Mechanical equipment fault feature extraction and model training method and device and fault diagnosis method and device

Abstract

The invention discloses a mechanical equipment fault feature extraction, model training and fault diagnosis method and device. The method comprises the steps that mechanical equipment fault sample signals corresponding to different mechanical equipment real fault types are acquired, noise signals are few, and the quality of the acquired mechanical equipment fault signals is high; decomposing the mechanical fault signal to obtain a first specified number of IMF components; for each IMF component, extracting a second specified number of effective time-frequency features from the fault sample signal to obtain an effective feature set of the fault sample signal; inputting the effective feature set of the fault sample signal into a pre-established self-encoding model for model training, fusing different IMF components of each effective feature in the effective feature set to obtain a fused feature of the effective features, and obtaining a fused feature of the effective features; after the trained model accords with the preset rule, the trained mechanical equipment fault feature extraction model is obtained, the fault type accuracy based on fusion feature prediction is improved, and the precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical equipment fault diagnosis, and particularly relates to a method and device for extracting fault characteristics of mechanical equipment, training a model, and diagnosing faults. Background Art

[0002] With the development of mechanical equipment technology, the maintenance system of mechanical equipment has gradually developed, and the fault diagnosis technology of mechanical equipment has emerged as the times require. The research results of the fault diagnosis technology of mechanical equipment show that applying the fault diagnosis technology of real-time condition detection to large-scale mechanical equipment can largely prevent and reduce the incidence of safety accidents, effectively eliminate potential safety hazards, and provide sufficient guarantee for safe production. Mechanical equipment fault diagnosis mainly includes three parts: signal acquisition, feature extraction, and fault identification and prediction. Among them, feature extraction is an important link in the field of fault diagnosis. By extracting features from the vibration signals of mechanical equipment obtained, the fault identification and prediction of mechanical equipment are realized. Summary of the Invention

[0003] Good feature extraction can effectively extract data features that objectively reflect normal and fault states from redundant vibration signals. The extracted sensitive features can not only more effectively characterize the health state of the equipment, but also reduce the amount of data. It can improve the accuracy and learning rate of the subsequent diagnosis model. Therefore, how to quickly and effectively extract fault-sensitive features is an important research topic in fault diagnosis. In actual situations, mechanical equipment will inevitably be affected by various factors in a complex working environment during operation, resulting in a large amount of noise in the collected vibration signals. However, the traditional fault-sensitive feature extraction methods used in mechanical equipment fault diagnosis do not consider the noise problem in vibration signals. The fault-sensitive features extracted by them are usually submerged by noise, resulting in too little effective information in the features and unable to extract feature data that objectively reflects normal and fault states, seriously affecting the subsequent diagnosis accuracy. The fault-sensitive features extracted by the existing fault-sensitive feature extraction methods for mechanical equipment fault diagnosis cannot obtain better diagnosis accuracy and have a low diagnosis accuracy rate. In order to improve the diagnosis accuracy rate, it is necessary to perform more effective fault-sensitive feature extraction.

[0004] In view of the above problems, the present invention is proposed to provide a method and device for extracting fault characteristics of mechanical equipment, training a model, and diagnosing faults that can overcome or at least partially solve the above problems.

[0005] An embodiment of the present invention provides a method for training a mechanical equipment fault feature extraction model, including:

[0006] Obtain a training set from mechanical equipment fault sample data, where the fault sample data includes mechanical equipment fault sample signals corresponding to different true fault types of mechanical equipment;

[0007] Decompose the mechanical equipment fault sample signals in the training set to obtain a first specified number of IMF components; extract a second specified number of effective time-frequency features from the fault sample signals for each IMF component to obtain an effective feature set of the fault sample signals;

[0008] Input the effective feature set of the fault sample signals into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fused feature of the effective feature;

[0009] Adjust the model parameters of the autoencoder model according to the comparison result between the decoded effective feature obtained by decoding based on the fused feature and the effective feature of the input fault sample signal. After iteratively performing model training for a preset number of times, obtain a trained fused feature extraction model.

[0010] In a preferred embodiment, the process of obtaining mechanical equipment fault sample data includes:

[0011] Collect mechanical equipment fault sample signals of mechanical equipment under different set fault types to obtain mechanical equipment fault sample data; the fault types include at least one of looseness fault, imbalance fault, cavitation fault, and rubbing fault.

[0012] In a preferred embodiment, decomposing the mechanical equipment fault sample signals to obtain a first specified number of IMF components; extracting a second specified number of effective time-frequency features from the fault sample signals for each IMF component to obtain an effective feature set of the fault sample signals includes:

[0013] Perform empirical mode decomposition on the mechanical equipment fault sample signals, and select a first specified number of IMF components from the decomposed IMF components according to a preset rule;

[0014] For each selected IMF component, select a second specified number of effective features from the time-domain features and frequency-domain features of the mechanical equipment fault sample signals based on the previously determined effective features of the fault sample signals to obtain an effective feature set for each IMF component.

[0015] In a preferred embodiment, a method for training a mechanical equipment fault feature extraction model further includes:

[0016] Use compensation distance evaluation technology to pre-select the top-ranked second specified number of features from the time-domain features and frequency-domain features of the fault sample signals as the effective features of the fault sample signals.

[0017] In a preferred embodiment, the effective feature set of the fault sample signal is input into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, different IMF components of the effective feature are fused to obtain the fused feature of the effective feature, including:

[0018] The first effective feature of each IMF component in the effective feature set of the fault sample signal is input into the first autoencoder for training, and the first autoencoder performs fusion and dimensionality reduction on the first effective feature to obtain the first fused feature;

[0019] The second effective feature of each IMF component in the effective feature set of the fault sample signal is input into the second autoencoder for training, and the second autoencoder performs fusion and dimensionality reduction on the second effective feature to obtain the second fused feature;

[0020] And so on, to obtain the fused features of each selected effective feature.

[0021] In a preferred embodiment, the model parameters of the autoencoder model are adjusted according to the comparison result between the decoded effective feature obtained by decoding the fused feature and the effective feature of the input fault sample signal. After iteratively executing the model training for a preset number of times, a trained fused feature extraction model is obtained, including:

[0022] The effective encoded feature output by the encoding layer and the decoded effective feature output by the decoding layer are compared, and the preset model parameters of the autoencoder model are adjusted according to the matching degree between the effective encoded feature output by the encoding layer and the decoded effective feature output by the decoding layer;

[0023] It is judged whether the number of iterative training times reaches the preset training times. If not, return to continue to execute the step of inputting the effective feature set of the fault sample signal into the pre-built autoencoder model for model training. If so, a trained fused feature extraction model is obtained.

[0024] In a preferred embodiment, a method for training a mechanical equipment fault feature extraction model further includes:

[0025] Obtain a test set from the mechanical equipment fault sample data;

[0026] Decompose the mechanical equipment fault sample signal in the training set to obtain a first specified number of IMF components; extract a first specified number of effective time-frequency features from the fault sample signal for each IMF component to obtain the effective feature set of the fault sample signal;

[0027] Input the effective feature set of the fault sample signal into the pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain the fused feature of the effective feature;

[0028] Determine the predicted fault type of the mechanical equipment according to the fused feature of the effective feature. According to the comparison result between the predicted fault type and the true fault type of the mechanical equipment included in the test set, determine the prediction accuracy, prediction precision and recall rate of the fused feature extraction model;

[0029] If the prediction precision and recall rate do not meet the preset requirements, continue to execute the steps of model training; if the prediction precision and recall rate meet the preset requirements, obtain the trained fused feature extraction model.

