CN116776284A - Fault diagnosis method for electromechanical device, computer device, and storage medium - Google Patents
Fault diagnosis method for electromechanical device, computer device, and storage medium Download PDFInfo
- Publication number
- CN116776284A CN116776284A CN202310781186.8A CN202310781186A CN116776284A CN 116776284 A CN116776284 A CN 116776284A CN 202310781186 A CN202310781186 A CN 202310781186A CN 116776284 A CN116776284 A CN 116776284A
- Authority
- CN
- China
- Prior art keywords
- fault diagnosis
- data
- dimensional feature
- feature matrix
- encoder
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 87
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000003860 storage Methods 0.000 title claims description 8
- 239000011159 matrix material Substances 0.000 claims abstract description 67
- 238000012549 training Methods 0.000 claims abstract description 28
- 230000004927 fusion Effects 0.000 claims abstract description 21
- 238000005516 engineering process Methods 0.000 claims abstract description 6
- 230000008569 process Effects 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 5
- 230000015654 memory Effects 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 230000009977 dual effect Effects 0.000 claims description 4
- 230000002547 anomalous effect Effects 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 4
- 239000000284 extract Substances 0.000 abstract description 4
- 238000010801 machine learning Methods 0.000 abstract description 4
- 230000003595 spectral effect Effects 0.000 description 9
- 230000036541 health Effects 0.000 description 6
- 238000010606 normalization Methods 0.000 description 6
- 238000012795 verification Methods 0.000 description 6
- 238000010276 construction Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000013139 quantization Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 230000003321 amplification Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 239000004816 latex Substances 0.000 description 1
- 229920000126 latex Polymers 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Landscapes
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The application relates to the technical field of detection of electromechanical equipment, in particular to a fault diagnosis method for the electromechanical equipment. The method comprises the steps of obtaining multi-source data; obtaining fusion data by utilizing a fusion technology; constructing a three-dimensional feature matrix based on the fusion data; inputting the three-dimensional feature matrix into a fault diagnosis model for training to obtain a trained fault diagnosis model; and inputting the three-dimensional feature matrix into a trained fault diagnosis model to obtain a diagnosis result of the electromechanical equipment. The three-dimensional feature matrix of the application can dynamically and simultaneously describe and extract various characteristics of all signals and integrate the characteristics into a three-dimensional structure, so that the time sequence relation of the input signals can be modeled by only one diagnosis model. Compared with the traditional characteristic data, the three-dimensional characteristic matrix is used as input, so that not only can a plurality of signals and characteristics be mined and learned at the same time, but also the training cost of a subsequent machine learning model can be effectively reduced.
Description
Technical Field
The present application relates to the field of electromechanical device detection technologies, and in particular, to a fault diagnosis method for an electromechanical device, a computer device, and a storage medium.
Background
Along with the continuous improvement of the intelligentization, the complexity and the maximization degree of the electromechanical equipment, the damage of the electromechanical equipment is larger and larger, the difficulty of equipment maintenance is also larger and larger, any fault or failure occurring in the operation process can cause serious consequences, timely and proper equipment maintenance is an important means for ensuring the stability, the reliability and the availability of an equipment system, and the accurate judgment of the current health state of the equipment has important significance for equipment maintenance.
The rise of the neural network model provides further effective guarantee for intelligent monitoring of the electromechanical equipment. In conventional applications, the health of an electromechanical device is generally determined based on a single signal or a single characteristic. However, electromechanical systems often cannot be characterized by a single signal or feature. But evaluates the health of the electromechanical device through various signals and the connection and association between the signals.
Disclosure of Invention
The application aims to provide a fault diagnosis method for electromechanical equipment, computer equipment and a storage medium.
In order to achieve the above object, the present application provides the following technical solutions:
the first aspect of the present disclosure is a fault diagnosis method for an electromechanical device, including:
acquiring multi-source data of a multi-source signal;
time synchronization and alignment of data sequences of different signals in the multi-source data are carried out by utilizing a fusion technology, so that fusion data are obtained;
creating three-dimensional feature data based on the fusion data, and constructing a three-dimensional feature matrix;
inputting the three-dimensional feature matrix into a fault diagnosis model for training to obtain a trained fault diagnosis model;
and inputting the three-dimensional feature matrix into a trained fault diagnosis model to obtain a diagnosis result of the electromechanical equipment.
In some embodiments, the constructing three-dimensional feature data based on the fused data, constructing a three-dimensional feature matrix, comprises:
dividing each signal data in the fusion data according to a preset window to obtain a window data set;
and extracting key features of each data in the window data set by utilizing the multi-domain features, creating three-dimensional feature data, and constructing a three-dimensional feature matrix.
