Pneumatic system intelligent locator fault diagnosis method, system and equipment based on neural network and medium

Abstract

The invention discloses a pneumatic system intelligent locator fault diagnosis method, system and device based on a neural network and a medium, and belongs to the technical field of power system energy storage equipment, and the method comprises the steps: obtaining the operation data of a pneumatic system through external sensors disposed at a plurality of monitoring positions in the pneumatic system; determining feature data for fault diagnosis according to the operation data; according to the feature data, performing deep feature mining on the time sequence data through a pre-trained neural network model, and determining a fault diagnosis result of the pneumatic system; and determining a fault diagnosis report according to the fault diagnosis result. According to the invention, data acquisition and processing are carried out through the intelligent diagnosis equipment independent of the intelligent locator, interference of a fault diagnosis function on a control function of the intelligent locator is completely avoided, corresponding preprocessing strategies are adopted for different data types, and reliability and timeliness of data acquisition are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power system energy storage equipment, and in particular to a neural network-based pneumatic system intelligent positioner fault diagnosis method, system, equipment and medium. Background Art

[0002] Pneumatic actuators are widely used in industrial production due to their high durability, safety, and responsiveness, especially in harsh operating environments. Intelligent positioners are the core control unit in pneumatic actuator systems. Their popularity has steadily increased in recent years. As the core control structure in pneumatic systems, they are crucial to the stable operation of the system. Consequently, fault diagnosis after positioner failure has become increasingly important.

[0003] Taking the common SIEMENS and Fisher intelligent positioners as an example, in addition to the conventional control functions, they have added fault diagnosis functions for the positioner itself, which can provide information such as abnormal gas source pressure alarm, abnormal command signal alarm, system shock alarm, etc.

[0004] In addition, the core function of the intelligent positioner is control, but it is limited by the power consumption of the 4-20mA power supply. Too many or overly complex additional functions may affect the real-time performance of the control algorithm, making it difficult to allocate more processor resources to diagnostic functions.

[0005] Also due to power limitations, the smart locator itself cannot add additional sensors, so it cannot cover too much system information and cannot perform more comprehensive fault diagnosis. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is: how to achieve comprehensive fault diagnosis of the intelligent positioner of the pneumatic system through independent intelligent diagnostic equipment and neural network deep feature mining technology, avoid the impact of power consumption limitations on control functions, improve the accuracy of fault diagnosis, and expand the scope of sensor information collection.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: a neural network-based pneumatic system intelligent positioner fault diagnosis method, which includes the following steps: obtaining operating data of the pneumatic system through external sensors set at multiple monitoring positions in the pneumatic system; determining characteristic data for fault diagnosis based on the operating data; based on the characteristic data, performing deep feature mining on time series data through a pre-trained neural network model to determine the fault diagnosis results of the pneumatic system; and determining a fault diagnosis report based on the fault diagnosis results.

[0009] As a preferred embodiment of the neural network-based pneumatic system intelligent positioner fault diagnosis method described in the present invention, the method includes obtaining the operating data of the pneumatic system by means of external sensors arranged at multiple monitoring positions in the pneumatic system, and includes the following steps: controlling the external sensors to collect data by means of an intelligent diagnostic device independent of the intelligent positioner; when the data types collected by the external sensors include pressure data and displacement data, performing time synchronization processing on the pressure data and the displacement data; and when the data types collected by the external sensors include signal data, performing signal integrity verification on the signal data.

[0010] The beneficial effects of this preferred technical solution are: by performing data collection and processing on intelligent diagnostic equipment independent of the intelligent locator, the interference of the fault diagnosis function on the control function of the intelligent locator is completely avoided. At the same time, corresponding preprocessing strategies are adopted for different data types to ensure the reliability and timeliness of data collection.

[0011] As a preferred solution of the neural network-based pneumatic system intelligent positioner fault diagnosis method described in the present invention, wherein: determining the characteristic data for fault diagnosis based on the operating data includes the following steps: performing principal component analysis on the operating data; extracting time domain characteristic parameters and frequency domain characteristic parameters from the operating data after principal component analysis; when the variance of the time domain characteristic parameters exceeds a first preset threshold and the frequency variance of the frequency domain characteristic parameters exceeds a second preset threshold, performing feature fusion on the time domain characteristic parameters and the frequency domain characteristic parameters to generate the characteristic data.

[0012] As a preferred solution of the neural network-based pneumatic system intelligent positioner fault diagnosis method described in the present invention, wherein: based on the feature data, deep feature mining is performed on time series data through a pre-trained neural network model to determine the fault diagnosis result of the pneumatic system, including the following steps: time series feature extraction of the feature data through a long short-term memory network to generate a deep feature vector; nonlinear classification processing of the deep feature vector through a support vector machine; when the distribution of the deep feature vector in the feature space meets the fault separation condition, outputting a fault type identifier; when the distribution of the deep feature vector in the feature space meets the health status condition, outputting a health status identifier.

