Intelligent state detection system applied to water conservancy valve

Abstract

The application relates to the technical field of water conservancy engineering equipment monitoring, and discloses a state intelligent detection system applied to a water conservancy valve, which comprises a data sensing module, an edge collection module and a state analysis and diagnosis module. The data sensing module is used for collecting electrical signals corresponding to multi-physical field state parameters; the edge collection module is used for converting the electrical signals into digital signals, filtering and denoising the digital signals and extracting feature frequency bands, so as to form a bottom layer data stream; and the state analysis and diagnosis module is used for receiving the bottom layer data stream, extracting key parameters representing a valve mechanical state, a fluid state and a dynamics state, determining the running state of the water conservancy valve, performing a smooth extrapolation calculation, calculating cumulative cavitation damage and reconstructing a preset health working condition benchmark pressure difference to form a disease prediction benchmark. The system can reduce data transmission load and calculation load, and improve the diagnosis capability for the running state and the deterioration trend of the water conservancy valve.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering equipment monitoring technology, specifically to an intelligent status detection system for water conservancy valves. Background Technology

[0002] Hydraulic valves are fundamental control components in water conservancy pipeline systems, and their operational status directly affects the safety and stability of fluid transport. Currently, online monitoring of hydraulic valves often uses single physical quantity sensors to collect data, and the unprocessed raw signals are directly uploaded to a central server for analysis. This centralized processing method not only generates a large data transmission load, but also allows background noise from the fluid and machinery in the raw signals to easily mask the true fault characteristics, reducing the accuracy of condition determination.

[0003] Furthermore, existing valve condition diagnostics typically rely on pre-set static thresholds, lacking a comprehensive consideration of the coupling relationship between mechanical and fluid conditions. As valves operate over time, their internal structures experience wear, aging, and performance drift, making it difficult for static thresholds to adapt to changes in operating baselines, easily leading to false alarms or missed alarms. Especially under high pressure differentials or complex flow field conditions, cavitation damage inside valves often has the characteristics of being minute in quantity but accumulating over a long period. Existing monitoring methods struggle to continuously and accurately characterize this type of damage, thus affecting the effective assessment of early, hidden defects and operational deterioration trends in hydraulic valves.

[0004] Therefore, this invention proposes an intelligent status detection system for hydraulic valves to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent status detection system for hydraulic valves. This system solves the problems of inaccurate diagnosis of early hidden defects and deviations in long-term service life assessment caused by the lack of edge noise reduction processing, reliance on static monitoring thresholds, and neglect of the cumulative calculation errors of minor cavitation damage in existing hydraulic valve monitoring systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent status detection system for hydraulic valves, comprising:

[0007] The data sensing module is located at the preset monitoring points of the hydraulic valve and is used to collect electrical signals of multi-physical field state parameters.

[0008] An edge acquisition module, connected to the data sensing module, includes a synchronous acquisition unit and a local computing component. The synchronous acquisition unit is used to acquire electrical signals and convert them into raw digital signals. The local computing component is used to filter and denoise the raw digital signals to obtain denoised digital signals, extract the characteristic frequency bands of the denoised digital signals, and concatenate the denoised digital signals with the characteristic frequency bands to form a bottom-level data stream.

[0009] The status analysis and diagnosis module receives the underlying data stream and includes a feature extraction unit, a status monitoring unit, a trend prediction unit, and a cascade diagnosis unit. The feature extraction unit is used to extract key parameters characterizing the mechanical, fluid, and dynamic states of the hydraulic valve from the underlying data stream. The status monitoring unit is used to determine the operating status of the hydraulic valve based on the key parameters. The trend prediction unit is used to perform smooth extrapolation calculations on the key parameters. The cascade diagnosis unit is used to calculate cumulative cavitation damage based on the key parameters and perform defect prediction benchmark reconstruction.

[0010] Preferably, the data sensing module includes:

[0011] A vibration sensing unit is installed at preset mechanical monitoring points of the hydraulic valve to acquire electrical signals of high-frequency mechanical state characteristics.

[0012] Pressure sensing units are installed in the pipe sections before and after the hydraulic valve to acquire electrical signals that indicate the fluid state characteristics.

[0013] A torque sensing unit is installed at the drive mechanism of the hydraulic valve to acquire electrical signals that reflect the mechanical dynamics characteristics.

[0014] An opening degree sensing unit is connected to the output control shaft of the actuator of the hydraulic valve and is used to acquire an electrical signal of the absolute opening degree value.

[0015] Preferably, the local computing component uses a wavelet transform algorithm to perform wavelet decomposition on the original digital signal to separate the original digital signal into high-frequency detail coefficients and low-frequency approximation coefficients, uses a hard thresholding method to process the high-frequency detail coefficients, and performs inverse wavelet transform on the processed high-frequency detail coefficients and low-frequency approximation coefficients to reconstruct the denoised digital signal.

[0016] The local computing component uses a short-time Fourier transform algorithm to map the denoised digital signal from a one-dimensional time domain to a two-dimensional time-frequency domain, and extracts high-frequency intervals in the time-frequency domain matrix whose energy amplitude exceeds a preset multiple of the global average energy amplitude as the characteristic frequency band.

[0017] Preferably, the feature extraction unit extracts the energy centroid frequency, root mean square amplitude, and temporal kurtosis parameter within the feature frequency band to construct a feature vector, and calculates the Mahalanobis distance between the feature vector and the normal reference spectrum space as a key parameter characterizing the mechanical state of the hydraulic valve.

[0018] The feature extraction unit uses the Fast Fourier Transform algorithm to deconstruct the frequency domain of the denoised digital signal after the electrical signal acquired by the pressure sensing unit, obtain the frequency domain complex amplitude, calculate the power spectral density, and integrate to obtain the transient turbulent kinetic energy. The transient turbulent kinetic energy is used as a key parameter characterizing the fluid state of the hydraulic valve.

[0019] Preferably, the feature extraction unit performs time-domain difference decomposition on the torque state variable within a single action cycle obtained by converting the electrical signal acquired by the torque sensing unit to obtain the torque change rate and torque fluctuation amplitude. It then calculates the average steady-state driving torque during the fully open stroke and the average steady-state driving torque during the fully closed stroke to construct an asymmetric discrimination index for opening and closing torque. The torque change rate, torque fluctuation amplitude, and opening and closing torque asymmetric discrimination index are used as key parameters characterizing the dynamic state of the hydraulic valve.

[0020] Preferably, the status monitoring unit uses the evidence synthesis formula of DS evidence theory to perform primary modal fusion on the basic probability allocation function of multiple independent sensing units in the data sensing module, and when the set conflict coefficient is greater than or equal to the conflict determination threshold, it uses the arithmetic average weighting method of multi-source evidence to replace the evidence synthesis formula and outputs a preliminary determination result for the operating status of the hydraulic valve.

[0021] Preferably, the state monitoring unit uses a support vector machine algorithm to map the multidimensional features contained in the key parameters to a linearly separable space through a radial basis kernel function to isolate abnormal noise samples, and uses a random forest algorithm to perform feature dimensionality reduction on the original feature matrix formed by splicing the key parameters by calculating the Gini importance of each dimension of the multidimensional features to obtain the core feature matrix.

[0022] The status monitoring unit converts the core feature matrix into a one-dimensional sequence and inputs it into a convolutional neural network for fitting and modeling. The output of the convolutional neural network is then used to output the predicted degradation value of the hydraulic valve.

[0023] Preferably, the trend prediction unit uses a cubic spline interpolation algorithm to numerically fill in the missing time node data contained in the key parameters to construct a one-dimensional time series;

[0024] The trend prediction unit uses a quadratic exponential smoothing algorithm to smooth the one-dimensional time series, calculates the first and second exponential smoothing values ​​at the current time node, solves the trend fitting baseline value and trend fitting degradation rate at the current time node based on the first and second exponential smoothing values, and uses the trend fitting baseline value and trend fitting degradation rate combined with a fixed duration parameter smoothing extrapolation logic to calculate the extrapolated prediction value for future time nodes.

[0025] Preferably, the cascaded diagnostic unit uses the Kahan tail compensation algorithm to calculate the cumulative cavitation damage of a single minor cavitation damage to the hydraulic valve over multiple operating cycles, separates the truncated part of the floating-point number generated in the calculation to generate a truncation error compensation amount, and passes it to the next operating cycle for compensation accumulation.

[0026] The cascaded diagnostic unit calculates the dynamic physical residual between the actual measured pressure difference and the preset healthy operating condition benchmark pressure difference based on the electrical signal obtained by the pressure sensing unit. It combines a dimensionless damage factor and obtains a physical residual penalty term by performing polynomial online fitting using the recursive least squares method. The physical residual penalty term is then used to translate and reconstruct the preset healthy operating condition benchmark pressure difference to generate a disease prediction benchmark.

[0027] Preferably, the cascaded diagnostic unit is also used to perform serial verification and hardware-level security circuit breaking;

[0028] The cascaded diagnostic unit calculates the cross-modal coupling degradation rate based on the asymmetric discrimination index of opening and closing torque and the cumulative cavitation damage.

[0029] When the cross-modal coupling degradation rate is detected to exceed the preset degradation rate safety threshold, the cascaded diagnostic unit opens a disturbance attenuation verification window containing multiple consecutive action cycles. If the cross-modal coupling degradation rate does not drop to the preset normal threshold range within the verification window, the operation status weighting logic based on multi-source information fusion is activated to output a comprehensive degradation diagnosis conclusion.