Further, the multi-domain features include at least one of a time domain, a frequency domain, and a statistical domain, and any one of the domains includes at least one feature.
In some embodiments, the training the three-dimensional feature matrix into the fault diagnosis model to obtain a trained fault diagnosis model includes evaluating a training result, and if the evaluation result meets an expectation, obtaining the trained fault diagnosis model; if the evaluation result does not meet the expectations, the fault diagnosis model is continuously trained.
In some embodiments, the time synchronization and alignment of the data sequences of different signals in the multi-source data using a fusion technique to obtain fused data includes:
preprocessing the fusion data by utilizing a preprocessing means to obtain preprocessed fusion data;
the preprocessing means comprises at least one processing means of filtering, amplifying, sampling and quantizing, normalizing, trending, window function and noise reduction.
In some embodiments, the fault diagnosis model is a ConvLSTM based neural network comprising:
an encoder and a decoder, wherein the encoder comprises a ConvLSTM network, a Pooling layer and an FC layer; the decoder comprises a PC layer, an upsping layer and a ConvLSTM network; the three-dimensional feature matrix is input into a first encoder and then is input into a decoder;
wherein,,
and calculating to obtain reconstruction loss by using the input three-dimensional feature matrix and the data output by the decoder, and evaluating whether the input data is abnormal or not through the reconstruction loss.
In some embodiments, the fault diagnosis model is a ConvLSTM based dual encoder neural network comprising:
the first encoder and the second encoder comprise ConvLSTM units, a Pooling layer and an FC layer; the decoder comprises a PC layer, an Upsamping layer and ConvLSTM units; the three-dimensional feature matrix is input into a first encoder and is input into a second encoder after passing through a decoder;
wherein,,
calculating by using the input three-dimensional feature matrix and the data output by the decoder to obtain reconstruction loss; the potential loss is obtained after calculation by utilizing the data output by the first encoder and the data output by the second encoder; and carrying out weighted summation on the reconstruction loss and the potential loss to obtain a final loss function-anomaly score, and evaluating whether the input data is anomalous or not through the anomaly score.
In some embodiments, the multi-source signal comprises multiple signal sources or/and multiple signal sources of the same signal.
A second aspect of the present disclosure provides a computer device comprising:
a memory for storing a computer program;
a processor for implementing the steps of the fault diagnosis method for an electromechanical device as described above when executing the computer program.
A third aspect of the present disclosure is a readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the fault diagnosis method for an electromechanical device as described above.
Compared with the prior art, the application has the advantages that:
firstly, the traditional technology generally adopts a single signal or a single characteristic, and the sampled data is limited; or, the traditional characteristic data lack of correlation consideration between different signal data and characteristic variables in construction; furthermore, it is often necessary to build a diagnostic model separately for multi-source signals or multi-dimensional features. The three-dimensional feature matrix can dynamically and simultaneously describe and extract various characteristics of all signals and integrate the characteristics into a three-dimensional structure, so that the time sequence relation of the input signals can be modeled by only one diagnosis model, and meanwhile, the spatial characteristics (between signals and between feature variables) of the input signals can be related. Compared with the traditional characteristic data, the three-dimensional characteristic matrix is used as input, so that not only can a plurality of signals and characteristics be mined and learned at the same time, but also the training cost of a subsequent machine learning model can be effectively reduced.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. Wherein:
FIG. 1 is a flow chart of a fault diagnosis method according to an embodiment of the present application;
FIG. 2 is a schematic diagram of three-dimensional feature matrix construction according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a fault diagnosis model according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a fault diagnosis model training in accordance with an embodiment of the present application;
FIG. 5 is a schematic diagram of a fault diagnosis system according to an embodiment of the present application.
Detailed Description
The application will be described in detail below with reference to the drawings in connection with embodiments. The examples are provided by way of explanation of the application and not limitation of the application. Indeed, it will be apparent to those skilled in the art that modifications and variations can be made in the present application without departing from the scope or spirit of the application. For example, features illustrated or described as part of one embodiment can be used on another embodiment to yield still a further embodiment. Accordingly, it is intended that the present application encompass such modifications and variations as fall within the scope of the appended claims and their equivalents.
In the description of the present application, the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", etc. refer to the orientation or positional relationship based on that shown in the drawings, merely for convenience of description of the present application and do not require that the present application must be constructed and operated in a specific orientation, and thus should not be construed as limiting the present application. The terms "coupled," "connected," and "configured" as used herein are to be construed broadly and may be, for example, fixedly connected or detachably connected; can be directly connected or indirectly connected through an intermediate component; either a wired electrical connection, a radio connection or a wireless communication signal connection, the specific meaning of which terms will be understood by those of ordinary skill in the art as the case may be.
