A method and device for identifying faults of reducer gears in a scraper conveyor

Abstract

The present invention discloses a fault identification method and device for a reducer gear in a scraper conveyor, and relates to the technical field of mechanical fault diagnosis. In order to solve the technical problem in the prior art that the two-dimensional convolution kernel is difficult to capture the long-distance relationship in the time-frequency spectrum, resulting in low accuracy of the fault identification result, the present invention divides the time-frequency spectrum according to the time dimension and the frequency dimension; then uses two one-dimensional feature pyramid modules to extract features from the time-frequency graph along the frequency dimension and the time dimension respectively: then the high-level features are upsampled layer by layer, and the information of the two dimensions flows during the upsampling operation. Since the information transmission starts based on the highest-level features, the high-level information of the two dimensions flows, and the extraction of the long-distance dependency relationship in the time-frequency spectrum is completed. Finally, the outputs of the two one-dimensional feature pyramid modules are fused, thereby improving the accuracy of fault identification.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical fault diagnosis, and in particular to a fault identification method and device for a reducer gear in a scraper conveyor. Background Art

[0002] Scraper conveyor is a conveying equipment used to transport solid materials. It pushes the material to move by moving the scraper fixed on the chain in a cycle along a fixed track. The reducer gear reduces the rotation speed of the motor in the scraper conveyor to the working speed required by the scraper chain through meshing. Since the operating condition of the scraper conveyor is directly affected by the working condition of the gear, the fault identification of the reducer gear and early treatment of the reducer gear problem can extend the service life of the scraper conveyor; and any abnormality of the gear can be reflected by the timing characteristics and frequency characteristics of the motor current, so the time-frequency spectrum of the motor current signal can be used to identify the reducer gear fault.

[0003] In the prior art, Chinese patent CN114048682B discloses an intelligent diagnosis method for rolling bearing acoustic emission based on the fusion of optimal wavelet basis and multi-dimensional deep features. The method first collects the rolling bearing acoustic emission signal, then selects the optimal wavelet basis of continuous wavelet transform, uses the selected optimal wavelet basis to perform continuous wavelet transform, and converts the one-dimensional acoustic emission signal into a two-dimensional time-frequency graph; divides the training set, verification set and test set, and defines the fault type; inputs the rolling bearing acoustic emission signal into a one-dimensional convolutional neural network, and the time-frequency graph into a two-dimensional convolutional neural network, and fuses the features of different dimensions and depths extracted by the convolutional neural network through a feature pyramid structure.

[0004] However, the currents corresponding to two time-frequency points that are far apart in the time-frequency spectrum may have a dependency relationship, and the two-dimensional convolution kernel is difficult to directly capture such long-distance or global dependencies due to the limitation of its local receptive field, resulting in the need to improve the accuracy of fault identification results. Summary of the invention

[0005] Based on this, in order to solve the technical problem in the prior art that the fault identification accuracy of the reducer gear in the scraper conveyor is not high, the present invention provides a method and device for fault identification of the reducer gear in the scraper conveyor.

[0006] The present invention provides a method for identifying a fault of a reducer gear in a scraper conveyor, comprising:

[0007] Constructing a deep learning network, comprising: a segmentation layer, a first one-dimensional feature pyramid and a second one-dimensional feature pyramid connected in parallel to an output end of the segmentation layer, a fusion layer connected to both the output end of the first one-dimensional feature pyramid and the output end of the second one-dimensional feature pyramid, and an output layer connected to the output end of the fusion layer; wherein the upsampling layer in the first one-dimensional feature pyramid and the upsampling layer in the second one-dimensional feature pyramid are connected to each other;

[0008] The fault time-frequency spectrum of the reducer gear in the scraper conveyor and the gear fault type corresponding to the time-frequency spectrum are collected to construct a gear fault data set; the fault time-frequency spectrum is a two-dimensional matrix, each row in the two-dimensional matrix corresponds to each frequency in the current signal when the gear fails; each column in the two-dimensional matrix corresponds to each time sampling point when the gear fails; the gear fault data set is used to train the deep learning network to obtain a reducer gear fault recognition model for mapping the gear fault time-frequency spectrum to the gear fault type;

[0009] The time-frequency spectrum of the gear fault to be identified is input into the gear fault identification model of the reducer. The time-frequency spectrum of the gear fault to be identified is segmented according to the time dimension and the frequency dimension respectively through the segmentation layer to obtain the gear fault time series and the gear fault frequency series; the features of the gear fault time series and the gear fault frequency series are extracted respectively through the first one-dimensional feature pyramid and the second one-dimensional feature pyramid in parallel. In the process of feature extraction, the feature map output by the upsampling layer in the first one-dimensional feature pyramid and the feature map output by the upsampling layer in the second one-dimensional feature pyramid are circulated with each other, so as to capture the long-distance dependency in the time-frequency spectrum of the gear fault to be identified by cross-fusion of information of different dimensions, and obtain the gear fault time series feature and the gear fault frequency feature; the gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature; the gear fault fusion feature is classified through the output layer, and the gear fault type corresponding to the time-frequency spectrum of the gear fault to be identified is output.