[0030] When the edge acquisition module reports that the hydraulic valve has reached the fully closed limit state, if the absolute cosine distance between the current vibration feature vector extracted by the feature extraction unit and the preset healthy fully closed state benchmark vector falls below the preset similarity lower limit threshold, and the fully closed state sealing feature index and the inlet and outlet pressure stability index simultaneously exceed the set safety envelope, the cascade diagnosis unit issues a safety fuse command to the actuator of the hydraulic valve.

[0031] The inlet and outlet pressure stability index is characterized by the fluctuation amplitude or standard deviation of the inlet and outlet static pressure within the fully closed maintenance time window, and the safety envelope is obtained by pre-calibration based on the sealing characteristic index of the fully closed working condition under healthy conditions and the statistical upper limit of the inlet and outlet pressure stability index.

[0032] This invention provides an intelligent status monitoring system for hydraulic valves. It offers the following advantages:

[0033] 1. This invention sets up a local computing component in the edge acquisition module and uses wavelet transform and short-time Fourier transform to filter and reduce noise and extract characteristic frequency bands from the original digital signal. This helps to reduce the impact of environmental background noise at the acquisition end, improve the signal-to-noise ratio of the underlying data stream, and reduce the communication and computing load of the subsequent state analysis and diagnosis modules, thereby improving detection efficiency and analysis accuracy.

[0034] 2. The state monitoring unit of the present invention utilizes DS evidence theory to perform primary modal fusion of multi-physics sensing data, and combines support vector machine, random forest and convolutional neural network for feature processing and fitting modeling, which helps to reduce the risk of misjudgment caused by abnormal interference from a single sensor and improve the comprehensive identification ability of the mechanical state, fluid state and dynamic state of hydraulic valves.

[0035] 3. The cascaded diagnostic unit of the present invention uses the Kahan tail compensation algorithm to compensate and accumulate the amount of minor cavitation damage in multiple operation cycles, and reconstructs the preset healthy operating condition benchmark pressure difference based on dynamic physical residuals. This helps to reduce the truncation error of minor damage in the long-term accumulation process and improve the diagnostic accuracy of the faulty operation status and deterioration trend of hydraulic valves. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0037] Figure 2 This is a schematic diagram of the method flow of the present invention;

[0038] Figure 3 This is a schematic diagram of the internal execution logic of the cascaded diagnostic unit of the present invention;

[0039] Figure 4 This is a schematic diagram of the time-domain waveform of the synchronous acquisition of multi-physics field state parameters according to the present invention; wherein, (a) is a waveform of high-frequency vibration acceleration along the Z-axis; (b) is a waveform of pressure pulsation on the outlet side; and (c) is a waveform of dynamic driving torque.

[0040] Figure 5 This is a schematic diagram of the degradation trend prediction and effect comparison curves of the present invention.

[0041] The components include: 110, Data Sensing Module; 111, Vibration Sensing Unit; 112, Pressure Sensing Unit; 113, Torque Sensing Unit; 114, Opening Sensing Unit; 120, Edge Acquisition Module; 121, Synchronous Acquisition Unit; 122, Local Computing Component; 130, State Analysis and Diagnosis Module; 131, Feature Extraction Unit; 132, State Monitoring Unit; 133, Trend Prediction Unit; and 134, Cascade Diagnosis Unit. Detailed Implementation

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] See attached document Figure 1 The present invention provides an intelligent status detection system for hydraulic valves, including a data sensing module 110, an edge acquisition module 120, and a status analysis and diagnosis module 130.

[0044] The data sensing module 110 is located on the valve body and actuator of the hydraulic valve. The data sensing module 110 includes a vibration sensing unit 111, a pressure sensing unit 112, a torque sensing unit 113, and an opening degree sensing unit 114. The data sensing module 110 acquires multi-physics field state parameters during the operation of the hydraulic valve and converts the physical quantities into electrical signals. The opening degree sensing unit 114 is integrated into the actuator and is used to acquire the absolute opening degree value of the hydraulic valve.

[0045] Edge acquisition module 120 is communicatively connected to data sensing module 110. Edge acquisition module 120 includes synchronous acquisition unit 121 and local computing component 122. Synchronous acquisition unit 121 performs analog-to-digital conversion of the signal. Local computing component 122 performs filtering calculations on the converted digital signal and packages it into a low-level data stream.

[0046] The state analysis and diagnosis module 130 receives the underlying data stream. The state analysis and diagnosis module 130 includes a feature extraction unit 131, a state monitoring unit 132, a trend prediction unit 133, and a cascaded diagnosis unit 134. The feature extraction unit 131 performs the extraction and classification of basic physical mode features. The state monitoring unit 132 performs limit comparison and feeds back the hardware communication status. The trend prediction unit 133 performs smooth extrapolation calculations of parameters over physical time. The cascaded diagnosis unit 134 performs cross-modal cascaded diagnosis and state reconstruction based on the extracted feature parameters.

[0047] See attached document Figure 2This invention provides a method for intelligent status detection of hydraulic valves, comprising the following steps:

[0048] S100 collects multi-physics field state parameters during the operation of hydraulic valves, performs synchronous analog-to-digital conversion and noise reduction preprocessing on the multi-physics field state parameters, and generates the underlying data stream;

[0049] S200 extracts vibration feature parameters from the underlying data stream to construct a benchmark spectrum library and calculates transient turbulent kinetic energy. Based on inlet and outlet pressure, pressure pulsation and valve opening, it constructs a benchmark pressure difference mapping relationship corresponding to different opening ranges and different inlet and outlet pressure conditions under healthy conditions. In this way, it establishes a preset healthy working condition benchmark pressure difference and outputs deviation parameters. At the same time, it calculates the torque change rate and dynamic asymmetry index of the actuator and quantifies the sealing state characteristics under the fully closed state.

[0050] S300 performs primary modal fusion and classification modeling on the underlying data stream to execute over-limit alarm control and hardware status inspection, and constructs the time series of detection parameters. It performs exponential smooth extrapolation calculation of parameters with physical time to obtain the degradation trend.

[0051] S400 maps the continuous time axis to the action cycle domain variable, calculates the cumulative cavitation damage including tail compensation, and calculates the degradation slope of the multidimensional feature cross-mapping to perform cross-domain interlocking determination of equipment physical wear.

[0052] The S500 calculates the dynamic physical residual between the measured pressure difference and the preset healthy operating condition benchmark pressure difference, reconstructs the equipment operation judgment benchmark, performs equipment state evolution prediction based on global damage variables, and triggers hardware-level safety protection actions according to preset structural extreme value boundaries.

[0053] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0054] See attached document Figure 1To construct a stable and reliable underlying physical data source, this system is equipped with a data sensing module 110. The data sensing module 110 is positioned on the valve body and actuator of the hydraulic valve to establish a low-level acquisition network for multi-physics field state parameters. As a preferred implementation, considering that hydraulic pipelines are often located in complex temperature and humidity interaction environments such as open-air or basement locations, the operating environment parameters of the data sensing module 110 are set to adapt to hydraulic working conditions, specifically including an operating temperature range controlled between -20℃ and 60℃. The physical basis for this temperature range design is that the lower limit setting can prevent the internal insulating silicone oil of the sensor from physically freezing and causing diaphragm failure in the frigid northern regions, while the upper limit setting can avoid nonlinear thermal drift of core electronic components in the hot southern environments. The external encapsulation of each sensing unit adopts the IP68 protection standard to resist continuous moisture intrusion or short-term flooding inside the pipe gallery.

[0055] In the mechanical state characteristic acquisition stage, the data sensing module 110 deploys vibration sensing units 111 at designated locations on the hydraulic valve to acquire high-frequency mechanical state characteristics. In this embodiment, the vibration sensing unit 111 includes three vibration acceleration sensors. To establish a unified and rigorous three-dimensional spatial reference coordinate system, this embodiment defines the Z-axis direction as the direction of the valve stem's central axis, and defines the X and Y directions as the axial flow direction of the fluid in the pipe and its perpendicular direction, respectively. The specific arrangement of the three vibration acceleration sensors is as follows: the first vibration acceleration sensor is arranged in the X direction of the valve stem, the second vibration acceleration sensor is arranged in the Y direction of the valve stem, and the third vibration acceleration sensor is arranged at the valve body. The X and Y directions are perpendicular to each other and are used to acquire the small radial yaw offset of the valve stem; the vibration acceleration sensor at the valve body is used to acquire the shell vibration generated by fluid-structure interaction. In the signal acquisition configuration, the effective analysis frequency upper limit of the vibration sensing unit 111 is limited to 3kHz. The physical basis for determining this upper frequency threshold is that the typical early minor wear excitation frequency of hydraulic valve mechanical transmission components and the fault characteristic frequency of support bearings are usually concentrated in the range of 1kHz to 3kHz. Using this upper frequency limit parameter can both cover the characteristic frequency band of mechanical degradation and filter out the extremely high frequency environmental background noise that exceeds the effective frequency band of hydraulic machinery.

