Signal enhancement method and device based on mixed variants

Abstract

The invention relates to a signal enhancement method and device based on mixed variants. The method comprises the following steps: acquiring a vibration signal of target equipment; randomly selecting a signal sample group from the vibration signal, wherein the signal sample group comprises a plurality of signal samples of different fault types; selecting a target strategy from a plurality of preset strategies by using the signal sample group; and performing enhancement processing on the vibration signal by using the target strategy to obtain an enhanced signal. According to the method, a traditional fault signal enhancement method based on mixed variants is improved, firstly, a signal sample group is selected according to configuration information of equipment, then enhancement processing is conducted on the signal sample group through multiple preset strategies, and therefore a target strategy most suitable for vibration signals in the preset strategies is selected; the signal enhancement accuracy is improved, the signal processing flexibility is enhanced, and useful fault information can be effectively extracted from the noise-containing signal, so that the fault detection accuracy and efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of signal processing, and in particular to a signal enhancement method and device based on hybrid variants. Background Art

[0002] In the current industrial field, automatic detection technology of bearing faults has become one of the research focuses. With the continuous advancement of technology, fault diagnosis methods based on deep learning have received widespread attention and application due to their excellent nonlinear learning ability; fault diagnosis methods based on deep learning mainly rely on advanced deep learning architectures such as convolutional neural networks (CNN) and recurrent neural networks (RNN).

[0003] However, the number of samples used for fault classification in public vibration datasets is often limited, and the operating conditions of rotating parts during actual operation, such as speed and load, vary greatly, resulting in a significant shift in data distribution. Therefore, in practical applications, bearing fault detection methods usually need to rely on manually designed time domain and frequency domain features of vibration signals as input, rather than directly using the original vibration signals. Although this feature engineering-based method can improve the performance of the classifier to a certain extent, in real scenarios, due to the shift in data distribution, the performance of fault classifiers based on vibration signals often cannot reach the level of testing in a controlled environment.

[0004] Therefore, there is an urgent need to develop a signal enhancement method and device based on hybrid variants to solve one or more of the above-mentioned problems. Summary of the invention

[0005] In view of this, in order to solve the above technical problems or part of the technical problems, the embodiments of the present invention provide a signal enhancement method and device based on hybrid variants.

[0006] In a first aspect, the present application provides a signal enhancement method based on hybrid variants, the method comprising:

[0007] Obtain vibration signals from the target device;

[0008] Randomly selecting a signal sample group from the vibration signal, wherein the signal sample group includes signal samples of multiple different fault types;

[0009] Using the signal sample group, selecting a target strategy from a plurality of preset strategies;

[0010] The vibration signal is enhanced using the target strategy to obtain an enhanced signal.

[0011] In a possible implementation manner, the randomly selecting a signal sample group from the vibration signal includes:

[0012] Obtaining a fault type corresponding to the vibration signal;

[0013] dividing the vibration signal into a plurality of vibration signal groups according to the fault type;

[0014] Vibration signals of a preset sample quantity are randomly selected from a plurality of vibration signal groups to form a signal sample group.

[0015] In a possible implementation, the using the signal sample group to select a target strategy from a plurality of preset strategies includes:

[0016] Using a plurality of preset strategies to perform enhancement processing on the signal sample group respectively, to obtain an enhanced signal of each preset strategy;

[0017] The signal feature extraction strategy is used to obtain the signal feature of each enhanced signal, and the strategy quality of each preset strategy is determined according to the feature score of the corresponding signal;

[0018] Based on the strategy quality, a target strategy corresponding to the signal sample group is determined from a plurality of preset strategies, wherein the target strategy is a preset strategy with the highest strategy quality.

[0019] In a possible implementation manner, the preset strategy includes a preset hybrid strategy and a deep learning strategy;

[0020] When the preset mixing strategy is the first mixing strategy, performing enhancement processing on the signal sample group to obtain an enhanced signal includes:

[0021] Performing mixed processing on vibration signals corresponding to different fault types in the signal sample group to obtain a mixed signal;

[0022] The mixed signal is enhanced using a deep learning strategy to obtain an enhanced signal.