One or more examples of the application are illustrated in the accompanying drawings. The detailed description uses numerical and letter designations to refer to features in the drawings. Like or similar designations in the drawings and description have been used to refer to like or similar parts of the application. As used herein, the terms "first," "second," and "third," etc. are used interchangeably to distinguish one component from another and are not intended to represent the location or importance of the individual components.
The application provides a fault diagnosis method for electromechanical equipment, which aims to acquire operation data of the electromechanical equipment in the operation process in real time and evaluate the health state of the electromechanical equipment. During operation of the electromechanical device, the device may fail due to anomalies in the electromechanical circuit, or anomalies between moving parts and components, or occasional anomalies in parts. In the running process of the electromechanical equipment, the health state of the equipment can be reflected in various state information. Therefore, it is necessary to monitor the status information of the electromechanical device itself and the operating environment, diagnose the information, and determine the health status of the device.
In conventional diagnostic methods, each signal and its respective features are often modeled independently, and this process ignores the inherent relationships that exist between signals and between features, which can lead to some potential device anomalies being unrecognized. Because for independent modeling, the numerical variation of individual signals or features does not accurately account for the problem. Relationships between other results need to be comprehensively considered to obtain comprehensive conclusions. However, it is difficult to achieve such comprehensive conclusions for conventional techniques. The application aims to construct a fault diagnosis model to diagnose multi-signal and multi-characteristic data of the electromechanical equipment, and simultaneously, the space characteristics (signals and characteristic variables) of the electromechanical equipment can be related, so that the diagnosis accuracy is improved.
Referring to fig. 1 and fig. 2 for a specific implementation process, fig. 1 is a flowchart of a fault diagnosis method for an electromechanical device according to an embodiment of the present application, and fig. 2 is a schematic diagram of three-dimensional feature matrix construction according to an embodiment of the present application, where the fault diagnosis method includes the following steps:
s1, acquiring multi-source data and fusing the multi-source data to obtain fused data.
In this embodiment, the data monitored by the method is multi-source data, i.e., the collected data is from multiple data sources. The source data includes a variety of signal sources including, but not limited to, current, voltage, torque of the motor; voltage and current of key nodes or devices; vibration, moving speed, inclination angle and acceleration of the electromechanical device; one or more of noise, temperature, humidity, air pressure, etc. of the operating environment of the electromechanical device. In addition, the source data also includes the number of any signal source, i.e. one or more signal sources can be set according to the actual situation. In this embodiment, the source data are acquired by corresponding sensors. The monitoring of a plurality of physical quantities is constructed, and a plurality of characteristics are extracted, so that the more comprehensive running state of the system can be reflected to the greatest extent. In fault diagnosis, different fault types may have different performances on different characteristics, and by comprehensively considering the characteristics, the detection rate and the recognition rate of faults can be improved.
The fusion here refers to time synchronization and alignment of the data sequences of different signals, that is, alignment of the collected signals of different frequencies, so as to ensure that the lengths (sampling frequencies) of all signals of the multi-source data are consistent. Thus, the length of each w is also consistent in the feature extraction stage below.
For example, in the present embodiment, a specific method of time synchronization and alignment is to select the signal of the lowest sampling frequency as a reference for matching. Assuming that sensor a (sampling frequency is 5000 Hz), then when a completes one sample, for sensor B (sampling frequency is 10000 Hz), the sample nearest to that moment is found, thus completing time synchronization and alignment of a and B data.
S2, constructing a three-dimensional feature matrix based on the fusion data.
The construction of a three-dimensional feature matrix is a focus of distinction from conventional techniques. The conventional feature data lacks consideration of relevance between different signal data and feature variables in construction, and in addition, a diagnosis model is often required to be built for multi-source signals or multi-dimensional features. The three-dimensional feature matrix can dynamically and simultaneously describe and extract various characteristics of all signals and integrate the characteristics into a three-dimensional structure, so that the time sequence relation of input signals can be modeled by only one diagnosis model, and the spatial characteristics (between signals and between feature variables) can be described. Compared with the traditional characteristic data, the three-dimensional characteristic matrix is used as input, so that not only can a plurality of signals and characteristics be mined and learned at the same time, but also the training cost of a subsequent machine learning model can be effectively reduced.
The fault diagnosis method comprises the following steps:
s201, dividing each data in the multi-source data processed in the S1 according to a preset window size to obtain a window data set. The purpose of segmentation is to reduce the dimensionality of the data, thereby improving the operational efficiency of the subsequent diagnostic model.
Here, it can be understood that, in the present application, a plurality of data sources contained in the multi-source data constitute a first dimension of the three-dimensional feature matrix, i.e., N in fig. 2; the segmented data forms a windowed dataset that forms the second dimension of the three-dimensional feature matrix, L in fig. 2.