[0010] Furthermore, the obtaining of the gear fault timing characteristics and the gear fault frequency characteristics specifically includes:

[0011] The gear fault time series is downsampled at different scales step by step through multiple downsampling operations in the first one-dimensional feature pyramid to obtain local features of the gear fault time series at different time granularities; the local features of the gear fault time series at the minimum scale are upsampled step by step through multiple upsampling layers in the first one-dimensional feature pyramid to obtain gear fault time series recovery features at different scales, and the gear fault time series recovery features at the maximum scale are used as the gear fault time series features;

[0012] The gear fault frequency sequence is downsampled step by step at different scales through multiple downsampling operations in the second one-dimensional feature pyramid to obtain local features of gear fault frequency at different frequency levels; the local features of gear fault frequency at the minimum scale are upsampled step by step through multiple upsampling layers in the second one-dimensional feature pyramid to obtain gear fault frequency recovery features at different scales, and the gear fault frequency recovery features at the maximum scale are used as the gear fault frequency features;

[0013] In the process of upsampling the first one-dimensional feature pyramid, the input of each upsampling layer is the time-frequency fusion feature obtained by fusing the local features of the gear fault timing output by the corresponding downsampling layer in the first one-dimensional feature pyramid, the gear fault timing recovery features output by the previous upsampling layer in the first one-dimensional feature pyramid, and the gear fault frequency recovery features output by the upsampling layer corresponding to the second one-dimensional feature pyramid;

[0014] During the upsampling process of the second one-dimensional feature pyramid, the input of each upsampling layer is the frequency-time fusion feature obtained by fusing the local features of the gear fault frequency output by the corresponding downsampling layer in the second one-dimensional feature pyramid, the gear fault frequency recovery features output by the previous upsampling layer in the second one-dimensional feature pyramid, and the gear fault timing recovery features output by the upsampling layer corresponding to the first one-dimensional feature pyramid.

[0015] Furthermore, the time-frequency fusion feature is the result of fusing the local features of the gear fault timing outputted by the corresponding downsampling layer in the first one-dimensional feature pyramid, the gear fault timing recovery features outputted by the previous upsampling layer in the first one-dimensional feature pyramid, and the gear fault frequency recovery features outputted by the upsampling layer corresponding to the second one-dimensional feature pyramid through principal component analysis, so as to eliminate redundant gear fault features in the process of obtaining the time-frequency fusion feature.

[0016] Furthermore, the gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature, which specifically includes:

[0017] Splicing the gear fault timing characteristics and the gear fault frequency to obtain a splicing characteristic diagram;

[0018] The gear fault fusion feature is obtained by introducing a compression and excitation attention module to expand the channel weights corresponding to the gear fault in the concatenated feature map.

[0019] The present invention provides a fault identification device for a reducer gear in a scraper conveyor, comprising:

[0020] A model construction module, for constructing a deep learning network, comprising: a segmentation layer, a first one-dimensional feature pyramid and a second one-dimensional feature pyramid connected in parallel to an output end of the segmentation layer, a fusion layer connected to the output end of the first one-dimensional feature pyramid and the output end of the second one-dimensional feature pyramid, and an output layer connected to the output end of the fusion layer; wherein the upsampling layer in the first one-dimensional feature pyramid and the upsampling layer in the second one-dimensional feature pyramid are connected to each other;

[0021] A model training module is used to collect the fault time-frequency spectrum of the reducer gear in the scraper conveyor and the gear fault type corresponding to the time-frequency spectrum to construct a gear fault data set; the fault time-frequency spectrum is a two-dimensional matrix, each row in the two-dimensional matrix corresponds to each frequency in the current signal when the gear fails; each column in the two-dimensional matrix corresponds to each time sampling point when the gear fails; the gear fault data set is used to train the deep learning network to obtain a reducer gear fault recognition model for mapping the gear fault time-frequency spectrum to the gear fault type;

[0022] A fault identification module is used to input the time-frequency spectrum of the gear fault to be identified into the reducer gear fault identification model, and segment the time-frequency spectrum of the gear fault to be identified according to the time dimension and the frequency dimension through the segmentation layer to obtain the gear fault time series and the gear fault frequency series; the gear fault time series and the gear fault frequency series are respectively extracted by the first one-dimensional feature pyramid and the second one-dimensional feature pyramid in parallel. During the feature extraction process, the feature map output by the upsampling layer in the first one-dimensional feature pyramid and the feature map output by the upsampling layer in the second one-dimensional feature pyramid are circulated with each other, so as to capture the long-distance dependency in the time-frequency spectrum of the gear fault to be identified by cross-fusion of information of different dimensions, and obtain the gear fault time series feature and the gear fault frequency feature; the gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature; the gear fault fusion feature is classified through the output layer, and the gear fault type corresponding to the time-frequency spectrum of the gear fault to be identified is output.

