Complex equipment key component fault diagnosis system based on LSTM coding network

Abstract

The application discloses a complex equipment key component fault diagnosis system based on an LSTM coding network, which comprises an equipment key operation data acquisition module, an equipment key operation data preprocessing module, an equipment key operation data feature extraction module and an equipment key component fault mode recognition module, the equipment key operation data acquisition module acquires key operation data in the operation process of an equipment key component, the equipment key operation data preprocessing module performs filtering, denoising, outlier rejection, missing value filling and normalization processing on the equipment key component operation data, the equipment key operation data feature extraction module performs feature extraction on the preprocessed equipment key component operation data, data extraction and transformation are performed to obtain an equipment key component data energy spectrum, and the equipment key component fault mode recognition module recognizes the fault mode of the equipment key component.

Description

Technical Field

[0001] This invention relates to the field of health management technology for complex equipment, and in particular to a fault diagnosis system for key components of complex equipment based on LSTM coding networks. Background Technology

[0002] Complex electromechanical equipment plays a vital and critical role in numerous fields, including industry and national defense. Achieving long-term stable, safe, and reliable operation of such equipment places high demands not only on its design and development, such as designing stable and reliable control systems and selecting appropriate platform materials, but also on the inherent possibility of system failures due to equipment performance degradation and various human and environmental factors. The causes and mechanisms of these failures are often highly complex, with failures typically occurring in critical components such as rotating mechanical parts. Therefore, intelligent fault diagnosis of critical components based on monitoring data such as equipment vibration signals is a pressing practical requirement for ensuring the smooth operation of such equipment.

[0003] Intelligent fault diagnosis of critical equipment components is a pattern recognition problem, involving three steps: data preprocessing, feature extraction and selection, and fault classification. Traditional fault mode classification methods, with their shallow structure, lack strong feature representation capabilities. Achieving good accuracy relies on extracting sensitive features reflecting the fault type based on prior knowledge. However, feature extraction is prone to losing crucial information, leading to a decrease in the accuracy of anomaly prediction.

[0004] In recent years, the rapid development and application of advanced technologies such as artificial intelligence and data mining have provided new ideas for fault diagnosis of key equipment components. Autoencoders are widely used models in deep learning, capable of extracting deep features from unlabeled data using a multi-layer encoding process. However, for many applications, model training is difficult, and performance is affected by the hidden layer structure and the number of iterations.

[0005] Vibration data of critical mechanical components of equipment can largely reflect the operating status of the equipment. For example, friction, impact, vibration, sudden changes in rotational speed caused by bearing failure, or deformation and fracture of the structure will all have different characteristics in vibration signals. Fault symptoms exhibited by vibration signals are generally distributed within different frequency bands, making the search for sensitive features through methods such as spectrum analysis and wavelet analysis a traditional but effective approach. However, in the early stages of health deterioration, fault symptoms may be very weak and may even be masked by vibrations from other components or environmental noise, increasing the difficulty of symptom identification. Therefore, the effective discovery and mining of hidden, weak defect features is an important research area in intelligent diagnostic technology for critical equipment components. Summary of the Invention

[0006] The purpose of this invention is to provide a fault diagnosis system for key components of complex equipment based on LSTM coding networks to solve the above-mentioned problems.

[0007] The present invention solves the technical problem by adopting the following technical solution:

[0008] A fault diagnosis system for critical components of complex equipment based on an LSTM coding network includes: a critical equipment operation data acquisition module, a critical equipment operation data preprocessing module, a critical equipment operation data feature extraction module, and a critical equipment component fault mode recognition module, wherein:

[0009] The key operational data acquisition module collects key operational data of the key components of the equipment during operation, and transmits and stores the data.

[0010] The key equipment operation data preprocessing module filters, denoises, removes outliers, fills in missing values, and normalizes the collected key equipment component operation data to obtain data for feature extraction.