To this end, two-dimensional data in the three-dimensional feature matrix has been formed.
S202, the application extracts key features from the multisource data processed in the step S201 by utilizing the multiservice features. The method aims at further reducing the dimension on the premise of keeping key characteristics, compressing the data size, facilitating data transmission and reducing the calculation amount in the later period. The multi-domain features include time domain features, frequency domain features, and statistical domain features. Wherein:
(1) Time domain features: the time domain features are extracted directly from the time series of the signal. They reflect the distribution and variation characteristics of the signal on the time axis. Common time domain features include rate of change, peak-to-peak distance, autocorrelation, zero crossing rate, signal energy, etc. They can reflect the basic characteristics of signal amplitude, fluctuation, asymmetry, sharpness, etc.
(2) Frequency domain characteristics: the frequency domain features are extracted from the spectrum of the signal. They reflect the distribution and variation characteristics of the signal on the frequency axis. Common frequency domain features include power spectral density, spectral peak, spectral center, spectral width, spectral tilt, spectral kurtosis, and the like. They can reflect the fundamental characteristics of the signal such as frequency content, frequency distribution, frequency ripple, frequency asymmetry, frequency sharpness, etc.
(3) Statistical domain features: statistical domain features are features that describe signals based on statistical principles, which may be statistical descriptions of time domain or frequency domain features. Common statistical domain features include mean, variance, standard deviation, skewness, kurtosis, correlation coefficient, covariance, etc. They can reflect statistical properties of the signal such as homogeneity, waviness, asymmetry, sharpness, correlation, etc.
The characteristics of these three domains describe the characteristics of the signal from different angles. In practical applications, it is generally necessary to select appropriate features for extracting key features after evaluating the computational complexity and correlation of each feature according to specific conditions and signal characteristics. Here, the computation complexity mainly considers the required computation time and the consumed computation power, and the correlation between features mainly considers the redundancy of the features. The features of each domain will screen out a number of relatively suitable features, the features of the three domains will be combined into a feature set, and each data of the source data will be calculated with each feature in the feature set.
The multi-domain feature in the present application includes at least one of the three domains, and each selected domain includes at least one feature.
After the key features of the multi-domain features are extracted from each data in the window dataset, a third dimension of the three-dimensional feature matrix, namely M in fig. 2, is formed.
It should be emphasized here that the purpose of constructing the three-dimensional feature matrix is to:
(1) The three-dimensional matrix form can more effectively capture the sample timing correlation and the correlation between features/signals. As shown in fig. 3, the three dimensions of the feature matrix are respectively: the number N of signal sources, the length L of the features and the number M of the features. Here we can understand the feature matrix as "image", the feature vector of the i-th signal at j time periods, i.e. the pixels of the image at j rows and i columns. This form is easier to handle in subsequent convolutional timing networks, the convolutional operation will learn the relationships between signal sources and between features, and long and short term memory networks capture mainly the relevant features in timing.
(2) The running portraits of the device over a period of time can be summarized from multiple dimensions. Traditionally, when judging the health of an electromechanical device, we generally rely on a single signal or a single characteristic, such as a common current signal and its spectral characteristics. However, electromechanical systems often cannot be characterized by a single signal or feature, and constructing a plurality of physical quantity monitors and extracting a plurality of features can reflect the more comprehensive operating state of the system to a maximum extent. In fault diagnosis, different fault types may have different performances on different characteristics, and by comprehensively considering the characteristics, the detection rate and the recognition rate of faults can be improved. In addition, the uncertainty and noise of each signal and feature are considered, and the sensitivity to a certain signal or feature loss and noise can be reduced by constructing the three-dimensional matrix of the feature, so that the robustness of the fault diagnosis model can be improved.
(3) Conventionally, for multi-source or multi-dimensional features, a corresponding fault diagnosis model, whether expert system or machine learning model, needs to be built for each signal or each feature. This operation not only ignores the correlation characteristics between features and in time sequence, but also brings about a huge calculation cost. The three-dimensional matrix form provided by the application is used as the direct input of the fault diagnosis model, and only one model is required to be established, so that a plurality of signals and features can be simultaneously excavated and learned, and the abnormal feature matrix can be rapidly and effectively identified.
S3, inputting the three-dimensional feature matrix into a fault diagnosis model for training, evaluating a training result, and stopping training and obtaining a trained fault diagnosis model if the result meets the expectation; otherwise, continuing to train the fault diagnosis model.