[0023] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0024] In the fault identification method of the reducer gear in the scraper conveyor provided by the present invention, the time-frequency spectrum is firstly segmented according to the time dimension and the frequency dimension; then two one-dimensional feature pyramid modules are used to extract features of the time-frequency spectrum along the frequency dimension and the time dimension respectively: the downsampling layer in the feature pyramid module gradually extracts high-level features, and then the high-level features are upsampled layer by layer. During the upsampling operation, information of the two dimensions flows. Since information transmission starts based on the highest-level features, high-level information of the two dimensions flows, and the extraction of long-distance dependencies in the time-frequency spectrum is completed. Finally, the outputs of the two one-dimensional feature pyramid modules are fused, thereby improving the accuracy of fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0026] Figure 1 A schematic flow chart of a method for identifying a fault of a reducer gear in a scraper conveyor provided by the present invention;

[0027] Figure 2 A schematic diagram of the deep learning network structure constructed by the present invention;

[0028] Figure 3 A schematic diagram of the network structure of the first one-dimensional feature pyramid and the second one-dimensional feature pyramid provided by the present invention;

[0029] Figure 4 A schematic diagram of the network structure of the fusion layer provided by the present invention;

[0030] Figure 5 A schematic diagram of a process for obtaining a gear fault time-frequency matrix provided by the present invention;

[0031] Figure 6 A schematic diagram of a computer device for implementing a fault identification method for a reducer gear in a scraper conveyor provided by the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in combination with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] The traditional method for fault diagnosis of reducer gear of scraper conveyor mainly relies on the sensor to obtain the vibration signal of the mechanical system, and the fault diagnosis is performed by analyzing and processing the vibration signal. However, due to the harsh working environment and small space, it is not realistic to obtain the vibration signal of the reducer of the scraper conveyor by installing sensors. The change of the working condition of the reducer gear can be fully reflected in the current signal, and the acquisition of the current signal does not require the installation of sensors. The current characteristic analysis method can be used to reveal the inherent relationship between the current characteristics and the gear fault, so as to realize the fault diagnosis of the reducer of the scraper conveyor. This method converts the current signal into a time-frequency spectrum through time-frequency analysis technology (the time-frequency spectrum can show the change of different frequency components in the current signal over time) to identify the gear fault. There is a close connection between the operating condition of the scraper conveyor, the gear working condition and its current characteristics, and the current characteristic analysis method can perform feature mining on the current signal, thereby revealing the inherent relationship between the current characteristics and the gear fault, so the gear fault identification problem can be converted into the classification problem of the current time-frequency spectrum.

[0034] However, when using deep learning networks to analyze time-frequency spectra, features are extracted by sliding convolution of two-dimensional convolution kernels on the time-frequency spectra. Two-dimensional convolution kernels perform well when processing grayscale images because the pixel values ​​in grayscale images usually have strong spatial correlation in local areas, such as edges, textures or other local features. The time-frequency spectra are generated by jointly representing the signal in two dimensions, time and frequency. When analyzing them, it is often necessary to consider the dependencies of time-frequency features that are far away, that is, it is necessary to analyze based on global features. Two-dimensional convolution kernels can usually only focus on features in local areas. For time-frequency features that are distributed at distant locations in the time-frequency spectra, it is difficult to effectively capture their global correlation. For example, if the fault mode is expressed as a scattered pattern in the time-frequency spectra, the two-dimensional convolution kernel may not be able to effectively associate these scattered patterns, resulting in fault identification errors. This limitation leads to the insufficient global pattern recognition ability of the two-dimensional convolution kernel for the time-frequency spectra, thus affecting the final fault identification accuracy.

[0035] Based on this, the present invention provides a fault identification method for the reducer gear in a scraper conveyor. The method first collects the engine current corresponding to different gear faults, and converts the current into corresponding fault time-frequency spectra to construct a data set; then a deep learning network is constructed to map the gear fault time-frequency spectra to the gear fault type; then the deep learning network is trained using the data set to obtain a fault identification model; finally, the fault time-frequency spectra corresponding to the gear to be identified are input into the fault identification model to obtain the gear fault type.

[0036] Figure 1The process of the fault identification method of the reducer gear in the scraper conveyor provided by this embodiment is shown. Figure 1 The method is described in detail and specifically comprises the following steps:

[0037] S1: Construct a deep learning network, including: a segmentation layer, a first one-dimensional feature pyramid and a second one-dimensional feature pyramid connected in parallel to an output end of the segmentation layer, a fusion layer connected to both the output end of the first one-dimensional feature pyramid and the output end of the second one-dimensional feature pyramid, and an output layer connected to the output end of the fusion layer; wherein the upsampling layer in the first one-dimensional feature pyramid and the upsampling layer in the second one-dimensional feature pyramid are connected to each other. Figure 2 The structure of the constructed deep learning network is shown. The constructed deep learning network specifically includes:

[0038] S101: Segmentation layer, which performs convolution operation on the input time-frequency spectrum through convolution kernels of sizes T×1 and 1×H respectively. The elements in the two convolution kernels of sizes T×1 and 1×H are all 1, so as to segment the input time-frequency spectrum according to the time dimension (row) and frequency dimension (column), and obtain H T-dimensional gear fault time series and T H-dimensional gear fault frequency series.