[0056] Accurate capture of the flow field operating conditions is also the foundation for subsequent cross-modal diagnostics. The data sensing module 110 deploys pressure sensing units 112 in the pipe sections before and after the hydraulic valve to acquire fluid state characteristics. The pressure sensing unit 112 includes a pressure pulsation sensor and a static pressure transmitter. Two pressure pulsation sensors are installed, one on the inlet side and the other on the outlet side of the hydraulic valve, respectively. The effective frequency upper limit of the pressure pulsation sensor is set to 2kHz, which is used to cover the broadband turbulence characteristic frequency band generated during the initial cavitation stage of the fluid. The static pressure transmitter is deployed at two points, inlet and outlet, respectively connected to the pressure taps upstream and downstream of the hydraulic valve. To avoid physical interference from local turbulence generated behind the valve orifice on the static pressure measurement, the upstream pressure tap is located at a steady-flow straight pipe section 5 times the nominal pipe diameter before the hydraulic valve inlet, and the downstream pressure tap is located at a steady-flow straight pipe section 3 times the nominal pipe diameter after the hydraulic valve outlet. The maximum pressure range limit for the inlet and outlet static pressure transmitters is set at 1.8 MPa. The logic behind this range limit is that the rated operating pressure of conventional municipal and water conservancy networks is mostly between 0.6 MPa and 1.0 MPa. Considering the transient water hammer effect caused by abnormal power outages or rapid valve closures, the peak water hammer pressure is typically 1.5 to 2 times the rated pressure. Setting an upper limit of 1.8 MPa reliably covers the theoretical peak pressure of water conservancy networks under extreme water hammer conditions. For the pressure tapping hole opening and pressure tapping pipe welding construction of the pressure pulsation sensor and static pressure transmitter, those skilled in the art can follow conventional fluid instrumentation installation specifications based on the pipe material and wall thickness. The specific installation process is well-known in the field and will not be elaborated here.

[0057] For dynamic monitoring of the actuator, the data sensing module 110 deploys a torque sensing unit 113 at the drive mechanism of the hydraulic valve to obtain mechanical dynamic state characteristics. The torque sensing unit 113 adopts a dynamic torque sensor. The torque sensing unit 113 is assembled and positioned at the direct connection between the output shaft of the actuator and the valve shaft of the hydraulic valve, and is connected in series in the transmission chain. This assembly and positioning method can directly obtain the real driving torque applied to the valve shaft by the actuator, avoiding signal distortion caused by the backlash of the intermediate reduction gearbox. According to the mechanical characteristics of the selected hydraulic valve actuator, the maximum measurement range of the torque sensor matched with the actuator is limited to 1.2 to 1.5 times greater than the maximum stall torque of the actuator. In this embodiment, the specific value of this range parameter is 2600 N·m. The principle of setting this range threshold is to ensure that the torque reading under normal opening and closing conditions is within the linear response range of the sensor, while also retaining sufficient safety margin to prevent irreversible overload damage to the torque sensing unit 113 under conditions of severe jamming or foreign object blockage of the hydraulic valve. For monitoring the mechanical opening and closing positions, the data sensing module 110 is also equipped with an opening degree sensing unit 114. Specifically, the opening degree sensing unit 114 uses a high-resolution absolute encoder or angular displacement sensor, and outputs an electrical signal corresponding to the absolute opening degree after signal conditioning. It is coaxially rigidly connected to the top of the valve stem of the hydraulic valve or the output control shaft of the actuator to acquire the absolute opening degree value of the hydraulic valve from 0% to 100% in real time. The physical quantities acquired by each sensing unit are transmitted through shielded cables to complete the physical layer construction of the data sensing module 110.

[0058] To achieve efficient acquisition and front-end processing of multiphysics state parameters, this system is equipped with an edge acquisition module 120. The edge acquisition module 120 and the data sensing module 110 establish a hardware communication connection via a shielded cable. In this embodiment, the edge acquisition module 120 includes a synchronous acquisition unit 121 and a local computing component 122. By offloading data acquisition and preliminary signal cleaning to the edge device, the edge acquisition module 120 effectively reduces the computational load on the central analysis node and minimizes network bandwidth usage.

[0059] In the physical connection and analog-to-digital conversion stage, the synchronous acquisition unit 121 uses a multi-channel synchronous acquisition component to acquire the electrical signals transmitted by each sensing unit. To prevent time phase deviations caused by inconsistent data timestamps during subsequent cross-modal diagnostics, each channel of the synchronous acquisition unit 121 is equipped with an independent analog-to-digital converter that shares the same high-precision oven-controlled crystal oscillator clock source. As a preferred underlying alignment logic, the multi-channel synchronous acquisition component integrates a synchronous sample-and-hold circuit, which ensures that all sensing channels are latched at the same microsecond-level physical clock edge, eliminating phase differences between multi-source data at the hardware level. In terms of specific parameter settings, the vertical resolution of the analog-to-digital converter is limited to 24 bits. This vertical resolution parameter is used to meet the dynamic range requirements when weak pressure pulsation signals and large-amplitude water hammer impact signals coexist. According to Shannon's sampling theorem and the aforementioned upper limit of the effective analysis frequency of 3kHz, the synchronous acquisition unit 121 fixes the sampling rate of the multi-channel synchronous acquisition component to 10kHz. The sampling rate parameter is designed so that a sampling frequency of 10kHz can avoid spectral aliasing between high-frequency vibration signals and fluid pulsation signals, and also prevent unnecessary data redundancy that exceeds the system's processing limits. After receiving the electrical signal, the synchronous acquisition unit 121 uses a front-end hardware anti-aliasing low-pass filter to filter out high-frequency interference above the Nyquist frequency, preferably filtering out high-frequency electromagnetic interference above 5kHz. Each channel synchronously performs analog-to-digital conversion and outputs discrete raw digital signals to the local computing component 122.

[0060] The original digital signal typically contains periodic fluid interference generated by the rotation of the pump impeller and power frequency electrical noise. To improve data quality, the local computing component 122 uses a wavelet transform algorithm to perform high-frequency filtering and noise reduction on the original digital signal. Considering the obvious non-stationary transient impact characteristics of the hydraulic valve vibration signal, the local computing component 122 selects the db4 wavelet basis function from the Daubechies wavelet family. The physical reason for choosing the db4 wavelet basis is that it has good compact support and smoothness, which can closely fit the energy attenuation waveform generated by mechanical rigid impact. The local computing component 122 performs a four-level wavelet decomposition on the original digital signal, separating it into high-frequency detail coefficients and low-frequency approximation coefficients. When performing threshold denoising, to avoid the limitations of setting a fixed threshold based on human experience, the local computing component 122 extracts the first-level ultra-high-frequency detail coefficients and calculates the median absolute deviation of the coefficients in this level to estimate the standard deviation of the environmental background noise without bias. Based on this noise standard deviation and the signal sampling length, a general hard threshold is derived. The local computing component 122 uses the adaptively derived hard thresholding method to zero out the high-frequency detail coefficients to eliminate Gaussian white noise, and then performs an inverse wavelet transform on the processed high-frequency detail coefficients and the original low-frequency approximation coefficients to reconstruct the denoised digital signal. For the specific mathematical derivation of wavelet transform decomposition and reconstruction, those skilled in the art can refer to standard signal processing theory; the specific derivation process is well-known in the field and will not be elaborated here.

[0061] After acquiring the denoised digital signal, the local computing component 122 further applies a short-time Fourier transform to map it from a one-dimensional time domain to a two-dimensional time-frequency domain to identify and extract feature frequency bands containing device degradation information. During this transformation, the local computing component 122 sets the sliding time window length of the short-time Fourier transform to 256 sampling points, and a Hamming window is selected as the window function. The length of this sliding time window is determined because 256 sampling points at a sampling rate of 10 kHz correspond to a physical time span of 25.6 milliseconds, which can simultaneously consider the temporal resolution of transient impact events in the time domain and the frequency resolution of characteristic spectral lines in the frequency domain. The Hamming window is used to suppress the spectral leakage effect caused by signal truncation at both ends. In the time-frequency domain matrix, the local computing component 122 filters and extracts high-frequency intervals whose energy amplitude exceeds a preset multiple of the global average energy amplitude as feature frequency bands. In this embodiment, the preset multiple is set to 3 times. The physical significance of this multiplier threshold lies in filtering the broadband background energy generated by normal fluid flow and accurately capturing concentrated high-energy frequency bands induced by mechanical wear or cavitation collapse. To improve the algorithm logic and avoid dead-zone misjudgment when the equipment is in no-load condition, the local computing component 122 pre-introduces an absolute energy lower limit benchmark for dual constraints before performing the multiplier comparison. The feature extraction operation of the 3-times threshold is only triggered when the global average energy amplitude is greater than or equal to the absolute energy lower limit benchmark; if the global average energy amplitude is lower than the absolute energy lower limit benchmark, the current time window is determined to be an invalid shutdown state or a pure background state, and the extraction logic is skipped directly, thereby avoiding the problem of abnormally amplified background electrical noise extraction caused by the comparison benchmark approaching 0.

[0062] After identifying and extracting the characteristic frequency bands, the local computing component 122 executes the packetization and uploading logic of the underlying data stream. To balance the integrity of the underlying data with network communication efficiency, the local computing component 122 uses a fixed time window of 10 seconds as a rule, and structurally splices the denoised digital signal and its corresponding characteristic frequency band data within this 10-second time window. The physical consideration for using a fixed 10-second time window is that the time span from receiving the action command to completing the mechanical stroke of a conventional medium and large-sized hydraulic valve, as well as the occurrence and attenuation period of a typical transient water hammer pressure wave in a water pipe network, are usually distributed within a physical range of 5 to 8 seconds. It is set that 10 seconds can completely cover an independent mechanical dynamic event. The data packet formed by splicing is the underlying data stream. The local computing component 122 uploads the underlying data stream to the status analysis and diagnosis module 130 through the industrial Ethernet communication protocol using a decentralized message publish-subscribe mechanism. This decentralized upload mechanism allows the detection system to smoothly connect more hydraulic valve edge nodes in the future, avoiding the single-point communication congestion bottleneck caused by the point-to-point polling architecture.