The fault diagnosis model adopts a neural network capable of processing a three-dimensional data matrix. Therefore, the architecture of the failure diagnosis model is first designed. Referring to fig. 3, in the present embodiment, the fault diagnosis model is a ConvLSTM-based dual encoder fault diagnosis model, i.e., a ConvLSTM-EDE (convolutional sequential dual encoder) architecture using anomaly feature matrix recognition. The architecture employs two encoders and one decoder. Wherein each encoder is provided with a first ConvLSTM network, a Pooling layer and an FC layer (Fully Connected Layer fully connected layer). There are two implementations of the Pooling layer, maxPooling (max Pooling layer) and MeanPooling (average Pooling layer), which are implemented in this embodiment. The decoder comprises an FC layer, an Upsamping layer matched with the Pooling layer and a second ConvLSTM network with the same structure as the first ConvLSTM network. Wherein the first ConvLSTM network at least comprises a ConvLSTM layer, any ConvLSTM layer comprises a ConvLSTM unit or a group of ConvLSTM units which are connected in series, the operation equation of each ConvLSTM unit is as follows,
more specifically, first, the first encoder processes the input feature matrix in time order with the ConvLSTM network, thus preserving the temporal and spatial dynamics of the input feature matrix. To achieve better feature extraction performance, multiple LSTM layers may be included in the ConvLSTM network. The characteristics coded by ConvLSTM network can pass through a MaxPooling layer to compress the size of the characteristic matrix, so that the number of parameters in the network is indirectly reduced, and the model overfitting is prevented. Finally, the FC layer is placed to further compress the size of the feature matrix and map the feature matrix into one-dimensional space. The decoder aims to reconstruct the input feature matrix from the encoded feature representation, the process of which is the inverse of the encoder. The decoder is obtained by connecting the FC layer, the Upsamping layer and the ConvLSTM unit in series. The Upsamping layer amplifies the compressed feature matrix size and finally attempts to restore the original input feature matrix through the ConvLSTM layer. By calculating the root mean square error of the reconstructed and input feature matrices, the reconstruction Loss (reconstruction Loss) can be deduced. The reconstructed feature matrix will be further input into a second encoder having the same configuration as the first encoder. The function of the second encoder is to re-encode the output of the decoder to obtain a feature representation of the reconstructed feature matrix. For an effectively trained self-encoder, the output of this encoder should be similar to the output of the first encoder. Thus, we introduce a potential Loss (latex Loss) to measure the difference in the two characteristic representations. Weighted summation of reconstructed and potential losses yields the final loss function-anomaly Score (extraction Score), i.e
S e ＝α 1 *ConstructLoss+α 2 *LatentLoss。
The larger anomaly score indicates that the corresponding feature matrix is more likely to have anomalies, and the feature sequences can be grouped by setting different anomaly score levels so as to distinguish different anomaly feature patterns.
After obtaining the anomaly score, we need to set a threshold value of anomaly score to classify the normal sample and the anomaly sample.
After the architecture of the fault diagnosis model is designed, various network parameters of the model, namely, determination of model super parameters (hyperparameters), need to be preset. The unit structure parameters in each encoder and decoder, including the number of layers of ConvLSTM, the number and size of filters, the number of layers of FC, neurons, etc., are specifically required to be set. In addition, parameters related to network training are required to be set, and the parameters mainly comprise an optimizer, a learning rate, a batch size, a maximum iteration round number and the like.
Of course, the architecture of the fault diagnosis model constructed by the one encoder and the decoder may be selected. However, the architecture contains only one reconstruction penalty. The reconstruction loss reflects the learning performance of the first encoder and decoder for normal samples. In general, a larger reconstruction error indicates a greater likelihood of an input sample anomaly.
S4, inputting the three-dimensional feature matrix into a trained fault diagnosis model to obtain a diagnosis result of the electromechanical equipment.
Of course, before formal application, the trained fault diagnosis model can be tested, and the reliability of the model can be judged through further evaluation of testers. If the assessment meets the expectations, the assessment will be formally applied online.
In addition, in the above S1, because the environment where the data sources are located is complex, there are many noises in the data collected by some data sources. Therefore, the obtained multi-source data needs to be preprocessed, which aims to remove noise in the data and improve the quality of the data, so that the data can be analyzed later. Here, some common processing means are listed, including but not limited to the following:
(1) And (3) filtering: by designing the appropriate filters, we can boost certain frequency signal components and attenuate others. For example, a low pass filter allows low frequency signals to pass while blocking high frequency signals; the high pass filter is the opposite. The band pass filter only allows signals of a certain frequency range to pass, while the band reject filter blocks signals of a certain frequency range.
(2) Amplifying: amplification is achieved by increasing the amplitude of the signal so that it maintains good signal quality when sampled and quantized. The amplifier may typically provide a fixed gain or may provide an adjustable gain to accommodate different signal conditions.
