, and d represents the expansion factor.
[0014] Preferably, the specific implementation of the iTransformer module in step S3 includes:
[0015] S31: The time series data Y of the nth parameter :,n Mapped to initial token through embedding layer:
[0016]
[0017] in, The token obtained after embedding the time series data of the nth parameter;
[0018] S32: iTransformer is used to capture the global features of time series data and the correlation between tokens. Specifically, each token is first processed by a multivariate attention layer to capture the correlation between variables; it is then processed by a feedforward neural network layer to represent the internal structure of the time series; and finally the global feature representation is output:
[0019] H l+1 =iTrans(H l )l=0,…,L-1
[0020] in, It contains N tokens and has a dimension of D, where the superscripts represent the neural network layer indices.
[0021] Preferably, the calculation process of the gated feature fusion module in step S4 includes:
[0022] S41: Output f to TCN module TCN and the output f of the iTransformer moduleiTr2ns Project them separately and get: F TcN =Project(f TCN )
[0023] F iTrans =Project(f iTrans );
[0024] S42: According to the projection result F TCN and F iTrans Calculate the gating weight G; the formula is as follows:
[0025]
[0026] Among them, σ is the Sigmoid activation function, [;] represents splicing, is the weight matrix, b1 and b2 are biases;
[0027] S43: Fusion output is performed based on the projection results and gating weights; the formula is as follows:
[0028] F out =G⊙F iTrans +(1-G)⊙F TCN +F iTrans +F TCN
[0029] Among them, F out In order to fuse the output results and avoid gradient problems and loss of original information, the process uses residual connections.
[0030] Preferably, in step S5, the predicted value is judged according to the operating parameter threshold. If it exceeds the threshold, it is determined that a corresponding fault may occur in the future; if it does not exceed the threshold, it is determined that no fault will occur in the future.
[0031] Compared with the prior art, the present invention has the following beneficial effects:
[0032] 1. This method addresses the problem that existing machine learning-based fault prediction methods are difficult to handle the nonlinear relationship of time series data and have high computational complexity. By introducing a temporal convolutional network (TCN) module, using dilated causal convolution and residual connection, it efficiently captures the local mutation characteristics of time series data, and combines the ReLU activation function to enhance the nonlinear modeling capability, significantly reducing the computational complexity and improving the adaptability to local time series patterns.
[0033] 2. This method addresses the existing problems of conventional deep learning models, which lack long-distance dependency modeling and have poor global feature extraction. By using an improved iTransformer module, this method independently encodes multivariate time series into tokens and captures the global correlation between parameters based on a multivariate attention mechanism. This method, combined with a feedforward neural network to strengthen the internal representation of time series, effectively addresses the limitations of long-distance dependency modeling, thereby improving the accuracy and generalization of fault prediction under complex working conditions. Furthermore, a gated feature fusion module dynamically weights and fuses local and global features, further avoiding the information loss problem of traditional single models and ensuring the robustness of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.
[0035] Figure 1 This is a flow chart of a belt conveyor fault prediction method based on deep learning provided by the present invention. DETAILED DESCRIPTION
[0036] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0037] See also Figure 1 The present invention provides a belt conveyor fault prediction method based on deep learning, comprising the following steps:
[0038] S1. Perform timestamp alignment, missing value filling and standardization on belt conveyor time series data;
[0039] S2. Use the TCN module to extract local features from the time series data of belt conveyor operating parameters; specifically,
[0040] S21. Use dilated causal convolution to extract local temporal features. The causal mechanism ensures that the output at time t depends only on the elements at and before that time, preventing future information leakage. The dilated convolution mechanism controls the dilation factor d to expand the receptive field, thereby capturing multi-scale features by stacking multiple convolutional layers. The formula is as follows:
[0041]
[0042] Among them, y t represents the convolution result corresponding to time step t, w iRepresents the i-th weight parameter of the convolution kernel, l represents the size of the convolution kernel, x represents the element in the input sequence, and d represents the expansion factor;
[0043] S22. Use the ReLU activation function to add nonlinearity to the output of the dilated causal convolution to model complex temporal patterns; use residual connections to build deeper networks and alleviate the gradient vanishing problem;
[0044] S3, using the iTransformer module to extract global features from the time series data of belt conveyor operating parameters; comprising the following steps:
[0045] S31: The time series data Y of the nth parameter :,n Mapped to initial token through embedding layer:
[0046]
[0047] in, The token obtained after embedding the time series data of the nth parameter;
[0048] S32: iTransformer is used to capture the global features of time series data and the correlation between tokens. Specifically, each token is first processed by a multivariate attention layer to capture the correlation between variables; it is then processed by a feedforward neural network layer to represent the internal structure of the time series; and finally the global feature representation is output:
[0049] H l+1 =iTrans(H l )l=0,…,L-1
[0050] in, It contains N tokens and has a dimension of D, where the superscript represents the neural network layer index;
[0051] S4. Using the gated feature fusion module, the local features extracted by the TCN module and the global features extracted by the iTransformer module are fused to generate the predicted values of the belt conveyor operating parameters;
[0052] S41, output f of TCN module TCN and the output f of the iTransformer module iTrans Projection is performed separately; the formula is as follows:
[0053] F TCN =Project(f TCN )
[0054] F iTrans =Project(f iTrans )
[0055] S42, according to the projection result F TCN and F iTrans Calculate the gating weight G; the formula is as follows:
[0056]
[0057] Among them, σ is the Sigmoid activation function, [;] represents splicing, is the weight matrix, b1 and b2 are biases;
[0058] S43. Perform fusion output based on the projection result and the gating weight; the formula is as follows:
[0059] F out =G⊙F iTrans +(1-G)⊙F TCN +F iTrans +F TCN
[0060] Among them, F out In order to fuse the output results and avoid gradient problems and loss of original information, the process uses residual connections;
[0061] S5. Generate corresponding fault prediction results based on the predicted values of the belt conveyor operating parameters and display them on the interactive interface.
