Pneumatic valve body fault diagnosis method and system based on multi-model fusion and Bayesian optimization
Through multi-model fusion and Bayesian optimization methods, combined with sensor data preprocessing and feature extraction, the model weights are dynamically determined to perform pneumatic valve body fault diagnosis, which solves the problems of inefficiency and insufficient accuracy of traditional methods, and achieves more efficient and reliable fault diagnosis.
Patent Information
- Application Number
- CN202510098611.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-22
- Publication Date
- 2025-05-23
AI Technical Summary
Traditional pneumatic valve body fault diagnosis methods rely on manual experience and regular maintenance, which are inefficient and diagnostic accuracy depends on the experience of the operator, making it difficult to ensure the consistency and reliability of the diagnosis. A single machine learning model is difficult to cope with complex and changing industrial environments, and the accuracy and robustness of fault diagnosis are limited.
The pneumatic valve body fault diagnosis method is adopted with multi-model fusion and Bayesian optimization. By obtaining the real-time operating status data collected by the sensor group, preprocessing and feature extraction are performed, and inputting them into the pre-trained LSTM, CNN and Transformer models. The Bayesian optimization algorithm is used to dynamically determine the model weight and weight sum to obtain the final fault diagnosis result.
It effectively improves the accuracy and robustness of fault diagnosis, enhances the adaptability and expansion of the diagnostic system, and can more accurately identify complex fault modes of the pneumatic valve body.
Smart Images
Figure CN120030409A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault diagnosis, and more specifically to a pneumatic valve body fault diagnosis method and system based on multi-model fusion and Bayesian optimization. Background Art
[0002] With the continuous improvement of industrial automation level, pneumatic valve bodies, as key actuators, have been widely used in various industrial production processes. Pneumatic valve bodies are mainly used to control the flow of gas or liquid, and their operating status directly affects the stability and efficiency of the entire production process. However, due to the complexity of the working environment, the diversity of working conditions and potential defects in the manufacturing process, pneumatic valve bodies are often affected by various external factors, resulting in failures. These failures will not only reduce production efficiency, but may also pose a potential threat to the safe operation of the equipment and even cause serious safety accidents.
[0003] Traditional methods for diagnosing pneumatic valve body faults mainly rely on manual experience and regular maintenance. These methods include operators judging the operating status of the pneumatic valve body by listening, observing, etc., or regularly disassembling and inspecting the valve body. However, this method is not only inefficient, but the accuracy of fault diagnosis depends largely on the experience and skills of the operator, making it difficult to ensure the consistency and reliability of the diagnosis. Especially in the context of increasing industrial production scale and increasing degree of automation, this traditional fault diagnosis method can no longer meet the needs of modern industry for efficient, accurate and intelligent fault diagnosis.
[0004] As industrial production continues to increase its requirements for automation and intelligence, data-driven fault diagnosis methods have gradually become a research hotspot. By installing various sensors on the pneumatic valve body, real-time collection of valve body operation data, combined with advanced data analysis technologies such as machine learning and deep learning, early diagnosis and early warning of pneumatic valve body failures can be performed, thereby helping operators to detect problems in a timely manner and avoid the impact of equipment failures on production and safety.
[0005] However, a single machine learning model is often unable to cope with complex and changing industrial environments, and the accuracy and robustness of fault diagnosis are limited. Therefore, in recent years, integrating multiple models to improve the performance of diagnostic systems has become a research focus. By integrating different types of models, such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), Transformer models, etc., the advantages of each model can be effectively combined to overcome the shortcomings of a single model and build a more efficient and reliable fault diagnosis system.
[0006] In addition, traditional multi-model fusion methods usually use fixed weights or simple weighting to fuse model results, which has great limitations in practical applications. The performance of different models may show significant differences in different scenarios, and fixed weights are difficult to adapt to changes in data distribution and working conditions.
[0007] Therefore, how to dynamically determine the fusion weights to improve the adaptability and accuracy of the fusion model has become an urgent problem to be solved. Summary of the invention
[0008] In view of this, the object of the present invention is to provide a pneumatic valve body fault diagnosis method and system based on multi-model fusion and Bayesian optimization.