(3) Sampling and quantization: sampling is the process of converting a continuous-time analog signal into a discrete-time digital signal, with the sampling rate determining the time resolution of the signal. Quantization is the process of converting an analog signal of continuous amplitude into a digital signal of discrete amplitude, the accuracy of which determines the amplitude resolution of the signal. Sampling and quantization is typically performed by an analog-to-digital converter (ADC).
(4) Normalization: normalization is the process of adjusting the amplitude of a signal to a standard range. Normalization generally facilitates signal processing and analysis, while also reducing problems in numerical calculations. Common normalization methods include max-min normalization, Z-Score normalization, and the like.
(5) Trending: detrending is the process of removing low frequency or long-term varying components from a signal. Detrending may help us to better focus on the rapid changes and dynamic behavior of the signal. Trending can be generally achieved by high-pass filtering or differential methods.
(6) Window function: in performing frequency domain analysis, the signal typically needs to be cut into small pieces of a segment. To reduce spectral leakage due to cutting and to improve spectral resolution, we will typically apply a window function to each segment. The window function may smoothly approach zero for the segment at both ends, thereby reducing the cut-induced abrupt changes. Common window functions include rectangular windows (no windows), hanning windows, hamming windows, and blackman windows, among others.
(7) Noise reduction: noise reduction is the process of removing noise components from a signal. In the present application, we use Kalman filtering to denoise and smooth the multi-source signal.
The application can preprocess the fused multi-source data by at least one processing means.
More specifically:
for example, the monitoring object of the embodiment is an elevator, and five signals are collected, including three current of a traction motor, vibration of a machine body and noise signals. Specific sampling criteria are as follows:
| original signal | Sampling frequency (Hz) | Sampling time (Tian) |
| U-phase current | 5000 | 183 |
| V-phase current | 5000 | 183 |
| W-phase current | 5000 | 183 |
| Vibration of the fuselage | 5000 | 183 |
| Noise signal | 5000 | 183 |
Firstly, various collected original signals are downsampled, and time synchronization and alignment of different signal data of multi-source data are performed.
Then, preprocessing the fused multi-source data, and selecting corresponding processing means for further processing. For example, filtering, amplification, normalization, and noise reduction are selected in the present embodiment.
Next, the preprocessed multi-source data needs to be divided according to a set window size. For example, the nth signal of original length TSplit into several consecutive sub-sequences of length S:wherein the size of the window can be set according to the actual situation. The original signal can be represented as a set of L sub-sequences, namely:
where l=t/S.
Then, a key feature extraction stage of the multi-domain features is entered. In this embodiment, three domains, namely, a time domain, a frequency domain and a statistical domain, are selected at this stage. Depending on the application, a suitable feature calculation method is selected in the three domains, for example, the following 18 features are selected in this embodiment:
for any sequence in any signalCalculating the +.o. of the characteristic vector for any sequence in any signal according to the formula in the table>Calculating the eigenvector +.> Wherein M is the total number of characteristic values, +.>Is the mth eigenvalue of the nth subsequence in the nth signal. Calculating all characteristic values of the N signals and the L subsequences to obtain a three-dimensional characteristic matrix F= [ F ] 1 ,F 2 ,…,F N ]。
For example, in the present embodimentIn the example, the number of signal sources of the multi-source data is 5 (n=5);the length of L after the sampling data is divided is 1200, namely, the feature matrix extracted every 1 minute is taken as an input sample of the model; the number of multi-domain features is 18 (m=18). Therefore, the three-dimensional feature matrix F is a matrix of (5×1200×18).
The obtained three-dimensional feature matrix F is input into the fault diagnosis model for training, wherein the fault diagnosis model adopts a double-encoder neural network architecture based on ConVLSTM, and the super parameters of the fault diagnosis model are set as shown in the following table.
Referring to fig. 4, in order to ensure the generalization performance of the model, cross validation is used for model training and verification. In this example, the data was sampled for 6 consecutive months (183 days) for a total of six exercises. One month of the data in 6 months is selected as a verification sample for each training, and the rest data are training samples. For example, for the first training, data from month 6 is a validation sample and data from 1 to 5 is a training sample; second training, month 5 data are validation samples, month 1-4 and month 6 data are training samples, and so on.
The model was evaluated using the F1 value (F1-Score). The F1 value is a harmonic average value of the Precision and Recall rates (Recall), and can comprehensively consider the misjudgment and missed judgment conditions of the model. The actual label of the sample is derived from the actual elevator service record and the manual confirmation of the technician. Model verification whether "up to standard" we will refer to the average F1 value over the course of 6 training, if F1 (avg) < 0.9 we will micro-scale model parametersTuning, and making tag corrections in the training samples. Up to model verification index F1 after multiple rounds of training (avg) Training was stopped > 0.9. Wherein F1 (avg) The formula of (2) is:
in the method, in the process of the application,
TP is the number of actual samples, and among all samples predicted as positive examples by the model, the actual number of samples is the number of positive examples; FP, number of false positive samples, number of samples actually being negative samples among all samples predicted as positive by the model;
FN, number of false-negative samples, number of samples actually being positive samples among all samples predicted as negative by the model.