[0013] The beneficial effects of this preferred technical solution are: it adopts an algorithm architecture that combines long short-term memory networks with support vector machines, fully leveraging the advantages of LSTM in time series feature extraction and the accuracy of SVM in nonlinear classification, achieving accurate fault identification through feature space distribution condition judgment, and significantly improving the diagnostic accuracy.

[0014] As a preferred solution of the neural network-based pneumatic system intelligent positioner fault diagnosis method described in the present invention, wherein: determining a fault diagnosis report based on the fault diagnosis result includes the following steps: when the fault diagnosis result indicates the existence of a fault, generating a detailed fault diagnosis report containing fault type information and fault location information; when the fault diagnosis result indicates a healthy state, generating a simplified fault diagnosis report containing a system operation status assessment; outputting the detailed fault diagnosis report or the simplified fault diagnosis report through a display interface or transmitting it to a distributed control system through bus communication.

[0015] As a preferred solution of the neural network-based pneumatic system intelligent positioner fault diagnosis method described in the present invention, the multiple monitoring positions include the air source position, the intelligent positioner air inlet position, the intelligent positioner first air outlet position, the intelligent positioner second air outlet position, the cylinder shaft side position, the cylinder non-shaft side position and the cylinder output shaft position; the external sensors include pressure sensors arranged at the air source position, the intelligent positioner air inlet position, the intelligent positioner first air outlet position, the intelligent positioner second air outlet position, the cylinder shaft side position and the cylinder non-shaft side position, a displacement sensor arranged at the cylinder output shaft position, and a signal sampling sensor arranged at the intelligent positioner command signal and feedback signal; the intelligent diagnostic equipment has an independent power supply module and data processing module, the power supply module does not occupy the power supply of the intelligent positioner, and the data processing module does not occupy the processor resources of the intelligent positioner.

[0016] The beneficial effects of this preferred technical solution are: by deploying different types of special sensors at seven key positions of the pneumatic system, all-round monitoring of the operating status of the pneumatic system is achieved. The independent power supply and processing module design ensures the autonomous operation capability of the diagnostic equipment, providing a complete data basis for comprehensive fault diagnosis.

[0017] As a preferred solution of the neural network-based pneumatic system intelligent positioner fault diagnosis method described in the present invention, wherein: the long short-term memory network includes an input gate, a forgetting gate and an output gate, the input gate controls the input amount of new information, the forgetting gate controls the retention amount of historical information, and the output gate controls the selection amount of output features; the support vector machine adopts a radial basis function as a kernel function, and maps the deep feature vector to a high-dimensional feature space for classification through the kernel function; the fault type identification includes at least one of an axial side leakage identification, a cylinder leakage identification, an air source fault identification and a piston jamming identification; wherein, the fault separation condition refers to that the similarity between the deep feature vector and the preset fault mode vector is greater than the fault judgment threshold, and the health status condition refers to that the similarity between the deep feature vector and the preset health mode vector is greater than the health judgment threshold.

[0018] The present invention provides a pneumatic system intelligent positioner fault diagnosis system based on neural network.

[0019] In order to solve the above technical problems, the present invention further provides the following technical solutions: a pneumatic system intelligent positioner fault diagnosis system based on a neural network, comprising: a data acquisition module, used to obtain the operating data of the pneumatic system through external sensors set at multiple monitoring positions in the pneumatic system; a feature extraction module, used to determine the feature data for fault diagnosis based on the operating data; a diagnostic analysis module, used to perform deep feature mining on time series data through a pre-trained neural network model based on the feature data to determine the fault diagnosis results of the pneumatic system; and a report generation module, used to determine a fault diagnosis report based on the fault diagnosis results.

[0020] The present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the neural network-based pneumatic system intelligent positioner fault diagnosis method are implemented.

[0021] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the neural network-based pneumatic system intelligent positioner fault diagnosis method are implemented.

[0022] The beneficial effects of the present invention are as follows: the present invention avoids the problem of traditional intelligent positioners being limited by 4-20mA power consumption through independent intelligent diagnostic equipment, ensuring that the fault diagnosis function does not affect the real-time control; the use of external sensors at multiple monitoring positions realizes comprehensive system information collection, making up for the defect of insufficient sensors in the intelligent positioner itself; the neural network algorithm combining LSTM and SVM can accurately mine fault characteristics in time series data, improve the accuracy and efficiency of fault diagnosis, and provide reliable protection for the safe and stable operation of the pneumatic system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 An overall flow chart of a neural network-based pneumatic system intelligent positioner fault diagnosis method according to one embodiment of the present invention;