[0009] In order to achieve the above object, the present invention adopts the following technical solution:
[0010] A first aspect provides a pneumatic valve body fault diagnosis method based on multi-model fusion and Bayesian optimization, comprising the following steps:
[0011] S1: Acquire the real-time operating status data of the pneumatic valve body collected by the sensor group and perform preprocessing;
[0012] S2: Perform time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed real-time operation status data;
[0013] S3: Input the time domain features, frequency domain features and wavelet features extracted in S2 into a pre-trained pneumatic valve body fault diagnosis model to obtain a predicted fault diagnosis result of the pneumatic valve body; wherein the predicted fault diagnosis result output by the pre-trained pneumatic valve body fault diagnosis model is a weighted summation result of the LSTM model, the CNN model and the Transformer model; the sum weights of the LSTM model, the CNN model and the Transformer model are obtained based on a Bayesian optimization algorithm.
[0014] Preferably, the pre-trained pneumatic valve body fault diagnosis model is obtained based on the following steps:
[0015] S31: Acquire historical operating status data of the pneumatic valve body collected by the sensor group and perform preprocessing; wherein the historical operating status data of the pneumatic valve body is marked with corresponding fault types;
[0016] S32: Performing time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed historical operating status data;
[0017] S33: using the time domain features, frequency domain features and wavelet features extracted in S32 to train the LSTM model, CNN model and Transformer model respectively, and using the Bayesian optimization algorithm to optimize and obtain the best weight combination of the LSTM model, CNN model and Transformer model;
[0018] The trained LSTM model, CNN model and Transformer model are weightedly summed using the optimal weight combination to obtain the pre-trained pneumatic valve body fault diagnosis model.
[0019] Preferably, the sensor group includes a pressure sensor, a temperature sensor, a flow sensor, an acceleration sensor and a stroke sensor;
[0020] Wherein, the pressure sensor is used to detect the input and output pressures of the pneumatic valve body;
[0021] The temperature sensor is used to monitor the temperature of the electromagnetic control mechanism of the pneumatic valve body;
[0022] The flow sensor is used to detect the fluid flow through the pneumatic valve body;
[0023] The acceleration sensor is used to detect the vibration signal of the pneumatic valve body;
[0024] The stroke sensor is used to measure the opening of the valve in real time.
[0025] Preferably, the preprocessing includes data cleaning, data standardization and data normalization.
[0026] Preferably, the time domain features include mean, variance, skewness and kurtosis; the frequency domain features include spectral density and main frequency; and the wavelet features include wavelet coefficients.
[0027] Preferably, the Bayesian optimization algorithm uses a Gaussian process as a proxy model of the objective function;
[0028] And the best weight combination is obtained by maximizing the expected improvement function;
[0029] The expression of the proxy model is: GP(μ(x),k(x,x′))(11);
[0030] Where x = (α, β, γ) represents the weight combination of the LSTM model, CNN model and Transformer model; μ(x) represents the mean of the predicted objective function; k(x, x′) represents the covariance function between the weight combination x and x′; GP represents Gaussian process;
[0031] The expression for maximizing the expected improvement function is: * = argmaxx EI(x;GP)(12);
[0032] Wherein, x* represents the optimal weight combination; EI(x; GP) represents the expected improvement function, that is, the expected improvement in the performance of the predicted fault diagnosis result that can be brought about by the currently selected weight combination x.
[0033] The second aspect provides a pneumatic valve body fault diagnosis system with multi-model fusion and Bayesian optimization, which applies any of the pneumatic valve body fault diagnosis methods described above, including a sensor group, a preprocessing unit, a feature extraction unit, and a pre-trained pneumatic valve body fault diagnosis model;
[0034] The sensor group is used to collect real-time operating status data of the pneumatic valve body;
[0035] The preprocessing unit is used to preprocess the real-time operating status data collected by the sensor group;
[0036] The feature extraction unit is used to perform time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed real-time operation status data;
[0037] The pre-trained pneumatic valve body fault diagnosis model is used to output the predicted fault diagnosis results of the pneumatic valve body according to the time domain features, frequency domain features and wavelet features extracted by the feature extraction unit; wherein the predicted fault diagnosis results output by the pre-trained pneumatic valve body fault diagnosis model are the results of weighted summation of the LSTM model, the CNN model and the Transformer model; the sum weights of the LSTM model, the CNN model and the Transformer model are obtained based on the Bayesian optimization algorithm.