[0030] In a preferred embodiment, determining the predicted fault type of the mechanical equipment according to the fused feature of the effective feature, and determining the prediction accuracy, prediction precision and recall rate of the fused feature extraction model according to the comparison result between the predicted fault type and the true fault type of the mechanical equipment included in the test set, includes:

[0031] Determine the predicted fault type of the mechanical equipment fault according to the fused feature of the extracted effective feature; and obtain the true fault type of the mechanical equipment included in the test set;

[0032] Use the following formula to determine the prediction accuracy of the fused feature extraction model:

[0033] Wherein, the total sample refers to the total number of samples in the test set, and all correctly predicted samples refer to the number of samples in which the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set.

[0034] Use the following formula to determine the prediction precision of the fused feature extraction model:

[0035] Wherein, the predicted positive class refers to the total number of samples of a fault type predicted by the model. Predicting the positive class as the positive class means that among the samples of the predicted fault type, the number of samples in which the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set;

[0036] Use the following formula to determine the recall rate of the fused feature extraction model:

[0037] Among them, the original positive class refers to the total number of samples of a fault type in the test set. Predicting the positive class as the positive class means that among the samples of a fault type, the number of samples in which the predicted fault type of the mechanical equipment by the model is consistent with the true fault type of the mechanical equipment included in the test set.

[0038] An embodiment of the present invention provides a method for extracting mechanical equipment fault features, including:

[0039] Obtain the mechanical equipment fault signal to be analyzed;

[0040] Decompose the mechanical equipment fault signal to be analyzed to obtain a first specified number of IMF components, and extract a second specified number of effective time-frequency features from the mechanical equipment fault signal to be analyzed for each IMF component to obtain an effective feature set of the signal to be analyzed;

[0041] Input the effective feature set of the signal to be analyzed into the trained fusion feature extraction model for effective feature fusion, and output the fault-sensitive fusion features of the mechanical equipment;

[0042] The fusion feature extraction model is trained by using the mechanical equipment fault feature extraction model training method as described above.

[0043] An embodiment of the present invention provides a method for diagnosing mechanical equipment faults, including:

[0044] Adopt the mechanical equipment fault feature extraction method as described above to extract the fault-sensitive fusion features of the mechanical equipment;

[0045] Based on the extracted sensitive fusion features of the mechanical equipment, determine whether the mechanical equipment has a fault and the type of fault.

[0046] An embodiment of the present invention provides a device for training a mechanical equipment fault feature extraction model, including: a data acquisition module, a feature acquisition module, and a model training module;

[0047] The data acquisition module: is used to obtain a training set from the mechanical equipment fault sample data, and the fault sample data includes mechanical equipment fault sample signals corresponding to different true fault types of mechanical equipment; it is also used to obtain a test set from the mechanical equipment fault sample data;

[0048] Feature acquisition module: It is used to decompose the mechanical equipment fault sample signal to obtain the first specified number of IMF components; for each IMF component, extract the second specified number of effective time-frequency features from the fault sample signal to obtain the effective feature set of the fault sample signal; it is also used to decompose the mechanical equipment fault sample signal in the training set to obtain the first specified number of IMF components; for each IMF component, extract the first specified number of effective time-frequency features from the fault sample signal to obtain the effective feature set of the fault sample signal;

[0049] Model training module: It is used to input the effective feature set of the fault sample signal into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain the fused feature of the effective feature; adjust the model parameters of the autoencoder model according to the comparison result between the decoded effective feature obtained by decoding based on the fused feature and the effective feature of the input fault sample signal. After iteratively performing model training for a preset number of times, obtain the trained fused feature extraction model.

[0050] Preferred embodiment A mechanical equipment fault feature extraction model training device further includes: a model testing module;

[0051] Data acquisition module: It is also used to obtain a test set from the mechanical equipment fault sample data;

[0052] Feature acquisition module: It is also used to decompose the mechanical equipment fault sample signal in the training set to obtain the first specified number of IMF components; for each IMF component, extract the first specified number of effective time-frequency features from the fault sample signal to obtain the effective feature set of the fault sample signal;

[0053] Model testing module: It is used to determine the predicted fault type of the mechanical equipment according to the fused feature of the effective feature, and determine the prediction accuracy, prediction precision and recall rate of the fused feature extraction model according to the comparison result between the predicted fault type and the actual fault type of the mechanical equipment included in the test set; if the prediction precision and recall rate do not meet the preset requirements, continue to execute the steps of model training; if the prediction precision and recall rate meet the preset requirements, then obtain the trained fused feature extraction model.

[0054] An embodiment of the present invention provides a mechanical equipment fault feature extraction device, including:

[0055] Signal acquisition module: It is used to acquire the mechanical equipment fault signal to be analyzed;

[0056] Signal analysis module: It is used to decompose the mechanical equipment fault signal to be analyzed, obtain the first specified number of IMF components, extract the second specified number of effective time-frequency features from the mechanical equipment fault signal to be analyzed for each IMF component, and obtain the effective feature set of the signal to be analyzed;

[0057] Feature extraction module: It is used to input the effective feature set of the signal to be analyzed into the trained fusion feature extraction model for effective feature fusion, and output the fault-sensitive fusion features of the mechanical equipment; the fusion feature extraction model is trained by using the mechanical equipment fault feature extraction model training method as described above.

[0058] An embodiment of the present invention provides a mechanical equipment fault diagnosis device, including:

[0059] The mechanical equipment fault feature extraction device as described above, which is used to extract the fault-sensitive fusion features of the mechanical equipment;

[0060] Mechanical fault diagnosis module: It is used to determine whether there is a fault in the mechanical equipment and the type of fault based on the extracted sensitive fusion features of the mechanical equipment.

[0061] An embodiment of the present invention provides a computer storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, at least one of the mechanical equipment fault feature extraction method as described above, the mechanical equipment fault feature extraction method as described above, and the mechanical equipment fault diagnosis method as described above is implemented.

[0062] An embodiment of the present invention provides a mechanical equipment signal processing device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, at least one of the mechanical equipment fault feature extraction method as described above, the mechanical equipment fault feature extraction method as described above, and the mechanical equipment fault diagnosis method as described above is implemented.

[0063] The beneficial effects of the above technical solutions provided by the embodiments of the present invention at least include:

[0064] Obtain mechanical equipment fault sample signals corresponding to different true fault types of mechanical equipment. The obtained mechanical equipment fault signals have less noise and high quality. Constructing a training set based on the obtained data can effectively improve the effect of model training; decompose the mechanical fault signal to obtain the first specified number of IMF components; extract the second specified number of effective time-frequency features from the fault sample signals for each IMF component to obtain an effective feature set of the fault sample signals; input the effective feature set of the fault sample signals into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fused feature of the effective feature, and the quality of the fused feature of the effective feature is good; adjust the model parameters of the autoencoder model according to the comparison result between the decoded effective feature obtained by decoding based on the fused feature and the effective feature of the input fault sample signal. After iteratively executing the model training until the preset number of times, obtain a trained fused feature extraction model. The sensitive fault features extracted based on the fused feature extraction model are closer to the true mechanical fault features, and the accuracy of fault diagnosis based on the sensitive fault features is greatly improved, significantly improving the diagnostic accuracy.

[0065] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the written specification, claims, and drawings.