[0011] The key operating data feature extraction module extracts features from the preprocessed operating data of key equipment components, and extracts and transforms the data to obtain the energy map of key equipment component data.

[0012] The fault mode identification module for key equipment components maps the data energy map of key equipment components to the health state space of key equipment components, thereby identifying the fault modes of key equipment components.

[0013] Furthermore, the key operational data acquisition module of the device adopts a layered architecture for data acquisition system deployment, consisting of an acquisition layer, a transmission layer, and a management layer from bottom to top. The acquisition layer collects relevant data, forms a network through the TCP / IP protocol, and sends the data to the transmission layer. The gateway converts the data format and sends the data to the processing layer via the local area network (LAN). The management layer manages and stores the data to support subsequent data analysis and processing.

[0014] Furthermore, the data collected by the acquisition layer consists of multi-dimensional data on key components of the equipment, including various types of data for key components, specifically operating status parameters of key components, including current, voltage, temperature, humidity, and vibration signal data.

[0015] Furthermore, the key operational data preprocessing module for the equipment includes a data segmentation module, a data cleaning module, a data transformation module, and a data reduction module, among which:

[0016] The data segmentation module is used to divide the overall data into smaller, independently managed data.

[0017] The data cleaning module is used to fill in missing values ​​in the data, filter noise, identify outliers, and correct data inconsistencies.

[0018] The data transformation module transforms and maps the raw data of key components of the equipment to other feature spaces, which can better highlight the inherent characteristics of the data and facilitate feature extraction and fault mode identification.

[0019] The data reduction module simplifies the original data of key components of the equipment, using as little data as possible to characterize the feature information of the entire system, and minimizing information loss.

[0020] Furthermore, the key operational data feature extraction module of the equipment extracts features by constructing a signal energy map of key components of the equipment through wavelet decomposition. For different types of signals, the wavelet with the most matching type of the desired signal features is selected for analysis to obtain reliable feature extraction results.

[0021] Furthermore, the fault mode identification module for key equipment components uses an LSTM coding network to classify and identify fault modes of key equipment components.

[0022] Furthermore, the basic unit of the LSTM coding network adopts a two-layer LSTM neural encoder; the first layer maps the original energy map to the initially extracted coding matrix to obtain the information representation of the energy of each layer; the second layer extracts the coding matrix obtained from the first layer into a coding vector to show the distribution characteristics of the health state; finally, the coding vector is mapped to the health state space.

[0023] The beneficial effects of this invention are:

[0024] This invention constructs a signal energy map from the device monitoring signal using wavelet transform, extracting the most effective original features of the data; it then uses an LSTM coding network to map the energy map to the device health state space, identifying the device's current health status; finally, it trains the model using a backpropagation algorithm. This approach improves feature extraction capabilities to a certain extent, requires fewer iterations, and achieves higher diagnostic accuracy. Attached Figure Description

[0025] Figure 1 The attached figure is a schematic diagram of the system composition structure of the fault diagnosis method for key components of equipment based on LSTM coding network provided by the present invention.

[0026] Figure 2 The attached figure is a schematic diagram of the equipment key operation data preprocessing method provided by the present invention.

[0027] Figure 3 The attached figure is a schematic diagram of the method for extracting key operating data features of equipment provided by the present invention.

[0028] Figure 4 The attached figure is a schematic diagram of the energy spectrum of key signals of the device provided by the present invention.

[0029] Figure 5 The attached figure is a schematic diagram of the fault mode identification method for key equipment components provided by the present invention.

[0030] Figure 6 The attached figure is a schematic diagram of the LSTM coding network structure provided by the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Reference Appendix Figure 1 This invention discloses a fault diagnosis system for critical components of complex equipment based on LSTM coding networks, comprising: a critical equipment operation data acquisition module, a critical equipment operation data preprocessing module, a critical equipment operation data feature extraction module, and a critical equipment component fault mode recognition module, wherein:

[0033] The key operational data acquisition module collects key operational data of the key components of the equipment during operation, and transmits and stores the data.