[0063] After the edge acquisition module 120 completes the front-end cleaning and encapsulation of multi-source data, the state analysis and diagnosis module 130 undertakes the task of multi-dimensional state deconstruction and feature reconstruction. In this embodiment, the state analysis and diagnosis module 130 is equipped with a feature extraction unit 131, which is used to extract key parameters characterizing the mechanical, fluid, and dynamic states of hydraulic valves from the structured underlying data stream.

[0064] For the extraction of vibration time-frequency domain features, feature extraction unit 131 receives feature frequency bands from the underlying data stream. To construct a mathematical space characterizing the mechanical operating state, feature extraction unit 131 extracts the energy centroid frequency, root mean square amplitude, and time-domain kurtosis parameter within the feature frequency bands to construct a three-dimensional feature vector. In specific implementations, the calculation methods for the aforementioned time-domain and frequency-domain basic statistical parameters can be found in standard mechanical vibration signal processing specifications by those skilled in the art, and will not be elaborated upon here. Under the healthy service state during the initial commissioning of the hydraulic valve, feature extraction unit 131 collects feature vectors from multiple normal opening and closing cycles, calculates the mean vector and covariance matrix of their multivariate Gaussian distribution, and thus constructs a normal baseline spectrum space. To prevent the covariance matrix from exhibiting singularity under specific low-frequency single operating conditions, leading to the inability to solve the inverse matrix, as a preferred underlying compensation method, feature extraction unit 131 introduces a fixed-value minimal regularization constant on the diagonal of the covariance matrix for inversion operations. Simultaneously, the feature extraction unit 131 incorporates typical mechanical fault data of hydraulic valves of the same model and specification under laboratory conditions to construct a fault map space covering valve stem bending and support bearing wear patterns. After offline initialization of the map space, during real-time operation, the feature extraction unit 131 calculates the Mahalanobis distance between the current feature vector and the normal baseline map space. If the Mahalanobis distance exceeds the three-standard-deviation confidence boundary determined by the Laida criterion, the mechanical state is determined to deviate from the normal range, and the cosine similarity matching of the vector angle between the current feature vector and the fault map space is triggered, thereby identifying the specific mechanical anomaly pattern with the highest spatial similarity.

[0065] In this embodiment, to achieve dynamic self-learning evolution of diagnostic benchmarks throughout the entire equipment lifecycle, the system further constructs a three-dimensional index system covering fault types, feature maps, and operating condition labels. It should be noted that when the system detects valve operating parameters exceeding normal thresholds (e.g., excessive vibration amplitude or abnormal torque fluctuations) or receives manual fault confirmation information, the system automatically triggers the fault learning and acquisition process. Specifically, the system automatically backtracks and extracts historical high-frequency vibration data from 72 hours before the fault occurred until the fault was resolved, and supplements data samples according to a preset high-frequency acquisition density (e.g., 1 minute / time). By introducing a spectrum comparison analysis tool, the system automatically extracts evolutionary features such as amplitude increases in specific frequency bands, the emergence of new frequency bands, or harmonic distortion, thereby segmenting and picking the differences in spectral features at the nascent, development, and outbreak stages of the fault. In the real-time warning stage, the system performs cosine similarity matching between the currently extracted spectral features and the initial fault map in the aforementioned three-dimensional index system. The system immediately outputs an early fault warning and simultaneously adds the evolution data of the current map to the fault database when the cosine similarity between the real-time map and the initial map of a specific type of fault (such as jamming, cavitation, bearing wear, etc.) is ≥85%, and this condition is met for three consecutive sampling periods. As an optional implementation, after each fault is handled, the system supports the distribution of manually labeled results from the maintenance terminal. Based on the actual confirmed fault type, misjudged maps are deleted and feature weights are optimized, thereby achieving self-learning iteration without deep manual intervention.

[0066] The assessment of flow field stability relies on in-depth analysis of pressure pulsation data. In this embodiment, the feature extraction unit 131 first employs a combination of Fast Fourier Transform (FFT), wavelet analysis, and order analysis to perform spectral analysis on the acquired fluid pulsation pressure signal. Specifically, the discrete calculation expression for the FFT is as follows:

[0067] ;

[0068] In the formula, This represents the total number of sampling points within the sampling time window; This refers to the time-domain sampling sequence number; The frequency index is the discrete frequency in the frequency domain. The imaginary unit; It is a time-domain discrete sampling sequence; The transformed frequency domain complex amplitude value. After obtaining the frequency domain distribution, the feature extraction unit 131 extracts the characteristic frequencies in the spectrum and cross-matches them with the pre-set pump blade passing frequency, pipeline structure resonance frequency, and valve throttling excitation frequency in the water network system. By locating the corresponding spectral peaks that match, the feature extraction unit 131 can accurately trace and determine the current dominant sources of vibration and noise. Furthermore, for the extracted high-frequency band (>1kHz) signal, the feature extraction unit 131 not only calculates its power spectral density and integrates to obtain the transient turbulent kinetic energy, but also identifies the white noise distribution characteristics in the high-frequency band that are wide-band, continuous, and without obvious discrete spectral lines through frequency band morphology scanning. This white noise feature is used as a physical criterion for judging the initial state of cavitation in the valve flow channel. Considering that the wide-band physical impact generated by the collapse of microbubbles during the initial stage of cavitation in hydraulic valves is mainly concentrated in the high-frequency region, the feature extraction unit 131 actively strips away the low-frequency background macroscopic fluctuations of the water flow and performs integral quantization operations on the transient turbulent kinetic energy in the extremely high-frequency band. This treatment mechanism aims to effectively isolate long-period macroscopic pressure fluctuations caused by pump impeller rotation or pipeline regulation. Its transient turbulent kinetic energy integral formula is as follows:

[0069] ;

[0070] In the formula, This is the integral value of the transient turbulent kinetic energy, with dimensions in joules; It is a frequency variable with the dimension of Hertz; This is the power spectral density function of the fluid pulsation signal at the corresponding frequency; To effectively analyze the upper frequency limit, in this embodiment, by extracting concentrated energy in the ultra-high frequency range, the feature extraction unit 131 can eliminate the physical interference of conventional macroscopic pressure fluctuations in the pipeline network and accurately extract the high-frequency destructive energy induced by local abnormal fluid events.

[0071] The extraction of dynamic parameters focuses on capturing abrupt changes in stiffness and evolution of frictional resistance between the actuator and the transmission chain. Feature extraction unit 131 performs time-domain difference decomposition on the torque state variables within a single action cycle of the actuator to obtain the torque change rate and torque fluctuation amplitude.

[0072] ;

[0073] ;

[0074] In the formula, The rate of change of torque; For time span The physical increment of driving torque within; This refers to the torque fluctuation amplitude of the actuator during its operating cycle. and These represent the maximum and minimum driving torques recorded within the corresponding period. To evaluate the difference in mechanical resistance of the transmission component under forces during forward and reverse motion, feature extraction unit 131 further calculates the asymmetry discrimination index of opening and closing torque:

[0075] ;

[0076] In the formula, This serves as an indicator for discriminating the asymmetry of opening and closing torque. This represents the average steady-state drive torque during the full-stroke process; This is the average steady-state drive torque during the full closing stroke. Under no-load commissioning conditions... If the value is too low, the system will also introduce a constant lower limit benchmark to replace the denominator to prevent the generation of distorted indicators that have no physical meaning.

[0077] After the hydraulic valve reaches the fully closed physical position, the feature extraction unit 131 calculates the sealing characteristic index of the fully closed state based on the steady-state fluid parameters within the fully closed maintenance time window:

[0078] ;

[0079] In the formula, This refers to the sealing characteristic index in the fully closed state; This is the actual measured pressure difference under the fully closed state; A preset healthy operating condition reference pressure difference is established. By quantifying the percentage deviation of the actual pressure difference from the preset healthy operating condition reference pressure difference, the system can intuitively reflect the degree of development of sealing abnormalities inside the sealing pair. As a preferred approach, the state analysis and diagnosis module 130 performs threshold determination based on the sealing characteristic indicators of the fully closed state, and performs auxiliary verification in combination with the pressure pulsation stability under the fully closed state, so as to avoid one-sided misjudgment caused by transient static pressure changes in a unilateral local pipeline network.

[0080] In this embodiment, to accurately determine the dynamic continuity of the start-stop action of the hydraulic valve and the unobstructed flow status, the feature extraction unit 131 simultaneously calculates the differential pressure change rate. First, the feature extraction unit 131 extracts the time points during the start-stop process. and For the corresponding differential pressure value, define and calculate the differential pressure change rate:

[0081] ;

[0082] In the formula, and They are time points and The measured inlet and outlet differential pressure is in Pascals. The differential pressure change rate is expressed in Pascals per second. The condition monitoring unit 132 monitors... Compliance determination of the numerical range of the action: If If the valve movement is below the set lower threshold, the status monitoring unit 132 determines that the valve is sluggish, stuck, or the actuator output torque is insufficient; if If the set upper limit threshold is exceeded, the system determines that there is a risk of water hammer impact and abnormal pipeline vibration.