[0066] The technical solutions of the present invention will be further described in detail below through the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0068] Figure 1 is a flowchart of a method for training a mechanical equipment fault feature extraction model in Embodiment 1 of the present invention;

[0069] Figure 2 is a flowchart of the model training process of a method for training a mechanical equipment fault feature extraction model in Embodiment 2 of the present invention;

[0070] Figure 3 is a flowchart of the model testing process of a method for training a mechanical equipment fault feature extraction model in Embodiment 2 of the present invention;

[0071] Figure 4 is a flowchart of sensitive fault feature extraction in the embodiments of the present invention;

[0072] Figure 5 Schematic diagram of part of the structure of the autoencoder model in an embodiment of the present invention;

[0073] Figure 6 Diagram showing the precision rate of mechanical fault prediction and the recall rate of mechanical fault prediction in an embodiment of the present invention;

[0074] Figure 7a Normal time domain waveform diagram collected in an embodiment of the present invention;

[0075] Figure 7b Time domain waveform diagram of cavitation fault collected in an embodiment of the present invention;

[0076] Figure 7c Normal frequency spectrum diagram in an embodiment of the present invention;

[0077] Figure 7d Cavitation fault frequency spectrum diagram in an embodiment of the present invention;

[0078] Figure 8 Flowchart of a method for extracting mechanical equipment fault features in Embodiment 3 of the present invention;

[0079] Figure 9 Flowchart of a method for diagnosing mechanical equipment faults in Embodiment 4 of the present invention;

[0080] Figure 10 Schematic diagram of the structure of a training device for a mechanical equipment fault feature extraction model in an embodiment of the present invention;

[0081] Figure 11 Schematic diagram of the structure of a mechanical equipment fault feature extraction device in an embodiment of the present invention;

[0082] Figure 12 Schematic diagram of the structure of a mechanical equipment fault diagnosis device in an embodiment of the present invention;

[0083] Figure 13a Schematic diagram of the accuracy rate of fault prediction based on initial time-frequency features in an embodiment of the present invention;

[0084] Figure 13b Schematic diagram of the precision rate and recall rate of fault prediction based on initial time-frequency features in an embodiment of the present invention;

[0085] Figure 13c Schematic diagram of the accuracy rate of fault prediction based on fused features in an embodiment of the present invention;

[0086] Figure 13d Schematic diagram of the precision rate and recall rate of fault prediction based on fused features in an embodiment of the present invention;

[0087] Figure 14This is a comparison chart showing the accuracy of predicting faults using fusion features and other features in the embodiments of the present invention. Detailed implementation manners

[0088] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0089] To solve the problems existing in the prior art, embodiments of the present invention provide a method and device for extracting mechanical equipment fault features, training a model, and diagnosing faults.

[0090] Embodiment 1

[0091] Embodiment 1 of the present invention provides a method for training a mechanical equipment fault feature extraction model, and its process is as Figure 1 shown, including the following steps:

[0092] Step S101: Obtain a training set from mechanical equipment fault sample data, where the fault sample data includes mechanical equipment fault sample signals corresponding to different true fault types of mechanical equipment.

[0093] Obtain mechanical equipment fault sample data, collect mechanical equipment fault sample signals under different set fault types of the mechanical equipment to obtain mechanical equipment fault sample data; the fault types include at least one of looseness fault, imbalance fault, cavitation fault, and rubbing fault; select specified sample data from the obtained mechanical equipment fault sample data as the training set. The training set obtained in this way contains fault sample data under various possible fault types, and using the obtained training set for subsequent model training can make the training effect of the model better.

[0094] Step S102: Decompose the mechanical equipment fault sample signals in the training set to obtain a first specified number of IMF components; extract a second specified number of effective time-frequency features from the fault sample signals for each IMF component to obtain an effective feature set of the fault sample signals.

[0095] When decomposing the mechanical equipment fault sample signals in the training set, different decomposition methods can be selected, for example, the empirical mode decomposition method can be used to first decompose the IMF components and then make the selection. Optionally, the mechanical equipment fault sample signals are subjected to empirical mode decomposition, and a first specified number of IMF components are selected from the decomposed IMF components according to preset rules; for each selected IMF component, based on the predetermined effective features of the fault sample signal, a second specified number of effective features are selected from the time domain features and frequency domain features of the mechanical equipment fault sample signal to obtain an effective feature set for each IMF component, and based on the effective feature set of each IMF component, an effective feature set of the fault sample signal is obtained.

[0096] Optionally, when selecting effective features, the time domain features and frequency domain features of the mechanical equipment fault sample signal can be sorted first, and the top-ranked features can be selected based on the sorting results. For example, the compensated distance evaluation technology can be used to pre-select a second specified number of top-ranked features in the time domain features and frequency domain features of the fault sample signal as effective features of the fault sample signal.

[0097] Step S103: input the effective feature set of the fault sample signal into the pre-built autoencoder model for model training, and for each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fusion feature of the effective feature.

[0098] The autoencoder in the autoencoder model includes an encoding layer, a hidden layer, and a decoding layer. The effective feature set of the fault sample signal is input into the pre-built autoencoder model for model training, including: inputting the effective features in the effective feature set of the fault sample signal into the autoencoder, encoding them in the encoding layer to obtain effective encoding features; fusing the effective encoding features in the hidden layer to obtain fusion features of the effective features; decoding the fusion features in the decoding layer to obtain the decoded effective features. Multiple autoencoders can be set, each of which is responsible for processing the data of an effective feature, and multiple effective feature fusion features can be obtained through multiple autoencoders.

[0099] For example: the first effective feature of each IMF component in the effective feature set of the fault sample signal is input into the first autoencoder for training, and the first effective feature is fused and reduced in dimension by the first autoencoder to obtain the first fused feature; the second effective feature of each IMF component in the effective feature set of the fault sample signal is input into the second autoencoder for training, and the second effective feature is fused and reduced in dimension by the second autoencoder to obtain the second fused feature; and so on, the fused feature of each selected effective feature is obtained.

[0100] Step S104: Adjust the model parameters of the autoencoder model according to the comparison result between the decoded effective features obtained by decoding the fusion features and the effective features of the input fault sample signal. After iteratively performing model training until the preset number of times is reached, a trained fusion feature extraction model is obtained.

[0101] Compare the effective encoded features output by the encoding layer with the decoded effective features output by the decoding layer, and adjust the preset model parameters of the autoencoder model according to the matching degree between the effective encoded features output by the encoding layer and the decoded effective features output by the decoding layer; determine whether the number of iterative training times reaches the preset training times. If not, return to continue executing the step of inputting the effective feature set of the fault sample signal into the pre-built autoencoder model for model training. If so, a trained fusion feature extraction model is obtained.

[0102] In the above method of the embodiment of the present invention, mechanical equipment fault sample signals corresponding to different real fault types of mechanical equipment are obtained. The mechanical equipment fault signals obtained with less noise have high quality. Constructing a training set based on the obtained data can effectively improve the effect of model training; during the training process, the effective features are decomposed and fused, and the sensitive fault features extracted based on the fusion feature extraction model are closer to the real mechanical fault features, so that the features extracted by the trained model are more accurate. Based on the extracted sensitive fault features for fault diagnosis, the diagnostic accuracy is greatly improved, and the diagnostic accuracy rate is greatly improved.