[0023] In a possible implementation manner, when the preset mixing strategy is the second mixing strategy, performing enhancement processing on the signal sample group to obtain an enhanced signal includes:

[0024] Using a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types;

[0025] Performing mixed processing on enhanced signals of different fault types to obtain a mixed enhanced signal;

[0026] The mixed enhanced signal is input into the fully connected layer and summed to obtain the enhanced signal.

[0027] In a possible implementation, the deep learning strategy includes:

[0028] Multilayer Perceptron strategy, two variants of convolutional network strategy, A-Net architecture strategy.

[0029] In a possible implementation, the using a deep learning strategy to perform enhancement processing on the mixed signal to obtain an enhanced signal includes:

[0030] According to the signal quality of the mixed signal, the mixed signal is divided into a plurality of mixed signal groups, and a corresponding quality label is added to each mixed signal group;

[0031] Using the quality labels, performing different degrees of data cleaning processing on multiple mixed signal groups to obtain multiple enhanced mixed signal groups;

[0032] Normalization processing is performed on the enhanced mixed signals in the plurality of enhanced mixed signal groups to obtain enhanced signals.

[0033] In a possible implementation, the use of a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types includes:

[0034] Acquire signal characteristics of vibration signals of different fault types in the signal sample group;

[0035] According to the signal characteristics, the vibration signals of different fault types are amplified to obtain enhanced signals of different fault types.

[0036] In a possible implementation, the feature extraction strategy includes a time domain extraction strategy, a real-valued fast Fourier transform strategy, and a short-time Fourier transform strategy.

[0037] In a second aspect, the present application provides a signal enhancement device based on a hybrid variant, the device comprising:

[0038] An acquisition module, used to acquire a vibration signal of a target device;

[0039] A first selection module, used for randomly selecting a signal sample group from the vibration signal;

[0040] A second selection module is used to select a target strategy from a plurality of preset strategies using the signal sample group;

[0041] The enhancement module is used to perform enhancement processing on the vibration signal using the target strategy to obtain an enhanced signal.

[0042] The above-mentioned technical scheme provided by the embodiment of the present application has the following advantages compared with the prior art: the method provided by the embodiment of the present application improves the traditional fault signal enhancement method based on hybrid variants. First, a signal sample group is selected according to the configuration information of the equipment, and then the signal sample group is enhanced using multiple preset strategies, so as to select the target strategy that is most suitable for the vibration signal among the preset strategies, and then the vibration signal is enhanced using the target strategy, which not only improves the accuracy of signal enhancement, but also enhances the flexibility of signal processing; useful fault information can be effectively extracted from noisy signals, thereby improving the accuracy and efficiency of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0046] Figure 1 A schematic diagram of a flow chart of a signal enhancement method based on hybrid variants provided in an embodiment of the present application;

[0047] Figure 2 A schematic flow chart of another signal enhancement method based on hybrid variants provided in an embodiment of the present application;

[0048] Figure 3 A Mixup diagram of the vibration waveform provided in the embodiment of the present application;

[0049] Figure 4 A schematic diagram of the steps of the first mixing strategy provided in an embodiment of the present application;

[0050] Figure 5 A schematic diagram of the steps of the second mixing strategy provided in an embodiment of the present application;

[0051] Figure 6 Different failure waveforms and their RFFT and STFT characteristic diagrams provided in the embodiments of the present application;

[0052] Figure 7 A schematic diagram of scores of different training data percentages provided in an embodiment of the present application;

[0053] Figure 8 A schematic structural diagram of a signal enhancement device based on a hybrid variant provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0055] The disclosure below provides many different embodiments or examples to implement different structures of the present invention. In order to simplify the disclosure of the present invention, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present invention. In addition, the present invention can repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0056] In order to solve the problem that the number of samples used for fault classification in the prior art is usually limited when processing public vibration data sets, and the working conditions of rotating parts in the actual operation process, such as speed and load, change greatly, resulting in a significant deviation in data distribution. And the bearing fault detection method relies on the manually designed time domain and frequency domain features of the vibration signal as input, and the deviation of the data distribution causes the performance of the fault classifier of the vibration signal to often fail to reach the level of the test under a controlled environment. The present application proposes a signal enhancement method and device based on hybrid variants to make innovative improvements to the traditional fault signal enhancement method based on hybrid variants. First, a suitable signal sample group is selected according to the configuration information of the equipment, and then these signal sample groups are enhanced using a variety of preset strategies. Through this method, the target strategy that best suits the vibration signal can be selected, and the vibration signal is enhanced using the selected target strategy, which not only improves the accuracy of signal enhancement, but also enhances the flexibility of signal processing, effectively extracts useful fault information from the noisy signal, and thus significantly improves the accuracy and efficiency of fault detection.