[0083] After the feature extraction unit 131 completes the stripping of key parameters across various dimensions, the state analysis and diagnosis module 130 uses its internally configured state monitoring unit 132 to comprehensively determine the multi-dimensional parameters and monitor the overall operating status of the equipment. In this embodiment, the state monitoring unit 132 employs multi-level progressive execution logic to ensure the comprehensiveness and reliability of the determination.

[0084] When performing multi-parameter comprehensive judgment, the condition monitoring unit 132 utilizes DS evidence theory to perform primary mode fusion and anomaly interference suppression on multi-source sensor information. To overcome the deficiency of single sensors being highly susceptible to external electromagnetic or mechanical vibration interference and generating false alarms in complex water conservancy network environments, the condition monitoring unit 132 introduces evidence theory to integrate the preliminary judgment results of different physical modes. Considering the significant differences in sampling frequencies of heterogeneous sensors, and to ensure the logical correlation of multi-source evidence in the same physical space-time, as a preferred approach, the condition monitoring unit 132 pre-introduces a timestamp alignment mechanism based on a unified timing system. The system extracts synchronous time window data under the same steady-state operating conditions, and based on the vibration feature vector, transient turbulent kinetic energy, and asymmetric discrimination index of opening and closing torque output by the feature extraction unit 131, it maps the deviation of each key parameter from the normal benchmark to the basic probability allocation values ​​corresponding to the normal state, early deterioration state, and severe fault state through preset threshold segmentation rules or membership functions, and constructs mutually independent basic probability allocation functions respectively. The identification framework is defined as a set of three mutually exclusive elements: normal state, early degradation state, and severe fault state. For any two independent sensor information sources corresponding to the basic probability assignment function, the state monitoring unit 132 adopts a pairwise successive fusion strategy, using the evidence synthesis formula for fusion:

[0085] ;

[0086] ;

[0087] In the formula, For the synthesized target event The base probability allocation value; and These are subsets of the first information source and the second information source, respectively. and The base probability allocation value; , is the conflict coefficient, a dimensionless parameter used to measure the degree of logical contradiction between two sources of evidence; This represents the empty set. To avoid the conflict coefficient resulting from completely contradictory extreme evidence output when a single point of hardware failure occurs in the sensor. The denominator approaches 1, which in turn triggers an algorithmic logic dead zone where the denominator approaches 0. In this embodiment, the state monitoring unit 132 sets a threshold when the conflict coefficient approaches 1. When the above evidence synthesis formula is used, the system automatically discards it and forces the use of the arithmetic average weighting method of multi-source evidence for alternative calculation, thereby effectively suppressing extreme abnormal interference.

[0088] For complex multidimensional time-frequency domain feature matrices, directly setting thresholds for judgment makes it difficult to uncover deep physical correlations. Therefore, the condition monitoring unit 132 adopts a hybrid diagnostic architecture with multiple algorithms, utilizing support vector machines (SVMs), random forests, and convolutional neural networks (CNNs) for data classification, feature dimensionality reduction, noise enhancement, and fitting modeling. Specifically, the condition monitoring unit 132 is equipped with SVM and random forest algorithms as pre-processing modules. The SVM and random forest algorithms are trained offline based on historical monitoring data of similar hydraulic valves and corresponding operating status labels, and the trained model parameters are called during the deployment phase. The SVM uses radial basis function kernels to map multidimensional features to a high-dimensional linearly separable space, achieving preliminary data classification and isolating significantly deviating abnormal noise samples. Simultaneously, the condition monitoring unit 132 concatenates the vibration feature vectors, transient turbulent kinetic energy, fully closed sealing feature indicators, and dynamic parameters extracted by the feature extraction unit 131 column-wise to construct the original feature matrix. The random forest algorithm performs feature dimensionality reduction and noise enhancement on the original feature matrix by constructing multiple decision trees and calculating the Gini importance of each dimension of features. As a preferred approach, the condition monitoring unit 132 sorts the feature set in descending order according to the Gini importance score, retaining only the core feature subset with a cumulative importance contribution rate of 90%, thereby amplifying key fault features while eliminating redundant parameters.

[0089] After completing the preprocessing steps of data classification, dimensionality reduction, and denoising, the state monitoring unit 132 converts the simplified core feature matrix into a one-dimensional sequence through a row-wise flattening operation, and inputs it into the backend convolutional neural network for deep fitting modeling. In this embodiment, the input of the convolutional neural network is configured to receive a formatted one-dimensional feature tensor. The network hierarchy consists of two one-dimensional convolutional layers, one global average pooling layer, and two fully connected layers. The kernel sizes of the one-dimensional convolutional layers are set to 1x5 and 1x3, respectively, to capture local transient patterns in the sequence features, and a ReLU activation function is used after the convolution operation to introduce nonlinear fitting capability. The global average pooling layer is used to compress the length of the feature mapping sequence to prevent overfitting. The output fully connected layer contains one neuron and finally outputs the predicted degradation value for the current operating state of the hydraulic valve. During the model training phase, the state monitoring unit 132 collects historical full-dimensional datasets of the same type of hydraulic valve in each operating cycle as samples. To construct a numerical supervision label that matches the one-dimensional feature tensor, the state monitoring unit 132 extracts the wear depth of the hydraulic valve stem and the increase in the sealing clearance of the valve, measured manually, from the equipment maintenance records. After normalizing these values ​​to their maximum extreme values, the values ​​are weighted and summed using pre-determined weights based on correlation analysis of historical labeled samples, resulting in a single numerical value that serves as the supervision label representing the actual degradation. The training process uses mean squared error as the loss function and employs the Adam optimization algorithm with an initial learning rate of 0.001 for backpropagation and iterative updates of the network weights until the validation set loss value converges.

[0090] After acquiring the basic numerical parameters and the output results of the aforementioned model, the status monitoring unit 132 executes a preset threshold comparison trigger logic based on the configuration delay time, constructing a multi-level preset threshold comparison trigger mechanism. Taking the actuator torque change rate as an example, the status monitoring unit 132 presets a first-level warning threshold and a second-level alarm threshold for the torque change rate. The determination of these preset thresholds is based on the statistical extreme value distribution range of the same model of hydraulic valve under factory no-load and rated load tests. To prevent malfunctions caused by instantaneous electromagnetic interference or local fluid pulsation, this embodiment introduces an anti-jitter mechanism in the alarm triggering stage. When the monitoring parameters calculated in real time by the status monitoring unit 132 continuously exceed the set limits, and the duration of the over-limit state reaches the system's preset delay time (e.g., set to 3 seconds in this embodiment), the status monitoring unit 132 officially triggers the response command. Specifically, after reaching the configured delay time, if the parameter is in the first-level warning range, a system pop-up window is triggered and an early degradation log is recorded; if the parameter reaches or exceeds the second-level alarm range, an emergency shutdown command is immediately triggered and the upstream and downstream bypass valves of the pipeline are activated. This hard-triggered logic, acting as a bottom-level protective barrier beyond model prediction, ensures the safety of the device hardware even before the complex computational model has converged.

[0091] In addition to judging the data flow at the software algorithm level, to ensure the continuity and physical reliability of the data transmission link between the edge acquisition module 120 and the status analysis and diagnosis module 130, the status monitoring unit 132 also implements a top-down hardware communication status inspection mechanism. In this embodiment, the status monitoring unit 132 sends inspection commands to the underlying sensors, edge acquisition module 120, and communication links at a fixed cycle (this cycle can be customized by the user, and the default setting is 1 minute per cycle). After receiving the response data from the lower-level machine or waiting for a timeout, the status monitoring unit 132 comprehensively judges the real-time operating status of the underlying hardware by detecting the signal feedback strength of the sensors, statistically analyzing the data packet transmission success rate of the communication link in the current cycle, and analyzing the channel response status of the edge acquisition module 120. When the signal amplitude, signal-to-noise ratio, or output stability of a sensor loop is detected to be lower than a preset anti-interference threshold, or the data packet transmission success rate of the communication link is lower than a preset communication reliability threshold (e.g., a transmission success rate of less than 95% within a single cycle), or the channel of the edge acquisition module 120 is unresponsive, the status monitoring unit 132 determines that the corresponding hardware node is at risk of disconnection or physical damage and pushes a hardware fault alarm to the system maintenance personnel. This mechanism effectively avoids the logical risk that the upper-layer algorithm may use incorrect or outdated data for invalid calculations due to the failure of the underlying hardware.