[0034] The key equipment operation data preprocessing module filters, denoises, removes outliers, fills in missing values, and normalizes the collected key equipment component operation data to obtain data for feature extraction.

[0035] The key operating data feature extraction module extracts features from the preprocessed operating data of key equipment components, and extracts and transforms the data to obtain the energy map of key equipment component data.

[0036] The fault mode identification module for key equipment components maps the data energy map of key equipment components to the health state space of key equipment components, thereby identifying the fault modes of key equipment components.

[0037] The data acquisition system of the key equipment operation data acquisition module adopts a layered architecture, consisting of an acquisition layer, a transmission layer, and a management layer from bottom to top. This architecture enables the acquisition, transmission, storage, and management of various key test parameters of the equipment. The acquisition layer collects relevant data, networks it via TCP / IP protocol, and sends the data to the transmission layer. The gateway converts the data format and sends the data to the processing layer via a local area network (LAN). Data acquisition is achieved through various sensors, and the sensor-collected data is stored in a database for management, supporting subsequent data analysis and fault mode identification. The data collected by the acquisition layer consists of multi-dimensional data on key equipment components, including various types of data for different key components, specifically operating status parameters of key components, including current, voltage, temperature, humidity, and vibration signal data.

[0038] The equipment critical operation data preprocessing module includes a data segmentation module, a data cleaning module, a data transformation module, and a data reduction module. Specifically: the data segmentation module divides the overall data into smaller, independently managed data sets; the data cleaning module fills in missing values, filters noise, identifies outliers, and corrects data inconsistencies; the data transformation module transforms and maps the raw data of key equipment components to other feature spaces, highlighting the inherent characteristics of the data and facilitating feature extraction and fault mode identification; and the data reduction module simplifies the raw data of key equipment components, using as little data as possible to represent the feature information of the entire system while minimizing information loss.

[0039] The data segmentation, data cleaning, data transformation, and data reduction modules specifically include data binning, missing value imputation, outlier analysis, and principal component analysis (PCA). The workflow is as follows: Figure 2 As shown. Wherein:

[0040] Data binning smooths ordered data values ​​by examining their nearest neighbors. These ordered values ​​are distributed into bins, and for bin mean smoothing, each value in a bin is replaced with the bin mean.

[0041] Missing value imputation methods fill missing values ​​by using the centrality measure (mean and median) of the attribute; or by using the mean or median of the attribute of all samples belonging to the same class as the given tuple.

[0042] Outlier analysis uses clustering to detect outliers, grouping similar values ​​into clusters and intuitively considering values ​​that fall outside the cluster set as outliers.

[0043] Principal component analysis (PCA) performs spatial mapping on highly correlated multi-sensor data to reduce dimensionality, using as little data as possible to characterize the features of the entire system and minimizing information loss.

[0044] The critical equipment operation data preprocessing module utilizes data mining and neural networks to preprocess data from key equipment components. This enables the cleaning and transformation of large amounts of monitoring data from these components, improving data quality and enhancing the accuracy and efficiency of feature extraction and fault mode identification. Data cleaning methods are employed to fill in missing values, filter noise, identify outliers, and correct data inconsistencies. Principal component analysis (PCA) is used to reduce the dimensionality of multidimensional data, mapping the original data to a new data space and extracting key information. This allows for the representation of the entire system's state information with less data while minimizing information loss. Furthermore, for specific data, neural networks can be used for more efficient and intelligent data preprocessing.

[0045] The key operational data feature extraction module of the equipment extracts features by constructing a signal energy map of key components of the equipment through wavelet decomposition. For different types of signals, the wavelet with the most matching type of the desired signal features is selected for analysis to obtain reliable feature extraction results.