[0039] S102: The first one-dimensional feature pyramid and the second one-dimensional feature pyramid connected in parallel to the output end of the segmentation layer have the same structure, and are obtained by replacing the two-dimensional convolution in the original feature pyramid network with a one-dimensional convolution. The Feature Pyramid Network (FPN) includes a first downsampling layer, a second downsampling layer, and a third downsampling layer connected in sequence, and each downsampling layer outputs a feature map with a gradually reduced size, thereby extracting features of different scales; then, the first upsampling layer, the second upsampling layer, and the third upsampling layer connected in sequence perform an upsampling operation on the feature map with the smallest size, and gradually restore the size of the feature map. During the restoration process, the input of each upsampling layer includes not only the input from the previous upsampling layer, but also the output of the corresponding downsampling layer. This structure enables the upsampling layer to utilize feature information from different scales.

[0040] The network structures of the first one-dimensional feature pyramid and the second one-dimensional feature pyramid are as follows Figure 3As shown. The convolution kernel size of the first downsampling layer in the first one-dimensional feature pyramid and the second one-dimensional feature pyramid is 5×1, the convolution kernel size of the second downsampling layer is 5×1, and the convolution kernel size of the third downsampling layer is 3×1; the convolution step size of the first, second and third downsampling layers is 1; the convolution kernel size of the first upsampling layer is 3×1, the convolution kernel size of the second upsampling layer is 5×1, and the convolution kernel size of the third upsampling layer is 5×1; the deconvolution step size of the first, second and third upsampling layers is 1. The structure of the downsampling layer is shown in Table 1:

[0041] Table 1 Downsampling layer structure

[0042]

[0043] Where m1, m2 and m3 can be 32, 64 and 128 respectively. The upsampling layer and downsampling layer in the first one-dimensional feature pyramid and the second one-dimensional feature pyramid are similar, but the results are opposite. Upsampling is used to increase the spatial size of the data. And the output end of the first upsampling layer, the output end of the second upsampling layer and the output end of the third upsampling layer in the first one-dimensional feature pyramid are respectively connected to the input end of the first upsampling layer, the input end of the second upsampling layer and the input end of the third upsampling layer in the second one-dimensional feature pyramid; the output end of the first upsampling layer, the output end of the second upsampling layer and the output end of the third upsampling layer in the second one-dimensional feature pyramid are respectively connected to the input end of the first upsampling layer, the input end of the second upsampling layer and the input end of the third upsampling layer in the first one-dimensional feature pyramid.

[0044] The gear fault time series is downsampled at different scales step by step through multiple downsampling operations in the first one-dimensional feature pyramid to obtain local features of the gear fault time series with different time granularities; the local features of the gear fault time series with the smallest scale are upsampled step by step through multiple upsampling layers in the first one-dimensional feature pyramid to obtain gear fault time series recovery features of different scales, and the gear fault time series recovery features with the largest scale are used as the gear fault time series features.

[0045] The gear fault frequency sequence is downsampled at different scales step by step through multiple downsampling operations in the second one-dimensional feature pyramid to obtain local features of gear fault frequency at different frequency levels; the local features of gear fault frequency at the minimum scale are upsampled step by step through multiple upsampling layers in the second one-dimensional feature pyramid to obtain gear fault frequency recovery features at different scales, and the gear fault frequency recovery features at the maximum scale are used as the gear fault frequency features.

[0046] In the process of upsampling the first one-dimensional feature pyramid, the input of each upsampling layer is the time-frequency fusion feature obtained by fusing the local features of the gear fault timing output by the corresponding downsampling layer in the first one-dimensional feature pyramid, the gear fault timing recovery features output by the previous upsampling layer in the first one-dimensional feature pyramid, and the gear fault frequency recovery features output by the upsampling layer corresponding to the second one-dimensional feature pyramid.

[0047] During the upsampling process of the second one-dimensional feature pyramid, the input of each upsampling layer is the frequency-time fusion feature obtained by fusing the local features of the gear fault frequency output by the corresponding downsampling layer in the second one-dimensional feature pyramid, the gear fault frequency recovery features output by the previous upsampling layer in the second one-dimensional feature pyramid, and the gear fault timing recovery features output by the upsampling layer corresponding to the first one-dimensional feature pyramid.

[0048] The above time-frequency fusion features and frequency-time fusion features are the results of fusing the three input features through PCA (Principal Component Analysis). The fusion of features from different sources through principal component analysis is actually to reduce the dimension of the feature vector while retaining the main information about the gear fault in the original data as much as possible. Specifically, the following steps are included:

[0049] The three features input to a certain upsampling layer are transposed to the same dimension and are denoted as X1, X2 and X3, where X1, X2 and X3 are matrices of n×d1, n×d2 and n×d3 respectively, n is the number of feature channels, and d1, d2 and d3 are the dimensions of X1, X2 and X3 respectively.