[0025] Figure 2 A schematic diagram of data collection points for a pneumatic actuator in a pneumatic system intelligent positioner fault diagnosis method based on a neural network according to an embodiment of the present invention;

[0026] Figure 3 A diagram showing the main fault classification of a pneumatic actuator according to a neural network-based pneumatic system intelligent positioner fault diagnosis method provided by one embodiment of the present invention;

[0027] Figure 4 A simulated fault flow chart of a neural network-based pneumatic system intelligent positioner fault diagnosis method according to one embodiment of the present invention;

[0028] Figure 5 This is a structural diagram of a full-dimensional unknown input observer for a neural network-based pneumatic system intelligent positioner fault diagnosis method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0029] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0030] Example 1, reference Figure 1, which is the first embodiment of the present invention, provides a pneumatic system intelligent positioner fault diagnosis method based on a neural network, comprising:

[0031] S100: Obtaining the operating data of the pneumatic system through external sensors installed at multiple monitoring locations in the pneumatic system;

[0032] S200: Determining characteristic data for fault diagnosis based on the operating data;

[0033] S300: Based on the feature data, a pre-trained neural network model is used to perform deep feature mining on the time series data to determine the fault diagnosis result of the pneumatic system;

[0034] S400: Determine a fault diagnosis report according to the fault diagnosis result.

[0035] It should be noted that pneumatic actuators are widely used in industrial production due to their advantages such as high durability, safety and response speed, especially in harsh working conditions. As the core control unit in the pneumatic actuator system, the fault diagnosis function of the intelligent positioner faces many technical limitations. Traditional intelligent positioners are limited by the power consumption constraints of the 4-20mA power supply. Too many or overly complex additional functions will affect the real-time performance of the control algorithm, making it difficult to use more processor resources for diagnostic functions. At the same time, due to power limitations, the intelligent positioner itself cannot add additional sensors, so it cannot cover enough system information and it is difficult to achieve comprehensive fault diagnosis. In addition, existing fault diagnosis technologies still have shortcomings in accuracy and coverage, and cannot meet the diagnostic needs under complex working conditions.

[0036] Therefore, to address the above-mentioned problems such as power consumption limitations, incomplete information collection, and insufficient diagnostic accuracy, an intelligent diagnostic device independent of the intelligent positioner is established through steps S100-S400. Comprehensive data collection is achieved through external sensors at multiple monitoring positions to avoid affecting the control function of the intelligent positioner. Data preprocessing technology is used to extract and optimize the features of the operating data to provide high-quality feature data for fault diagnosis. The neural network algorithm combining long short-term memory networks and support vector machines is used to conduct deep feature mining on time series data to accurately identify various fault modes of the pneumatic system. At the same time, a diagnostic report containing the fault type and treatment suggestions is generated based on the fault diagnosis results, providing reliable technical support for the safe and stable operation of the pneumatic system.

[0037] Example 2, reference Figure 3 and Figure 4 , which is the second embodiment of the present invention, provides a pneumatic system intelligent positioner fault diagnosis method based on neural network.

[0038] In the embodiment of the present application, in step S100, external sensors are set at multiple monitoring positions in the pneumatic system to obtain the operating data of the pneumatic system, and data collection and processing are performed through independent intelligent equipment to avoid affecting the control function of the intelligent positioner.

[0039] In an optional embodiment, the operating data of the pneumatic system obtained in step S100 can also be collaboratively collected through a distributed sensor network, multiple sensor nodes are organized into a network topology structure, and the time consistency of the data of each node is ensured through timestamp synchronization technology, which is suitable for distributed monitoring of large pneumatic systems.

[0040] In another optional embodiment, the operating data of the pneumatic system obtained in step S100 can also be remotely collected through a wireless sensor network, combined with low-power wide area network communication technology to achieve wireless transmission and centralized processing of sensor data, which is particularly suitable for industrial environments where on-site wiring is difficult.

[0041] In the embodiment of the present application, step S100 obtains the operating data of the pneumatic system by using external sensors provided at multiple monitoring positions in the pneumatic system, including the following steps A1-A2:

[0042] A1: Data collection is performed by controlling external sensors through intelligent diagnostic equipment independent of the intelligent positioner;

[0043] A2: When the data types collected by the external sensor include pressure data and displacement data, time synchronization processing is performed on the pressure data and displacement data. When the data types collected by the external sensor include signal data, signal integrity verification is performed on the signal data.