[0103] In some optional embodiments, the above method further includes the process of testing the trained model using a test set, including:

[0104] Obtain a test set from the mechanical equipment fault sample data;

[0105] Decompose the mechanical equipment fault sample signals in the training set to obtain a first specified number of IMF components; extract a first specified number of effective time-frequency features from the fault sample signals for each IMF component to obtain an effective feature set of the fault sample signals;

[0106] Input the effective feature set of the fault sample signal into the pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fusion feature of the effective feature;

[0107] Determine the predicted fault type of the mechanical equipment according to the fusion feature of the effective feature, and determine the prediction accuracy rate, prediction precision rate and recall rate of the fusion feature extraction model according to the comparison result between the predicted fault type and the real fault type of the mechanical equipment included in the test set;

[0108] If the prediction precision rate and recall rate do not meet the preset requirements, continue to execute the steps of model training; if the prediction precision rate and recall rate meet the preset requirements, then obtain the trained fusion feature extraction model.

[0109] Embodiment 2

[0110] Embodiment 2 of the present invention provides a specific implementation process of a method for training a mechanical equipment fault feature extraction model, including a model training process and a model testing process, where the model training process is as Figure 2 shown, and the model testing process is as shown in 3. The principle block diagram of this method is shown in Figure 4 shown.

[0111] The model training process includes the following steps:

[0112] Step S201: Collect mechanical equipment fault sample signals of the mechanical equipment under different set fault types to obtain mechanical equipment fault sample data; the fault sample data includes mechanical equipment fault sample signals corresponding to different true fault types of the mechanical equipment, and the fault types include at least one of looseness fault, unbalance fault, cavitation fault, and rubbing fault.

[0113] Collect the mechanical equipment simulated fault vibration acceleration signals with an acceleration sensor, set the acceleration sensor on the outer shell of the mechanical equipment, set different fault conditions for the mechanical equipment, and collect a variety of different mechanical equipment fault signals, and the fault types include at least one of looseness fault, unbalance fault, cavitation fault, and rubbing fault.

[0114] Collect the fault vibration acceleration signals of the mechanical equipment by using a vibration acceleration sensor, and use a mechanical equipment that meets the preset conditions; use a data acquisition board that meets the preset conditions, for example, use a data acquisition board equivalent to the B&K data acquisition board; use an acceleration sensor measuring device that meets the preset conditions, for example, use an acceleration sensor measuring device equivalent to the PCB608A11 type acceleration sensor measuring device; set the mechanical equipment to a rotation speed that meets the preset conditions, for example, set the rotation speed to 2980 rpm, and the mechanical equipment can simulate preset mechanical faults, such as at least one of looseness fault, unbalance fault, cavitation fault, and rubbing fault; set the acceleration sensor on the outer shell of the mechanical equipment, specifically, it can be adsorbed on the outer shell of the mechanical equipment through a base, and set a sampling frequency that meets the conditions, for example, set the sampling frequency to 25600 Hz.

[0115] Among them, some of the collected signal waveforms and spectra are as Figure 7a shown as the collected normal time-domain waveform diagram, as Figure 7b shown as the collected cavitation fault time-domain waveform diagram, as Figure 7c shown as the collected normal spectrum, and as Figure 7d shown as the collected cavitation fault spectrum.

[0116] Step S202: Obtain a training set from mechanical equipment failure sample data.

[0117] Select a part of the sample data from the mechanical equipment failure sample data to form a training set. You can directly select part of the data from the mechanical equipment failure sample data to build a training set for model training, and then select part of the data to build a test set for model testing. You can also build the mechanical equipment failure sample data into a data set, and then divide a certain proportion of the data set as a training set and a certain proportion as a test set.

[0118] Step S203: performing empirical mode decomposition on the mechanical equipment fault sample signals in the training set, and selecting a first specified number of IMF components from the decomposed IMF components according to a preset rule.

[0119] like Figure 4 As shown, empirical mode decomposition is performed on the mechanical equipment fault sample signals of the training set to extract the first number of IMF components, for example, the first number of IMF components extracted are the first four IMF components; empirical mode decomposition is performed on each mechanical equipment fault sample signal of the training set, and the first three, first four, and first five IMF components are reconstructed in turn, and effective features of the three reconstructed signals are respectively extracted and input into the support vector machine for classification, and the first four IMFs are finally selected as the signals with the best denoising effect based on the classification results, and finally the first four IMF components of the fault sample signal are selected as the first specified number of IMF components.

[0120] Step S204: using the compensation distance evaluation technology, pre-selecting a second specified number of features that are ranked top in the time domain features and frequency domain features of the fault sample signal as effective features of the fault sample signal.

[0121] like Figure 4 As shown, the effective features of the fault sample signal are determined, and the time domain and frequency domain features of the fault sample signal are extracted. There are 18 feature indicators in total for the time domain and frequency domain features. Then, the top 10 feature indicators with the highest effectiveness among the 18 feature indicators are selected as effective features through the compensation distance evaluation technology.

[0122] For example, extract the time-domain features and frequency-domain features of each fault sample signal. The time-domain features and frequency-domain features are the maximum value, minimum value, average value, peak-to-peak value, rectified average value, peak value, variance, standard deviation, skewness, root mean square, waveform factor, peak factor, impulse factor, margin factor, center frequency, mean square frequency, and frequency variance. Then, through distance compensation technology, select the top ten time-domain features and frequency-domain features with the highest effectiveness as effective features. Finally, the effective features are determined to be the minimum value, rectified average value, peak value, variance, standard deviation, skewness, root mean square, center frequency, mean square frequency, and frequency variance, a total of 10 effective features, to obtain the effective features of the fault sample signal.

[0123] Step S205: For each selected IMF component, based on the pre-determined effective features of the fault sample signal, select a second specified number of effective features from the time-domain features and frequency-domain features of the mechanical equipment fault sample signal to obtain an effective feature set for each IMF component.

[0124] As Figure 4 shown, extract the effective feature sets of the four IMF components of each fault signal respectively, and extract the minimum value, rectified average value, peak value, variance, standard deviation, skewness, root mean square, center frequency, mean square frequency, and frequency variance of the four IMF components of each fault signal. The effective feature sets of the four IMF components are respectively denoted as effective feature set one (F1), effective feature set two (F2), effective feature set three (F3), and effective feature set four (F4) to obtain the effective feature set for each IMF component.

[0125] Step S206: Input the effective features in the effective feature set of the fault sample signal into the autoencoder, perform encoding in the encoding layer to obtain effective encoded features; perform feature fusion on the effective encoded features in the hidden layer to obtain the fused features of the effective features; perform decoding on the fused features in the decoding layer to obtain the decoded effective features.

[0126] Build an autoencoding model as Figure 5 shown. The autoencoder includes at least an encoder, a hidden layer, and a decoder; fuse the same type of features in the four effective feature sets, and save the trained autoencoder model. Set the hyperparameters for model training: the maximum training value (Epochs) is a preset value, for example, 150, the learning rate is a preset value, for example, 0.001, the model loss function is mean square error (MSE), and the number of model training rounds is a preset number, for example, 150 times. Input the effective features in the effective feature set of the fault sample signal into the autoencoder, perform encoding in the encoding layer to obtain effective encoded features; perform feature fusion on the effective encoded features in the hidden layer to obtain the fused features of the effective features; perform decoding on the fused features in the decoding layer to obtain the decoded effective features.