[0050] Concatenate X1, X2, and X3 along the feature channel dimension into a new matrix X with a dimension of n×(d1+d2+d3), and calculate the covariance matrix C of X or directly calculate its correlation coefficient matrix.

[0051] Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​and corresponding eigenvectors; select the eigenvectors corresponding to the first k largest eigenvalues ​​to form a projection matrix P; use the projection matrix P to project the original feature matrix X into a new low-dimensional space to obtain the reduced feature matrix Xreduced, which is the input feature of the upsampling layer.

[0052] S103: A fusion layer connected to the output of the first one-dimensional feature pyramid and the output of the second one-dimensional feature pyramid at the same time. The fusion layer concatenates the two feature maps of the gear fault timing feature and the gear fault frequency along the channel dimension through the concatenation layer to obtain a concatenated feature map; and expands the channel weights that respond to the gear fault in the concatenated feature map by introducing a squeeze and excitation attention module. The squeeze and excitation attention module (SE) is mainly divided into two parts: the squeeze module and the excitation module. The squeeze module compresses the spatial information of the input feature into a compressed feature map through global average pooling. The compressed feature map contains the global information of each channel. The excitation module processes the compressed feature map through a fully connected layer and a nonlinear activation function, captures the correlation between channels, and generates the weight of each channel to obtain a weighted feature map. These weights reflect the importance of different channels to the final output. The weight output by the excitation module is multiplied element by element with the feature map of the corresponding channel of the original input, thereby enhancing or suppressing the feature representation of the channel to emphasize the features that are beneficial to the gear fault identification task.

[0053] S104: an output layer connected to the output end of the fusion layer, the output layer includes a fully connected layer and a softmax activation function, the softmax activation function converts the output value of the fully connected layer into a probability distribution of each gear fault type, that is, the probability of normal working condition, wear working condition, pitting working condition and broken tooth working condition, and outputs the gear fault type with the highest probability as the recognition result.

[0054] S2: Collect the fault time-frequency spectrum of the reducer gear in the scraper conveyor and the gear fault type corresponding to the time-frequency spectrum to construct a gear fault data set; the fault time-frequency spectrum is a two-dimensional matrix, and each row in the two-dimensional matrix corresponds to each frequency in the current signal when the gear fails; each column in the two-dimensional matrix corresponds to each time sampling point when the gear fails; use the gear fault data set to train the deep learning network to obtain a reducer gear fault recognition model for mapping the gear fault time-frequency spectrum to the gear fault type.

[0055] S201: Collect the fault time-frequency spectrum of the reducer gear in the scraper conveyor.

[0056] Gear fault time-frequency spectrum is a signal processing technology used to analyze and diagnose gear faults. It combines information in the time domain and frequency domain to reveal possible faults that may occur in gears during operation. This spectrum is usually generated from the original current signal through time-frequency analysis technology. The current characteristics of the gear during operation are represented as a time-frequency spectrum. The time-frequency spectrum results after wavelet packet time-frequency analysis transformation of the current signal under various load conditions include the normal state of the gear, wear and pitting, and broken tooth state. The process of obtaining the gear fault time-frequency spectrum is as follows: Figure 5 As shown, the specific steps include:

[0057] (1): After anti-aliasing filtering of the scraper conveyor working current, the current data is obtained.

[0058] For low-speed heavy-load mechanical scraper conveyors, the working current amplitude is relatively large, so first, it is necessary to use current mutual induction equipment to adjust the amplitude of the three-phase working current and reduce it to a safe range. Then, the three-phase current is collected and saved. The collected current data is usually a time series: A={A1,....,A i ,....,A n}, where A1~A n It is between time T1 and T n The collected current data at different frequencies.

[0059] (2): When the speed gear of the tube sheet conveyor has pitting, wear, broken teeth and other fault conditions, the scraper conveyor will produce operating shocks during transportation. The collected current signal is decomposed into sub-band signals using the wavelet packet multi-scale decomposition method. The signal is then compared with the current signal of the scraper conveyor under normal working conditions. The sub-band signals with the largest current difference are compared to locate the fault frequency band of the scraper conveyor speed gear. The signals of other frequency bands with load interference are filtered out to obtain the fault frequency band signal.

[0060] (3): After decomposing the collected three-phase power frequency current to obtain the fault frequency band signal of the gear, the Hilbert transform is used to demodulate the amplitude of the fault frequency band signal to filter out the power frequency interference of the existing three-phase current power frequency component on the current signal, thereby improving the intensity of the fault information in the current signal, constructing the analytical signal of the frequency band where the gear fault is located, and obtaining the enhanced fault frequency band signal.