[0046] The key data feature extraction module for the equipment is completed using a wavelet packet energy decomposition method. The specific process is as follows: Figure 3 As shown, the original monitoring signal from the device is first processed through data segmentation, data cleaning, data transformation, and data reduction to obtain the wavelet packet decomposition input signal. A wavelet packet decomposes the original signal into a space derived from wavelet basis functions, and the original signal can be reconstructed from the coefficients. Decomposition coefficients for different frequency bands are obtained across all frequency ranges, combining the characteristics of time-domain and frequency-domain analysis. The signal energy of each frequency band obtained by wavelet packet decomposition contains rich features, which can be used to analyze non-stationary signals, extract weak features from the signal, and remove noise bands from the signal.

[0047] The equipment signal energy spectrum obtained by the equipment key data feature extraction module is as follows: Figure 4 As shown, the gray level of the color block in the figure represents the energy level of the color block. The larger the gray level of the color block, the higher the energy of that part of the signal. The right side of the figure shows the corresponding scale of the gray level and value of the color block.

[0048] The key operational data feature extraction module for equipment utilizes data mining and neural networks to extract features from critical components, enabling the extraction of system health and fault characteristics from large amounts of equipment monitoring data. Principal component analysis (PCA) is employed to reduce the dimensionality of multidimensional data, mapping the original data to a new feature space and transforming it into effective features. Similarity analysis is used to analyze time series of different lengths, extracting health features by comparing the similarity between real-time monitoring data and standard data. Furthermore, for specific data, neural networks can be used to achieve more efficient health feature extraction.

[0049] The fault mode identification module for key equipment components uses an LSTM-AE encoding network based on the basic structure of a Long Short-Term Memory (LSTM) network to classify and identify fault modes of key equipment components, thereby realizing the extraction and mapping from data energy maps to health state space.

[0050] The fault mode identification module for key equipment components is implemented based on an LSTM coding network. The specific process is as follows: Figure 5 As shown, the LSTM coding network is trained using equipment signal energy spectrum data, and the model is tested and optimized to identify fault modes of key equipment components.

[0051] The structure of an LSTM coding network (LSTM-AE) is as follows: Figure 6 As shown, the LSTM network is an improvement on the standard recurrent neural network. It is designed with four gate structures to enable neurons to have long-term memory. This is an effective gradient-based algorithm. This structure allows constant error to flow through the internal state of special units, thereby avoiding the problems of gradient explosion or vanishing.

[0052] Suppose the input to the encoding network is a sample set x = {x1, x2, ..., xn} containing D-dimensional samples. D}, x D This represents the D-th input vector. The encoding process maps the input vector to hidden layer vectors h = {h1, h2, ..., h...}. D}, h D The encoding process for the D-th hidden layer vector can be represented as follows:

[0053] h = s(Wx + b)

[0054] s(t)=1 / (1+e -t )

[0055] Where s is the activation function of the encoding network, t represents the input of the activation function, W is the m×n dimensional encoding weight matrix, and b is the m dimensional encoding bias vector.

[0056] The decoding process involves transforming the hidden layer vector h into a reconstructed dataset z = {z1, z2, ..., zn} with the same dimensions as the input data. D}, z D The decoding process for the D-th output vector can be represented as follows:

[0057] z = s(W′h + b′)

[0058] Where W′ is an n×m dimensional decoding weight matrix and b′ is an n-dimensional decoding bias vector.

[0059] The basic unit of the LSTM coding network adopts a two-layer LSTM neural encoder. The first layer maps the original energy map to the initially extracted coding matrix to obtain the information representation of the energy of each layer. The second layer extracts the coding matrix obtained from the first layer into a coding vector to show the distribution characteristics of the health state. Finally, the coding vector is mapped to the health state space to realize the identification of the fault modes of the key components of the equipment.