[0057] Figure 1 A schematic diagram of a signal enhancement method based on a hybrid variant provided in an embodiment of the present application, such as Figure 1 As shown, the method specifically includes:

[0058] S101, obtaining a vibration signal of a target device;

[0059] Advances in IoT technology, the availability of low-cost sensors, and the latest developments in machine learning have driven the development of predictive maintenance. Rolling bearings are key components of rotating machinery, and their failure can have a serious impact on the equipment. Vibration signals acquired through various sensors are widely used to monitor and evaluate the health of rotating systems. Timely fault detection can alert maintenance personnel, thereby avoiding productivity losses.

[0060] In this embodiment, in order to realize fault detection of the target device, a plurality of sensors are arranged on the target device, and the sensors are used to capture the tiny vibrations generated by the device during operation to obtain the vibration signals of the target device.

[0061] In order to achieve real-time monitoring of the operating status of the target device, multiple sensors are installed on the device to capture the tiny vibrations generated by the device during daily operation, and the vibration signals of the target device are obtained through the sensors.

[0062] S102, randomly selecting a signal sample group from the vibration signal, wherein the signal sample group includes signal samples of multiple different fault types;

[0063] In the traditional method, samples of any fault category are randomly sampled from the entire vibration signal set, which may result in samples coming from the same fault category and lack comprehensiveness;

[0064] In this embodiment, randomly selecting faults belonging to different fault categories, that is, mixed fault categories that do not belong to the same fault, from the vibration signal can compensate for the situation where there are fewer available samples in the fault category and provide a certain degree of robustness under different operating conditions.

[0065] S103, using the signal sample group, selecting a target strategy from a plurality of preset strategies;

[0066] According to the signal characteristics of various devices, a variety of preset signal enhancement strategies are provided to ensure the best signal reception effect in different usage environments. The preset signal enhancement strategy consists of hybrid strategy and deep learning strategy.

[0067] In this embodiment, the signal sample group is used to obtain the enhancement effect corresponding to each preset strategy, so as to select the signal enhancement strategy that best suits the vibration signal of the target device as the target strategy for the target device.

[0068] S104: Use the target strategy to perform enhancement processing on the vibration signal to obtain an enhanced signal.

[0069] In this embodiment, after determining the target strategy that is most suitable for the target device, the target strategy is used to enhance the vibration signal of the target device to obtain an enhanced signal. This enhanced signal can effectively improve the signal-to-noise ratio of the signal and reduce noise interference, thereby making the fault characteristics in the signal more obvious.

[0070] The embodiment of the present application provides a method of first selecting a suitable signal sample group according to the configuration information of the device, and then enhancing the signal sample groups using a plurality of preset strategies. Through this method, the target strategy that best suits the vibration signal can be selected, and the vibration signal can be enhanced using the selected target strategy, which not only improves the accuracy of signal enhancement, but also enhances the flexibility of signal processing, effectively extracting useful fault information from noisy signals, thereby significantly improving the accuracy and efficiency of fault detection.

[0071] In an embodiment of the present application, a group of suitable signal samples are first selected according to the configuration information of the device to lay the foundation for subsequent signal processing. After the signal sample group is selected, these signal samples are enhanced using a variety of preset strategies. According to the enhancement results, the target strategy that best suits the vibration signal is selected, and the vibration signal is enhanced using the strategy to obtain the enhancement strategy that best suits the vibration signal, which can significantly improve the signal-to-noise ratio of the signal and ensure the accuracy and efficiency of the enhancement processing.

[0072] In an optional solution of the embodiment of the present invention, the randomly selecting a signal sample group from the vibration signal includes:

[0073] Obtaining a fault type corresponding to the vibration signal;

[0074] dividing the vibration signal into a plurality of vibration signal groups according to the fault type;

[0075] Vibration signals of a preset sample quantity are randomly selected from a plurality of vibration signal groups to form a signal sample group.