[0061] (4): Using the wavelet time-frequency analysis method that integrates the time domain and frequency domain characteristics of the signal, the gear fault time-frequency spectrum is extracted from the enhanced fault frequency band signal. That is, for the fault frequency bands under normal, wear, pitting and broken gear conditions, a set of low-pass filters are used to filter the working condition signal to filter out the high-frequency component signal and retain the low-frequency component signal to complete the extraction of the gear fault characteristics in the signal. After the current signal is processed in this step, the time-frequency spectrum of the signal is obtained. The extracted gear fault time-frequency spectrum is a two-dimensional matrix of size T×H, where T represents the time dimension in the two-dimensional matrix: {t1, t2, ..., T}, where 1~T is in seconds; H represents the frequency dimension in the two-dimensional matrix: {h1, h2, ..., H}, where h1~H is in Hertz Hz. If each column in the gear fault time-frequency spectrum is regarded as a set of data, then each column represents the signal strength of different frequency ranges (0-10Hz, 10-20Hz, etc.) within a preset time window (such as 5 seconds, 10 seconds, etc.), and the entire matrix can be regarded as different current frequency distributions in the same time window. By observing each column of data, you can see which frequency components exist at a specific time point and how their intensity changes over time. If each row is regarded as a set of data, then each row represents the change of a specific frequency range (such as 0-10Hz) over time, and the entire matrix can be regarded as the trajectory of the change of different frequency components over time. By observing each row of data, you can track the change trend of a specific frequency over time.

[0062] S202: Use the gear fault dataset to train a deep learning network to obtain a fault identification model for the reducer gear in the scraper conveyor that maps the gear fault time-frequency spectrum to the gear fault type.

[0063] The loss function used in the training process is the cross entropy loss function; the optimization algorithm is the Adam optimization algorithm; the training experimental environment is: the server platform processor model is 24 vCPU AMD EPYC 7642 48-Core Processor, using the NVIDIA RTX3090 graphics card. The video memory is 24GB, the deep learning framework is Pytorch, version 1.9.0, CUDA version is 11.1, and Python version is 3.7. During the training process, the batch size is set to 16, the process works is set to 16, and the number of training rounds is set to 300 rounds. The same variables will also be maintained in subsequent comparison experiments of different algorithms. The experimental parameter settings are shown in Table 2.

[0064] Table 2 Experimental parameter settings

[0065]

[0066] S3: Input the time-frequency spectrum of the gear fault to be identified into the reducer gear fault identification model, and segment the time-frequency spectrum of the gear fault to be identified according to the time dimension and the frequency dimension respectively through the segmentation layer to obtain the gear fault time series and the gear fault frequency series; extract features of the gear fault time series and the gear fault frequency series respectively through the first one-dimensional feature pyramid and the second one-dimensional feature pyramid in parallel. In the process of feature extraction, the feature map output by the upsampling layer in the first one-dimensional feature pyramid and the feature map output by the upsampling layer in the second one-dimensional feature pyramid are circulated with each other, so as to capture the long-distance dependency in the time-frequency spectrum of the gear fault to be identified by cross-fusion of information of different dimensions, and obtain the gear fault time series feature and the gear fault frequency feature; fuse the gear fault time series feature and the gear fault frequency feature through the fusion layer to obtain the gear fault fusion feature; classify the gear fault fusion feature through the output layer, and output the gear fault type corresponding to the time-frequency spectrum of the gear fault to be identified.

[0067] based on Figure 1 The fault identification method of the reducer gear in the scraper conveyor shown in the figure can capture the reason why the interconnection of the two networks can capture long-distance dependencies. Feature fusion is used. When two feature pyramid networks (one processing row vectors and one processing column vectors) are connected to each other in the upsampling layer, they actually fuse time series features and frequency series features at different scales. This fusion enables the network to take into account dependencies in time and frequency at the same time, even if these dependencies are far away from the original data. And the feature pyramid network can capture information at different scales by extracting features at different levels. During the upsampling process, these multi-scale features are gradually fused; lower-level features capture local details, while higher-level features can capture more global structures. This multi-scale fusion helps capture long-distance dependencies because global structures often contain relationships between distant elements. In addition, the attention mechanism allows the network to learn the correlation between different parts of the data without being restricted by their physical distance in the data. Adding an attention module to the upsampling layer allows the network to pay more attention to features that are important for the current task and span longer distances. When two networks are interconnected, information in one network can be passed to the other network through the connection, which helps to combine the features learned by one network in a local area with the features learned by another network in another area; through interconnection, the network can integrate contextual information from different dimensions. For example, in audio signal processing, one network may capture rhythm changes in time, while another network may capture pitch changes in frequency. The combination of the two can provide a more comprehensive understanding of the signal, even if these changes are far apart in the original signal.

[0068] The above is a method for identifying a fault of a reducer gear in a scraper conveyor provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding method as follows: Figure 6 The fault identification device of the reducer gear in the scraper conveyor shown comprises:

[0069] A model construction module is used to construct a deep learning network, including: a segmentation layer, a first one-dimensional feature pyramid and a second one-dimensional feature pyramid connected in parallel to an output end of the segmentation layer, a fusion layer connected to both the output end of the first one-dimensional feature pyramid and the output end of the second one-dimensional feature pyramid, and an output layer connected to the output end of the fusion layer; wherein the upsampling layer in the first one-dimensional feature pyramid and the upsampling layer in the second one-dimensional feature pyramid are connected to each other.