[0092] After the status monitoring unit 132 completes the assessment of the real-time operating status of the equipment, the trend prediction unit 133 is responsible for predicting the trend of the degradation process of the hydraulic valve and pipeline system from a time dimension. During the long-term operation of the hydraulic valve and pipeline system, various mechanical wear and fluid erosion usually manifest as slow cumulative degradation. In this embodiment, the trend prediction unit 133 extracts specific historical monitoring parameters within a specific time span from the database. The aforementioned historical monitoring parameters specifically include historical time-series data of physical measurement points such as torque, inlet and outlet differential pressure, pipeline pressure, and pressure pulsation during the operation of the hydraulic valve. The physical basis for selecting the above measurement point data is that the continuous increase in opening and closing torque directly reflects the mechanical wear of the transmission mechanism or the degree of packing aging, while the continuous enhancement of abnormal inlet and outlet differential pressure and pressure pulsation indicates a trend of scaling blockage or cavitation damage in the internal flow channel of the valve body. To meet the requirements of subsequent prediction algorithms for equally spaced time series samples and avoid time series breaks caused by occasional sensor outages, the trend prediction unit 133 performs data cleaning logic before constructing the sequence. It uses the three-standard-deviation rule to remove isolated outliers and employs cubic spline interpolation to fill in missing time node data, thus constructing a one-dimensional time series strictly aligned to the physical time axis. As a preferred approach, the system uses a day as the base time step, extracting the average value of monitoring parameters under the same steady-state operating conditions each day to form an original feature vector containing data from multiple discrete time points. This mechanism effectively shields against parameter abrupt changes caused by load switching under different operating conditions.

[0093] After constructing the historical time series, to extract the inherent degradation trend of the series, the trend prediction unit 133 uses a quadratic exponential smoothing algorithm to smooth the time series and calculate the trend parameters. From the general technical principles of time series analysis, the evolution of physical parameters typically includes a base level component, a trend variation component, and a random fluctuation component. The exponential smoothing algorithm filters high-frequency random fluctuations by assigning weights to historical observation data that decrease exponentially with physical time; while quadratic exponential smoothing, based on this, specifically performs mathematical lag compensation for data with obvious linear growth or decline trends. To ensure that the recursive smoothing algorithm can be effectively started and to avoid divergence in the initial calculations, the trend prediction unit 133 sets the initial conditions of the algorithm at the initial zero point node of the physical time axis. Specifically, the initial values ​​of both the first and second exponential smoothing are set to the actual monitored parameter input values ​​of the first time node of the time series, i.e. and In the formula, The first exponential smoothing value of the initial zero node; Input the actual monitoring parameters for the first time point; The quadratic exponential smoothing value is the initial zero-point node.

[0094] After setting the initial values, the trend prediction unit 133 calculates the exponential smoothing value once:

[0095] ;

[0096] In the formula, For the current time node The first exponentially smoothed value; For the current time node The actual monitoring parameter input values; The previous time point The first exponentially smoothed value; This is the smoothing coefficient. The value range represents the sensitivity of the predictive model to recent data. In this embodiment, the smoothing coefficient... The specific values ​​are determined through grid search based on historical lifecycle data of hydraulic valves and the principle of minimizing the root mean square error of prediction. For mechanical wear parameters with relatively gradual changes, the smoothing coefficient... The default value is usually 0.3 to enhance the smoothing effect and filter out short-term fluctuations.

[0097] Based on the obtained first exponential smoothing value, the trend prediction unit 133 further calculates the second exponential smoothing value to compensate for the prediction lag bias caused by single smoothing when the sequence has a clear evolutionary trend:

[0098] ;

[0099] In the formula, For the current time node The quadratic exponential smoothed value; The previous time point The quadratic exponential smoothed value.

[0100] After completing the double smoothing process, the trend prediction unit 133 calculates the current time node based on the smoothing results. The trend-fit baseline value and the trend-fit degradation rate are calculated using the following formulas:

[0101] ;

[0102] ;

[0103] In the formula, For the current time node The trend fitting baseline value, whose physical meaning is the theoretical reference state quantity of the equipment at the current moment after eliminating high-frequency random fluctuations; For the current time node The trend fitting degradation rate characterizes the theoretical slope of equipment performance degradation over physical time. To ensure the denominator in the formula is not zero and to avoid algorithmic dead zones, in this embodiment, the system forcibly limits the smoothing coefficient during the parameter configuration stage. The maximum threshold is 0.95, thus preventing division overflow errors in the calculation program.

[0104] After obtaining the baseline state and degradation rate at the current time point, the trend prediction unit 133 performs a smooth extrapolation of fixed-duration parameters based on the physical time axis and an assessment of the equipment degradation trend. Specifically, the trend prediction unit 133 extrapolates into the future within the physical time domain, calculating the extrapolated prediction value after a fixed duration. The specific formula for the fixed-duration extrapolation prediction in the physical time domain is as follows:

[0105] ;

[0106] In the formula, For future time nodes Extrapolated predicted values; The number of look-ahead prediction time steps is a dimensionless parameter. As a preferred approach, the number of look-ahead prediction time steps... The sampling granularity is set according to the time series; when the step size is days, It can be set to 30; when the step size is the motion cycle, Multiple action cycles can be set according to predicted needs.

[0107] Through the aforementioned smooth extrapolation, the trend prediction unit 133 obtains the trajectory of equipment performance parameters over a future period. Subsequently, the system evaluates and compares the extrapolated predicted value with the equipment's preset performance failure threshold. If the extrapolated predicted value reaches or exceeds the performance failure threshold within the set forward prediction time window, the trend prediction unit 133 determines that the target hydraulic valve has a high potential failure risk and sends an alarm output list containing equipment maintenance guidance to the maintenance terminal. In this embodiment, the alarm output list includes detailed core elements such as the measurement point name, predicted over-limit time, current parameter value, predicted over-limit value, and alarm threshold. As a preferred engineering application scenario, when the system detects a sharp increase in the extrapolated predicted trend of the inlet and outlet differential pressure measurement points, combined with the specific over-limit time node calculated by the prediction formula, the system automatically outputs a warning message indicating that the differential pressure measurement point is expected to exceed the limit in 48 hours and suggesting checking the valve core scaling. This judgment logic transforms a post-fault alarm into a pre-fault status warning, providing sufficient time for on-site maintenance personnel to maintain equipment and replace spare parts.

[0108] In this embodiment, after completing the conventional time series prediction, the state analysis and diagnosis module 130 performs a deep cross-diagnosis of the multidimensional physical field coupling effect through the internally configured cascaded diagnosis unit 134. (See attached diagram.) Figure 3Since hydraulic valves typically face dual damage from mechanical wear and fluid cavitation in actual operating conditions, and the degradation processes of the two are highly coupled, the cascaded diagnostic unit 134 performs specific diagnostic tasks through the following internal logic.

[0109] To overcome the limitation of continuous physical time-based monitoring algorithms where performance derivatives approach zero when hydraulic valves remain stationary for extended periods, the cascaded diagnostic unit 134 maps the continuous time axis to discrete action cycle domain variables and introduces a maximum time slice constraint to prevent deadlock. Specifically, the system defines the complete mechanical stroke of the valve from opening to closing and then reopening as one action cycle. To prevent logical deadlocks in subsequent degradation rate calculation algorithms caused by prolonged equipment inactivity, the cascaded diagnostic unit 134 introduces a maximum time slice constraint mechanism. As a preferred approach, the system presets the maximum time slice to 72 hours, a value set based on the typical scheduling cycle of the water network. When the valve remains stationary for longer than this preset value, the system forcibly generates a virtual action cycle and collects the current static pressure-holding characteristics for time-domain occupancy, thereby ensuring the continuity of the discrete action cycle domain over time.

[0110] To address the need for accumulating minute damage within discrete operation cycles, the cascaded diagnostic unit 134 executes the Kahan tail compensation algorithm to prevent truncation errors in calculating extremely small cumulative cavitation damage. During the small valve opening adjustment process, the energy of a single cavitation damage event caused by localized fluid vaporization and rupture is extremely small. When the processor performs double-precision floating-point accumulation, this tiny increment is easily overwhelmed by the large existing accumulated value, leading to truncation distortion. To ensure accurate accumulation of cavitation damage throughout the entire equipment lifecycle, for any operation cycle... ( The formula for calculating the cumulative compensation is as follows:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] In the formula, To compensate for cavitation damage in the current cycle after eliminating the previous truncation error; For the first The single micro cavitation damage amount, measured in joules, is collected and calculated within each action cycle. In this embodiment, the single damage amount is obtained by the transient turbulent kinetic energy conversion obtained by the aforementioned feature extraction unit 131 based on the integral solution of the fluid pulsating pressure signal. To compensate for the truncation error left over from the previous cycle, in the initial state... ; This represents the total temporary cumulative damage for the current period. The new truncation error compensation amount generated by the cumulative calculation for the current cycle is used to pass on to the next cycle; For the updated number The final cumulative cavitation damage over each cycle, measured in joules. This algorithm uses independent variables... By temporarily storing the truncated mantissa of tiny floating-point numbers and returning it in the next calculation, the physical accuracy of the cumulative cavitation damage is maintained over a long period.

[0116] After obtaining the accurate cumulative cavitation damage, considering that the mechanical wear and fluid cavitation of the transmission mechanism evolve independently and worsen each other, the cascaded diagnostic unit 134 further calculates the degradation slope of the cross-mapping between the opening / closing torque asymmetry discrimination index and cavitation damage, and implements serial triple interlocking and reversible jamming verification logic. The cascaded diagnostic unit 134 extracts the opening / closing torque asymmetry discrimination index already solved by the aforementioned feature extraction unit 131 as a characterization of the transmission state. Subsequently, the system calculates the cross-modal coupling degradation rate of the opening / closing torque asymmetry discrimination index relative to the cumulative cavitation damage. To avoid the denominator of the division operation approaching zero due to the zero increment of single cavitation damage under certain operating conditions, the system introduces a minimum regularization constant in the divisor term of the calculation formula for hard lower limit protection. The corresponding cross-modal coupling degradation rate calculation formula is as follows:

[0117] ;

[0118] In the formula, The cross-modal coupling degradation rate characterizes the degree of mechanical structural asymmetry caused by unit cavitation damage; For the first The asymmetry discrimination index of the opening and closing torque extracted from each action cycle is a dimensionless percentage value. For the first The asymmetric index for the opening and closing torque of each cycle; To calculate the window span, a preset span is used in this embodiment. One action cycle; and These represent the cumulative cavitation damage for the corresponding period; As a minimum regularization constant, its value is 10 as an preferred approach. -8 joule.