[0127] Specifically, the first effective feature of each IMF component in the effective feature set of the fault sample signal can be input into the first autoencoder for training. Through the first autoencoder, the first effective feature is fused and dimension-reduced to obtain the first fused feature. The second effective feature of each IMF component in the effective feature set of the fault sample signal is input into the second autoencoder for training. Through the second autoencoder, the second effective feature is fused and dimension-reduced to obtain the second fused feature, and so on, to obtain the fused features of each selected effective feature.

[0128] The input layer value of the autoencoder is a preset value, and the hidden layer value is a preset value. For example, the value of the input layer is 4, and the value of the hidden layer is 1. The autoencoder can compress and fuse the feature vector of length 4 into a feature vector of length 1 through the transfer function (tanh) function. The first feature in the four effective feature sets is used as the input of the first autoencoder. After the first autoencoder is trained, the encoder one is used for fusion and dimension reduction to obtain the fused feature. The second feature in the four effective feature sets is used as the input of the second autoencoder. After the second autoencoder is trained, the encoder two is used for fusion and dimension reduction to obtain the fused feature. Repeat the above operations to complete the dimension reduction and fusion of ten features in sequence, obtain the feature subset, and save the ten trained encoders. The dimension reduction and fusion of the ten features in the effective feature set are performed in sequence to obtain the fused features of the effective features.

[0129] Step S207: Compare the effective encoded features output by the encoding layer with the decoded effective features output by the decoding layer, and adjust the preset model parameters of the autoencoding model according to the matching degree between the effective encoded features output by the encoding layer and the decoded effective features output by the decoding layer.

[0130] According to the model loss function, compare the effective encoded features output by the encoding layer with the decoded effective features output by the decoding layer, calculate the mean square error (MSE) of the effective encoded features output by the encoding layer and the decoded effective features output by the decoding layer, and adjust the preset model parameters of the autoencoding model according to the mean square error (MSE).

[0131] Step S208: Determine whether the number of iterative training times reaches the preset training times. If not, return to continue executing the step of inputting the effective feature set of the fault sample signal into the pre-built autoencoding model for model training. If so, obtain the trained fused feature extraction model.

[0132] Judge the number of training iterations of the autoencoding model according to the preset training times of the autoencoding model. If the number of training iterations of the autoencoding model is less than the preset training times, continue to train the autoencoding model. If the number of training iterations of the autoencoding model is equal to the preset training times, obtain the trained fused feature extraction model.

[0133] The model testing process includes the following steps:

[0134] Step S211: Obtain a test set from mechanical equipment failure sample data.

[0135] A part of sample data is selected from the mechanical equipment failure sample data to form a training set.

[0136] Step S212: Performing empirical mode decomposition on the mechanical equipment fault sample signals of the test set, and selecting a first specified number of IMF components from the decomposed IMF components according to a preset rule.

[0137] like Figure 4 As shown, empirical mode decomposition is performed on the mechanical equipment failure sample signal of the test set to extract the first number of IMF components, for example, the first number of IMF components extracted are the first four IMF components; empirical mode decomposition is performed on each mechanical equipment failure sample signal of the test set, and the first three, first four, and first five IMF components are reconstructed in turn, and effective features of the three reconstructed signals are respectively extracted and input into the support vector machine for classification, and the first four IMFs are finally selected as the signals with the best denoising effect based on the classification results, and finally the first four IMF components of the mechanical equipment failure sample signal of the test set are selected as the first specified number of IMF components.

[0138] Step S213: using the compensation distance evaluation technology, pre-selecting a second specified number of features that are ranked top in the time domain features and frequency domain features of the fault sample signal as effective features of the fault sample signal.

[0139] Step S214: for each selected IMF component, based on the predetermined effective features of the fault sample signal, effective features are selected from the time domain features and frequency domain features of the mechanical equipment fault sample signal to obtain an effective feature set for each IMF component.

[0140] For each selected IMF component, based on the predetermined effective features of the fault sample signal, effective features are selected from the time domain features and frequency domain features of the mechanical equipment fault sample signal to obtain an effective feature set of each IMF component, and the effective features of the fault sample signal are obtained based on the effective feature set of each IMF component.

[0141] Step S215: input the effective feature set of the fault sample signal into the pre-built autoencoder model for model training, and for each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fusion feature of the effective feature;

[0142] Inputting the effective feature set of the fault sample signal into the above-trained fusion feature extraction model, fusing different IMF components of the effective features to obtain fusion features of the effective features;

[0143] Step S216: Determine the predicted fault type of the mechanical equipment according to the fusion feature of the effective features, and determine the prediction accuracy, prediction precision and recall rate of the fusion feature extraction model according to the comparison result between the predicted fault type and the true fault type of the mechanical equipment included in the test set.

[0144] Determine the predicted fault type of the mechanical equipment fault according to the fusion feature of the extracted effective features; and obtain the true fault type of the mechanical equipment included in the test set;

[0145] Determine the prediction accuracy of the fusion feature extraction model by using the following formula:

[0146]

[0147] Among them, the accuracy reflects the proportion of the samples predicted correctly by the model in the total samples. The total samples refer to the total number of samples in the test set, and all the samples predicted correctly refer to the number of samples in which the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set. See Figure 6 As shown, 5 fault types are exemplified. The number of samples of each fault type in the test set is 150, and the total number of samples of 5 categories is 750. Among them, the correctly predicted fault types are 146 + 123 + 150 + 137 + 150 = 706, and the prediction accuracy is 706 / 750 = 94.13%.

[0148] Use the following formula to determine the prediction precision of the fusion feature extraction model:

[0149]

[0150] The precision refers to the proportion of the number of samples predicted correctly in the samples of each fault type predicted by the model in the total number of samples of the fault type predicted by the model. Among them, the predicted positive class refers to the total number of samples of a fault type predicted by the model. Predicting the positive class as the positive class means that among the samples of the predicted fault type, the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set. For example Figure 6As shown, for example, for fault type 1, the number of positive classes predicted as positive is 146, the predicted positive classes are 146, the accuracy is 100%, and the false alarm rate is 0; for fault type 2, the number of positive classes predicted as positive is 123, the predicted positive classes are 123 + 13, the accuracy is 90.4%, and the false alarm rate is 9.6%; for fault type 3, the number of positive classes predicted as positive is 150, the predicted positive classes are 105 + 4, the accuracy is 97.4%, and the false alarm rate is 2.6%; for fault type 4, the number of positive classes predicted as positive is 137, the predicted positive classes are 137 + 27, the accuracy is 83.5%, and the false alarm rate is 16.5%; for fault type 5, the number of positive classes predicted as positive is 150, the predicted positive classes are 150, the accuracy is 100%, and the false alarm rate is 0.

[0151] Determine the recall rate of the fusion feature extraction model using the following formula:

[0152]

[0153] The recall rate refers to the proportion of the number of correctly predicted samples to the total number of samples of each fault type in the test set. Among them, the original positive class refers to the total number of samples of a fault type in the test set, and predicting the positive class as positive means the number of samples in which the predicted fault type of the mechanical equipment by the model is consistent with the true fault type of the mechanical equipment included in the test set among the samples of a fault type.

[0154] As Figure 6 shown, for example, for fault type 1, the number of positive classes predicted as positive is 146, the original positive class is 146 + 4, the recall rate is 97.3%, and the missed alarm rate is 2.7%; for fault type 2, the number of positive classes predicted as positive is 123, the original positive class is 123 + 27, the recall rate is 82%, and the missed alarm rate is 18%; for fault type 3, the number of positive classes predicted as positive is 150, the original positive class is 150, the recall rate is 100%, and the missed alarm rate is 0; for fault type 4, the number of positive classes predicted as positive is 137, the original positive class is 137 + 13, the recall rate is 91.3%, and the missed alarm rate is 8.7%; for fault type 5, the number of positive classes predicted as positive is 150, the original positive class is 150, the recall rate is 100%, and the missed alarm rate is 0.