[0044] Specifically, in A1, the intelligent device features intelligent algorithm calculations, data acquisition, and information display. This device is completely separated from the intelligent positioner through independent power and data processing modules. The power module does not consume the intelligent positioner's 4-20mA power supply, and the data processing module does not occupy the intelligent positioner's processor resources. The specific layout of external sensors includes: pressure sensors at the air source, positioner air inlet, positioner OUT1 air outlet, positioner OUT2 air outlet, the cylinder's shaft side, and the cylinder's non-shaft side; a displacement sensor at the cylinder's output shaft; signal sampling sensors at the positioner's command and feedback signal outputs; and vibration and temperature sensors at the positioner, air source, and actuator. Data acquisition is achieved through a dedicated data acquisition circuit board, which integrates multiple signal input channels and a data processing unit. It supports the simultaneous acquisition of multiple signal types, including pressure, displacement, current, vibration, and temperature, and transmits the collected data to the CPU for processing.

[0045] Specifically, in A2, time synchronization processing ensures the time correspondence between different types of sensor data, and signal integrity verification uses digital signal processing technology to detect abnormal conditions during signal transmission to ensure the reliability of data acquisition.

[0046] For example, a common expression for a pneumatic system is:

[0047]

[0048] in, represents the state vector, represents the system input, Represents the system output, is a known real constant matrix with a certain dimension.

[0049] This system expression provides a mathematical basis for the state space description of the pneumatic system. The state vector x(t) describes the internal state of the system, the input vector u(t) describes the control signal, and the output vector y(t) describes the measurable system response. It provides a theoretical model support for subsequent fault diagnosis algorithms, enabling intelligent diagnostic equipment to perform accurate state estimation and fault detection based on mathematical models.

[0050] In an optional embodiment, the data collection of the external sensor in step S100 can also be pre-processed on-site through edge computing technology. A microprocessor is deployed at the sensor node to perform preliminary processing such as filtering and denoising on the collected raw data, thereby reducing the amount of data transmission and improving processing efficiency.

[0051] In the embodiment of the present application, determining characteristic data for fault diagnosis based on the operating data in step S200 includes the following steps B1-B3:

[0052] B1: Perform principal component analysis on the operating data;

[0053] B2: Extract time domain characteristic parameters and frequency domain characteristic parameters from the operating data processed by principal component analysis;

[0054] B3: When the variance of the time domain feature parameter exceeds a first preset threshold and the frequency variance of the frequency domain feature parameter exceeds a second preset threshold, the time domain feature parameter and the frequency domain feature parameter are subjected to feature fusion to generate feature data.

[0055] Specifically, in B1, data preprocessing begins with principal component analysis (PCA) to reduce and compress the data, reducing the data volume and increasing processing speed. PCA extracts the main direction of change in the data, removes redundant information, and retains the most valuable characteristic components for fault diagnosis.

[0056] Specifically, in B2, the operational status characteristic quantity evaluation process is then carried out. Characteristic quantities sensitive to fault occurrence are screened from the data preprocessed by principal component analysis. Time or frequency domain analysis can then be performed. This analysis then analyzes the historical operational status and the operational status evaluation characteristic vectors. Time domain characteristic parameters include variance, effective value, absolute mean, root mean square amplitude, skewness, and kurtosis; frequency domain characteristic parameters include mean square frequency and frequency variance. These characteristic parameters can reflect different aspects of the pneumatic system's operational status, providing multi-dimensional information for fault identification.

[0057] Specifically, in B3, the feature fusion process sets threshold conditions to ensure that feature fusion is performed only when both time-domain and frequency-domain features exhibit abnormal changes, preventing accidental fluctuations in a single feature from influencing the diagnostic results. Feature fusion employs a weighted fusion strategy, assigning different weights based on the importance of time-domain and frequency-domain feature parameters. The fused feature vectors are normalized to ensure that feature parameters of different dimensions are effectively combined to form a unified feature representation for subsequent neural network analysis.

[0058] It should be noted that in the technical solution for fault diagnosis of intelligent positioners in pneumatic systems of the present invention, the first preset threshold is the determination threshold for the variance of the time-domain characteristic parameters, and the second preset threshold is the determination threshold for the frequency variance of the frequency-domain characteristic parameters. Both are determined based on the statistical laws of historical data during normal operation of the pneumatic system. Specifically, the first preset threshold is generated by collecting time-domain data such as pressure and displacement (such as the mean value of the pressure signal and the peak value of the displacement) from monitoring points such as the gas source position and the cylinder shaft side, calculating their variance and taking the mean + 3 times the standard deviation to determine whether abnormal fluctuations in the time-domain signal occur (such as sudden changes in the displacement of the cylinder output shaft); the second preset threshold is generated by Fourier transforming the frequency-domain data such as the intelligent positioner command signal, extracting the frequency variance, and using the same statistical method to identify whether there is abnormal energy distribution in the frequency domain (such as gas source pressure mixed with mechanical vibration frequency). For example, when the time domain variance of the cylinder's non-axial pressure signal exceeds 0.08MPa2 (healthy mean 0.05MPa2, standard deviation 0.01MPa2), and the frequency domain variance of the intelligent positioner's air inlet pressure signal exceeds 0.035Hz2 (healthy mean 0.02Hz2, standard deviation 0.005Hz2), the system determines that there are fault characteristics in both the time domain and the frequency domain, triggering feature fusion after principal component analysis to generate composite feature data containing time domain trends and frequency domain components, providing multi-dimensional input for the LSTM network to mine timing fault modes, thereby accurately identifying complex faults such as axial side leakage and piston sticking.