And after the model verification index accords with the expectation, obtaining a trained fault diagnosis model.
So far, the fault diagnosis method can be applied online. And evaluating the input three-dimensional feature matrix by the trained fault diagnosis model to obtain a diagnosis result of the electromechanical equipment.
To further verify the stability of the model, the model may be further tested. Here, the test environment where the verification "up to standard" is deployed on the cloud server can be tested for 3 months, and the test process and the data acquisition are synchronously performed on line in real time. After the test is passed, the trained fault diagnosis model is deployed in the production environment of the cloud server for application.
In the generation environment, the three-dimensional feature matrix is input into a trained fault diagnosis model, the fault diagnosis model outputs a diagnosis result according to a threshold value of an abnormality score of the three-dimensional feature matrix, for example, the threshold value is a certain preset value, if the abnormality score is larger than the preset value, the situation that the electromechanical equipment is abnormal is judged, the threshold value can be thinned, and different abnormal information can be fed back through different threshold values.
Referring to fig. 5, an embodiment of the present application further provides an electronic device, which includes:
a memory for storing a computer program; and the processor is used for realizing the fault diagnosis method steps when executing the computer program.
The electronic device may be configured or configured differently, may include at least one processor and memory, and may be temporarily stored or permanently stored.
The electronic device may further comprise at least one power source, at least one wired or wireless network interface, at least one input output interface. The input/output interfaces include, for example, analog interfaces, UART interfaces, RS485 interfaces, IO-Link interfaces, etc., and support most of sensor interfaces in the market. And a high-precision and high-reliability sensor can be selected, and all mechanical, noise and electric signals of the electromechanical system can be accurately acquired in a contact or/and non-contact mode.
The steps in the above-described fault diagnosis method may be implemented by the structure of the above-described electronic apparatus.
The embodiment of the present application also provides a readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the above-described fault diagnosis method.
The readable storage medium may be a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, and the like.
The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, but various modifications and variations can be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims (10)
1. A fault diagnosis method for an electromechanical device, comprising:
acquiring multi-source data of a multi-source signal;
time synchronization and alignment of data sequences of different signals in the multi-source data are carried out by utilizing a fusion technology, so that fusion data are obtained;
creating three-dimensional feature data based on the fusion data, and constructing a three-dimensional feature matrix;
inputting the three-dimensional feature matrix into a fault diagnosis model for training to obtain a trained fault diagnosis model;
and inputting the three-dimensional feature matrix into a trained fault diagnosis model to obtain a diagnosis result of the electromechanical equipment.
2. The fault diagnosis method according to claim 1, wherein,
the constructing three-dimensional feature data based on the fusion data, and constructing a three-dimensional feature matrix, includes:
dividing each signal data in the fusion data according to a preset window to obtain a window data set;
and extracting key features of each data in the window data set by using the multi-domain features to obtain a three-dimensional feature matrix.
3. The fault diagnosis method according to claim 2, wherein,
the multi-domain features include at least one of a time domain, a frequency domain, and a statistical domain, and any one of the domains includes at least one feature.
4. The fault diagnosis method according to claim 1, wherein,
inputting the three-dimensional feature matrix into a fault diagnosis model for training to obtain a trained fault diagnosis model, wherein the training process comprises the steps of evaluating a training result, and obtaining the trained fault diagnosis model if the evaluation result accords with the expectation; if the evaluation result does not meet the expectations, the fault diagnosis model is continuously trained.
5. The fault diagnosis method according to claim 1, wherein,
the method for synchronizing and aligning the time of the data sequences of different signals in the multi-source data by using the fusion technology to obtain fusion data comprises the following steps:
preprocessing the fusion data by utilizing a preprocessing means to obtain preprocessed fusion data;
the preprocessing means comprises at least one processing means of filtering, amplifying, sampling and quantizing, normalizing, trending, window function and noise reduction.
6. The fault diagnosis method according to claim 1, wherein,
the fault diagnosis model is a ConvLSTM-based neural network, comprising:
an encoder and a decoder, wherein the encoder comprises a ConvLSTM network, a Pooling layer and an FC layer; the decoder comprises a PC layer, an upsping layer and a ConvLSTM network; the three-dimensional feature matrix is input into a first encoder and then is input into a decoder;
wherein,,
and calculating to obtain reconstruction loss by using the input three-dimensional feature matrix and the data output by the decoder, and evaluating whether the input data is abnormal or not through the reconstruction loss.