[0038] Preferably, the sensor group includes a pressure sensor, a temperature sensor, a flow sensor, an acceleration sensor and a stroke sensor;
[0039] Wherein, the pressure sensor is used to detect the input and output pressures of the pneumatic valve body;
[0040] The temperature sensor is used to monitor the temperature of the valve body electromagnetic control mechanism;
[0041] The flow sensor is used to detect the fluid flow through the pneumatic valve body;
[0042] The acceleration sensor is used to detect the vibration signal of the pneumatic valve body;
[0043] The stroke sensor is used to measure the opening of the valve in real time.
[0044] Preferably, the preprocessing unit includes a data cleaning module, a data standardization module and a data normalization module;
[0045] The data cleaning module is used to clean the real-time operating status data collected by the sensor group;
[0046] The data standardization module is used to standardize the real-time operation status data after data cleaning;
[0047] The data normalization module is used to normalize the real-time operating status data after the standardization process.
[0048] Preferably, the feature extraction unit includes a time domain feature extraction module, a frequency domain feature extraction module and a wavelet feature extraction module;
[0049] The time domain feature extraction module is used to extract the mean, variance, skewness and kurtosis of the pre-processed real-time operation status data;
[0050] The frequency domain feature extraction module is used to extract the spectrum density and main frequency of the pre-processed real-time operation status data;
[0051] The wavelet feature extraction module is used to extract the wavelet coefficients of the pre-processed real-time running status data.
[0052] It can be seen from the above technical solution that compared with the prior art, the present invention discloses a pneumatic valve body fault diagnosis method and system based on multi-model fusion and Bayesian optimization. The present invention can effectively improve the accuracy and robustness of fault diagnosis, and has better adaptability and scalability than the traditional single model method. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0054] Figure 1 A flow chart of a pneumatic valve body fault diagnosis method based on multi-model fusion and Bayesian optimization provided by the present invention. DETAILED DESCRIPTION
[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0056] like Figure 1 As shown, the embodiment of the present invention discloses a pneumatic valve body fault diagnosis method based on multi-model fusion and Bayesian optimization, comprising the following steps:
[0057] S1: Acquire the real-time operating status data of the pneumatic valve body collected by the sensor group and perform preprocessing;
[0058] S2: Perform time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed real-time operation status data;
[0059] S3: Input the time domain features, frequency domain features and wavelet features extracted in S2 into a pre-trained pneumatic valve body fault diagnosis model to obtain a predicted fault diagnosis result of the pneumatic valve body; wherein the predicted fault diagnosis result output by the pre-trained pneumatic valve body fault diagnosis model is a weighted summation result of the LSTM model, the CNN model and the Transformer model; the sum weight of the LSTM model, the CNN model and the Transformer model is obtained based on a Bayesian optimization algorithm.
[0060] It is understandable that:
[0061] The LSTM (Long Short-Term Memory) model is suitable for processing time series data and can capture long-term and short-term dependencies in the data. The core of LSTM is the memory cell (CellState), which controls the transmission and retention of information through the input gate, forget gate, and output gate, thereby processing complex dynamic changes in time series.
[0062] The specific formula is as follows:
[0063]
[0064] Among them, X t represents the input data at time t; h t and h t-1 Represent the hidden states at time t and time t-1 respectively; C t and C t-1 Respectively represent the memory state at time t and time t-1; W f ,W i ,W c ,W o represents the weight matrix; σ is the activation function. Through these mechanisms, LSTM can effectively capture the time-dependent failure modes in the operation of pneumatic valve bodies.
[0065] The CNN (Convolutional Neural Network) model is used to extract local spatial features of data, and is particularly suitable for detecting local patterns or features in data. The convolution operation enables CNN to effectively identify local anomalies in signals, such as local vibration patterns or sudden failures, through a weight sharing mechanism.
[0066] The mathematical expression of the convolution operation is:
[0067]
[0068] in, is the output of the kth convolution kernel at position (i, j), x i+m-1,j+n-1 is the local region of the input feature map, is the weight of the convolution kernel, b k is the bias, and σ is the activation function. Physically, this local feature extraction can identify the local abnormal behavior of specific components in the pneumatic valve body, such as wear or damage of a certain component; M and N represent the height and width of the convolution kernel, respectively.
[0069] The Transformer model uses the self-attention mechanism to capture the dependencies between different positions in the sequence, and is suitable for processing time series data with long-distance dependencies. The Transformer model is particularly suitable for identifying complex systemic faults in the operation of pneumatic valve bodies, which may manifest as abnormal correlations between different signals.
[0070] The calculation formula of the self-attention mechanism is:
[0071] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key vector. Through this mechanism, the Transformer model can effectively model the complex correlation between the signals of various parts of the pneumatic valve body and identify potential systematic faults.
[0072] In one embodiment, the pre-trained pneumatic valve body fault diagnosis model is obtained based on the following steps:
[0073] S31: Acquire historical operating status data of the pneumatic valve body collected by the sensor group and perform preprocessing; wherein the historical operating status data of the pneumatic valve body is marked with corresponding fault types;
[0074] S32: Performing time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed historical operating status data;
[0075] S33: using the time domain features, frequency domain features and wavelet features extracted in S32 to train the LSTM model, CNN model and Transformer model respectively, and using the Bayesian optimization algorithm to optimize and obtain the best weight combination of the LSTM model, CNN model and Transformer model;
[0076] The trained LSTM model, CNN model and Transformer model are weightedly summed using the optimal weight combination to obtain the pre-trained pneumatic valve body fault diagnosis model.
[0077] In one embodiment, the sensor group includes a pressure sensor, a temperature sensor, a flow sensor, an acceleration sensor and a travel sensor;
[0078] Wherein, the pressure sensor is used to detect the input and output pressure of the pneumatic valve body and monitor whether there is leakage inside the valve body;
[0079] The temperature sensor is used to monitor the temperature of the electromagnetic control mechanism of the pneumatic valve body to determine whether it is working within the normal temperature range;
[0080] The flow sensor is used to detect the fluid flow through the pneumatic valve body and determine whether the stroke of the pneumatic valve body is normal;
[0081] The acceleration sensor (installed on the pneumatic valve body housing) is used to detect the vibration signal of the pneumatic valve body to determine whether abnormal vibration occurs in the mechanical part;
[0082] The stroke sensor is used to measure the opening of the valve in real time and determine whether the valve operates according to a predetermined trajectory.
[0083] It can be understood that the above-mentioned sensor group is installed on the pneumatic valve body to monitor and collect the operating status data of the pneumatic valve body in real time.
[0084] In one embodiment, the preprocessing includes data cleaning, data standardization and data normalization.
[0085] 1) Data cleaning: Use mean filtering and median filtering to remove high-frequency noise and outliers in the data collected by the sensor.
[0086] The calculation formula of mean filtering is:
[0087] Among them, x i (t) represents the unfiltered original data at time t, and L represents the size of the mean filter window.
[0088] Median filtering removes outliers by selecting the median of the data in the window to ensure the smoothness of the data.
[0089] 2) Standardization: In order to eliminate the dimensional differences between different sensor data, all data are z-score standardized.
[0090] The normalization formula is:
[0091] Among them, z ij represents the standardized data, x ij is the data before normalization, μ j and σ j represent the mean and standard deviation of the j-th sensor data before normalization.
[0092] Normalization ensures that all sensor data are concentrated on the same scale, avoiding model training bias caused by magnitude differences.
[0093] Normalization processing: The present invention further normalizes the standardized data to the range of [0,1] to improve the convergence speed and stability of model training.
[0094] The normalization formula is:
[0095] Among them, z' ij is the normalized data, z ij Indicates the standardized data; min(z j ) and max(z j ) represent the minimum and maximum values of the j-th sensor data before normalization.
[0096] In one embodiment, the time domain features include mean, variance, skewness and kurtosis; the frequency domain features include spectrum density and main frequency; and the wavelet features include wavelet coefficients.
[0097] The time domain characteristics, frequency domain characteristics and wavelet characteristics reflect different operating states of the pneumatic valve body and provide a rich source of information for the model.
[0098] Specific:
[0099] 1) Mean: reflects the central trend of sensor data, represents the average value of sensor data, and is used to measure the overall level of sensor data.