[0059] For example, for a sensor failure, the system model can be expressed as:

[0060]

[0061] in, is the sensor fault direction vector, is a characteristic data vector representing a certain fault.

[0062] The sensor fault model is constructed by introducing the fault term f in the output equation. s ζ(t) can describe changes in sensor conditions such as stuck sensors, constant gain changes, constant deviations, and time-varying deviations. It provides a mathematical basis for intelligent diagnostic equipment to identify sensor-related faults, enabling the system to distinguish between normal operating conditions and abnormal sensor conditions.

[0063] In an optional embodiment, the determination of the characteristic data in step S200 can also be performed by blind source separation through independent component analysis to decompose the mixed sensor signal into independent source signals, which is particularly suitable for complex working conditions where multiple sensor signals interfere with each other.

[0064] In another optional embodiment, the determination of the characteristic data in step S200 can also be performed by performing time-frequency domain analysis through wavelet transform, decomposing the signal into different time scales and frequency scales, and extracting local features of the signal, which is suitable for the analysis of non-stationary signals.

[0065] In the embodiment of the present application, step S300 performs deep feature mining on the time series data using a pre-trained neural network model based on the feature data to determine the fault diagnosis result of the pneumatic system, including the following steps C1-C3:

[0066] C1: Extract time series features from feature data through a long short-term memory network to generate a deep feature vector;

[0067] C2: Perform nonlinear classification processing on the deep feature vector through support vector machine;

[0068] C3: When the distribution of the deep feature vector in the feature space meets the fault separation condition, the fault type identification is output; when the distribution of the deep feature vector in the feature space meets the health state condition, the health state identification is output.

[0069] Specifically, in C1, the Long Short-Term Memory (LSTM) network is a recurrent neural network with memory capabilities specifically designed for processing time series data. LSTM not only addresses the information loss problem in traditional BP networks and CNN algorithms, which rely solely on current input data to determine the output state, but also addresses the shortcomings of traditional RNNs, which are prone to vanishing and exploding gradients when processing long time series data due to their long-term dependency mechanisms. The LSTM network consists of an input gate, a forget gate, and an output gate. The input gate controls the amount of new information input, the forget gate controls the amount of historical information retained, and the output gate controls the amount of output features selected. Through its gating mechanism, the LSTM achieves selective memorization and forgetting of time series information, enabling it to capture subtle changes in fault states as they evolve over time. This makes it particularly suitable for processing the temporal dependencies of fault signals in pneumatic systems.

[0070] Specifically, such as Figure 4 As shown in Figure 2, a new LSTM-based SVM algorithm is designed to address the shortcomings of LSTM networks and traditional shallow machine learning algorithms, enabling rapid fault diagnosis using multi-sensor data fusion. This diagnostic framework consists of three components: the top layer is the raw data input layer; the middle layer is the LSTM feature extraction layer; and the bottom layer is the SVM output layer. By constructing a multi-layer stacked LSTM as a deep feature extractor, deep temporal and spatial feature mining is performed on the raw one-dimensional time series data from multiple sensors. The LSTM memory unit and forgetting mechanism capture subtle changes in the fault state over time. A nonlinear SVM is used as the final classification discriminator, replacing the softmax function in the traditional LSTM, further improving fault diagnosis accuracy. The diagnostic results are directly output by the SVM classifier.

[0071] Specifically, in C3, a fault classification model was established, defining five states: healthy, shaft-side air leakage, cylinder leakage, air source failure, and piston sticking. The fault separation condition requires that the similarity between the deep feature vector and the preset fault pattern vector exceed the fault judgment threshold. The healthy state condition requires that the similarity between the deep feature vector and the preset health pattern vector exceed the health judgment threshold. For each fault state, 30 sets of relevant data were collected under different conditions for model training. The judgment threshold was determined through statistical analysis of the training data, ensuring high diagnostic accuracy and low false alarm rate.

[0072] For example, for the actuator failure case, the system model can be expressed as:

[0073]

[0074] in, is the fault function of the control system, is the fault direction vector of the actuator, is the fault signal vector, for different faults, f i There are many ways to represent the function of (t).

[0075] It should be noted that in the technical solution for fault diagnosis of the intelligent positioner of the pneumatic system of the present invention, the first preset threshold is the judgment threshold of the variance of the time domain characteristic parameter, and the second preset threshold is the judgment threshold of the frequency variance of the frequency domain characteristic parameter. Both are determined based on the statistical laws of historical data during normal operation of the pneumatic system. Specifically, the first preset threshold is generated by collecting time domain data such as pressure and displacement of monitoring points such as the gas source position and the shaft side of the cylinder (such as the mean value of the pressure signal and the peak value of the displacement), calculating its variance and taking "mean + 3 times the standard deviation" to judge whether the time domain signal has abnormal fluctuations (such as sudden changes in the displacement of the cylinder output shaft); the second preset threshold performs Fourier transform on the frequency domain data such as the intelligent positioner command signal, extracts the frequency variance and uses the same statistical method to set it, which is used to identify whether there is abnormal energy distribution in the frequency domain (such as the gas source pressure mixed with the mechanical vibration frequency). For example, when the time domain variance of the cylinder non-shaft side pressure signal exceeds 0.08MPa 2 (Healthy state average 0.05MPa 2 , standard deviation 0.01MPa 2 ), and the frequency domain frequency variance of the intelligent positioner inlet pressure signal exceeds 0.035Hz 2 (Healthy average 0.02Hz 2 , standard deviation 0.005Hz 2 ), the system determines that there are fault features in both the time domain and the frequency domain, triggers the feature fusion after principal component analysis, and generates composite feature data containing time domain trends and frequency domain components, providing multi-dimensional input for the LSTM network to mine timing fault modes, thereby accurately identifying complex faults such as shaft side leakage and piston sticking.

[0076] The actuator fault model introduces the fault term f into the state equation. i n(t) can describe the situation where an actuator fails to operate normally according to commands, providing a theoretical basis for the LSTM-SVM combination algorithm to identify actuator-related faults, enabling the neural network model to learn and identify different types of actuator failure modes, and improving the comprehensiveness and accuracy of fault diagnosis.

[0077] In an optional embodiment, the neural network model in step S300 can also adopt a hybrid architecture of convolutional neural network and recurrent neural network, extracting local features through the convolution layer and capturing temporal dependencies through the recurrent layer, which is suitable for sensor data with dual spatial and temporal characteristics.

[0078] In another optional embodiment, the neural network model in step S300 can also adopt an observer-based fault diagnosis method, by detecting the measurable input and output signals of the system, designing a corresponding state observer, and comparing the output value of the observer with the original output value of the system to obtain a residual signal that is indicative of the fault.

[0079] For example, the mathematical model of the observer can be expressed as:

[0080]

[0081] in, is the observer state vector, is the state estimation vector, Design the matrix for the observer.

[0082] It should be noted that this formula can be compared to the mathematical model of the state observer (or Kalman filter) in the present invention, which is used to solve the problem that some state variables in the pneumatic system (such as cylinder internal pressure and piston displacement) cannot be directly measured. The specific analysis is as follows: In the state observation equation (first row), z(t) is the state vector of the observer, and the internal dynamics of the observer are described by the matrix F (state transfer matrix, related to the system matrix A). The TBu(t) term reflects the input effect of the control signal u(t) (such as the intelligent positioner command signal) on the observer, where T is the input transformation matrix (which can be understood as the gain adjustment of the control quantity), Ky(t) is the feedback correction term, and K is the observer gain matrix. The observation error is corrected in real time through the system measurable output y(t) (such as the measured signal of the pressure sensor and displacement sensor), which corresponds to the data correction mechanism after the "signal integrity verification" in the present invention; in the state estimation equation (second row), It is an estimate of the system state x(t), which is generated by a linear combination (weight matrix H) of the observer state z(t) and the output y(t), to achieve reconstruction of unmeasurable states (such as the pressure difference between the shaft side and the non-shaft side of the cylinder). In the pneumatic system fault diagnosis of the present invention, if some key states (such as displacement deviation caused by piston jamming) cannot be directly obtained through external sensors, this model can be used to estimate the internal state of the system based on measurable signals (such as cylinder output shaft displacement y(t), gas source pressure u(t)), providing more complete input features for the LSTM network, thereby improving the diagnostic ability of hidden faults such as "cylinder leakage" and "piston jamming". The parameters of the model (F, T, K, H) can be determined by the "historical health data statistics" or "neural network training" in the present invention, which complements the "feature fusion after principal component analysis" step to enhance the accuracy of system modeling.

[0083] like Figure 5As shown in the figure, the basic idea of ​​the observer model is to establish new variable values ​​by combining the measurable signals of the system with the model, compare the actual state matrix vector of the measured system with the estimated matrix vector value to generate a deviation value, detect and separate faults through the deviation, and finally identify the features for fault diagnosis. It provides a mathematical tool for traditional model-based fault diagnosis methods and can be combined with neural network methods to form a hybrid diagnosis strategy.

[0084] In the embodiment of the present application, determining a fault diagnosis report according to the fault diagnosis result in step S400 includes the following steps D1-D3:

[0085] D1: When the fault diagnosis result indicates that a fault exists, a detailed fault diagnosis report including fault type information and fault location information is generated;

[0086] D2: When the fault diagnosis result indicates a healthy state, a simplified fault diagnosis report including a system operation status assessment is generated;

[0087] D3: Output the detailed fault diagnosis report or the simplified fault diagnosis report through the display interface or transmit it to the distributed control system through bus communication.

[0088] Specifically, such as Figure 3 As shown in Figure 1, in D1, the detailed fault diagnosis report includes the fault type (including shaft-side leakage, cylinder leakage, air source failure, piston sticking, etc.), fault confidence, occurrence time, possible cause analysis, and fault handling suggestions. In D2, the simplified fault diagnosis report includes an assessment of the overall system health, the status of key operating parameters, and preventive maintenance recommendations. In D3, diagnostic results and diagnostic information are processed through screen prompts and bus communication, and operating information and fault details are presented in a detailed analysis report. Bus communication uses standard industrial communication protocols to ensure reliable data exchange with the DCS system and timely transmission and processing of fault information.

[0089] like Figure 2As shown in the figure, a common double-acting pneumatic actuator system is typically composed of an intelligent positioner, a double-acting cylinder, a controlled valve, and other pneumatic components. Pressure sensors are installed at the air source, positioner air inlet, positioner OUT1 outlet, positioner OUT2 outlet, and the cylinder's shaft and non-shaft sides. A displacement sensor is installed at the cylinder's output shaft. Signal sampling sensors are installed at the positioner's command and feedback signal outputs. Vibration and temperature sensors are installed at the positioner, air source, and actuator. Appropriate pressure, displacement, and current sensors are placed at these data collection points. Data is collected by intelligent devices and transmitted to the CPU for processing. Using real-time sampled data and raw factory data from the equipment, combined with a neural network algorithm, a process involving feature extraction, data analysis, and algorithm calculation is used to determine system status thresholds and determine whether a fault or non-fault state exists. If the algorithm determines a system fault, the neural network algorithm is then used to determine the fault type. Information is displayed on the screen of an external intelligent device, or the diagnostic results are sent to the DCS system via a bus for fault notification.

[0090] For example, 30 sets of relevant data were collected under different conditions for each fault state. The improved LSTM algorithm was used for fault diagnosis, and confusion matrix verification results were obtained. The trained LSTM model achieved high fault classification accuracy and was generally able to effectively distinguish between different states of pneumatic actuators, accurately identifying five states: healthy, shaft-side air leakage, cylinder leakage, air source failure, and piston sticking.

[0091] In an optional implementation, the physical pneumatic system and the virtual model can be synchronized in real time through digital twin technology, and a digital mapping can be established by combining historical data and real-time monitoring data to achieve more accurate fault prediction and health assessment.

[0092] In another optional implementation, an artificial intelligence expert system can be combined to encode the experiential knowledge of domain experts into a rule base, which can be combined with a data-driven neural network model to provide more comprehensive and reliable fault diagnosis capabilities.

[0093] In summary, the present invention effectively solves the problem of power consumption limitation of traditional intelligent positioners by establishing an independent intelligent diagnostic device; realizes comprehensive monitoring of the operating status of the pneumatic system through multi-sensor fusion technology; and adopts advanced neural network algorithms to improve the accuracy and reliability of fault diagnosis, providing important technical support for the safe and stable operation of the pneumatic system.

[0094] Example 3 is the third embodiment of the present invention, which differs from the first two embodiments in that:

[0095] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0096] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a computer-readable medium can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0097] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0098] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0099] Example 4 is the fourth embodiment of the present invention, which provides a pneumatic system intelligent positioner fault diagnosis system based on a neural network, including:

[0100] A data acquisition module is used to obtain operating data of the pneumatic system through external sensors set at multiple monitoring positions in the pneumatic system;

[0101] A feature extraction module is used to determine feature data for fault diagnosis based on the operating data;

[0102] The diagnostic analysis module is used to perform deep feature mining on time series data using a pre-trained neural network model based on feature data to determine the fault diagnosis results of the pneumatic system;

[0103] The report generation module is used to determine the fault diagnosis report according to the fault diagnosis result.

[0104] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A neural network-based fault diagnosis method for intelligent positioners in pneumatic systems, characterized by: include, Acquiring operating data of the pneumatic system through external sensors arranged at multiple monitoring positions in the pneumatic system; determining characteristic data for fault diagnosis based on the operating data; Based on the characteristic data, deep feature mining is performed on the time series data using a pre-trained neural network model to determine a fault diagnosis result of the pneumatic system; A fault diagnosis report is determined according to the fault diagnosis result.

2. The neural network-based pneumatic system intelligent positioner fault diagnosis method according to claim 1, characterized in that: The method of obtaining the operating data of the pneumatic system by using external sensors arranged at multiple monitoring positions in the pneumatic system comprises the following steps: Controlling the external sensor to collect data through an intelligent diagnostic device independent of the intelligent locator; When the data types collected by the external sensor include pressure data and displacement data, performing time synchronization processing on the pressure data and the displacement data; When the data type collected by the external sensor includes signal data, signal integrity verification is performed on the signal data.

3. The neural network-based pneumatic system intelligent positioner fault diagnosis method according to claim 2, characterized in that: Determining characteristic data for fault diagnosis based on the operating data includes the following steps: performing principal component analysis on the operating data; Extracting time domain characteristic parameters and frequency domain characteristic parameters from the operating data processed by principal component analysis; When the variance of the time domain feature parameter exceeds a first preset threshold and the frequency variance of the frequency domain feature parameter exceeds a second preset threshold, the time domain feature parameter and the frequency domain feature parameter are feature-fused to generate the feature data.

4. The neural network-based pneumatic system intelligent positioner fault diagnosis method according to claim 3, characterized in that: The method of performing deep feature mining on time series data using a pre-trained neural network model based on the feature data to determine the fault diagnosis result of the pneumatic system includes the following steps: Performing time series feature extraction on the feature data through a long short-term memory network to generate a deep feature vector; Performing nonlinear classification processing on the depth feature vector by using a support vector machine; When the distribution of the deep feature vector in the feature space meets the fault separation condition, outputting a fault type identifier; When the distribution of the deep feature vector in the feature space meets the health state condition, a health state identifier is output.

5. The neural network-based pneumatic system intelligent positioner fault diagnosis method according to claim 4, characterized in that: Determining a fault diagnosis report according to the fault diagnosis result includes the following steps: When the fault diagnosis result indicates that a fault exists, generating a detailed fault diagnosis report including fault type information and fault location information; When the fault diagnosis result indicates a healthy state, generating a simplified fault diagnosis report including a system operation status assessment; The detailed fault diagnosis report or the simplified fault diagnosis report is output through a display interface or transmitted to a distributed control system through bus communication.

6. The neural network-based pneumatic system intelligent positioner fault diagnosis method according to claim 5, characterized in that: The multiple monitoring positions include the air source position, the air inlet position of the intelligent positioner, the first air outlet position of the intelligent positioner, the second air outlet position of the intelligent positioner, the axial side position of the cylinder, the non-axial side position of the cylinder and the output shaft position of the cylinder; the external sensors include pressure sensors arranged at the air source position, the air inlet position of the intelligent positioner, the first air outlet position of the intelligent positioner, the second air outlet position of the intelligent positioner, the axial side position of the cylinder and the non-axial side position of the cylinder, a displacement sensor arranged at the output shaft position of the cylinder, and a signal sampling sensor arranged at the command signal and feedback signal of the intelligent positioner; the intelligent diagnostic equipment has an independent power supply module and a data processing module, the power supply module does not occupy the power supply of the intelligent positioner, and the data processing module does not occupy the processor resources of the intelligent positioner.

7. The neural network-based pneumatic system intelligent positioner fault diagnosis method according to claim 6, characterized in that: The long short-term memory network includes an input gate, a forget gate and an output gate, the input gate controls the input amount of new information, the forget gate controls the retention amount of historical information, and the output gate controls the selection amount of output features; the support vector machine uses a radial basis function as a kernel function, and maps the deep feature vector to a high-dimensional feature space for classification through the kernel function; the fault type identification includes at least one of an axial side leakage identification, a cylinder leakage identification, an air source fault identification and a piston jam identification; wherein, the fault separation condition refers to that the similarity between the deep feature vector and the preset fault mode vector is greater than the fault judgment threshold, and the health status condition refers to that the similarity between the deep feature vector and the preset health mode vector is greater than the health judgment threshold.

8. A pneumatic system intelligent positioner fault diagnosis system based on a neural network, applying the pneumatic system intelligent positioner fault diagnosis method based on a neural network according to any one of claims 1 to 7, characterized in that: include: A data acquisition module, configured to obtain operating data of the pneumatic system through external sensors disposed at multiple monitoring locations in the pneumatic system; A feature extraction module, configured to determine feature data for fault diagnosis based on the operating data; A diagnostic analysis module, configured to perform deep feature mining on time series data using a pre-trained neural network model based on the feature data to determine a fault diagnosis result for the pneumatic system; The report generating module is used to determine the fault diagnosis report according to the fault diagnosis result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the neural network-based pneumatic system intelligent positioner fault diagnosis method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the neural network-based pneumatic system intelligent positioner fault diagnosis method according to any one of claims 1 to 7 are implemented.

Images