[0076] In this embodiment, the fault type of the device corresponding to the vibration signal is first obtained, and then the vibration signal is classified according to the fault type to obtain vibration signal groups corresponding to different fault types, which is convenient for subsequent analysis and processing of the vibration signal. Finally, a preset number of vibration signal samples are randomly selected from the divided vibration signal groups to form a signal sample group, so that the sample data in the signal sample group includes all fault types of the target device and is more representative.

[0077] Exemplarily, the training data is represented as (X, Y), where represents n input feature vectors, Represents the corresponding one-hot encoded label vector. Input vector X iRepresents features extracted from raw vibration signals based on domain knowledge, or spectral transforms of accelerometer time series.

[0078] Mixup generates a new training sample by randomly extracting two samples from the training data, and then uses the input X i , X j and its corresponding label Y i , Y j Get new samples (X', Y'), where:

[0079] X′=λgX i +(1-λ)gX j

[0080] Y′=λgY i +(1-λ)gY j

[0081] λ∈[0,1], λ: Bate(α,α) and α∈(0,∞).

[0082] like Figure 3 As shown in the figure, in the context of bearing fault prediction, Mixup is used to train a multi-class classifier, taking the vibration signal recorded by the vibration sensor as input to achieve the distinction between the normal state and different fault categories.

[0083] Figure 2 A schematic diagram of another signal enhancement method based on hybrid variants provided in an embodiment of the present application, such as Figure 2 As shown, the method specifically includes:

[0084] S201, respectively using a plurality of preset strategies to perform enhancement processing on the signal sample group to obtain an enhanced signal of each preset strategy;

[0085] In this embodiment, a plurality of preset enhancement strategies (preset strategies) are first obtained, and then the sample signals in the signal sample group are enhanced in turn using the plurality of enhancement strategies to obtain the enhancement result of each enhancement strategy (preset strategy), i.e., the enhanced signal.

[0086] S202, using a signal feature extraction strategy to obtain a signal feature of each enhanced signal, and determining a strategy quality of each preset strategy according to a feature score of the corresponding signal;

[0087] In this embodiment, after obtaining the enhanced signal of each preset strategy, the characteristic score of the enhanced signal of each preset strategy is obtained, and then the strategy quality of each preset strategy is determined based on the characteristic score of the enhanced signal of each preset strategy.

[0088] Specifically, the feature score is an important indicator for measuring signal quality, which can further determine the strategy quality of each preset strategy, that is, evaluate the performance and effect of each strategy in actual application. The feature score of the enhanced signal is extracted by enhancing the key signal features of the signal, such as frequency, amplitude, phase, etc., and a comprehensive score is calculated based on these features to reflect the quality of the signal.

[0089] Furthermore, the evaluation of each preset strategy enhancement signal is based on the correlation between the signal characteristics and the strategy effect, which ensures the accuracy and reliability of the evaluation results, and then effectively screens out the optimal signal enhancement strategy to improve the overall performance of signal processing.

[0090] S203: Determine a target strategy corresponding to the signal sample group from a plurality of preset strategies based on the strategy quality, wherein the target strategy is a preset strategy with the highest strategy quality.

[0091] In this embodiment, based on the evaluation result of the strategy quality, the strategy that best suits the current signal sample group is selected from the preset strategies to ensure that the target strategy finally selected can meet the enhancement requirements of the vibration signal to the greatest extent.

[0092] The signal enhancement method based on hybrid variants provided in the embodiment of the present application adopts multiple preset enhancement strategies to enhance the signal sample group to obtain multiple enhancement results. By judging the enhancement quality of the multiple enhancement results, the enhancement strategy that best suits the vibration signal of the target device is selected from the multiple preset enhancement strategies, which not only improves the efficiency of signal processing, but also enhances the feature score of the signal, thereby improving the overall quality of the signal.

[0093] In an optional solution of the embodiment of the present invention, the preset strategy includes a preset hybrid strategy and a deep learning strategy;

[0094] When the preset mixing strategy is the first mixing strategy, performing enhancement processing on the signal sample group to obtain an enhanced signal includes:

[0095] Performing mixed processing on vibration signals corresponding to different fault types in the signal sample group to obtain a mixed signal;

[0096] The mixed signal is enhanced using a deep learning strategy to obtain an enhanced signal.