[0100] The calculation formula is:
[0101] Among them, x i represents the i-th incoming data; N' represents the number of incoming data.
[0102] 2) Variance: reflects the volatility of sensor data and indicates the degree of dispersion of sensor data. The larger the variance, the more drastic the fluctuation of sensor data.
[0103] The calculation formula is:
[0104] 3) Skewness and Kurtosis: They describe the symmetry and peak characteristics of sensor data respectively. Skewness is used to identify whether there is bias in sensor data, and kurtosis is used to detect whether there are abnormal peaks in sensor data.
[0105] 4) Power Spectral Density (PSD): Through Fourier transform, the time series sensor data is converted to the frequency domain, and the power spectrum density is used to analyze the frequency composition of the sensor data. It reveals the energy distribution of different frequency components in the sensor data.
[0106] The calculation formula is: Power_Spectrum(f)=|X(f)| 2 (9);
[0107] Among them, X(f) is the Fourier transform result of the sensor data. Frequency domain characteristics are of great significance for detecting periodic vibration and mechanical resonance in the operation of pneumatic valve bodies.
[0108] 5) Dominant Frequency: It is the frequency component with the most concentrated energy in the sensor data, which is used to identify the main vibration mode in the sensor data. Changes in the dominant frequency often correspond to signs of failure of mechanical components.
[0109] 6) Wavelet coefficients: Extract the characteristics of different frequency bands of sensor data through multi-scale wavelet decomposition. Wavelet transform can not only capture the frequency information of sensor data, but also retain the time domain information.
[0110] The calculation formula is:
[0111] Among them, x(t) is the data before wavelet transformation; ψ is the mother wavelet, a is the scale factor, b is the translation factor, and W(a,b) is the wavelet coefficient. Wavelet features can effectively capture sudden events and short-term changes in signals, which is particularly important for diagnosing sudden faults of pneumatic valve bodies.
[0112] In one embodiment, the Bayesian optimization algorithm uses a Gaussian process as a proxy model of the objective function; and obtains the best weight combination by maximizing the expected improvement function;
[0113] The expression of the proxy model is: GP(μ(x),k(x,x′))(11);
[0114] Where x = (α, β, γ) represents the weight combination of the LSTM model, CNN model and Transformer model; μ(x) represents the mean of the predicted objective function; k(x, x′) represents the covariance function between the weight combination x and x′; GP represents Gaussian process;
[0115] The expression for maximizing the expected improvement function is: * = argmax x EI(x;GP)(12);
[0116] Among them, x* represents the optimal weight combination; EI(x; GP) represents the expected improvement function (the strategy function used to select the next sampling point in the optimization process), that is, the expected improvement in the performance of the predicted fault diagnosis results that can be brought about by the current selection of weight combination x.
[0117] It should be noted that the present invention uses Gaussian Process as a proxy model of the objective function, aiming to find the optimal model fusion weight combination with minimal calculations. Specifically, Gaussian Process uses previous experimental results (such as diagnostic accuracy under each weight combination) to build a prediction model, and selects the next weight combination by maximizing the expected improvement function (Expected Improvement, EI). Through repeated calculations and updates, Bayesian optimization ultimately determines the weight combination that can achieve the best effect of model fusion.
[0118] Once the optimal weight combination is determined, the present invention adopts a weighted voting method to weighted sum the prediction results of the LSTM model, CNN model and Transformer model as the final prediction result.
[0119] The specific calculation formula for weighted voting is as follows:
[0120] y=α·y LSTM +β·y CNN +γ·y Transformer (13);
[0121] Among them, y LSTM ,y CNN and Transfomer These are the prediction results of the LSTM model, CNN model, and Transformer model. The weights α, β, and γ are the optimal weight combinations determined by Bayesian optimization, with the goal of achieving the best diagnostic effect under various fault modes. In a physical sense, weighted fusion can combine the strengths of different models, thereby providing more accurate fault diagnosis when dealing with complex and changing working conditions.