[0155] Step S217: If the prediction precision rate and recall rate do not meet the preset requirements, continue to execute the steps of model training; if the prediction precision rate and recall rate meet the preset requirements, then obtain the trained fusion feature extraction model.

[0156] As Figure 6 shown, according to the prediction precision rate and recall rate, when the preset errors of the prediction precision rate and recall are not met, continue to execute steps S201 - S208 of model training.

[0157] The fusion feature extraction model extracts 10 sensitive features through the fusion of time-frequency analysis and autoencoding. To verify the effectiveness of the proposed sensitive features, a support vector machine is used as the diagnostic model. The traditional time-frequency features and sensitive features are respectively used as the inputs of the diagnostic model, and the fault recognition rates are compared. It is found that the fault recognition accuracy is as shown by Figure 13a with an accuracy of 91.07% for the traditional time-frequency features, and the accuracy of the fusion feature prediction is as shown by Figure 13c with an accuracy of 94.13%, an increase of 3.06%.

[0158] The precision rates of faults 1, 2, 3, 4, and 5 predicted based on the initial time-frequency features are as shown by Figure 13b 100%, 85%, 94.3%, 77.8%, and 99.3% respectively, and the recall rates are as shown by Figure 13b 100%, 75.3%, 99.3%, 86.7%, and 94.0% respectively;

[0159] The precision rates of faults 1, 2, 3, 4, and 5 predicted based on the fusion features are as shown by Figure 13d 100%, 90.4%, 97.4%, 83.5%, and 100% respectively, and the recall rates are as shown by Figure 13d 97.3%, 82.0%, 100%, 91.3%, and 100% respectively.

[0160] Among them, types 1 to 5 respectively represent 5 different fault states (1 - normal, 2 - looseness fault, 3 - cavitation fault, 4 - imbalance fault, 5 - rubbing fault).

[0161] According to the comparison results shown in Figure 13, which meet the preset error, a trained fusion feature extraction model is obtained, achieving a reduction in the interference of noise on the extraction of sensitive fault features in a strong noise working condition environment, effectively extracting fault sensitive features, and improving the fault recognition accuracy and the precision of fault diagnosis.

[0162] See Figure 14 shown, the fault accuracies predicted based on the initial time-frequency features, the fault accuracies predicted based on the effective feature types, and the fault accuracies predicted based on the fusion features are 91.07%, 91.87%, and 94.13% respectively; it can be seen from this that the fault accuracy predicted based on the fusion features is the highest and the effect is good.

[0163] The above method extracts the fault-sensitive features of rotating machinery under strong noise background. Signals of different fault types of mechanical equipment are collected using vibration acceleration sensors. First, the compensation distance evaluation technology is used to select effective features. Then, the signal is subjected to empirical mode decomposition, and the first four IMF components are taken out. The effective features of the four components are extracted in sequence to obtain an effective feature set. Then, the autoencoder is used to reduce the dimension and fuse the effective feature set to obtain a fault-sensitive feature set and an autoencoder fusion model. Similarly, the fault signal to be analyzed is subjected to empirical mode decomposition, the first four IMF components are taken, and the effective feature sets of the components are extracted in sequence. Finally, the autoencoder fusion model is used to reduce the dimension and fuse the effective feature set to complete the extraction of fault-sensitive features. Thus, the extraction of fault-sensitive features of rotating machinery under strong noise background is completed, which can effectively remove the interference of noise, select the sensitive feature set, and then improve the accuracy of fault diagnosis.

[0164] Embodiment III

[0165] Embodiment III of the present invention provides a method for extracting fault features of mechanical equipment, and its process is as Figure 8 shown, including the following steps:

[0166] Step S301: Obtain the fault signal of the mechanical equipment to be analyzed.

[0167] Step S302: Decompose the fault signal of the mechanical equipment to be analyzed to obtain the first specified number of IMF components. For each IMF component, extract the second specified number of effective time-frequency features from the fault signal of the mechanical equipment to be analyzed to obtain the effective feature set of the signal to be analyzed.

[0168] Perform empirical mode decomposition on the fault signal of the mechanical equipment to be analyzed, take the first four IMF components of each fault signal, and extract the effective feature sets of the first four components of each fault signal in sequence as the effective features of the signal to be analyzed.

[0169] Step S303: Input the effective feature set of the signal to be analyzed into the trained fusion feature extraction model for effective feature fusion, and output the fault-sensitive fusion features of the mechanical equipment.

[0170] Input the effective feature sets of the first four components of each fault signal into the fusion feature extraction model for effective feature fusion. During the fusion process, the ten features of the effective feature set are reduced in dimension and fused in sequence to obtain the final fault-sensitive fusion features of the signal to be analyzed. Among them, the fusion feature extraction model can be trained using the mechanical equipment fault feature extraction model training methods described in Embodiment I and II above.

[0171] Embodiment IV

[0172] Embodiment IV of the present invention provides a method for diagnosing faults of mechanical equipment, and its process is asFigure 9 As shown in the figure, it includes the following steps:

[0173] Step S401: Extract the fault-sensitive fusion features of the mechanical equipment.

[0174] The fault-sensitive fusion features of the mechanical equipment can be extracted by using the mechanical equipment fault feature extraction method described in Embodiment 3.

[0175] Step S402: Based on the extracted sensitive fusion features of the mechanical equipment, determine whether the mechanical equipment has a fault and the type of the fault.

[0176] Analyze according to the sensitive fusion features of the mechanical equipment. For example, compare the sensitive fusion features with the sensitive fusion features of type looseness faults, unbalance faults, cavitation faults, and rubbing faults. If the comparison result is within the preset range that meets the requirements, output the corresponding fault type; if the comparison result is not within the preset range that meets the requirements, it is considered that there is no fault.

[0177] Based on the same inventive concept, an embodiment of the present invention further provides a mechanical equipment fault feature extraction model training device. The structure of the device is as Figure 10 shown, including: a data acquisition module 111, a feature acquisition module 112, and a model training module 113;

[0178] The data acquisition module 111 is used to obtain a training set from the mechanical equipment fault sample data, and the fault sample data includes mechanical equipment fault sample signals corresponding to different real fault types of mechanical equipment.

[0179] The feature acquisition module 112 is used to decompose the mechanical equipment fault sample signal to obtain a first specified number of IMF components; extract a second specified number of effective time-frequency features from the fault sample signal for each IMF component to obtain an effective feature set of the fault sample signal.

[0180] The model training module 113 is used to input the effective feature set of the fault sample signal into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fusion feature of the effective feature; adjust the model parameters of the autoencoder model according to the comparison result between the decoded effective feature obtained by decoding according to the fusion feature and the effective feature of the input fault sample signal. After iteratively performing model training until a preset number of times, a trained fusion feature extraction model is obtained.

[0181] In a preferred embodiment, the mechanical equipment fault feature extraction model training device further includes a model testing module 114;

[0182] Optionally, the above data acquisition module 111 is further configured to obtain a test set from mechanical equipment failure sample data.

[0183] The feature acquisition module 112 is further configured to decompose the mechanical equipment failure sample signals in the training set to obtain a first specified number of IMF components; extract a first specified number of effective time-frequency features from the failure sample signals for each IMF component to obtain an effective feature set of the failure sample signals.

[0184] The model testing module 114 is configured to determine the predicted failure type of the mechanical equipment according to the fusion features of the effective features, and determine the prediction accuracy, prediction precision, and recall rate of the fusion feature extraction model according to the comparison result between the predicted failure type and the actual failure type of the mechanical equipment included in the test set; if the prediction precision and recall rate do not meet the preset requirements, continue to execute the steps of model training; if the prediction precision and recall rate meet the preset requirements, then obtain the trained fusion feature extraction model.

[0185] Based on the same inventive concept, an embodiment of the present invention further provides a mechanical equipment failure feature extraction device, and the structure of the device is as Figure 11 shown, including: a signal acquisition module 211, a signal analysis module 212, and a feature extraction module 213;

[0186] The signal acquisition module 211 is configured to acquire mechanical equipment failure signals to be analyzed.

[0187] The signal analysis module 212 is configured to decompose the mechanical equipment failure signals to be analyzed to obtain a first specified number of IMF components, and extract a second specified number of effective time-frequency features from the mechanical equipment failure signals to be analyzed for each IMF component to obtain an effective feature set of the signals to be analyzed.

[0188] The feature extraction module 213 is configured to input the effective feature set of the signals to be analyzed into the trained fusion feature extraction model for effective feature fusion, and output the failure-sensitive fusion features of the mechanical equipment; the fusion feature extraction model is trained by using the mechanical equipment failure feature extraction model training method described above.

[0189] Based on the same inventive concept, an embodiment of the present invention further provides a mechanical equipment failure diagnosis device, and the structure of the device is as Figure 12 shown, including: a mechanical equipment failure feature extraction device 311 and a mechanical failure diagnosis module 312;

[0190] The mechanical equipment failure feature extraction device described above is used to extract the failure-sensitive fusion features of the mechanical equipment.

[0191] A mechanical fault diagnosis module 312 is configured to determine whether there is a fault in the mechanical equipment and the type of the existing fault based on the extracted sensitive fusion features of the mechanical equipment.

[0192] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, at least one of the following is implemented: a method for extracting fault features of a mechanical equipment as shown above, a method for extracting fault features of a mechanical equipment as described above, and a method for diagnosing faults of a mechanical equipment as described above.

[0193] Based on the same inventive concept, an embodiment of the present invention further provides a signal processing device for mechanical equipment, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, at least one of the following is implemented: a method for extracting fault features of a mechanical equipment as shown above, a method for extracting fault features of a mechanical equipment as described above, and a method for diagnosing faults of a mechanical equipment as described above.

[0194] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0195] Unless otherwise specifically stated, terms such as processing, calculating, computing, determining, displaying, etc. may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, which operate on and transform data represented as physical (such as electronic) quantities in the registers or memories of the processing system into other data similarly represented as physical quantities in the memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals can be represented using any of a variety of different technologies and methods. For example, the data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description can be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[0196] It should be understood that the specific order or hierarchy of the steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of the steps in the process can be rearranged without departing from the scope of the present disclosure. The appended method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy.

[0197] In the foregoing detailed description, various features are combined in a single embodiment to simplify the present disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are expressly recited in each claim. On the contrary, as reflected in the appended claims, the invention lies in less than all of the features of a single disclosed embodiment. Accordingly, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0198] Those skilled in the art should also understand that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments herein can be implemented as electronic hardware, computer software, or combinations thereof. To clearly illustrate the interchangeability of hardware and software, the various illustrative components, blocks, modules, circuits, and steps have been generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in a flexible manner for each particular application, but such implementation decisions should not be interpreted as departing from the scope of the present disclosure.

[0199] The steps of a method or algorithm described in connection with the embodiments herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination thereof. The software modules may be located in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be integral to the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also exist as discrete components in a user terminal.

[0200] For a software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit may be implemented within the processor or outside the processor, and in the latter case, it is communicatively coupled to the processor by various means, which are well known in the art.

[0201] The foregoing description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but those of ordinary skill in the art should recognize that the various embodiments can be further combined and arranged. Accordingly, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. In addition, with respect to the term "comprising" as used in the specification or claims, the term is inclusive in a manner similar to the term "including", as is interpreted when "including" is used as a transitional word in a claim. Further, any use of the term "or" in the specification or claims is to be meant "non-exclusive or".

Claims

1. A method for training a fault feature extraction model of mechanical equipment, characterized in that, it includes: Obtain a training set from the mechanical equipment fault sample data, where the fault sample data includes mechanical equipment fault sample signals corresponding to different real fault types of mechanical equipment; Decompose the mechanical equipment fault sample signals in the training set to obtain a first specified number of IMF components; for each IMF component, extract a second specified number of effective time-frequency features from the fault sample signals to obtain an effective feature set of the fault sample signals; Input the effective feature set of the fault sample signals into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, fuse different IMF components of the effective feature to obtain a fused feature of the effective feature; Adjust the model parameters of the autoencoder model according to the comparison result between the decoded effective feature obtained by decoding based on the fused feature and the effective feature of the input fault sample signal. After iteratively performing model training for a preset number of times, obtain a trained fused feature extraction model.

2. The method according to claim 1, characterized in that, The process of obtaining the mechanical equipment fault sample data includes: Collect mechanical equipment fault sample signals of mechanical equipment under different set fault types to obtain mechanical equipment fault sample data; the fault types include at least one of looseness fault, imbalance fault, cavitation fault, and rubbing fault.

3. The method according to claim 1, characterized in that, Decompose the mechanical equipment fault sample signals to obtain a first specified number of IMF components; for each IMF component, extract a second specified number of effective time-frequency features from the fault sample signals to obtain an effective feature set of the fault sample signals, including: Perform empirical mode decomposition on the mechanical equipment fault sample signals, and select a first specified number of IMF components from the decomposed IMF components according to a preset rule; For each selected IMF component, select a second specified number of effective features from the time-domain features and frequency-domain features of the mechanical equipment fault sample signals based on the pre-determined effective features of the fault sample signals to obtain an effective feature set for each IMF component.

4. The method according to claim 1, characterized in that, It further includes: Use the compensation distance evaluation technique to pre-select the second specified number of features with higher rankings in the time-domain features and frequency-domain features of the fault sample signals as the effective features of the fault sample signals.

5. The method according to claim 1, characterized in that, The step of inputting the effective feature set of the fault sample signals into a pre-built autoencoder model for model training, and for each effective feature in the effective feature set, fusing different IMF components of the effective feature to obtain a fused feature of the effective feature, includes: Input the first effective feature of each IMF component in the effective feature set of the fault sample signals into the first autoencoder for training, and fuse and reduce the dimension of the first effective feature through the first autoencoder to obtain a first fused feature; Input the second valid feature of each IMF component in the valid feature set of the fault sample signal into the second autoencoder for training. Through the second autoencoder, fuse and reduce the dimension of the second valid feature to obtain the second fused feature; And so on, obtain the fused features of each selected valid feature.

6. The method according to claim 1, wherein, the autoencoder in the autoencoding model includes an encoding layer, a hidden layer, and a decoding layer; the step of inputting the valid feature set of the fault sample signal into the pre-built autoencoding model for model training includes: Input the valid features in the valid feature set of the fault sample signal into the autoencoder, and obtain valid encoded features through encoding in the encoding layer; Perform feature fusion on the valid encoded features in the hidden layer to obtain the fused features of the valid features; Decode the fused features in the decoding layer to obtain the decoded valid features.

7. The method according to claim 6, wherein, Adjust the model parameters of the autoencoding model according to the comparison result between the decoded valid features obtained by decoding the fused features and the valid features of the input fault sample signal. After iteratively executing the model training for a preset number of times, obtain the trained fused feature extraction model, including: Compare the valid encoded features output by the encoding layer with the decoded valid features output by the decoding layer, and adjust the preset model parameters of the autoencoding model according to the matching degree between the valid encoded features output by the encoding layer and the decoded valid features output by the decoding layer; Judge whether the number of iterative training times reaches the preset training times. If not, return to continue executing the step of inputting the valid feature set of the fault sample signal into the pre-built autoencoding model for model training. If so, obtain the trained fused feature extraction model.

8. The method according to claim 1, wherein, further includes: Obtain a test set from the mechanical equipment fault sample data; Decompose the mechanical equipment fault sample signal in the training set to obtain a first specified number of IMF components; extract a first specified number of valid time-frequency features from the fault sample signal for each IMF component to obtain the valid feature set of the fault sample signal; Input the valid feature set of the fault sample signal into the pre-built autoencoding model for model training. For each valid feature in the valid feature set, fuse different IMF components of the valid feature to obtain the fused feature of the valid feature; Determine the predicted fault type of the mechanical equipment according to the fused feature of the valid feature, and determine the prediction accuracy, prediction precision, and recall rate of the fused feature extraction model according to the comparison result between the predicted fault type and the actual fault type of the mechanical equipment included in the test set; If the prediction precision and recall rate do not meet the preset requirements, continue to execute the model training step; if the prediction precision and recall rate meet the preset requirements, then obtain the trained fused feature extraction model.

9. The method according to claim 8, wherein, Determine the predicted fault type of the mechanical equipment according to the fused features of the effective features, and determine the prediction accuracy, prediction precision and recall rate of the fused feature extraction model according to the comparison result between the predicted fault type and the true fault type of the mechanical equipment included in the test set, including: Determine the predicted fault type of the mechanical equipment fault according to the fused features of the extracted effective features; and obtain the true fault type of the mechanical equipment included in the test set; Use the following formula to determine the prediction accuracy of the fused feature extraction model: Among them, the total sample refers to the total number of samples in the test set, and all correctly predicted samples refer to the number of samples in which the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set; Use the following formula to determine the prediction precision of the fused feature extraction model: Among them, the predicted positive class refers to the total number of samples of a fault type predicted by the model. Predicting the positive class as the positive class means that among the samples of the predicted fault type, the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set; Use the following formula to determine the recall rate of the fused feature extraction model: Among them, the original positive class refers to the total number of samples of a fault type in the test set. Predicting the positive class as the positive class means that among the samples of a fault type, the predicted fault type of the mechanical equipment predicted by the model is consistent with the true fault type of the mechanical equipment included in the test set.

10. A method for extracting mechanical equipment fault features Characterized in that Including: Obtain the mechanical equipment fault signal to be analyzed; Decompose the mechanical equipment fault signal to be analyzed to obtain a first specified number of IMF components, and extract a second specified number of effective time-frequency features from the mechanical equipment fault signal to be analyzed for each IMF component to obtain an effective feature set of the signal to be analyzed; Input the effective feature set of the signal to be analyzed into the trained fused feature extraction model for effective feature fusion, and output the fault-sensitive fused features of the mechanical equipment; The fused feature extraction model is trained by using the mechanical equipment fault feature extraction model training method described in any one of claims 1-9.

11. A method for diagnosing mechanical equipment faults Characterized in that Including: Adopt the mechanical equipment fault feature extraction method described in claim 10 to extract the fault-sensitive fused features of the mechanical equipment; Based on the extracted sensitive fused features of the mechanical equipment, determine whether the mechanical equipment has faults and the types of faults.

12. A device for training a mechanical equipment fault feature extraction model, characterized in that Including: Data acquisition module: used to obtain a training set from mechanical equipment fault sample data, and the fault sample data includes mechanical equipment fault sample signals corresponding to different true fault types of mechanical equipment; Feature acquisition module: used to decompose the mechanical equipment fault sample signal to obtain a first specified number of IMF components; extract a second specified number of effective time-frequency features from the fault sample signal for each IMF component to obtain an effective feature set of the fault sample signal; Model training module: It is used to input the effective feature set of the fault sample signal into a pre-built autoencoder model for model training. For each effective feature in the effective feature set, different IMF components of the effective feature are fused to obtain the fused feature of the effective feature; the model parameters of the autoencoder model are adjusted according to the comparison result between the decoded effective feature obtained by decoding according to the fused feature and the effective feature of the input fault sample signal. After iteratively performing model training for a preset number of times, a trained fused feature extraction model is obtained.

13. The device according to claim 12, wherein, it further includes: Model testing module; Data acquisition module: It is further used to obtain a test set from the mechanical equipment fault sample data; Feature acquisition module: It is further used to decompose the mechanical equipment fault sample signal in the training set to obtain a first specified number of IMF components; for each IMF component, a first specified number of effective time-frequency features are extracted from the fault sample signal to obtain the effective feature set of the fault sample signal; Model testing module: It is used to determine the predicted fault type of the mechanical equipment according to the fused feature of the effective feature, and determine the prediction accuracy, prediction precision and recall rate of the fused feature extraction model according to the comparison result between the predicted fault type and the actual fault type of the mechanical equipment included in the test set; if the prediction precision and recall rate do not meet the preset requirements, continue to execute the steps of model training; if the prediction precision and recall rate meet the preset requirements, then obtain a trained fused feature extraction model.

14. A mechanical equipment fault feature extraction device, wherein, it includes: Signal acquisition module: It is used to acquire the mechanical equipment fault signal to be analyzed; Signal analysis module: It is used to decompose the mechanical equipment fault signal to be analyzed to obtain a first specified number of IMF components, and for each IMF component, a second specified number of effective time-frequency features are extracted from the mechanical equipment fault signal to be analyzed to obtain the effective feature set of the signal to be analyzed; Feature extraction module: It is used to input the effective feature set of the signal to be analyzed into the trained fused feature extraction model for effective feature fusion, and output the fault-sensitive fused feature of the mechanical equipment; the fused feature extraction model is trained by using the mechanical equipment fault feature extraction model training method as described above.

15. A mechanical equipment fault diagnosis device, wherein, it includes: The mechanical equipment fault feature extraction device according to claim 14, which is used to extract the fault-sensitive fused feature of the mechanical equipment; Mechanical fault diagnosis module: It is used to determine whether there is a fault in the mechanical equipment and the type of the existing fault based on the extracted sensitive fused feature of the mechanical equipment.

16. A computer storage medium, wherein, The computer storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, at least one of the methods for extracting mechanical equipment fault features described in any one of claims 1-9, the method for extracting mechanical equipment fault features described in claim 10, and the method for diagnosing mechanical equipment faults described in claim 11 is implemented.

17. A signal processing device for mechanical equipment, characterized in that it includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, at least one of the methods for extracting mechanical equipment fault features described in any one of claims 1-9, the method for extracting mechanical equipment fault features described in claim 10, and the method for diagnosing mechanical equipment faults described in claim 11 is implemented.

Images