[0070] The model training module is used to collect the fault time-frequency spectrum of the reducer gear in the scraper conveyor and the gear fault type corresponding to the time-frequency spectrum to construct a gear fault data set; the fault time-frequency spectrum is a two-dimensional matrix, and each row in the two-dimensional matrix corresponds to each frequency in the current signal when the gear fails; each column in the two-dimensional matrix corresponds to each time sampling point when the gear fails; the gear fault data set is used to train the deep learning network to obtain a reducer gear fault identification model for mapping the gear fault time-frequency spectrum to the gear fault type.

[0071] A fault identification module is used to input the time-frequency spectrum of the gear fault to be identified into the reducer gear fault identification model, and segment the time-frequency spectrum of the gear fault to be identified according to the time dimension and the frequency dimension through the segmentation layer to obtain the gear fault time series and the gear fault frequency series; the gear fault time series and the gear fault frequency series are respectively extracted by the first one-dimensional feature pyramid and the second one-dimensional feature pyramid in parallel. During the feature extraction process, the feature map output by the upsampling layer in the first one-dimensional feature pyramid and the feature map output by the upsampling layer in the second one-dimensional feature pyramid are circulated with each other, so as to capture the long-distance dependency in the time-frequency spectrum of the gear fault to be identified by cross-fusion of information of different dimensions, and obtain the gear fault time series feature and the gear fault frequency feature; the gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature; the gear fault fusion feature is classified through the output layer, and the gear fault type corresponding to the time-frequency spectrum of the gear fault to be identified is output.

[0072] The specific definition of the fault identification device for the reducer gear in the scraper conveyor can be found in the definition of the fault identification method for the reducer gear in the scraper conveyor mentioned above, which will not be repeated here. Each module in the above-mentioned fault identification device for the reducer gear in the scraper conveyor can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0073] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 A fault identification method for the reducer gear in a scraper conveyor is provided.

[0074] The present invention also provides a computer device structure. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 A fault identification method for the reducer gear in a scraper conveyor is provided.

[0075] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0076] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A method for identifying a fault of a reducer gear in a scraper conveyor, characterized in that: include: Constructing a deep learning network, comprising: a segmentation layer, a first one-dimensional feature pyramid and a second one-dimensional feature pyramid connected in parallel to an output end of the segmentation layer, a fusion layer connected to both the output end of the first one-dimensional feature pyramid and the output end of the second one-dimensional feature pyramid, and an output layer connected to the output end of the fusion layer; wherein the upsampling layer in the first one-dimensional feature pyramid and the upsampling layer in the second one-dimensional feature pyramid are connected to each other; The fault time-frequency spectrum of the reducer gear in the scraper conveyor and the gear fault type corresponding to the time-frequency spectrum are collected to construct a gear fault data set; the fault time-frequency spectrum is a two-dimensional matrix, each row in the two-dimensional matrix corresponds to each frequency in the current signal when the gear fails; each column in the two-dimensional matrix corresponds to each time sampling point when the gear fails; the gear fault data set is used to train the deep learning network to obtain a reducer gear fault recognition model for mapping the gear fault time-frequency spectrum to the gear fault type; The time-frequency spectrum of the gear fault to be identified is input into the gear fault identification model of the reducer. The time-frequency spectrum of the gear fault to be identified is segmented according to the time dimension and the frequency dimension respectively through the segmentation layer to obtain the gear fault time series and the gear fault frequency series; the features of the gear fault time series and the gear fault frequency series are extracted respectively through the first one-dimensional feature pyramid and the second one-dimensional feature pyramid in parallel. In the process of feature extraction, the feature map output by the upsampling layer in the first one-dimensional feature pyramid and the feature map output by the upsampling layer in the second one-dimensional feature pyramid are circulated with each other, so as to capture the long-distance dependency in the time-frequency spectrum of the gear fault to be identified by cross-fusion of information of different dimensions, and obtain the gear fault time series feature and the gear fault frequency feature; the gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature; the gear fault fusion feature is classified through the output layer, and the gear fault type corresponding to the time-frequency spectrum of the gear fault to be identified is output.