[0062] In step S5, the predicted value is judged according to the operating parameter threshold. If it exceeds the threshold, it is determined that a corresponding fault may occur in the future. If it does not exceed the threshold, it is determined that no fault will occur in the future.
[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A belt conveyor fault prediction method based on deep learning, characterized in that: The following steps are involved: S1: Preprocessing the time series data of belt conveyor operating parameters collected by sensors, including timestamp alignment, missing value filling and standardization; S2: Use the TCN module to extract local features from time series data. The TCN module expands the receptive field by dilated causal convolution to capture multi-scale local time series features, and uses residual connections and ReLU activation functions to build a deep network; S3: Extract global features from time series data using the iTransformer module. The iTransformer module encodes the time series of each parameter into tokens independently through the embedding layer and captures global correlations based on a multivariate attention mechanism and a feedforward neural network. S4: fusing the local features and the global features through a gated feature fusion module, generating a fused feature representation through a projection operation and gated weight calculation, and outputting a predicted value of the belt conveyor operating parameter using a residual connection; S5: Generate fault prediction information based on the comparison result between the predicted value and the preset threshold value, and visualize it through the interactive interface.
2. A belt conveyor fault prediction method based on deep learning according to claim 1, characterized in that: The operation of dilated causal convolution in step S2 is defined as: Among them, y t represents the convolution result corresponding to time step t, w i Represents the i-th weight parameter of the convolution kernel, l represents the size of the convolution kernel, x represents the element in the input sequence, and d represents the expansion factor.
3. A belt conveyor fault prediction method based on deep learning according to claim 1, characterized in that: The specific implementation of the iTransformer module in step S3 includes: S31: The time series data Y of the nth parameter :,n Mapped to initial token through embedding layer: in, The token obtained after embedding the time series data of the nth parameter; S32: iTransformer is used to capture the global features of time series data and the correlation between tokens. Specifically, each token is first processed by a multivariate attention layer to capture the correlation between variables; it is then processed by a feedforward neural network layer to represent the internal structure of the time series; and finally the global feature representation is output: H l+1 =iTrans(H l ) l=0,…,L-1 in, It contains N tokens and has a dimension of D, where the superscripts represent the neural network layer indices.
4. A belt conveyor fault prediction method based on deep learning according to claim 1, characterized in that: The calculation process of the gated feature fusion module in step S4 includes: S41: Output f to TCN module TCN and the output f of the iTransformer module iTrans Project them separately and get: F TCN =Project(f TCN ) F iTrans =Project(f iTrans ); S42: According to the projection result F TCN and F iTrans Calculate the gating weight G; the formula is as follows: Among them, σ is the Sigmoid activation function, [;] represents splicing, is the weight matrix, b1 and b2 are biases; S43: Fusion output is performed based on the projection results and gating weights; the formula is as follows: F out =G⊙F iTrans +(1-G)⊙F TCN +F iTrans +F TCN Among them, F out In order to fuse the output results and avoid gradient problems and loss of original information, the process uses residual connections.
5. The belt conveyor fault prediction method based on deep learning according to claim 1, characterized in that: In step S5, the predicted value is judged according to the operating parameter threshold. If it exceeds the threshold, it is determined that a corresponding fault may occur in the future. If it does not exceed the threshold, it is determined that no fault will occur in the future.