[0097] In this embodiment, the first mixing strategy is a feature level mixing variant (Feature level Mixup). When the current preset mixing strategy is set to the first mixing strategy, the signal sample group is enhanced to obtain an enhanced signal, which specifically includes the following steps: first, the vibration signals representing different fault types in the signal sample group are mixed based on the mixing strategy (Mixup) to obtain a mixed signal; then, the mixed signal obtained above is further enhanced using the deep learning strategy (Deep learning model (D)) to finally obtain the desired enhanced signal.

[0098] like Figure 4 As shown in the figure, the spectral characteristics of bearing fault categories are different from other fields such as computer vision. Therefore, different variants of corresponding label selection data points based on Mixup (mixing strategy) are studied, and the impact of different variants on signal enhancement performance is studied using feature-level mixing variants.

[0099] In an optional solution of the embodiment of the present invention, when the preset mixing strategy is the second mixing strategy, performing enhancement processing on the signal sample group to obtain an enhanced signal includes:

[0100] Using a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types;

[0101] Performing mixed processing on enhanced signals of different fault types to obtain a mixed enhanced signal;

[0102] The mixed enhanced signal is input into the fully connected layer and summed to obtain the enhanced signal.

[0103] In this embodiment, the second hybrid strategy is a representation hybrid variant (Representation Mixup). First, the vibration signals corresponding to various fault types in the signal sample group are effectively enhanced through the deep learning strategy (Deep learning model (D)), so as to obtain enhanced signals for different fault types; then, based on Mixup (hybrid strategy), the enhanced signals of different fault types are mixed to obtain a hybrid enhanced signal that integrates multiple fault features; finally, the hybrid enhanced signal is input into the fully connected layer, and by performing a summation process, a signal enhanced by the deep learning strategy is finally obtained.

[0104] like Figure 5 As shown in Figure 2, another variant of Mixup does not mix the inputs, but instead passes the inputs through a neural network to elicit a representation D(x i ) and D(x j), and then mix them. For deep learning models, a Mixup layer is introduced before the fully connected layer to achieve representation mixing, and then the model is trained end-to-end.

[0105] It should be noted that Figure 4 or Figure 5 In the example, the input vector X i and X j Represents features obtained from the raw vibration signal (time) based on domain knowledge, or the spectral transform of the accelerometer time series (RFFT, SFFT).

[0106] In an optional solution of the embodiment of the present invention, the deep learning strategy includes:

[0107] Multilayer perceptron strategy, variant convolutional network strategy, A-Net architecture strategy.

[0108] In this example, four different neural network architectures were used: a multilayer perceptron (MLP), two variants of convolutional networks (CNN1 and CNN2), and an AlexNet-inspired architecture (A-Net) that combines temporal, real-valued fast Fourier transform, and short-time Fourier transform features. The proposed MLP is a 4-layer fully connected network with half the number of input features, including 256, 128, and 6 neurons, and uses batch normalization and ReLU activation functions. The short-time Fourier transform features of the entire time dimension are averaged as input to the MLP, while for CNN1, CNN2, and A-Net, a two-dimensional convolutional layer is used to process the two-dimensional short-time Fourier transform features. CNN1 includes 4 convolutional layers and 3 fully connected layers, while CNN2 only includes 1 fully connected layer. A-Net has 5 convolutional layers and 2 fully connected layers.

[0109] In an optional solution of the embodiment of the present invention, the step of using a deep learning strategy to perform enhancement processing on the mixed signal to obtain an enhanced signal includes:

[0110] According to the signal quality of the mixed signal, the mixed signal is divided into a plurality of mixed signal groups, and a corresponding quality label is added to each mixed signal group;

[0111] Using the quality labels, performing different degrees of data cleaning processing on multiple mixed signal groups to obtain multiple enhanced mixed signal groups;

[0112] Normalization processing is performed on the enhanced mixed signals in the plurality of enhanced mixed signal groups to obtain enhanced signals.

[0113] In this embodiment, the mixed signal is enhanced using a deep learning strategy to obtain an enhanced signal, which mainly includes the following steps:

[0114] Step 1: According to the signal quality of the mixed signal, the mixed signal is divided into multiple mixed signal groups, and a corresponding quality label is added to each mixed signal group to ensure that the signals in each signal group have similar characteristics in quality, thereby facilitating subsequent data processing.

[0115] Step 2: Use the quality labels to perform different degrees of data cleaning on multiple mixed signal groups to obtain multiple enhanced mixed signal groups. Different cleaning strategies can be adopted for signal groups with different quality levels to optimize data accuracy and reliability.

[0116] Step 3: Normalize the enhanced mixed signals in the multiple enhanced mixed signal groups to obtain enhanced signals. Adjusting the signals to a standard range through normalization helps to eliminate the dimensional differences between different signal groups, so that the signals can be more effectively used for subsequent analysis and processing.

[0117] In an optional solution of the embodiment of the present invention, the use of a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types includes:

[0118] Acquire signal characteristics of vibration signals of different fault types in the signal sample group;

[0119] According to the signal characteristics, the vibration signals of different fault types are amplified to obtain enhanced signals of different fault types.

[0120] In this embodiment, using a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group mainly includes the following steps:

[0121] Step 1: Extract characteristic information of vibration signals of different fault types from the signal sample group;

[0122] Step 2: Based on the extracted signal features, the vibration signals of different fault types are amplified to obtain enhanced signals for each fault type, thereby improving the fault detection efficiency.

[0123] In an optional solution of the embodiment of the present invention, the feature extraction strategy includes a time domain extraction strategy, a real-valued fast Fourier transform strategy, and a short-time Fourier transform strategy.

[0124] In this embodiment, each waveform in the data set is divided into 2048-point windows with 25% overlap between windows; on each window, three feature extraction methods are used: the first method is to extract different time domain features (time), including mean, absolute median, standard deviation, skewness, kurtosis, peak factor, energy, root mean square value, number of peaks, number of zero crossings, Shapiro test and KL divergence; the second method is to use real-value Fast Fourier Transform (RFFT); the third method is to use short-time Fourier Transform (STFT).

[0125] It should be noted that when using the same evaluation index for model comparison, if there are large differences in the quality of the features and the data distribution, the training results may be significantly different. However, in the experiments corresponding to the embodiments, the data distribution of the time domain features, real-valued fast Fourier transform (FFT) and short-time Fourier transform (STFT) features used has little change, and these features are relatively close in description, thereby effectively avoiding the above problems.

[0126] Figure 6 As shown in the figure, the data set consists of vibration data recorded by accelerometers. The data comes from 15 bearings, which are operated to failure under three different load × force operating conditions. At the end of the experiment, the bearing failure type is recorded. The data set contains inner ring (IR), outer ring (OR), inner ring-outer ring (IR and OR), cage (cage), inner ring-ball-cage (COBI) bearing faults and normal data. The vibration signal is recorded at a sampling rate of 25.6kHz for 1.28 seconds per minute, and each waveform produces 32768 data points; five fault class and normal class sample waveform windows and their corresponding RFFT and average STFT features are shown. The time and RFFT features are vectors, while the STFT features lead to matrix results; in this experimental data set, there are 4928 OR samples, 1152 IR samples, 32 IR and OR samples, 2832 COBI samples, 3136 fault samples and 135376 normal samples, where Amplitude is amplitude and Frequency is frequency.

[0127] Furthermore, the experiment of multi-class classifier of bearing faults compared the performance of the baseline model with the effect of the proposed hybrid variant method. At the same time, the performance of the baseline model and the model trained using the proposed method were studied, and the impact of different percentages of training data on the model performance was analyzed. In the experimental test, the training set, test set and validation set were divided by ensuring that all waveforms from the same bearing belonged to the same partition. Due to the imbalance of the dataset, the scale F was used. macro (Fmac ), precision (Rec), recall (Rec) and other indicators.

[0128] Table 1 shows the results of training 6 types of bearing fault classifiers using different architectures. It can be observed that on each model and dataset, the performance of RFFT and STFT features is generally better than that of temporal features. As shown in Table 1, among these combinations, CNN2 with real fast Fourier transform features has the highest FFT performance. mac score (0.314), and thus it is used as a baseline for comparison with the proposed method. The performance of the baseline classifier is lower compared to the random partitioning, which indicates that performance evaluation based on bearing partitioning is more reflective of real-world scenarios.