[0122] The embodiment of the present invention discloses a pneumatic valve body fault diagnosis system with multi-model fusion and Bayesian optimization, which applies any of the pneumatic valve body fault diagnosis methods described above, including a sensor group, a preprocessing unit, a feature extraction unit, and a pre-trained pneumatic valve body fault diagnosis model;
[0123] The sensor group is used to collect real-time operating status data of the pneumatic valve body;
[0124] The preprocessing unit is used to preprocess the real-time operating status data collected by the sensor group;
[0125] The feature extraction unit is used to perform time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed real-time operation status data;
[0126] The pre-trained pneumatic valve body fault diagnosis model is used to output the predicted fault diagnosis results of the pneumatic valve body according to the time domain features, frequency domain features and wavelet features extracted by the feature extraction unit; wherein the predicted fault diagnosis results output by the pre-trained pneumatic valve body fault diagnosis model are the results of weighted summation of the LSTM model, the CNN model and the Transformer model; the sum weights of the LSTM model, the CNN model and the Transformer model are obtained based on the Bayesian optimization algorithm.
[0127] In one embodiment, the sensor group includes a pressure sensor, a temperature sensor, a flow sensor, an acceleration sensor and a travel sensor;
[0128] Wherein, the pressure sensor is used to detect the input and output pressure of the pneumatic valve body (monitor whether there is leakage inside the valve body);
[0129] The temperature sensor is used to monitor the temperature of the valve body electromagnetic control mechanism (to determine whether it is working within the normal temperature range);
[0130] The flow sensor is used to detect the fluid flow through the pneumatic valve body (to determine whether the stroke of the pneumatic valve body is normal);
[0131] The acceleration sensor is used to detect the vibration signal of the pneumatic valve body (to determine whether abnormal vibration occurs in the mechanical part);
[0132] The stroke sensor is used to measure the opening of the valve in real time (to determine whether the valve is running according to a predetermined trajectory).
[0133] In one embodiment, the preprocessing unit includes a data cleaning module, a data standardization module and a data normalization module;
[0134] The data cleaning module is used to clean the real-time operating status data collected by the sensor group;
[0135] The data standardization module is used to perform standardization processing on the real-time operation status data after data cleaning;
[0136] The data normalization module is used to normalize the real-time operating status data after the standardization process.
[0137] In a certain embodiment, the feature extraction unit includes a time domain feature extraction module, a frequency domain feature extraction module and a wavelet feature extraction module;
[0138] The time domain feature extraction module is used to extract the mean, variance, skewness and kurtosis of the pre-processed real-time operation status data;
[0139] The frequency domain feature extraction module is used to extract the spectrum density and main frequency of the pre-processed real-time operation status data;
[0140] The wavelet feature extraction module is used to extract the wavelet coefficients of the pre-processed real-time running status data.
[0141] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0142] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A pneumatic valve body fault diagnosis method based on multi-model fusion and Bayesian optimization, characterized in that: The following steps are involved: S1: Acquire the real-time operating status data of the pneumatic valve body collected by the sensor group and perform preprocessing; S2: Perform time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed real-time operation status data; S3: Input the time domain features, frequency domain features and wavelet features extracted in S2 into a pre-trained pneumatic valve body fault diagnosis model to obtain a predicted fault diagnosis result of the pneumatic valve body; wherein the predicted fault diagnosis result output by the pre-trained pneumatic valve body fault diagnosis model is a weighted summation result of the LSTM model, the CNN model and the Transformer model; the sum weights of the LSTM model, the CNN model and the Transformer model are obtained based on a Bayesian optimization algorithm.
2. The pneumatic valve body fault diagnosis method according to claim 1, characterized in that: The pre-trained pneumatic valve body fault diagnosis model is obtained based on the following steps: S31: Acquire historical operating status data of the pneumatic valve body collected by the sensor group and perform preprocessing; wherein the historical operating status data of the pneumatic valve body is marked with corresponding fault types; S32: Performing time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed historical operating status data; S33: using the time domain features, frequency domain features and wavelet features extracted in S32 to train the LSTM model, CNN model and Transformer model respectively, and using the Bayesian optimization algorithm to optimize and obtain the best weight combination of the LSTM model, CNN model and Transformer model; The trained LSTM model, CNN model and Transformer model are weightedly summed using the optimal weight combination to obtain the pre-trained pneumatic valve body fault diagnosis model.