[0060] This invention discloses a fault diagnosis method for key equipment components based on LSTM coding networks. Addressing the difficulties and low accuracy in discovering and mining hidden weak defect features of key equipment components based on vibration signals, this method achieves feature extraction and fault mode identification from multi-frequency vibration data of key equipment components. It boasts stronger feature extraction capabilities, higher diagnostic accuracy, and better stability. Based on data cleaning and principal component analysis, a large amount of detection data from key equipment components is preprocessed to construct a data space reflecting the state of the key components, facilitating feature extraction and fault mode identification. Based on wavelet decomposition, the vibration data of key equipment components is decomposed and integrated multiple times to construct a signal energy spectrum and extract the most effective features of the vibration signal. Finally, based on LSTM coding networks, the signal energy spectrum is extracted and mapped layer by layer to identify fault modes of key equipment components.

[0061] This invention constructs a signal energy map from the device monitoring signal using wavelet transform, extracting the most effective original features of the data; it then uses an LSTM coding network to map the energy map to the device health state space, identifying the device's current health status; finally, it trains the model using a backpropagation algorithm. This approach improves feature extraction capabilities to a certain extent, requires fewer iterations, and achieves higher diagnostic accuracy.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault diagnosis system for key components of complex equipment based on LSTM coding networks, characterized in that, include: The system comprises a key equipment operation data acquisition module, a key equipment operation data preprocessing module, a key equipment operation data feature extraction module, and a key equipment component fault mode identification module, wherein: The key operational data acquisition module collects key operational data of the key components of the equipment during operation, and transmits and stores the data. The key equipment operation data preprocessing module filters, denoises, removes outliers, fills in missing values, and normalizes the collected key equipment component operation data to obtain data for feature extraction. The key operating data feature extraction module extracts features from the preprocessed operating data of key equipment components, and extracts and transforms the data to obtain the energy map of key equipment component data. The key operational data feature extraction module of the equipment extracts features by constructing a signal energy map of key components of the equipment through wavelet decomposition. For different types of signals, the wavelet with the most matching type of the desired signal features is selected for analysis to obtain reliable feature extraction results. The fault mode identification module for key equipment components maps the data energy map of key equipment components to the health status space of key equipment components, thereby identifying the fault modes of key equipment components. The equipment's key operational data preprocessing module includes a data segmentation module, a data cleaning module, a data transformation module, and a data reduction module, among which: The data segmentation module is used to divide the overall data into smaller, independently managed data. The data cleaning module is used to fill in missing values ​​in the data, filter noise, identify outliers, and correct data inconsistencies. The data transformation module transforms and maps the raw data of key components of the equipment to other feature spaces, which can better highlight the inherent characteristics of the data and facilitate feature extraction and fault mode identification. The data reduction module simplifies the raw data of key components of the equipment, uses less data to characterize the feature information of the entire system, and minimizes information loss. The fault mode identification module for key equipment components uses an LSTM coding network to classify and identify fault modes of key equipment components. The basic unit of the LSTM coding network adopts a two-layer LSTM neural encoder. The first layer maps the original energy spectrum into an initially extracted coding matrix to obtain the information representation of the energy of each layer. The second layer extracts the coding matrix obtained from the first layer into a coding vector to show the distribution characteristics of the health state. Finally, the coding vector is mapped to the health state space.

2. The fault diagnosis system for key components of complex equipment based on LSTM coding networks according to claim 1, characterized in that, The key operational data acquisition module of the device adopts a layered architecture, consisting of an acquisition layer, a transmission layer, and a management layer from bottom to top. The acquisition layer collects relevant data, forms a network using the TCP / IP protocol, and sends the data to the transmission layer. The gateway converts the data format and sends the data to the processing layer via a local area network (LAN). The management layer manages and stores the data, providing support for subsequent data analysis and processing.

3. The fault diagnosis system for key components of complex equipment based on LSTM coding networks according to claim 2, characterized in that, The data collected by the acquisition layer consists of multi-dimensional data on key components of the equipment, including various types of data for key components, specifically operating status parameters of key components, including current, voltage, temperature, humidity and vibration signal data.

Images