[0129] Table 1

[0130]

[0131]

[0132] The corresponding usage feature mixed variants and representation mixed variant results, Figure 2 The results of different feature mixing variants (first mixing strategy) are shown, as described for CNN1 in Table 2 (Table 2(a), Table 2(b), Table 2(c)). It can be observed that the mixed variants perform better than the baseline method, where the STFT feature of A-Net is better in F mac Among all feature-level mixing variants, traditional Mixup performs better, F mac The score is 0.376, which is 6.2% higher than the baseline (about 20% improvement). Other feature-level hybrid variants such as fault-only, same fault, and no fault also perform best, with F mac The scores are 0.369, 0.352, and 0.341, respectively, which are 5.5%, 3.8%, and 2.7% higher than the baseline.

[0133] Table 2(a)

[0134]

[0135]

[0136] Table 2(b)

[0137]

[0138]

[0139] Table 2(c)

[0140]

[0141]

[0142] The results of using representation mixture variants are shown in Table 3. A-Net with STF performs well in various combinations, F mac The score is 0.351, which is 3.7% higher than the baseline. Although the representation hybrid variant outperforms the baseline, it is not as effective as the feature-level hybrid variant under the current architecture setting.

[0143] Table 3

[0144]

[0145]

[0146] In addition, if Figure 7 As shown, we hypothesize that data augmentation strategies will be particularly useful in situations where only a small amount of data is available to train the model. To test this, we used different parts of the training dataset while keeping the evaluation set unchanged and trained the A-Net architecture with STFT features as well as without mixup. The best feature-level mixup (Traditional Mixup) and representation-level mixup variants (Percentage of Data Used in Training) consistently outperformed Figure 7 Thus, using Mixup on just over 10% of the training data, the model trained with Mixup is able to achieve similar or higher performance than the baseline classifier trained using 100% of the training data.

[0147] Therefore, the hybrid variants (Table 2(a), Table 2(b), Table 2(c)), and Table 3) effectively improve the performance of the baseline classifier (Table 1) under challenging conditions that simulate real-world scenarios, especially in terms of precision and recall, overcoming the lower performance of the baseline model under these conditions.

[0148] Figure 8 A schematic diagram of the structure of a signal enhancement device based on a hybrid variant provided in an embodiment of the present application is shown in FIG. Figure 8 As shown, the device specifically includes:

[0149] An acquisition module 801 is used to acquire a vibration signal of a target device;

[0150] A first selection module 802 is used to randomly select a signal sample group from the vibration signal;

[0151] A second selection module 803 is used to select a target strategy from a plurality of preset strategies using the signal sample group;

[0152] The enhancement module 804 is used to perform enhancement processing on the vibration signal using the target strategy to obtain an enhanced signal.

[0153] In a possible implementation, the first selection module 802 is further used to obtain the fault type corresponding to the vibration signal; divide the vibration signal into multiple vibration signal groups according to the fault type; and randomly select a preset sample amount of vibration signals from the multiple vibration signal groups to form a signal sample group.

[0154] In one possible implementation, the first selection module 802 is further used to perform enhancement processing on the signal sample group using a plurality of preset strategies respectively to obtain an enhanced signal for each preset strategy; obtain the signal feature of each enhanced signal using a signal feature extraction strategy, and determine the strategy quality of each preset strategy based on the feature score of the corresponding signal; based on the strategy quality, determine the target strategy corresponding to the signal sample group from a plurality of preset strategies, wherein the target strategy is the preset strategy with the highest strategy quality.

[0155] In one possible embodiment, the first selection module 802 is also used to enhance the signal sample group to obtain an enhanced signal when the preset mixing strategy is the first mixing strategy, including: mixing the vibration signals corresponding to different fault types in the signal sample group to obtain a mixed signal; using a deep learning strategy to enhance the mixed signal to obtain an enhanced signal.

[0156] In one possible implementation, the first selection module 802 is further used to use a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types; to mix the enhanced signals of different fault types to obtain a mixed enhanced signal; and to input the mixed enhanced signal into a fully connected layer for summing to obtain an enhanced signal.

[0157] In a possible implementation, the first selection module 802 is further used to divide the mixed signal into multiple mixed signal groups according to the signal quality of the mixed signal, and add a corresponding quality label in each mixed signal group; use the quality labels to perform different degrees of data cleaning processing on the multiple mixed signal groups to obtain multiple enhanced mixed signal groups; normalize the enhanced mixed signals in the multiple enhanced mixed signal groups to obtain enhanced signals.

[0158] In a possible implementation, the first selection module 802 is further used to obtain signal characteristics of vibration signals of different fault types in the signal sample group; and amplify the vibration signals of different fault types according to the signal characteristics to obtain enhanced signals of different fault types.

[0159] The signal enhancement device based on the hybrid variant provided in this embodiment can be as follows Figure 8 The signal enhancement device based on the hybrid variant shown in FIG. Figure 1-5 All steps of hybrid variant-based signal enhancement in Figure 1-5 The technical effect of signal enhancement based on hybrid variants is shown, please refer to Figure 1-5 For the sake of brevity, the relevant description is not repeated here.

[0160] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0161] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0162] It should be understood that the terms used in the text are only for the purpose of describing specific example embodiments, and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used in the text may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0163] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A signal enhancement method based on hybrid variants, characterized in that: include: Obtain vibration signals from the target device; Randomly selecting a signal sample group from the vibration signal, wherein the signal sample group includes signal samples of multiple different fault types; Using the signal sample group, selecting a target strategy from a plurality of preset strategies; The vibration signal is enhanced using the target strategy to obtain an enhanced signal.

2. The method according to claim 1, characterized in that The randomly selecting a signal sample group from the vibration signal comprises: Obtaining a fault type corresponding to the vibration signal; dividing the vibration signal into a plurality of vibration signal groups according to the fault type; Vibration signals of a preset sample quantity are randomly selected from a plurality of vibration signal groups to form a signal sample group.

3. The method according to claim 2, characterized in that The method of using the signal sample group to select a target strategy from a plurality of preset strategies includes: Using a plurality of preset strategies to perform enhancement processing on the signal sample group respectively, to obtain an enhanced signal of each preset strategy; The signal feature extraction strategy is used to obtain the signal feature of each enhanced signal, and the strategy quality of each preset strategy is determined according to the feature score of the corresponding signal; Based on the strategy quality, a target strategy corresponding to the signal sample group is determined from a plurality of preset strategies, wherein the target strategy is a preset strategy with the highest strategy quality.

4. The method according to claim 3, characterized in that The preset strategy includes a preset hybrid strategy and a deep learning strategy; When the preset mixing strategy is the first mixing strategy, performing enhancement processing on the signal sample group to obtain an enhanced signal includes: Performing mixed processing on vibration signals corresponding to different fault types in the signal sample group to obtain a mixed signal; The mixed signal is enhanced using a deep learning strategy to obtain an enhanced signal.

5. The method according to claim 4, characterized in that When the preset mixing strategy is the second mixing strategy, performing enhancement processing on the signal sample group to obtain an enhanced signal includes: Using a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types; Performing mixed processing on enhanced signals of different fault types to obtain a mixed enhanced signal; The mixed enhanced signal is input into the fully connected layer and summed to obtain the enhanced signal.

6. The method according to claim 5, characterized in that The deep learning strategy includes: Multilayer perceptron strategy, variant convolutional network strategy, A-Net architecture strategy.

7. The method according to claim 4, characterized in that The method of using a deep learning strategy to enhance the mixed signal to obtain an enhanced signal includes: According to the signal quality of the mixed signal, the mixed signal is divided into a plurality of mixed signal groups, and a corresponding quality label is added to each mixed signal group; Using the quality labels, performing different degrees of data cleaning processing on multiple mixed signal groups to obtain multiple enhanced mixed signal groups; Normalization processing is performed on the enhanced mixed signals in the plurality of enhanced mixed signal groups to obtain enhanced signals.

8. The method according to claim 5, characterized in that The method of using a deep learning strategy to enhance the vibration signals corresponding to different fault types in the signal sample group to obtain enhanced signals of different fault types includes: Acquire signal characteristics of vibration signals of different fault types in the signal sample group; According to the signal characteristics, the vibration signals of different fault types are amplified to obtain enhanced signals of different fault types.

9. The method according to claim 3, characterized in that: The feature extraction strategies include a time domain extraction strategy, a real-valued fast Fourier transform strategy, and a short-time Fourier transform strategy.

10. A signal enhancement device based on a hybrid variant, characterized in that: include: An acquisition module, used to acquire a vibration signal of a target device; A first selection module, used for randomly selecting a signal sample group from the vibration signal; A second selection module is used to select a target strategy from a plurality of preset strategies using the signal sample group; The enhancement module is used to perform enhancement processing on the vibration signal using the target strategy to obtain an enhanced signal.

Images