[0119] Solve for the cross-modal coupling degradation rate Subsequently, to avoid false alarms caused by sudden changes in non-degradable parameters due to brief intrusion of foreign objects into the pipeline, the cascaded diagnostic unit 134 executes a serial verification process. When a... When the preset degradation rate safety threshold is exceeded, where this safety threshold is set based on the upper limit of the confidence interval of the device's historical health log data, the system opens a disturbance attenuation verification window containing three consecutive action cycles. Within the verification window, if... The value shows a step drop and falls back into the normal baseline range. The system determines that the current operating condition is a reversible blockage in the flow channel. The cascaded diagnostic unit 134 only records the disturbance event in the equipment log and outputs flushing maintenance suggestions to the maintenance terminal. Conversely, if If the damage remains high or continues to worsen within the verification window, the system will officially activate the subsequent DS evidence interlocking and confirmation logic based on multi-source information fusion to comprehensively confirm the true coupled degradation state of mechanical damage and cavitation damage, and output the final comprehensive deterioration diagnosis conclusion.

[0120] As the service life of equipment extends, hydraulic valves gradually enter a stable and controllable stage of operation with defects. Continuing to use the factory-set absolute health benchmark for exceeding limits will result in continuous redundant alarm output. Therefore, the cascaded diagnostic unit 134 performs dynamic physical residual stripping and adaptive reconstruction of the defect benchmark based on dimensionless damage polynomial fitting. The cascaded diagnostic unit 134 calculates the dynamic physical residual between the actual measured pressure drop and the preset healthy operating condition benchmark pressure difference. The corresponding dynamic physical residual calculation formula is:

[0121] ;

[0122] In the formula, For the first Dynamic physical residuals for each cycle; The actual measured pressure drop through the hydraulic valve; The baseline differential pressure for the health condition is pre-calibrated or updated online based on the valve's historical health operation data, current opening degree, and corresponding health condition label. Simultaneously, the cascaded diagnostic unit 134 calculates a dimensionless damage factor to characterize the current degree of equipment degradation. The corresponding formula for calculating the dimensionless damage factor is:

[0123] ;

[0124] In the formula, For the first Dimensionless damage factor for each cycle; The reference cavitation critical energy constant, in Joules, is used to calibrate the hydraulic valve during the design phase. The system further employs a recursive least squares method to perform online fitting of the attenuation compensation polynomial based on historical data, obtaining the physical residual penalty term used for benchmark compensation. The corresponding attenuation compensation polynomial is:

[0125] ;

[0126] In the formula, The attenuation compensation amount, calculated based on the dimensionless damage factor, is directly used as the physical residual penalty. Participate in calculation; and These are the polynomial fitting coefficients. Using the obtained physical residual penalty, the system reconstructs the original preset healthy operating condition baseline pressure difference by translation:

[0127] ;

[0128] In the formula, This serves as the baseline for predicting defects after superimposing physical residuals. Through the aforementioned reconstruction, the system's compensation equipment undergoes reasonable normal degradation, allowing the residual judgment logic to focus on sudden, severe failures.

[0129] Based on the reconstructed prediction benchmark, the cascaded diagnostic unit 134 performs homogeneous evolution prediction based on global damage variables and triggers hardware-level safety circuit breaker action according to the multidimensional mathematical feature boundary. The cascaded diagnostic unit 134 constructs a homogeneous evolution prediction equation mapped to global damage variables:

[0130] ;

[0131] In the formula, Global damage independent variable Increase The predicted values ​​of the evolution of the target parameters; This is a composite global damage independent variable constructed by linearly weighting discrete action periodic variables and dimensionless damage factors according to pre-calibrated weight coefficients. This represents the baseline state value of the parameters at the current node. This represents the degeneracy slope.

[0132] While relying on the above equations to extrapolate the steady-state trend, to prevent pipeline pressure loss caused by severe internal leakage, the cascade diagnostic unit 134 configures an extreme value boundary determination strategy based on underlying mathematical characteristics. When the edge acquisition module 120 reports that the hydraulic valve has reached the fully closed limit state, the cascade diagnostic unit 134 compares the frequency domain data collected by the current sensor array in real time and calculates the absolute cosine distance between the current vibration feature vector and the preset healthy fully closed state reference vector. The system simultaneously extracts the sealing feature indicators of the fully closed state that have been solved in the previous step. The system performs a joint over-limit assessment based on the inlet and outlet pressure stability under the fully closed state. The inlet and outlet pressure stability is characterized by the fluctuation amplitude or standard deviation of the inlet and outlet static pressure within the fully closed maintenance time window. If the absolute cosine distance falls below the preset similarity lower limit threshold, which is set based on the allowable drift bandwidth of the structure's natural frequency under the valve's intact sealing state, and the sealing characteristic index β and the inlet and outlet pressure stability index simultaneously exceed the set safety envelope under the fully closed state, the cascade diagnostic unit 134 determines that the equipment has encountered irreversible severe internal leakage. The safety envelope is obtained by pre-calibrating the sealing characteristic index and the statistical upper limit of inlet and outlet pressure stability under the fully closed state under healthy conditions. At this time, the system directly issues a safety fuse command to the actuator controller, forcibly locking the current servo control loop and triggering a local audible and visual alarm, thereby ensuring the extreme operational safety of the physical pipeline network.

[0133] To further verify the engineering effectiveness of the intelligent status detection system and method for hydraulic valves provided by this invention, the following detailed explanation is provided in conjunction with specific field application test data and experimental verification process.

[0134] In this embodiment, the test object is an electric butterfly valve with a nominal diameter of DN800 that has been continuously in service in a municipal water supply pumping station. The rated operating static pressure of the pipeline network where this hydraulic valve is located is 0.8 MPa. The data sensing module 110 is installed according to the aforementioned rules, and the range of the torque sensing unit 113 is set to 3000 N·m. The experimental period is set to 180 consecutive days, during which 85 complete opening and closing cycle data of the target hydraulic valve are recorded.

[0135] See attached document Figure 4 , attached Figure 4 This visually presents a segment of the underlying data stream received and processed by the edge acquisition module 120 during the 45th actuation cycle of the hydraulic valve. (From the attached...) Figure 4 It can be clearly observed that within the time interval of 2.5 to 3.8 seconds (corresponding to the small throttling stage of the hydraulic valve opening to 15% to 25%), significant high-frequency broadband spikes appear in the pressure pulsation waveform, while the amplitude of the vibration acceleration waveform is synchronously amplified. The local computing component 122 successfully extracted the characteristic frequency band with an energy amplitude exceeding 3.5 times the global average energy amplitude by performing a short-time Fourier transform on the original digital signal within this interval. The feature extraction unit 131, based on the attached... Figure 4 The transient turbulent kinetic energy obtained by the integral calculation of the fluid pulsating pressure signal shown reaches 0.045 joules, thus accurately locating the small cavitation initiation event that occurred within this action cycle.

[0136] To verify the technical advantages of this invention in predicting equipment degradation trends and accumulating minor damage, a control group experiment was set up in this embodiment. The control group adopted a traditional fixed threshold alarm strategy and a conventional single-order exponential smoothing prediction algorithm, without introducing Kahan tail compensation logic and dynamic physical residual reconstruction mechanism. The experimental group of this invention fully executed the aforementioned state analysis and diagnosis module 130 and the cascaded diagnosis and smoothing extrapolation logic of its internal units.

[0137] See attached document Figure 5 , attached Figure 5 It contains three data curves: the solid black line represents the actual degradation baseline determined by post-disassembly and inspection; the dashed black line represents the traditional prediction curve output by the control group; and the dotted black line represents the comprehensive degradation prediction curve output by the trend prediction unit 133 and the cascaded diagnostic unit 134 in the experimental group of this invention.

[0138] Combining experimental test data and appendix Figure 5 The curve comparison shows that during the early, stable operation phase from the 1st to the 30th operation cycle, the energy of micro-cavitation is extremely low. Due to the truncation error of double-precision floating-point accumulation, the cumulative cavitation damage recorded by the control group remained at 0 for a long time, resulting in a horizontal dead zone on the traditional prediction curve (black dashed line), completely failing to reflect the latent wear of the equipment. In contrast, the experimental group of this invention, under the action of the Kahan tail compensation algorithm of the cascaded diagnostic unit 134, accurately retained the energy as low as 10 for each operation. -6 The minute cavitation increments at the joule level allow the overall degradation prediction curve (black dotted line) to closely match the initial slow upward trend of the actual degradation baseline (black solid line).

[0139] After the 60th operating cycle, the sealing pair of the target hydraulic valve experienced an asymmetric increase in opening and closing resistance due to the peeling off caused by previous cavitation. At this point, the asymmetric discrimination index of the opening and closing torque calculated by the feature extraction unit 131 exceeded the set deviation of 8%. The control group, due to the failure to remove normal physical residuals, frequently triggered false alarms in the 65th operating cycle. In contrast, the cascaded diagnostic unit 134 of the experimental group of this invention, by introducing a dimensionless damage factor to calculate the attenuation compensation amount, reconstructed the defect prediction benchmark through translation, effectively suppressing redundant alarms.

[0140] As degradation intensifies, the trend prediction unit 133 of this invention, in the 70th operation cycle, performs a smooth extrapolation with 15 look-ahead prediction time steps based on the trend fitting baseline value and trend fitting degradation rate calculated by the aforementioned quadratic exponential smoothing algorithm. The trend prediction unit 133 accurately predicts 15 operation cycles in advance that the target hydraulic valve will reach the cross-modal coupling degradation rate limit threshold of 0.15 in the 85th operation cycle, and sends an alarm output list containing suggestions for replacing the sealing ring and polishing the valve core to the maintenance terminal in advance. Comparison of quantitative evaluation indicators shows that, for the core parameter of cross-modal coupling degradation rate, the root mean square error (RMSE) of the traditional control group prediction model is 0.038, while the RMSE of the experimental group of this invention is only 0.012, improving the prediction accuracy by 68.4%. The above experimental data and comparison curves demonstrate that the intelligent condition detection system and method of this invention can identify minor damage under complex hydraulic operating conditions and improve the accuracy of predicting long-term operational degradation trends.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart status detection system for hydraulic valves, characterized in that, include: The data sensing module is located at the preset monitoring points of the hydraulic valve and is used to collect electrical signals of multi-physical field state parameters. An edge acquisition module, connected to the data sensing module, includes a synchronous acquisition unit and a local computing component. The synchronous acquisition unit is used to acquire electrical signals and convert them into raw digital signals. The local computing component is used to filter and denoise the raw digital signals to obtain denoised digital signals, extract the characteristic frequency bands of the denoised digital signals, and concatenate the denoised digital signals with the characteristic frequency bands to form a bottom-level data stream. The status analysis and diagnosis module receives the underlying data stream and includes a feature extraction unit, a status monitoring unit, a trend prediction unit, and a cascade diagnosis unit. The feature extraction unit is used to extract key parameters characterizing the mechanical, fluid, and dynamic states of the hydraulic valve from the underlying data stream. The status monitoring unit is used to determine the operating status of the hydraulic valve based on the key parameters. The trend prediction unit is used to perform smooth extrapolation calculations on the key parameters. The cascade diagnosis unit is used to calculate cumulative cavitation damage based on the key parameters and perform defect prediction benchmark reconstruction.

2. The intelligent status detection system for hydraulic valves according to claim 1, characterized in that, The data sensing module includes: A vibration sensing unit is installed at preset mechanical monitoring points of the hydraulic valve to acquire electrical signals of high-frequency mechanical state characteristics. Pressure sensing units are installed in the pipe sections before and after the hydraulic valve to acquire electrical signals that indicate the fluid state characteristics. A torque sensing unit is installed at the drive mechanism of the hydraulic valve to acquire electrical signals that reflect the mechanical dynamics characteristics. An opening degree sensing unit is connected to the output control shaft of the actuator of the hydraulic valve and is used to acquire an electrical signal of the absolute opening degree value.

3. The intelligent status detection system for hydraulic valves according to claim 1, characterized in that, The local computing component uses a wavelet transform algorithm to decompose the original digital signal into high-frequency detail coefficients and low-frequency approximation coefficients. It then uses a hard thresholding method to process the high-frequency detail coefficients and performs inverse wavelet transform on the processed high-frequency detail coefficients and low-frequency approximation coefficients to reconstruct the denoised digital signal. The local computing component uses a short-time Fourier transform algorithm to map the denoised digital signal from a one-dimensional time domain to a two-dimensional time-frequency domain, and extracts high-frequency intervals in the time-frequency domain matrix whose energy amplitude exceeds a preset multiple of the global average energy amplitude as the characteristic frequency band.

4. The intelligent status detection system for hydraulic valves according to claim 2, characterized in that, The feature extraction unit extracts the energy centroid frequency, root mean square amplitude, and temporal kurtosis parameter within the feature frequency band to construct a feature vector, and calculates the Mahalanobis distance between the feature vector and the normal reference spectrum space as a key parameter characterizing the mechanical state of the hydraulic valve. The feature extraction unit uses the Fast Fourier Transform algorithm to deconstruct the frequency domain of the denoised digital signal after the electrical signal acquired by the pressure sensing unit, obtain the frequency domain complex amplitude, calculate the power spectral density, and integrate to obtain the transient turbulent kinetic energy. The transient turbulent kinetic energy is used as a key parameter characterizing the fluid state of the hydraulic valve.

5. The intelligent status detection system for hydraulic valves according to claim 2, characterized in that, The feature extraction unit performs time-domain difference decomposition on the torque state variables within a single action cycle obtained by converting the electrical signal acquired by the torque sensing unit to obtain the torque change rate and torque fluctuation amplitude. It then calculates the average steady-state driving torque during the fully open stroke and the average steady-state driving torque during the fully closed stroke to construct an asymmetric discrimination index for opening and closing torque. The torque change rate, torque fluctuation amplitude, and opening and closing torque asymmetric discrimination index are used as key parameters characterizing the dynamic state of the hydraulic valve.

6. The intelligent status detection system for hydraulic valves according to claim 2, characterized in that, The status monitoring unit uses the evidence synthesis formula of DS evidence theory to perform primary modal fusion on the basic probability allocation function of multiple independent sensing units in the data sensing module. When the set conflict coefficient is greater than or equal to the conflict determination threshold, the evidence synthesis formula is replaced by the arithmetic average weighting method of multi-source evidence to output a preliminary judgment result for the operating status of the hydraulic valve.

7. The intelligent status detection system for hydraulic valves according to claim 1, characterized in that, The state monitoring unit uses the support vector machine algorithm to map the multidimensional features contained in the key parameters to a linearly separable space through the radial basis kernel function in order to isolate abnormal noise samples. It uses the random forest algorithm to perform feature dimensionality reduction on the original feature matrix formed by splicing the key parameters by calculating the Gini importance of each dimension of the multidimensional features to obtain the core feature matrix. The status monitoring unit converts the core feature matrix into a one-dimensional sequence and inputs it into a convolutional neural network for fitting and modeling. The output of the convolutional neural network is then used to output the predicted degradation value of the hydraulic valve.

8. The intelligent status detection system for hydraulic valves according to claim 1, characterized in that, The trend prediction unit uses a cubic spline interpolation algorithm to numerically fill in the missing time node data contained in the key parameters in order to construct a one-dimensional time series. The trend prediction unit uses a quadratic exponential smoothing algorithm to smooth the one-dimensional time series, calculates the first and second exponential smoothing values ​​at the current time node, solves the trend fitting baseline value and trend fitting degradation rate at the current time node based on the first and second exponential smoothing values, and uses the trend fitting baseline value and trend fitting degradation rate combined with a fixed duration parameter smoothing extrapolation logic to calculate the extrapolated prediction value for future time nodes.

9. The intelligent status detection system for hydraulic valves according to claim 2, characterized in that, The cascaded diagnostic unit uses the Kahan tail number compensation algorithm to calculate the cumulative cavitation damage of a single minor cavitation damage to a hydraulic valve over multiple operating cycles. It separates the truncated part of the floating-point number tail number generated in the calculation to generate a truncation error compensation amount and passes it to the next operating cycle for compensation accumulation. The cascaded diagnostic unit calculates the dynamic physical residual between the actual measured pressure difference and the preset healthy operating condition benchmark pressure difference based on the electrical signal obtained by the pressure sensing unit. It combines a dimensionless damage factor and obtains a physical residual penalty term by performing polynomial online fitting using the recursive least squares method. The physical residual penalty term is then used to translate and reconstruct the preset healthy operating condition benchmark pressure difference to generate a disease prediction benchmark.

10. The intelligent status detection system for hydraulic valves according to claim 9, characterized in that, The cascaded diagnostic unit is also used to perform serial verification and hardware-level security circuit breaking; The cascaded diagnostic unit calculates the cross-modal coupling degradation rate based on the asymmetric discrimination index of opening and closing torque and the cumulative cavitation damage. When the cross-modal coupling degradation rate is detected to exceed the preset degradation rate safety threshold, the cascaded diagnostic unit opens a disturbance attenuation verification window containing multiple consecutive action cycles. If the cross-modal coupling degradation rate does not drop to the preset normal threshold range within the verification window, the operation status weighting logic based on multi-source information fusion is activated to output a comprehensive degradation diagnosis conclusion. When the edge acquisition module reports that the hydraulic valve has reached the fully closed limit state, if the absolute cosine distance between the current vibration feature vector extracted by the feature extraction unit and the preset healthy fully closed state benchmark vector falls below the preset similarity lower limit threshold, and the fully closed state sealing feature index and the inlet and outlet pressure stability index simultaneously exceed the set safety envelope, the cascade diagnosis unit issues a safety fuse command to the actuator of the hydraulic valve. The inlet and outlet pressure stability index is characterized by the fluctuation amplitude or standard deviation of the inlet and outlet static pressure within the fully closed maintenance time window, and the safety envelope is obtained by pre-calibration based on the sealing characteristic index of the fully closed working condition under healthy conditions and the statistical upper limit of the inlet and outlet pressure stability index.

Images