3. The pneumatic valve body fault diagnosis method according to claim 1 or 2, characterized in that: The sensor group includes a pressure sensor, a temperature sensor, a flow sensor, an acceleration sensor and a stroke sensor; Wherein, the pressure sensor is used to detect the input and output pressures of the pneumatic valve body; The temperature sensor is used to monitor the temperature of the electromagnetic control mechanism of the pneumatic valve body; The flow sensor is used to detect the fluid flow through the pneumatic valve body; The acceleration sensor is used to detect the vibration signal of the pneumatic valve body; The stroke sensor is used to measure the opening of the valve in real time.
4. The pneumatic valve body fault diagnosis method according to claim 1 or 2, characterized in that: The preprocessing includes data cleaning, data standardization and data normalization.
5. The pneumatic valve body fault diagnosis method according to claim 1 or 2, characterized in that: The time domain features include mean, variance, skewness and kurtosis; the frequency domain features include spectrum density and main frequency; and the wavelet features include wavelet coefficients.
6. The pneumatic valve body fault diagnosis method according to claim 1, characterized in that: The Bayesian optimization algorithm uses a Gaussian process as a proxy model for the objective function; and obtains the best weight combination by maximizing the expected improvement function; The expression of the proxy model is: GP(μ(x),k(x,x′))(11); Where x = (α, β, γ) represents the weight combination of the LSTM model, CNN model and Transformer model; μ(x) represents the mean of the predicted objective function; k(x, x′) represents the covariance function between the weight combination x and x′; GP represents Gaussian process; The expression for maximizing the expected improvement function is: * = argmax x EI(x;GP)(12); Wherein, x* represents the optimal weight combination; EI(x; GP) represents the expected improvement function, that is, the expected improvement in the performance of the predicted fault diagnosis result that can be brought about by the currently selected weight combination x.
7. A pneumatic valve body fault diagnosis system based on multi-model fusion and Bayesian optimization, characterized in that: The pneumatic valve body fault diagnosis method according to any one of claims 1 to 6 comprises a sensor group, a preprocessing unit, a feature extraction unit, and a pre-trained pneumatic valve body fault diagnosis model; The sensor group is used to collect real-time operating status data of the pneumatic valve body; The preprocessing unit is used to preprocess the real-time operating status data collected by the sensor group; The feature extraction unit is used to perform time domain feature extraction, frequency domain feature extraction and wavelet feature extraction on the pre-processed real-time operation status data; The pre-trained pneumatic valve body fault diagnosis model is used to output the predicted fault diagnosis results of the pneumatic valve body according to the time domain features, frequency domain features and wavelet features extracted by the feature extraction unit; wherein the predicted fault diagnosis results output by the pre-trained pneumatic valve body fault diagnosis model are the results of weighted summation of the LSTM model, the CNN model and the Transformer model; the sum weights of the LSTM model, the CNN model and the Transformer model are obtained based on the Bayesian optimization algorithm.
8. The pneumatic valve body fault diagnosis system according to claim 7, characterized in that: The sensor group includes a pressure sensor, a temperature sensor, a flow sensor, an acceleration sensor and a stroke sensor; Wherein, the pressure sensor is used to detect the input and output pressures of the pneumatic valve body; The temperature sensor is used to monitor the temperature of the valve body electromagnetic control mechanism; The flow sensor is used to detect the fluid flow through the pneumatic valve body; The acceleration sensor is used to detect the vibration signal of the pneumatic valve body; The stroke sensor is used to measure the opening of the valve in real time.
9. The pneumatic valve body fault diagnosis system according to claim 8, characterized in that: The preprocessing unit includes a data cleaning module, a data standardization module and a data normalization module; The data cleaning module is used to clean the real-time operating status data collected by the sensor group; The data standardization module is used to perform standardization processing on the real-time operation status data after data cleaning; The data normalization module is used to normalize the real-time operating status data after the standardization process.
10. The pneumatic valve body fault diagnosis system according to claim 8, characterized in that: The feature extraction unit includes a time domain feature extraction module, a frequency domain feature extraction module and a wavelet feature extraction module; The time domain feature extraction module is used to extract the mean, variance, skewness and kurtosis of the pre-processed real-time operation status data; The frequency domain feature extraction module is used to extract the spectrum density and main frequency of the pre-processed real-time operation status data; The wavelet feature extraction module is used to extract the wavelet coefficients of the pre-processed real-time running status data.