7. The fault diagnosis method according to claim 1, wherein,
the fault diagnosis model is a ConvLSTM-based neural network of dual encoders, comprising:
the first encoder and the second encoder comprise ConvLSTM units, a Pooling layer and an FC layer;
the decoder comprises a PC layer, an Upsamping layer and ConvLSTM units;
the three-dimensional feature matrix is input into a first encoder and is input into a second encoder after passing through a decoder;
wherein,,
calculating by using the input three-dimensional feature matrix and the data output by the decoder to obtain reconstruction loss; the potential loss is obtained after calculation by utilizing the data output by the first encoder and the data output by the second encoder; and carrying out weighted summation on the reconstruction loss and the potential loss to obtain a final loss function-anomaly score, and evaluating whether the input data is anomalous or not through the anomaly score.
8. The fault diagnosis method according to claim 1, wherein,
the multi-source signal includes multiple signal sources or/and multiple signal sources of the same signal.
9. A computer device, comprising:
a memory for storing a computer program;
processor for implementing the steps of the fault diagnosis method for an electromechanical device according to any one of claims 1 to 8 when executing the computer program.
10. A readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the fault diagnosis method for an electromechanical device according to any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310781186.8A CN116776284A (en) | 2023-06-28 | 2023-06-28 | Fault diagnosis method for electromechanical device, computer device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310781186.8A CN116776284A (en) | 2023-06-28 | 2023-06-28 | Fault diagnosis method for electromechanical device, computer device, and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116776284A true CN116776284A (en) | 2023-09-19 |
Family
ID=87992757
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310781186.8A Pending CN116776284A (en) | 2023-06-28 | 2023-06-28 | Fault diagnosis method for electromechanical device, computer device, and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116776284A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117056865A (en) * | 2023-10-12 | 2023-11-14 | 北京宝隆泓瑞科技有限公司 | Method and device for diagnosing operation faults of machine pump equipment based on feature fusion |
-
2023
- 2023-06-28 CN CN202310781186.8A patent/CN116776284A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117056865A (en) * | 2023-10-12 | 2023-11-14 | 北京宝隆泓瑞科技有限公司 | Method and device for diagnosing operation faults of machine pump equipment based on feature fusion |
| CN117056865B (en) * | 2023-10-12 | 2024-01-23 | 北京宝隆泓瑞科技有限公司 | Method and device for diagnosing operation faults of machine pump equipment based on feature fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111259921B (en) | Transformer sound abnormality detection method based on improved wavelet packet and deep learning | |
| CN111291614B (en) | Child epileptic syndrome classification method based on migration learning multimode decision fusion | |
| CN110705456A (en) | Micro motor abnormity detection method based on transfer learning | |
| CN113158814B (en) | Bearing health state monitoring method based on convolution self-encoder | |
| CN113537328B (en) | Rotary machine fault diagnosis method and device based on deep learning | |
| CN110991471B (en) | Fault diagnosis method for high-speed train traction system | |
| CN113449473B (en) | Wind power bearing degradation trend prediction method, system and device based on DFCAE-BiLSTM | |
| CN115409131B (en) | Production line abnormity detection method based on SPC process control system | |
| CN113673346A (en) | Motor vibration data processing and state recognition method based on multi-scale SE-Resnet | |
| CN116776284A (en) | Fault diagnosis method for electromechanical device, computer device, and storage medium | |
| CN113865868A (en) | Rolling bearing fault diagnosis method based on time-frequency domain expression | |
| CN113538353B (en) | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network | |
| CN116401532B (en) | Method and system for recognizing frequency instability of power system after disturbance | |
| CN113568774A (en) | Real-time anomaly detection method for multi-dimensional time sequence data by using unsupervised deep neural network | |
| CN110192864B (en) | Cross-domain electrocardiogram biological characteristic identity recognition method | |
| CN113705405B (en) | Nuclear pipeline fault diagnosis method | |
| CN117809164A (en) | Substation equipment fault detection method and system based on multi-mode fusion | |
| CN111610026B (en) | Rotary machine fault diagnosis method based on deep clustering | |
| CN117421616A (en) | Mine shaft detection system and method | |
| CN111967507A (en) | Discrete cosine transform and U-Net based time sequence anomaly detection method | |
| CN116955947A (en) | Rainfall-induced power transmission line slope displacement nonlinear prediction method and system | |
| CN116595465A (en) | High-dimensional sparse data outlier detection method and system based on self-encoder and data enhancement | |
| CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
| CN116720063B (en) | Rolling bearing consistency degradation characteristic extraction method based on DTW-CAE | |
| CN117361256B (en) | Elevator safety management method and system based on artificial intelligence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |