A solenoid valve fault detection system

Abstract

The present invention relates to a solenoid valve fault detection system, which comprises a plurality of sensor modules and a processing unit, and is used for accurately monitoring and analyzing the working state of the solenoid valve; the system integrates an acoustic sensor, a vibration sensor, a current sensor and an environmental sensor, and is used for respectively collecting acoustic signals, vibration signals, current signals and ambient environmental parameters of the solenoid valve during operation; a data processing module filters and denoises the above signals and extracts characteristic signals, and then analyzes them in combination with the changing characteristics of the environmental parameters; a fault analysis module identifies the fault type of the solenoid valve based on these characteristic signals through a pre-trained neural network model, and triggers an alarm and records relevant fault information when a fault is detected; the system can significantly improve the accuracy and real-time performance of solenoid valve fault detection through multi-sensor fusion and intelligent analysis.

Description

Technical Field

[0001] The invention relates to the technical field of automated detection, and in particular to a solenoid valve fault detection system. Background Art

[0002] As an important control element, solenoid valves are widely used in industrial automation systems. In order to ensure the stable operation of the system, the fault detection technology of solenoid valves has been widely studied. Traditional fault detection methods mainly rely on the monitoring of pressure signals, displacement signals and current signals, and judge the working status of the solenoid valve by the changes of these signals. These detection methods can detect the faults of solenoid valves to a certain extent, but in complex industrial environments, especially in the case of high environmental noise or changeable working conditions, the accuracy and timeliness of detection are limited.

[0003] However, traditional fault detection methods often rely on a single or a few physical signals, which makes it difficult to fully capture the diversity and complexity of solenoid valve faults. In addition, existing systems are easily disturbed in complex environments, resulting in false alarms or missed alarms, which affects the overall reliability and safety of the system. In particular, for some potential fault types, such as mechanical wear or impurity intrusion inside the solenoid valve, traditional detection methods are difficult to effectively identify, resulting in increased maintenance costs and increased system downtime.

[0004] Therefore, it is necessary to develop a new solenoid valve fault detection system. Summary of the invention

[0005] The present application provides a solenoid valve fault detection system to improve the accuracy of solenoid valve fault detection.

[0006] The present application provides a solenoid valve fault detection system, comprising:

[0007] An acoustic sensor module, used for collecting acoustic signals generated by the solenoid valve during operation;

[0008] A vibration sensor module, used to collect vibration signals generated by the solenoid valve during operation;

[0009] Current sensor module, used to monitor the current signal of the solenoid valve coil in real time;

[0010] An environmental sensor module, used for monitoring environmental parameters around the solenoid valve;

[0011] A data processing module, electrically connected to the acoustic sensor module, the vibration sensor module, the current sensor module and the environmental sensor module, for filtering and denoising the acoustic signal, the vibration signal and the current signal, and extracting features from the processed acoustic signal, the vibration signal and the current signal to obtain acoustic signal features, vibration signal features and current signal features; and analyzing the environmental parameters to obtain change features of the environmental parameters;

[0012] A fault analysis module is electrically connected to the data processing module and is used to identify the fault type of the solenoid valve based on the acoustic signal characteristics, vibration signal characteristics, current signal characteristics and change characteristics of environmental parameters using a pre-trained neural network model; trigger an alarm and record fault information when a fault is detected.

[0013] Furthermore, the acoustic sensor module includes a plurality of acoustic sensors distributed around the solenoid valve, and each acoustic sensor is configured to collect acoustic signals in different directions to ensure that various types of sound waves generated by the solenoid valve during operation can be accurately captured.

[0014] Furthermore, the vibration sensor module includes at least two vibration sensors with different sensitivities, wherein one vibration sensor is configured to detect small vibration changes for capturing early fault signals, and the other vibration sensor is configured to detect vibrations with larger amplitudes for identifying serious mechanical failures or impacts.

[0015] Furthermore, the current sensor module includes a temperature compensation unit for correcting the current detection signal when the ambient temperature changes to ensure the accuracy of current monitoring.

[0016] Furthermore, the environmental sensor module includes a self-calibration functional unit for self-calibrating the measurement accuracy of the sensor according to a preset time interval or environmental change, so as to improve the long-term stability and accuracy of environmental parameter measurement.

[0017] Furthermore, the data processing module includes a multi-stage filter unit configured to perform hierarchical filtering on the acoustic signal, vibration signal and current signal, first removing high-frequency noise and then further processing the low-frequency signal to ensure that the final output signal characteristics have a high signal-to-noise ratio.

[0018] Furthermore, the neural network model includes:

[0019] An input layer, used to receive and process the acoustic signal features, vibration signal features, current signal features and environmental parameter change features, and output a corresponding feature vector; wherein the corresponding feature vector includes an acoustic vector, a vibration signal vector, a current signal vector and an environmental parameter feature vector;

[0020] A feature fusion layer, used for receiving the corresponding feature vector provided by the input layer; the feature fusion layer includes four parallel self-attention sub-networks, each of which processes a corresponding feature vector respectively, so as to capture the long-range dependency relationship within each signal feature and output a corresponding enhanced feature vector;

[0021] A cross attention layer receives the corresponding enhanced feature vector output by the feature fusion layer, is used to learn the mutual influence and correlation between different types of signal features, and outputs a fused feature vector;

[0022] A time series modeling layer receives the fused feature vector output by the cross attention layer, wherein the time series modeling layer includes a bidirectional gated recurrent unit for capturing the time series variation pattern of the signal feature and outputting a time series feature vector containing time series information;

[0023] The environmental adaptation layer is used to receive the feature vector output by the time series modeling layer and the change characteristics of the environmental parameters in the input layer; dynamically adjust the time series characteristics according to the change characteristics of the environmental parameters, and output an environmentally adapted feature vector;

[0024] The fault classification layer receives the environmental adaptation feature vector output by the environmental adaptation layer. The fault classification layer includes a fully connected layer and a softmax activation function, and is used to output the probability distribution of the fault type of the solenoid valve.

[0025] Furthermore, each self-attention sub-network of the feature fusion layer implements adaptive weighting within the feature through the following formula 1, with the input being a single type of feature vector and the output being an enhanced feature vector of the same dimension:

[0026] ;

[0027] in, Represents the feature vector and The attention weight between represents the total number of input feature vectors; and Represent the input feature vectors and The query matrix and key matrix of ; Represents the input feature vector The autocorrelation coefficient of yes The standardized function of and It is a weighting factor that is dynamically adjusted according to different signal types;

[0028] According to the following formula 2, the enhanced feature vector is calculated :

[0029] ;

[0030] in, is the input feature vector corresponding to The value matrix of Represents the input feature vector The updated version of is the enhanced feature vector after adaptive weighting, and the output enhanced feature vector maintains the same dimension as the input feature vector.

[0031] Furthermore, the cross-attention layer uses the following formula 3 to calculate the cross-attention weights between different enhanced feature vectors to achieve the fusion of multimodal information:

[0032] ;

[0033] in, represents the enhanced feature vector and The cross attention weights between ; and They represent the output of the feature fusion layer. The enhanced eigenvector and the Enhanced feature vectors; is the enhanced eigenvector The dimension of is used to scale the dot product calculation; is the total number of enhanced eigenvectors; and is the dynamic weighting factor; is the cross-correlation function, which is implemented using the following formula 4:

[0034] ;

[0035] in is the adjustment factor used to control the sensitivity of the difference in the nonlinear function; Represents the dimension index of the enhanced feature vector; To enhance the dimension of feature vector;

[0036] According to the following formula 5, the fusion feature vector is calculated :

[0037] ;

[0038] in, Represents the fused feature vector; is the total number of enhanced eigenvectors; Indicates An enhanced feature vector.

[0039] Furthermore, the bidirectional gated recurrent unit of the timing modeling layer captures the bidirectional timing dependency of signal features by updating hidden states in both forward and backward directions, with the input being a fused feature vector sequence and the output being a feature vector containing bidirectional timing information.

[0040] The beneficial effects of the present application mainly include: (1) The system combines acoustic signals, vibration signals, current signals and environmental parameters, and through comprehensive analysis of multi-dimensional signals, effectively improves the accuracy of identifying solenoid valve faults, can comprehensively capture various possible fault types, and avoid the limitations of single signal detection. (2) The current sensor and data processing module equipped with the system can monitor the working status of the solenoid valve in real time, and quickly process and analyze the signal. When a fault is detected, the system can trigger an alarm and record fault information in time, thereby greatly shortening the fault detection and response time and reducing system downtime. (3) By using a pre-trained neural network model for fault analysis, the system can automatically identify complex fault modes without manual intervention, significantly improving the efficiency and accuracy of fault diagnosis, while reducing the possibility of false alarms and missed alarms. (4) The system's environmental sensor module can monitor the environmental parameters around the solenoid valve, and the data processing module can adaptively adjust other signals based on environmental changes, so that in a complex environment with large noise and vibration, it can still stably and reliably perform fault detection and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of a solenoid valve fault detection system provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0042] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0043] The first embodiment of the present application provides a solenoid valve fault detection system. Figure 1 , which is a schematic diagram of the first embodiment of the present application. Figure 1 A solenoid valve fault detection system is provided in detail in accordance with the first embodiment of the present application.

[0044] The solenoid valve fault detection system includes an acoustic sensor module 101 , a vibration sensor module 102 , a current sensor module 103 , an environmental sensor module 104 , a data processing module 105 and a fault analysis module 106 .

[0045] The acoustic sensor module 101 is used to collect acoustic signals generated by the solenoid valve during operation.

[0046] The acoustic sensor module 101 is a key component in the solenoid valve fault detection system, which is designed to collect the acoustic signals generated by the solenoid valve during operation, thereby providing the necessary data support for accurate detection and diagnosis of solenoid valve faults. The module usually includes one or more highly sensitive acoustic sensors that can accurately capture the sound wave signals generated by the solenoid valve during operation within a wide frequency range. The acoustic sensor can be a piezoelectric sensor, an electret microphone, or other suitable sound wave capture device, and the specific selection depends on the noise characteristics and signal requirements of the solenoid valve working environment.

[0047] During the operation of the solenoid valve, the acoustic sensor module 101 is installed on or near the external structure of the solenoid valve so as to accurately obtain the acoustic signals generated by the mechanical actions inside the valve (such as valve core movement, spring compression or relaxation). These acoustic signals usually include the impact sound when the valve is opened and closed, the mechanical friction sound, and possible abnormal sounds, such as noise caused by sticking, vibration, looseness or valve core deformation. By converting these acoustic signals into electrical signals and transmitting them to the data processing module 105, the system can further filter and denoise the signals to extract characteristic information related to the solenoid valve failure.

[0048] The design of the acoustic sensor module 101 not only considers the sensitivity of signal acquisition, but also pays special attention to anti-interference ability to ensure that accurate acoustic data can be captured in the presence of complex background noise in industrial environments. The module can integrate noise shielding technology or adopt a multi-sensor array structure to enhance the accuracy of signal acquisition and improve the adaptability of the system under different working conditions.

[0049] By combining the acoustic signal with other signals (such as vibration signals and current signals), the acoustic sensor module 101 provides the fault analysis module 106 with a more comprehensive fault diagnosis basis.

[0050] Furthermore, the acoustic sensor module includes a plurality of acoustic sensors distributed around the solenoid valve, and each acoustic sensor is configured to collect acoustic signals in different directions to ensure that various types of sound waves generated by the solenoid valve during operation can be accurately captured.

[0051] In the solenoid valve fault detection system, the design of the acoustic sensor module is particularly critical, especially in complex industrial environments. Accurately capturing and analyzing various acoustic signals generated during the operation of the solenoid valve can greatly improve the accuracy of fault detection. The acoustic sensor module includes multiple acoustic sensors distributed around the solenoid valve, each of which is configured to collect acoustic signals in different directions to ensure that the system can comprehensively and accurately capture all acoustic characteristics of the solenoid valve during operation.

[0052] To achieve this goal, multiple sensors in the acoustic sensor module are usually arranged at different positions or orientations of the solenoid valve, such as on the front, side and back of the solenoid valve, or above and below the solenoid valve. This arrangement ensures that the sensor can capture sound waves propagating from all directions, which is crucial for detecting different types of acoustic signals, especially in complex mechanical structures, where sounds at different positions may reflect different operating conditions or fault types.

[0053] Each acoustic sensor is carefully configured with a specific sensitivity and frequency response range to suit the acoustic characteristics of a specific location. For example, a sensor near the high-frequency vibration area of ​​the solenoid valve can be configured with a higher frequency response, while a sensor near the low-frequency vibration area can be configured with a wider frequency response to ensure that different types of acoustic signals can be effectively captured and transmitted to the subsequent signal processing unit.

[0054] These sensors not only simply collect acoustic signals, but may also integrate pre-processing functions, such as preliminary signal filtering and amplification processing, to improve signal quality and reduce interference from environmental noise. During the signal transmission process, these pre-processed signals will be further transmitted to the system's central data processing module for comprehensive analysis. Through such multi-point and multi-angle signal collection and processing, the acoustic sensor module can provide rich and high-quality acoustic data for fault detection, greatly improving the accuracy and reliability of solenoid valve fault analysis.

[0055] The vibration sensor module 102 is used to collect vibration signals generated by the solenoid valve during operation.

[0056] The vibration sensor module 102 is an indispensable component in the solenoid valve fault detection system. Its main function is to collect the vibration signal generated by the solenoid valve during operation, thereby providing key data for fault detection. The vibration sensor is usually installed outside or near the solenoid valve so that it can directly sense the vibration generated by the mechanical action inside the valve. The module uses high-precision vibration sensors, which can be piezoelectric accelerometers, MEMS sensors or other devices suitable for vibration measurement. The specific selection depends on the working environment and detection requirements of the solenoid valve.

[0057] The vibration sensor module 102 is designed with special consideration of its sensitivity and frequency response to ensure that it can capture various vibration signals generated by the solenoid valve under different operating conditions, including small vibrations in normal operation and larger vibrations caused by faults or abnormalities. Normally, the solenoid valve will produce specific vibration patterns when it is opened, closed or maintained, and these patterns are directly related to the mechanical state of the solenoid valve. When the solenoid valve fails, such as stuck, valve core wear, spring fatigue or loose internal components, the vibration characteristics will change significantly. By monitoring these changes in real time, the vibration sensor module 102 can detect potential fault signals at an early stage.

[0058] The vibration signal collected by this module is transmitted in the form of an electrical signal to the data processing module 105, where the vibration signal is first filtered and denoised to remove background noise and irrelevant vibrations. Subsequently, the system extracts features from the processed signal, including amplitude, frequency, waveform, and time domain analysis. These features are compared with pre-stored normal operating data to identify any abnormal conditions.

[0059] The design of the vibration sensor module 102 also focuses on anti-interference capability to ensure that the vibration characteristics of the solenoid valve can still be accurately captured in a noisy industrial environment. Through comprehensive analysis with other signal sources (such as acoustic signals and current signals), the vibration sensor module 102 provides important reference data for the fault analysis module 106, which helps to improve the accuracy and reliability of fault diagnosis.

[0060] Furthermore, the vibration sensor module includes at least two vibration sensors with different sensitivities, wherein one vibration sensor is configured to detect small vibration changes for capturing early fault signals, and the other vibration sensor is configured to detect vibrations with larger amplitudes for identifying serious mechanical failures or impacts.

[0061] In the solenoid valve fault detection system, the design of the vibration sensor module is one of the key factors to achieve efficient and accurate fault detection. The vibration sensor module includes at least two vibration sensors with different sensitivities, and the configuration and purpose of each sensor have its specific functions to ensure that the system can detect different types of vibration signals during the operation of the solenoid valve and effectively identify potential faults.

[0062] First, the first vibration sensor configured in the vibration sensor module has high sensitivity and is specifically designed to detect tiny vibration changes. This highly sensitive sensor is able to capture subtle vibration signals generated during the operation of the solenoid valve, especially in the early stages of failure. For example, slight wear, looseness, or mechanical misalignment within a small range of the internal components of the solenoid valve may only cause very subtle vibrations, and these subtle vibrations are often the manifestation of the early stages of failure. If these signals are detected in time, the system can issue an alarm and perform maintenance before the fault worsens, thereby avoiding more serious mechanical damage and system downtime. Highly sensitive vibration sensors are typically designed to detect extremely small acceleration changes and may respond at microgravity levels.

[0063] Secondly, the vibration sensor module also includes a vibration sensor with lower sensitivity but wider range, which is used to detect vibration signals with larger amplitudes. Such vibration signals are usually associated with more serious mechanical failures or severe mechanical shocks. For example, when there is obvious component failure, mechanical shock or severe loosening of parts inside the solenoid valve, these faults will cause the solenoid valve to produce large vibrations. This type of vibration signal has a large amplitude and a wide frequency range. The low-sensitivity sensor can effectively capture these signals and quickly identify the fault type through the system's analysis module, thereby triggering the corresponding fault handling measures. This sensor is usually able to detect large amplitude vibrations and can remain stable even in harsh industrial environments.

[0064] By configuring these two vibration sensors with different sensitivities, the vibration sensor module can achieve multi-level vibration monitoring, which can not only keenly capture early subtle fault signals, but also effectively detect obvious mechanical faults or impacts. This multi-level monitoring strategy enables the solenoid valve fault detection system to more comprehensively cover faults of different types and severity, thereby improving the reliability and safety of the system.

[0065] The current sensor module 103 is used to monitor the current signal of the solenoid valve coil in real time.

[0066] The current sensor module 103 is one of the key components in the solenoid valve fault detection system, and is specifically used to monitor the current signal of the solenoid valve coil in real time. The module detects the current change in the solenoid valve coil through a high-precision current sensor, thereby reflecting the working status of the solenoid valve. The current sensor usually adopts a Hall effect sensor, a shunt or other suitable current detection technology, and the specific selection depends on the electrical characteristics of the solenoid valve and the system requirements.

[0067] During the operation of the solenoid valve, the current sensor module 103 is directly installed in the power supply circuit of the solenoid valve so that the current of the coil can be monitored in real time. When the solenoid valve is in normal working condition, the current of the coil appears as a stable working current and will fluctuate within a certain range according to the opening and closing state of the valve. However, when the solenoid valve fails, such as the coil short circuit, open circuit or overheating, the current signal will change abnormally. By capturing these current changes in real time, the current sensor module 103 can provide key diagnostic information, which helps to detect potential faults early.

[0068] The current signal collected by the current sensor module 103 is converted into a voltage signal and transmitted to the data processing module 105. In the data processing module, these signals are filtered and denoised to eliminate electromagnetic interference and other irrelevant noises to ensure that the obtained current characteristic signals are accurate and reliable. The processed current signal characteristics can reveal the working conditions of the solenoid valve under different operating conditions, including starting current, operating current, and possible fault current characteristics. By analyzing these characteristic signals, the system can determine whether the solenoid valve is in a normal working state or whether there are abnormalities that may cause failures.

[0069] The design of the current sensor module 103 also takes into account its high reliability and long-term stable working ability, especially in industrial environments with large current fluctuations, it can still maintain accurate current detection performance. By working in conjunction with other modules (such as the acoustic sensor module 101 and the vibration sensor module 102), the current sensor module 103 provides important electrical signal basis for system fault analysis, significantly improving the accuracy and timeliness of fault diagnosis.

[0070] Furthermore, the current sensor module includes a temperature compensation unit for correcting the current detection signal when the ambient temperature changes to ensure the accuracy of current monitoring.

[0071] The current sensor module design includes a temperature compensation unit to correct the current detection signal when the ambient temperature changes to ensure the accuracy of current monitoring.

[0072] The core function of the temperature compensation unit is to monitor the changes in ambient temperature in real time and dynamically adjust the output signal of the current sensor according to these changes. Specifically, the current sensor module usually contains one or more precision temperature sensors, which are responsible for detecting the temperature of the environment around the solenoid valve. When the ambient temperature rises or falls, the temperature compensation unit will correct the current detection signal according to the preset correction curve or algorithm to offset the impact of temperature changes on the measurement accuracy of the current sensor. For example, in a high temperature environment, the current sensor may cause measurement errors due to thermal expansion or changes in the temperature coefficient of electronic components. At this time, the temperature compensation unit adjusts the gain or offset of the current signal to ensure that the output signal remains accurate.

[0073] In addition, the temperature compensation unit can be tightly integrated with the internal electronic circuit of the current sensor to achieve a fast-response signal correction mechanism. Through the combination of real-time temperature monitoring and automatic correction mechanism, the current sensor module can maintain high-precision current measurement over a wide temperature range, especially in industrial environments with frequent temperature changes. This compensation mechanism is particularly important. In this way, even if the working environment of the solenoid valve has large temperature fluctuations, the system can still accurately monitor current changes and promptly detect potential current anomalies or fault signals.

[0074] By configuring the temperature compensation unit, it can ensure that the current sensor module can provide stable and accurate signal output under various temperature conditions, thereby greatly improving the reliability and accuracy of the solenoid valve fault detection system. This design is not only suitable for standard industrial environments, but can also be extended to special working conditions with extreme temperatures or large fluctuations, thus meeting a wider range of application needs.

[0075] The environmental sensor module 104 is used to monitor environmental parameters around the solenoid valve.

[0076] The environmental sensor module 104 plays a key role in the solenoid valve fault detection system. Its main function is to monitor the environmental parameters around the solenoid valve in real time and provide necessary background data for fault detection and analysis. This module usually includes a variety of sensors to measure temperature, humidity, pressure and other environmental factors that may affect the working state of the solenoid valve. The selection and configuration of the sensor depends on the specific application scenario. For example, in a high temperature environment, the accuracy and tolerance of the temperature sensor will become the focus of the design, while in a humid or corrosive gas environment, the protection of the humidity and gas sensors will receive special attention.

[0077] In terms of installation location, the environmental sensor module 104 is usually arranged near the solenoid valve or in the equipment structure connected to the solenoid valve to ensure that the microenvironmental conditions of the solenoid valve can be accurately reflected. By monitoring the ambient temperature, the system can determine whether there is a risk of solenoid valve performance degradation or failure due to overheating; the humidity sensor can detect humidity changes in the environment, especially in the presence of condensation or corrosive gases, humidity changes may cause damage to the internal components of the solenoid valve. The pressure sensor can be used to monitor external pressure changes to help identify abnormal valve operation caused by external pressure fluctuations.

[0078] All data collected by the environmental sensor module 104 will be transmitted to the data processing module 105, which analyzes the data to extract the change characteristics of the environmental parameters. The data processing module can comprehensively analyze these environmental parameters with the operating signals of the solenoid valve (such as acoustic signals, vibration signals, and current signals), thereby improving the accuracy of fault detection. For example, when the ambient temperature exceeds the designed operating temperature range of the solenoid valve, even if the current and vibration signals do not show abnormalities, the system can warn of potential overheating fault risks.

[0079] In addition, the design of the environmental sensor module 104 must also take into account the durability and long-term stability of the sensor, especially when the solenoid valve may be exposed to extreme environmental conditions. The sensor should have anti-interference capabilities to ensure that it can continue to provide reliable data in complex industrial environments. By working in conjunction with other sensor modules, the environmental sensor module 104 provides important environmental background data for system fault analysis, helps identify potential faults caused by environmental changes, and ensures the long-term stable operation of the solenoid valve.

[0080] Furthermore, the environmental sensor module includes a self-calibration functional unit for self-calibrating the measurement accuracy of the sensor according to a preset time interval or environmental change, so as to improve the long-term stability and accuracy of environmental parameter measurement.

[0081] The environmental sensor module is specially designed with a self-calibration function unit, which is used to self-calibrate the measurement accuracy of the sensor according to preset time intervals or environmental changes to ensure the long-term stability and accuracy of environmental parameter measurements.

[0082] The self-calibration function unit is designed to overcome the measurement errors caused by drift, aging or environmental factors in traditional sensors during long-term use. Usually, sensors may gradually deviate from their initial calibration state during long-term operation, especially in complex or harsh environments, which will affect the accuracy of the measurement results. To cope with this situation, the self-calibration function unit automatically adjusts the measurement accuracy of the sensor regularly.

[0083] The self-calibration process usually includes the following steps. First, the system triggers the self-calibration function unit at a preset time interval, or automatically activates the self-calibration function when a significant change in environmental parameters is detected (such as a sudden increase or decrease in temperature or humidity). During the calibration process, the self-calibration function unit compares the current output of the sensor with an internal reference standard (for example, a preset reference value or a known environmental parameter). Based on the comparison results, the system adjusts the output signal of the sensor to bring it back to the standard calibration state.

[0084] This automated calibration process ensures that the environmental sensor module can maintain high-precision measurement performance during long-term use, reducing measurement errors caused by sensor drift or environmental changes. This function is particularly important in industrial environments that require continuous monitoring, and can effectively improve the reliability and accuracy of the entire solenoid valve fault detection system.

[0085] Through this design, it can be ensured that the environmental sensor module maintains its measurement stability and accuracy in long-term operation. Even under harsh working conditions, the system can provide reliable data support, providing a solid foundation for fault detection and preventive maintenance of the solenoid valve.

[0086] The data processing module 105 is electrically connected to the acoustic sensor module, the vibration sensor module, the current sensor module and the environmental sensor module, and is used to filter and denoise the acoustic signal, the vibration signal and the current signal, and respectively extract features from the processed acoustic signal, the vibration signal and the current signal to obtain acoustic signal features, vibration signal features and current signal features; and analyze the environmental parameters to obtain change features of the environmental parameters.

[0087] The data processing module 105 is the core component of the solenoid valve fault detection system, which is responsible for processing the signals collected from multiple sensor modules to extract useful feature information and provide accurate input data for fault analysis. This module is connected to the acoustic sensor module 101, the vibration sensor module 102, the current sensor module 103 and the environmental sensor module 104, and can receive and process the original signals collected by these sensors.

[0088] In specific operation, the data processing module 105 first pre-processes the signals from each sensor, including filtering and denoising, to eliminate possible environmental interference and noise, and ensure the clarity and accuracy of the signal. The filtering process usually uses a digital filter, such as a low-pass filter or a band-pass filter, which is selected according to the characteristics of the signal and the type of noise to be eliminated. De-noising can be achieved through various algorithms, such as the average method, wavelet transform or other signal processing techniques, with the aim of retaining useful information in the signal while removing irrelevant noise components.

[0089] After signal preprocessing, the data processing module 105 continues to extract features from the processed signal. This process includes extracting acoustic features from acoustic signals, such as spectral features and sound pressure level; extracting vibration features from vibration signals, such as amplitude, frequency, and acceleration; extracting current features from current signals, such as current amplitude and waveform changes; and extracting temperature, humidity, or pressure change features from environmental parameters. Each feature can reflect different states of the solenoid valve during operation and provide early warning information about potential failures.

[0090] In the data processing module 105, all extracted signal features will be integrated and transmitted to the fault analysis module 106 for further fault identification and diagnosis. These signal features can not only reflect the current working state of the solenoid valve, but also identify the gradually evolving fault trend through historical data comparison to support predictive maintenance.

[0091] The design of the data processing module 105 also needs to consider real-time performance and processing efficiency, especially in complex industrial environments, where the system must be able to complete signal processing and feature extraction in a short time to ensure the timeliness of fault analysis. In addition, the algorithm of this module should be highly robust to cope with signal fluctuations under different working conditions.

[0092] Furthermore, the data processing module includes a multi-stage filter unit configured to perform hierarchical filtering on the acoustic signal, vibration signal and current signal, first removing high-frequency noise and then further processing the low-frequency signal to ensure that the final output signal characteristics have a high signal-to-noise ratio.

[0093] The data processing module includes a multi-stage filter unit, which is configured to perform hierarchical filtering on acoustic signals, vibration signals and current signals to ensure that the final output signal features have a high signal-to-noise ratio. The goal of this design is to effectively remove noise from interference signals while retaining and extracting useful information related to the working status of the solenoid valve.

[0094] The multi-stage filter unit first performs preliminary high-frequency noise removal on the received signal. In practical applications, sensor signals from industrial environments are often interfered with by various high-frequency noises, such as high-frequency components from electromagnetic interference (EMI) or mechanical vibration. These high-frequency noises may mask or distort the effective signals collected by the sensor, affecting the system's accurate assessment of the working status of the solenoid valve. To solve this problem, the first layer of the multi-stage filter unit usually uses a low-pass filter, which can effectively suppress these high-frequency interferences, thereby retaining the effective signals in the lower frequency range.

[0095] After the high-frequency noise is suppressed, the signal enters the next stage of the multi-stage filter unit, where the filter further processes the low-frequency signal. The goal of low-frequency signal processing is to remove low-frequency interference in the signal, which may include low-frequency mechanical noise or slowly changing background signals caused by the solenoid valve structure or working environment. This step ensures that the final output signal is clear and contains key information related to the working status of the solenoid valve. This hierarchical filtering strategy can effectively improve the signal-to-noise ratio of the signal, allowing the data processing module to extract more accurate and reliable feature information from the sensor signal.

[0096] Through the design of multi-stage filter units, the data processing module can significantly improve the performance of the entire solenoid valve fault detection system, especially in complex industrial environments. It not only reduces the interference of environmental noise on the signal, but also ensures the accurate extraction of key signal features, thus providing a solid data foundation for subsequent fault analysis.

[0097] The fault analysis module 106 is electrically connected to the data processing module, and is used to identify the fault type of the solenoid valve based on the acoustic signal characteristics, vibration signal characteristics, current signal characteristics and change characteristics of environmental parameters using a pre-trained neural network model; trigger an alarm and record fault information when a fault is detected.

[0098] The fault analysis module 106 is one of the core components in the solenoid valve fault detection system. Its main task is to accurately identify the fault type of the solenoid valve using a pre-trained neural network model based on the various signal features obtained from the data processing module 105. The module processes the acoustic signal features, vibration signal features, current signal features, and environmental parameter change features, and comprehensively analyzes these data to determine the operating status of the solenoid valve.

[0099] The fault analysis module 106 first receives processed signal features from the data processing module, including acoustic features (such as spectrum analysis, sound pressure changes), vibration features (such as acceleration, frequency fluctuations), current features (such as current waveform, amplitude) and environmental parameter features (such as temperature, humidity changes). These features are generated by filtering, denoising, feature extraction and other steps of the raw data collected by multiple sensor modules, providing the working performance of the solenoid valve in different dimensions.

[0100] During the analysis process, the fault analysis module 106 uses a pre-trained neural network model, which is trained with a large amount of historical data to enable it to identify various types of faults that may occur in the solenoid valve, such as valve core sticking, coil overheating, mechanical wear or abnormalities caused by environmental influences. The neural network model can process complex signal feature combinations through multi-layer nonlinear mapping, thereby achieving high-precision fault diagnosis.

[0101] When the module detects that the signal characteristics match a known fault pattern, the fault analysis module 106 immediately triggers a fault alarm and records detailed fault information. This information includes the fault type, occurrence time, involved signal characteristics, etc., which are stored in the system database to facilitate subsequent fault analysis and maintenance decisions. The real-time design of the system ensures that faults can be detected and warned at the early stage of occurrence, thereby reducing potential equipment damage and production downtime.

[0102] The design of the fault analysis module 106 also takes scalability into consideration. By regularly updating the neural network model and adding new fault modes, the system can continuously improve its diagnostic capabilities and adapt to the complex operating environment and changing working conditions of the solenoid valve.

[0103] Furthermore, the neural network model includes:

[0104] An input layer, used to receive and process the acoustic signal features, vibration signal features, current signal features and environmental parameter change features, and output a corresponding feature vector; wherein the corresponding feature vector includes an acoustic vector, a vibration signal vector, a current signal vector and an environmental parameter feature vector;

[0105] A feature fusion layer, used for receiving the corresponding feature vector provided by the input layer; the feature fusion layer includes four parallel self-attention sub-networks, each of which processes a corresponding feature vector respectively, so as to capture the long-range dependency relationship within each signal feature and output a corresponding enhanced feature vector;

[0106] A cross attention layer receives the corresponding enhanced feature vector output by the feature fusion layer, is used to learn the mutual influence and correlation between different types of signal features, and outputs a fused feature vector;

[0107] A time series modeling layer receives the fused feature vector output by the cross attention layer, wherein the time series modeling layer includes a bidirectional gated recurrent unit for capturing the time series variation pattern of the signal feature and outputting a time series feature vector containing time series information;

[0108] The environmental adaptation layer is used to receive the feature vector output by the time series modeling layer and the change characteristics of the environmental parameters in the input layer; dynamically adjust the time series characteristics according to the change characteristics of the environmental parameters, and output an environmentally adapted feature vector;

[0109] The fault classification layer receives the environmental adaptation feature vector output by the environmental adaptation layer. The fault classification layer includes a fully connected layer and a softmax activation function, and is used to output the probability distribution of the fault type of the solenoid valve.

[0110] In the solenoid valve fault detection system, the design and structure of the neural network model are crucial to achieve high-precision fault detection. The neural network model consists of multiple functional layers, each with specific functions and roles, ensuring that the system can effectively process and analyze multiple input signal features, thereby accurately identifying the fault type of the solenoid valve.

[0111] First, the input layer is the initial processing unit of the neural network model, and its main function is to receive signal features from different sensor modules. These features include acoustic signal features, vibration signal features, current signal features, and environmental parameter change features. The input layer does not simply pass these signals, it also preprocesses each feature, such as standardization or normalization, to ensure that the subsequent processing layer can analyze these features on a unified scale. The processed features are output in the form of vectors, each of which corresponds to an input signal feature, forming an acoustic vector, a vibration signal vector, a current signal vector, and an environmental parameter feature vector.

[0112] Next, the feature fusion layer receives and processes the individual feature vectors provided by the input layer. The feature fusion layer processes each type of feature vector separately through four parallel self-attention sub-networks. Each self-attention sub-network is dedicated to capturing the long-range dependencies within the respective signal features, which means that it is able to identify the long-range correlations between different elements in the feature vector. In this way, the network can better understand the complex patterns and changing trends in the signal. The self-attention mechanism highlights important features in a dynamic weighted manner and ultimately outputs enhanced feature vectors, which contain richer signal information and help to more accurately analyze the working status of the solenoid valve.

[0113] After the feature fusion layer is processed, the enhanced feature vector is passed to the cross-attention layer. The main function of this layer is to learn the mutual influence and correlation between different types of signal features. For example, changes in acoustic signals may be associated with changes in vibration signals, and fluctuations in environmental parameters may also affect the characteristics of current signals. The cross-attention layer calculates the interaction between these features and fuses the individual enhanced feature vectors into a comprehensive fused feature vector. This fused feature vector contains comprehensive information from multiple signal sources and is an important basis for the system to conduct further analysis.

[0114] The time series modeling layer receives the fused feature vector output by the cross-attention layer, and its task is to capture the time series change pattern of signal features. This layer uses a bidirectional gated recurrent unit (GRU) to process time series data. This design can simultaneously consider the front-end and back-end dependencies of signals in the time dimension. The bidirectional GRU can not only process the impact of past signals, but also capture the potential impact of future signals on the current state, thereby generating a time series feature vector containing time series information. This time series feature vector provides key dynamic information for subsequent fault classification.

[0115] After the time series modeling layer, the environmental adaptation layer receives the time series feature vector and dynamically adjusts the time series feature in combination with the environmental parameter change characteristics in the input layer. This design takes into account the impact of environmental factors on the working state of the solenoid valve. By adjusting the time series features according to the changes in environmental parameters, the system can more accurately reflect the state changes in the actual working environment. The feature vector after environmental adaptation more accurately represents the working state of the solenoid valve under specific environmental conditions.

[0116] Finally, the fault classification layer receives the adapted feature vector output by the adaptation layer. This layer includes a fully connected layer and a softmax activation function. The fully connected layer maps the high-dimensional feature vector to the fault category space, while the softmax activation function converts the output into a probability distribution of each fault type. In this way, the system can provide a probability value for each possible fault type, thereby determining the most likely fault of the solenoid valve.

[0117] Overall, this neural network model can extract key information related to solenoid valve failure from complex multi-source signals and accurately identify the specific fault type through in-depth processing and analysis. This design is not only highly accurate, but also highly adaptable, and can provide stable and reliable fault detection results in different working environments.

[0118] The following is the reference implementation code for the neural network model design.

[0119] import torch

[0120] import torch.nn as nn

[0121] import torch.nn.functional as F

[0122] # Define the neural network model class

[0123] class ValveFaultDetectionNN(nn.Module):

[0124] def __init__(self, input_dim, hidden_dim, output_dim, num_classes):

[0125] super(ValveFaultDetectionNN, self).__init__()

[0126] # Input layer, used to receive and process acoustic, vibration, current and environmental parameter features

[0127] self.input_layer = nn.Linear(input_dim, hidden_dim)

[0128] # Feature fusion layer, including four parallel self-attention sub-networks

[0129] self.attention_layers = nn.ModuleList([

[0130] SelfAttentionNetwork(hidden_dim) for _ in range(4) ])

[0132] # Cross attention layer, used to fuse different types of enhanced feature vectors

[0133] self.cross_attention = CrossAttentionLayer(hidden_dim)

[0134] # Temporal modeling layer, using bidirectional gated recurrent unit (GRU)

[0135] self.gru = nn.GRU(hidden_dim, hidden_dim, batch_first=True,bidirectional=True)

[0136] #Environmental adaptation layer, used to dynamically adjust the time series feature vector

[0137] self.environment_adaptation = EnvironmentAdaptationLayer(hidden_dim)

[0138] # Fault classification layer, including fully connected layer and softmax activation function

[0139] self.fc = nn.Linear(hidden_dim, num_classes)

[0140] self.softmax = nn.Softmax(dim=-1)

[0141] def forward(self, acoustic, vibration, current, environment):

[0142] # Input layer processes and generates corresponding feature vectors

[0143] x_acoustic = self.input_layer(acoustic)

[0144] x_vibration = self.input_layer(vibration)

[0145] x_current = self.input_layer(current)

[0146] x_environment = self.input_layer(environment)

[0147] # Feature fusion layer, which performs self-attention processing on different signal features

[0148] x_acoustic = self.attention_layers[0](x_acoustic)

[0149] x_vibration = self.attention_layers[1](x_vibration)

[0150] x_current = self.attention_layers[2](x_current)

[0151] x_environment = self.attention_layers[3](x_environment)

[0152] # Cross attention layer, fusion of each enhanced feature vector

[0153] fused_features = self.cross_attention(x_acoustic, x_vibration, x_current, x_environment)

[0154] # Time series modeling layer, capturing time series features

[0155] gru_output, _ = self.gru(fused_features.unsqueeze(0)) # Increase batch dimension

[0156] # Environmental adaptation layer, adjusts the timing characteristics according to environmental parameters

[0157] adapted_features = self.environment_adaptation(gru_output,environment)

[0158] # Fault classification layer, output fault type probability distribution

[0159] logits = self.fc(adapted_features.squeeze(0)) # Remove the batch dimension

[0160] probs = self.softmax(logits)

[0161] return probs

[0162] # Self-attention subnetwork to capture long-range dependencies

[0163] class SelfAttentionNetwork(nn.Module):

[0164] def __init__(self, hidden_dim):

[0165] super(SelfAttentionNetwork, self).__init__()

[0166] self.query = nn.Linear(hidden_dim, hidden_dim)

[0167] self.key = nn.Linear(hidden_dim, hidden_dim)

[0168] self.value = nn.Linear(hidden_dim, hidden_dim)

[0169] def forward(self, x):

[0170] # Calculate attention weights

[0171] Q = self.query(x)

[0172] K = self.key(x)

[0173] V = self.value(x)

[0174] attention_weights = F.softmax(Q @ K.transpose(-2, -1) / (hidden_dim 0.5), dim=-1)

[0175] enhanced_features = attention_weights@V

[0176] return enhanced_features

[0177] # Cross attention layer, used to learn the mutual influence between different features

[0178] class CrossAttentionLayer(nn.Module):

[0179] def __init__(self, hidden_dim):

[0180] super(CrossAttentionLayer, self).__init__()

[0181] self.hidden_dim = hidden_dim

[0182] def forward(self, args):

[0183] # Simplified Cross-Attention Mechanism

[0184] cross_attention_output = sum(args) / len(args)

[0185] return cross_attention_output

[0186] # Environmental adaptation layer, dynamically adjust the feature vector according to environmental parameters

[0187] class EnvironmentAdaptationLayer(nn.Module):

[0188] def __init__(self, hidden_dim):

[0189] super(EnvironmentAdaptationLayer, self).__init__()

[0190] self.fc = nn.Linear(hidden_dim, hidden_dim)

[0191] def forward(self, features, environment):

[0192] # Adjust the feature vector according to the environment parameters

[0193] adjustment = torch.sigmoid(self.fc(environment))

[0194] adapted_features = features adjustment

[0195] return adapted_features

[0196] The following are the detailed steps for training the neural network model of the solenoid valve fault detection system:

[0197] First, it is crucial to prepare the training data set. The data set should include signal features from multiple sources, including acoustic signal features, vibration signal features, current signal features, and environmental parameter change features. Each sample needs to be labeled with the corresponding fault type for supervised training. The data should be organized in the form of time series for subsequent time series modeling. The data preprocessing step needs to normalize these features to ensure that all input features are in the same numerical range, which helps the convergence of the model.

[0198] Next, initialize the neural network model. In the implementation code of the model, the structure of each layer is defined, including the input layer, feature fusion layer, cross attention layer, time series modeling layer, environment adaptation layer, and fault classification layer. After initializing the model, you need to select a suitable loss function and optimizer. For classification problems, you can use the cross entropy loss function. The optimizer can choose the Adam optimizer because it can provide better convergence performance in complex model structures.

[0199] Before formal training, the dataset is divided into a training set and a validation set, usually in a ratio of 80% and 20%. The validation set is used to monitor the performance of the model during training to prevent overfitting. During training, the dataset is batched to make more efficient use of GPU resources. Usually, the size of each batch can be selected based on the model complexity and hardware performance.

[0200] At the beginning of training, the training data is input into the model and forward propagated layer by layer. The input layer receives the processed acoustic, vibration, current and environmental feature signals and passes these feature vectors to the feature fusion layer. The feature fusion layer captures the long-range dependencies of each signal feature through the self-attention sub-network and generates enhanced feature vectors. These enhanced feature vectors then enter the cross-attention layer, which learns and fuses the mutual influence between different types of signal features. Next, the time series modeling layer captures the time series patterns of these fused features through a bidirectional GRU and generates a feature vector containing time information. The environmental adaptation layer dynamically adjusts the time series features in combination with environmental parameters, and finally obtains an environmentally adapted feature vector.

[0201] The next step is to calculate the loss function. The feature vector after environmental adaptation is passed to the fault classification layer, and the probability distribution of the fault type is output through the fully connected layer and the softmax function. It is compared with the actual fault label and the cross entropy loss is calculated. After that, back propagation is performed and the model parameters are updated using the Adam optimizer. After each epoch of model training (traversing all training data), the model performance is evaluated on the validation set, and indicators such as accuracy and loss are recorded to ensure that the model performs well on the validation set and avoid overfitting.

[0202] Repeat the above training steps until the model's performance converges, that is, the loss of the validation set no longer decreases significantly or reaches the preset training rounds. When the training is completed, save the model parameters. The last step is to test the model and evaluate the actual performance of the model with unseen test data to ensure that it has good generalization ability.

[0203] This training process requires multiple stages of monitoring and adjustment, including adjusting the learning rate, modifying the network structure, etc., to achieve optimal detection performance. In practical applications, the model may need to be further tuned and improved to meet the needs of specific environments and application scenarios.

[0204] Furthermore, each self-attention sub-network of the feature fusion layer implements adaptive weighting within the feature through the following formula 1, with the input being a single type of feature vector and the output being an enhanced feature vector of the same dimension:

[0205] ;

[0206] in, Represents the feature vector and The attention weight between represents the total number of input feature vectors; and Represents the input feature vectors and The query matrix and key matrix of ; Represents the input feature vector The autocorrelation coefficient of yes The standardized function of and It is a weighting factor that is dynamically adjusted according to different signal types;

[0207] According to the following formula 2, the enhanced feature vector is calculated :

[0208] ;

[0209] in, is the input feature vector corresponding to The value matrix of Represents the input feature vector The updated version of is the enhanced feature vector after adaptive weighting, and the output enhanced feature vector maintains the same dimension as the input feature vector.

[0210] In the solenoid valve fault detection system, the feature fusion layer processes each type of feature vector through the self-attention sub-network. The core of this layer is to use an adaptive weighting mechanism to finely process the internal features to generate enhanced feature vectors. This process is implemented through formulas 1 and 2.

[0211] Formula 1 describes how to calculate the attention weights between feature vectors The specific expression is:

[0212] ;

[0213] In this formula, Represents the input feature vector and The attention weight between . This weight reflects the feature vector For the feature vector The size of the weight determines the degree of attention paid to the feature vector in the final feature update. For the feature vector contribution.

[0214] In the formula Represents the total number of input feature vectors, which are different signal features from the input layer, such as acoustic, vibration, current, and environmental parameters. Feature vector The query matrix and the eigenvector The key matrix Respectively represent the specific mapping of the feature vector. The feature vector The result is mapped to the query space, and the key matrix The feature vector The result of mapping to the key space is obtained by dot product. , the eigenvector is calculated and This similarity, as part of the self-attention mechanism, determines the degree of association between the two feature vectors.

[0215] In the formula Represents the feature vector The autocorrelation coefficient reflects the similarity or consistency between the elements in the feature vector. , the autocorrelation coefficient can be normalized to a standard range to ensure that its impact on the attention weight is controllable.

[0216] Normalization function The specific form of can be realized in many ways, the purpose of which is to convert the input autocorrelation coefficient Normalize to a standard range to ensure that its impact is controllable when calculating attention weights. Here are several common forms of normalization functions:

[0217] 1. Min-Max Normalization:

[0218] This is the most common standardization method. Scale to a fixed range (usually [0, 1]).

[0219] 2. Z-score normalization:

[0220] This standardized approach will Converted to its distance from the mean in standard deviations.

[0221] 3. Sigmoid function:

[0222] This nonlinear normalization method will It is mapped to [0, 1] and has a strong inhibitory effect on extreme values.

[0223] The choice of these normalization functions depends on the specific application requirements. Simply scale to [0,1], you can choose Min-Max standardization; if you need to adjust the distribution so that its mean is 0, you can choose Z-score standardization; if you want to strongly compress large and small values, you can use the Sigmoid function. These methods can help control the autocorrelation coefficient The impact on the attention weight makes the weighting of the feature vector more reasonable and stable.

[0224] coefficient and are dynamically adjusted weighting factors that control the relative influence of query-key similarity and autocorrelation coefficient on attention weight. These weighting factors can be adjusted according to different signal types or feature requirements, so that the self-attention mechanism can better adapt to different application scenarios.

[0225] The attention weights calculated , which can be further used to calculate the enhanced feature vector . Equation 2 describes this process:

[0226] ;

[0227] In this formula, is the input feature vector corresponding to The value matrix represents the result after the input feature vector is mapped specifically. Multiply it by its corresponding attention weight , and summing these results, we can get the updated eigenvector This updated feature vector is the input feature vector It is an enhanced version of , which contains the information of other related feature vectors while retaining the dimension of the original feature vector.

[0228] Through the calculation of these formulas, the feature fusion layer can effectively combine the complex relationships between the input feature vectors to generate more accurate and rich enhanced feature vectors, providing a solid data foundation for subsequent signal analysis and fault detection. This adaptive weighting mechanism not only improves the performance of the model, but also enhances its adaptability in diverse application scenarios.

[0229] Furthermore, the cross-attention layer uses the following formula 3 to calculate the cross-attention weights between different enhanced feature vectors to achieve the fusion of multimodal information:

[0230] ;

[0231] in, represents the enhanced feature vector and The cross attention weights between ; and They represent the output of the feature fusion layer. The enhanced eigenvector and the Enhanced feature vectors; is the enhanced eigenvector The dimension of is used to scale the dot product calculation; is the total number of enhanced eigenvectors; and is the dynamic weighting factor; is the cross-correlation function, which is implemented using the following formula 4:

[0232] ;

[0233] in is the adjustment factor used to control the sensitivity of the difference in the nonlinear function; Represents the dimension index of the enhanced feature vector; To enhance the dimension of feature vector;

[0234] According to the following formula 5, the fusion feature vector is calculated :

[0235] ;

[0236] in, Represents the fused feature vector; is the total number of enhanced eigenvectors; Indicates An enhanced feature vector.

[0237] In the solenoid valve fault detection system, the design of the cross attention layer aims to achieve effective fusion of multimodal information by calculating the cross attention weights between different enhanced feature vectors. This process is implemented by Formula 3, which describes how to determine the interaction weights between each pair of enhanced feature vectors to reflect the correlation between them.

[0238] The specific expression of formula 3 is:

[0239] ;

[0240] In this formula, Represents the enhanced feature vector and The cross attention weight between . This weight determines the feature vector For the feature vector The calculation of weights takes into account two key factors: the similarity between feature vectors and their cross-correlation.

[0241] first, and is the enhanced feature vector output by the feature fusion layer, representing the and feature vectors. To calculate the similarity between them, their dot product is used in the formula The dot product represents the inner product between vectors, reflecting the directional similarity of the two vectors. To avoid the problem of too large or too small values ​​in the calculation process, the dot product result is scaled and divided by ,in Represents the dimension of the enhanced feature vector. This scaling operation can balance vectors of different dimensions and make the calculation more stable.

[0242] In addition, a cross-correlation function is introduced into the formula , which is implemented by formula 4:

[0243] ;

[0244] This function enhances the eigenvector by computing and The difference in each dimension , and input it into the hyperbolic tangent function The hyperbolic tangent function maps the input difference to the range of -1 to 1, which can effectively suppress the influence of excessive differences on the results. The results are then summed and averaged to obtain the overall difference of the enhanced feature vector in each dimension, which is called cross-correlation. It is an adjustment factor used to control the sensitivity of the difference in the nonlinear function, that is, to adjust the impact of the difference between different eigenvectors on the final result.

[0245] During the entire calculation process, the parameters and are dynamic weighting factors, which are used to adjust the influence of dot product similarity and cross correlation in calculating attention weights. By adjusting these two parameters, the performance of the model under different feature conditions can be optimized to adapt to various complex input signals.

[0246] Calculated cross attention weights It is then used in Formula 5 to calculate the fused feature vector :

[0247] ;

[0248] In this formula, the fused feature vector is obtained by adding all enhanced feature vectors The weight is the cross attention weight calculated previously. The final result is It is a feature vector that integrates multimodal information, which can more comprehensively reflect the working status of the solenoid valve under different signals and provide richer input information for subsequent fault analysis and detection.

[0249] Through the detailed calculation of these formulas, the cross-attention layer can effectively fuse the multimodal information of different feature vectors, providing a more accurate and comprehensive fault feature analysis for the solenoid valve fault detection system. This design not only improves the detection accuracy of the system, but also enhances the adaptability of the model when processing complex signals.

[0250] Furthermore, the bidirectional gated recurrent unit of the timing modeling layer captures the bidirectional timing dependency of signal features by updating hidden states in both forward and backward directions, with the input being a fused feature vector sequence and the output being a feature vector containing bidirectional timing information.

[0251] In the solenoid valve fault detection system, the timing modeling layer is designed to better capture the time series dependencies of signal features and ensure that the model can identify various possible fault modes during the operation of the solenoid valve. The core component of this layer is the bidirectional gated recurrent unit (GRU), which captures the bidirectional timing dependencies of signal features by processing the forward and backward information of the signal simultaneously.