2. The method for identifying a fault of a reducer gear in a scraper conveyor according to claim 1, characterized in that: The obtaining of the gear fault timing characteristics and the gear fault frequency characteristics specifically includes: The gear fault time series is downsampled at different scales step by step through multiple downsampling operations in the first one-dimensional feature pyramid to obtain local features of the gear fault time series at different time granularities; the local features of the gear fault time series at the minimum scale are upsampled step by step through multiple upsampling layers in the first one-dimensional feature pyramid to obtain gear fault time series recovery features at different scales, and the gear fault time series recovery features at the maximum scale are used as the gear fault time series features; The gear fault frequency sequence is downsampled step by step at different scales through multiple downsampling operations in the second one-dimensional feature pyramid to obtain local features of gear fault frequency at different frequency levels; the local features of gear fault frequency at the minimum scale are upsampled step by step through multiple upsampling layers in the second one-dimensional feature pyramid to obtain gear fault frequency recovery features at different scales, and the gear fault frequency recovery features at the maximum scale are used as the gear fault frequency features; In the process of upsampling the first one-dimensional feature pyramid, the input of each upsampling layer is the time-frequency fusion feature obtained by fusing the local features of the gear fault timing output by the corresponding downsampling layer in the first one-dimensional feature pyramid, the gear fault timing recovery features output by the previous upsampling layer in the first one-dimensional feature pyramid, and the gear fault frequency recovery features output by the upsampling layer corresponding to the second one-dimensional feature pyramid; During the upsampling process of the second one-dimensional feature pyramid, the input of each upsampling layer is the frequency-time fusion feature obtained by fusing the local features of the gear fault frequency output by the corresponding downsampling layer in the second one-dimensional feature pyramid, the gear fault frequency recovery features output by the previous upsampling layer in the second one-dimensional feature pyramid, and the gear fault timing recovery features output by the upsampling layer corresponding to the first one-dimensional feature pyramid.

3. The method for identifying a fault of a reducer gear in a scraper conveyor according to claim 2, characterized in that: The time-frequency fusion feature is the result of fusing the local features of the gear fault timing outputted by the corresponding downsampling layer in the first one-dimensional feature pyramid, the gear fault timing recovery features outputted by the previous upsampling layer in the first one-dimensional feature pyramid, and the gear fault frequency recovery features outputted by the upsampling layer corresponding to the second one-dimensional feature pyramid through principal component analysis, so as to eliminate redundant gear fault features in the process of obtaining the time-frequency fusion feature.

4. The method for identifying a fault of a reducer gear in a scraper conveyor according to claim 1, characterized in that: The gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature, which specifically includes: Splicing the gear fault timing characteristics and the gear fault frequency to obtain a splicing characteristic diagram; The gear fault fusion feature is obtained by introducing a compression and excitation attention module to expand the channel weights corresponding to the gear fault in the concatenated feature map.

5. The method for identifying a fault of a reducer gear in a scraper conveyor according to claim 1, characterized in that: The acquisition of the fault time-frequency spectrum of the reducer gear in the scraper conveyor specifically includes: Collect current data when the reducer gear in the scraper conveyor is working; Perform wavelet packet multi-scale decomposition on the current data to obtain a sub-band signal; locate the frequency band where the gear fault is located in the obtained sub-band signal by comparing the obtained sub-band signal with the sub-band signal under normal working conditions, filter out the load interference in the frequency band where the gear fault is located, and obtain the fault frequency band signal; The fault frequency band signal is amplitude demodulated using Hilbert transform to suppress power frequency interference and enhance gear fault characteristics to obtain enhanced fault frequency band signal; The wavelet packet time-frequency analysis method is used to extract the gear fault time-frequency spectrum from the enhanced fault frequency band signal.

6. A fault identification device for a reducer gear in a scraper conveyor, characterized in that: include: A model construction module, for constructing a deep learning network, comprising: a segmentation layer, a first one-dimensional feature pyramid and a second one-dimensional feature pyramid connected in parallel to an output end of the segmentation layer, a fusion layer connected to the output end of the first one-dimensional feature pyramid and the output end of the second one-dimensional feature pyramid, and an output layer connected to the output end of the fusion layer; wherein the upsampling layer in the first one-dimensional feature pyramid and the upsampling layer in the second one-dimensional feature pyramid are connected to each other; A model training module is used to collect the fault time-frequency spectrum of the reducer gear in the scraper conveyor and the gear fault type corresponding to the time-frequency spectrum to construct a gear fault data set; the fault time-frequency spectrum is a two-dimensional matrix, each row in the two-dimensional matrix corresponds to each frequency in the current signal when the gear fails; each column in the two-dimensional matrix corresponds to each time sampling point when the gear fails; the gear fault data set is used to train the deep learning network to obtain a reducer gear fault recognition model for mapping the gear fault time-frequency spectrum to the gear fault type; A fault identification module is used to input the time-frequency spectrum of the gear fault to be identified into the reducer gear fault identification model, and segment the time-frequency spectrum of the gear fault to be identified according to the time dimension and the frequency dimension through the segmentation layer to obtain the gear fault time series and the gear fault frequency series; the gear fault time series and the gear fault frequency series are respectively extracted by the first one-dimensional feature pyramid and the second one-dimensional feature pyramid in parallel. During the feature extraction process, the feature map output by the upsampling layer in the first one-dimensional feature pyramid and the feature map output by the upsampling layer in the second one-dimensional feature pyramid are circulated with each other, so as to capture the long-distance dependency in the time-frequency spectrum of the gear fault to be identified by cross-fusion of information of different dimensions, and obtain the gear fault time series feature and the gear fault frequency feature; the gear fault time series feature and the gear fault frequency feature are fused through the fusion layer to obtain the gear fault fusion feature; the gear fault fusion feature is classified through the output layer, and the gear fault type corresponding to the time-frequency spectrum of the gear fault to be identified is output.

Images