[0252] Specifically, the bidirectional GRU is an enhanced recurrent neural network (RNN) structure that includes two independent GRU networks: one for processing the forward sequence of the input signal, and the other for processing the backward sequence of the signal. The forward GRU receives the starting position of the signal sequence and processes it to the end position in sequence, while the backward GRU starts processing from the end position of the signal sequence and gradually traces back to the starting position. This bidirectional processing mechanism can capture the information of the forward and backward dependencies of the signal in the time dimension, thereby identifying the complex temporal patterns implicit in the signal.

[0253] In actual operation, the signal input to the time series modeling layer is a sequence of fused feature vectors after being processed by the previous layers. These sequences have integrated the information of multimodal signals and contain rich features of the working state of the solenoid valve. The bidirectional GRU continuously updates and adjusts its understanding of signal features through hidden state updates in both the forward and backward directions, and gradually generates feature vectors containing bidirectional time series information. This process not only enables the model to take into account the impact of past signals on the current state, but also captures the impact that future signal changes may have on the current state.

[0254] Through this bidirectional time series processing mechanism, the output feature vector not only retains the information of the signal in the time series, but also enhances the understanding of the context of the signal, thereby improving the accuracy and reliability of fault detection. The design of this bidirectional GRU ensures that the solenoid valve fault detection system can cope with complex time series data and effectively identify early fault signals or potential fault modes, thereby providing important support for timely fault warning and maintenance.

[0255] In summary, the timing modeling layer provides the system with comprehensive timing feature analysis capabilities through the implementation of bidirectional GRU, making solenoid valve fault detection more accurate and efficient.

[0256] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A solenoid valve fault detection system, characterized in that: include: An acoustic sensor module, used for collecting acoustic signals generated by the solenoid valve during operation; A vibration sensor module, used to collect vibration signals generated by the solenoid valve during operation; Current sensor module, used to monitor the current signal of the solenoid valve coil in real time; An environmental sensor module, used for monitoring environmental parameters around the solenoid valve; A data processing module, electrically connected to the acoustic sensor module, the vibration sensor module, the current sensor module and the environmental sensor module, for filtering and denoising the acoustic signal, the vibration signal and the current signal, and extracting features from the processed acoustic signal, the vibration signal and the current signal, respectively, to obtain acoustic signal features, vibration signal features and current signal features; Analyzing the environmental parameters to obtain change characteristics of the environmental parameters; A fault analysis module, electrically connected to the data processing module, for identifying the fault type of the solenoid valve based on the acoustic signal characteristics, vibration signal characteristics, current signal characteristics and change characteristics of environmental parameters using a pre-trained neural network model; triggering an alarm and recording fault information when a fault is detected; Wherein, the current sensor module includes a temperature compensation unit for correcting the current detection signal when the ambient temperature changes to ensure the accuracy of current monitoring; Wherein, the neural network model includes: An input layer, used to receive and process the acoustic signal features, vibration signal features, current signal features and environmental parameter change features, and output a corresponding feature vector; wherein the corresponding feature vector includes an acoustic vector, a vibration signal vector, a current signal vector and an environmental parameter feature vector; A feature fusion layer, used for receiving the corresponding feature vector provided by the input layer; the feature fusion layer includes four parallel self-attention sub-networks, each of which processes a corresponding feature vector respectively, so as to capture the long-range dependency relationship within each signal feature and output a corresponding enhanced feature vector; A cross attention layer receives the corresponding enhanced feature vector output by the feature fusion layer, is used to learn the mutual influence and correlation between different types of signal features, and outputs a fused feature vector; A time series modeling layer receives the fused feature vector output by the cross attention layer, wherein the time series modeling layer includes a bidirectional gated recurrent unit for capturing the time series variation pattern of the signal feature and outputting a time series feature vector containing time series information; The environmental adaptation layer is used to receive the feature vector output by the time series modeling layer and the change characteristics of the environmental parameters in the input layer; dynamically adjust the time series characteristics according to the change characteristics of the environmental parameters, and output an environmentally adapted feature vector; The fault classification layer receives the environmental adaptation feature vector output by the environmental adaptation layer. The fault classification layer includes a fully connected layer and a softmax activation function, and is used to output the probability distribution of the fault type of the solenoid valve.

2. The solenoid valve fault detection system according to claim 1, characterized in that: The acoustic sensor module includes a plurality of acoustic sensors distributed around the solenoid valve, and each acoustic sensor is configured to collect acoustic signals in different directions to ensure that various types of sound waves generated by the solenoid valve during operation can be accurately captured.

3. The solenoid valve fault detection system according to claim 1, characterized in that: The vibration sensor module includes at least two vibration sensors with different sensitivities, wherein one vibration sensor is configured to detect small vibration changes for capturing early fault signals, and the other vibration sensor is configured to detect vibrations with larger amplitudes for identifying serious mechanical faults or impacts.

4. The solenoid valve fault detection system according to claim 1, characterized in that: The environmental sensor module includes a self-calibration functional unit for self-calibrating the measurement accuracy of the sensor according to a preset time interval or environmental change, so as to improve the long-term stability and accuracy of environmental parameter measurement.

5. The solenoid valve fault detection system according to claim 1, characterized in that: The data processing module includes a multi-stage filter unit configured to perform hierarchical filtering on acoustic signals, vibration signals and current signals, first removing high-frequency noise and then further processing low-frequency signals to ensure that the final output signal characteristics have a high signal-to-noise ratio.

6. The solenoid valve fault detection system according to claim 1, characterized in that: Each self-attention sub-network of the feature fusion layer implements adaptive weighting within the feature through the following formula 1, with a single type of feature vector as input and an enhanced feature vector of the same dimension as output: ； in, Represents the feature vector and The attention weight between represents the total number of input feature vectors; and Represent the input feature vectors and The query matrix and key matrix of ; Represents the input feature vector The autocorrelation coefficient of yes The standardized function of and It is a weighting factor that is dynamically adjusted according to different signal types; According to the following formula 2, the enhanced feature vector is calculated : ； in, is the input feature vector corresponding to The value matrix of Represents the input feature vector The updated version of is the enhanced feature vector after adaptive weighting, and the output enhanced feature vector maintains the same dimension as the input feature vector.

7. The solenoid valve fault detection system according to claim 1, characterized in that: The cross attention layer uses the following formula 3 to calculate the cross attention weights between different enhanced feature vectors to achieve the fusion of multimodal information: ； in, represents the enhanced feature vector and The cross attention weights between ; and They represent the output of the feature fusion layer. The enhanced eigenvector and Enhanced feature vectors; is the enhanced eigenvector The dimension of is used to scale the dot product calculation; is the total number of enhanced eigenvectors; and is the dynamic weighting factor; is the cross-correlation function, which is implemented using the following formula 4: ； in is the adjustment factor used to control the sensitivity of the difference in the nonlinear function; Represents the dimension index of the enhanced feature vector; To enhance the dimension of feature vector; According to the following formula 5, the fusion feature vector is calculated : ； in, Represents the fused feature vector; is the total number of enhanced eigenvectors; Indicates An enhanced feature vector.

8. The solenoid valve fault detection system according to claim 1, characterized in that: The bidirectional gated recurrent unit of the timing modeling layer captures the bidirectional timing dependency of signal features by updating hidden states in both forward and backward directions, with the input being a fused feature vector sequence and the output being a feature vector containing bidirectional timing information.

Images