A valve fault monitoring method based on vibration and flow signal collaborative analysis

Abstract

The application provides a valve fault monitoring method based on vibration and flow signal collaborative analysis, and belongs to the technical field of industrial automation process control and equipment state monitoring, and comprises the following steps: synchronously collecting flow, vibration and opening degree signals, calculating fluid excitation force through a fluid mechanics lumped parameter model, stripping the flow-induced vibration component from the measured vibration in combination with a transfer function to obtain a pure mechanical vibration residual signal; extracting kurtosis values and fault characteristic frequency amplitudes for each axis component to generate mechanical state indicators, and simultaneously calculating the root mean square value of the fluid excitation force to generate hydraulic state indicators; inputting the three indicators into a two-level recursive mutual verification decision structure, dividing the field with the hydraulic indicators, and positioning the fault with the mechanical indicators. The application strips the flow-induced vibration from the physical layer, effectively distinguishes the flow field disturbance and the mechanical fault, and improves the accuracy and reliability of the valve online monitoring.

Description

Technical Field

[0001] This invention relates to the field of industrial automation process control and equipment condition monitoring technology, specifically to a valve fault monitoring method based on the collaborative analysis of vibration and flow signals. Background Technology

[0002] Butterfly valves are crucial actuators in industrial process control systems, and their reliable operation is key to ensuring the precise execution of control loops. If a mechanical fault in a butterfly valve is not detected in time, it will lead to control loop failure, affecting the accuracy of process control and the safe and stable operation of the system. Therefore, online monitoring and fault diagnosis of butterfly valve operation status is an important development direction for intelligent manufacturing and intelligent operation and maintenance of industrial equipment. Currently, fault monitoring of butterfly valves in industrial settings mainly relies on periodic disassembly and manual inspection. Periodic disassembly requires interrupting production processes, causing system downtime and consuming significant manpower and time. Manual inspection relies on the experience of maintenance personnel, is highly subjective, and makes it difficult to detect early faults. These passive maintenance modes cannot meet the needs of industrial automation process control for real-time perception and intelligent diagnosis of equipment status. Vibration analysis-based online valve monitoring methods collect vibration signals and extract characteristic parameters by installing accelerometers on the valve surface, comparing them with normal reference values ​​to determine faults. However, the vibration signal of a butterfly valve contains both flow-induced vibration and mechanical fault vibration components. Flow-induced vibrations are caused by fluid throttling, eddies, and pressure pulsations, and are normal fluid dynamic responses. Their intensity and frequency domain characteristics vary drastically with operating parameters such as flow rate and pressure, and overlap with mechanical fault vibrations in the frequency spectrum. The two types of vibrations overlap at the sensor measurement point, making it difficult to effectively separate them using existing methods. This leads to normal fluctuations in flow-induced vibrations being misjudged as mechanical faults.

[0003] Furthermore, the operating conditions of butterfly valves change frequently due to adjustments in the production process, resulting in non-stationary statistical characteristics of vibration signals. Traditional fixed threshold methods have limited adaptability under variable operating conditions and non-stationary signals, easily leading to missed or false alarms. When transient events such as changes in upstream pump speed occur in the pipeline system, the flow field itself exhibits abnormal fluctuations, making it difficult for existing methods to effectively distinguish between flow field disturbances and valve mechanical failures, further increasing the risk of false alarms. In industrial automation process control systems, false alarms not only increase unnecessary maintenance workload but may also lead to erroneous control decisions.

[0004] In summary, existing valve fault monitoring methods suffer from several problems, including false alarms due to the aliasing of flow-induced vibration and mechanical fault vibration, limited adaptability of fixed threshold methods under variable operating conditions and non-stationary signals, and difficulty in effectively distinguishing between flow field disturbances and mechanical faults. There is an urgent need for an online valve monitoring method capable of physically separating flow-induced vibration components from monitoring signals and effectively identifying flow field disturbances and mechanical faults, in order to meet the requirements of industrial automation process control systems for real-time equipment status perception, intelligent diagnosis, and high reliability. Summary of the Invention

[0005] The first aspect disclosed in this application provides a valve fault monitoring method based on the joint analysis of vibration and flow signals, including: A flow meter is installed upstream of the valve, a first triaxial vibration sensor is installed at the valve body, and a second triaxial vibration sensor is installed at the valve stem. Vibration signals output by the first and second triaxial vibration sensors, flow signals output by the flow meter, and valve opening signals are synchronously acquired at a preset sampling frequency during valve operation. The flow signals are low-pass filtered, and the two vibration signals are wavelet thresholded to obtain preprocessed flow time-series data, valve body triaxial vibration time-series data, and valve stem triaxial vibration time-series data. Using preprocessed flow time-series data and valve opening signal as input, the force exerted by the fluid on the valve plate under the current operating condition is calculated through a pre-constructed fluid dynamics lumped parameter model, yielding the fluid excitation force signal. Based on the current valve opening signal, the valve body vibration transfer function and valve stem vibration transfer function matching the valve opening signal are determined. Based on the fluid excitation force signal and the two transfer functions, the first flow-induced vibration response component at the valve body position and the second flow-induced vibration response component at the valve stem position are calculated. The preprocessed valve body triaxial vibration time-series data is synchronously subtracted from the first flow-induced vibration response component to obtain the valve body triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component. The preprocessed valve stem triaxial vibration time-series data is synchronously subtracted from the second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component. For each axis component of the valve body pure mechanical vibration residual signal and the valve stem pure mechanical vibration residual signal, the time-domain kurtosis value and the frequency-domain amplitude corresponding to the preset fault characteristic frequency are extracted as feature values. Each feature value is compared with the corresponding preset normal range. When any feature value exceeds its corresponding preset normal range, it is determined that the corresponding mechanical structure is abnormal, and a mechanical state flag representing the abnormal state is generated; otherwise, a mechanical state flag representing the normal state is generated. At the same time, the root mean square value of the fluid excitation force signal is calculated, and this root mean square value is compared with the preset hydraulic steady-state threshold range. When the root mean square value exceeds the hydraulic steady-state threshold range, it is determined that the flow field is unstable, and a hydraulic state flag representing the abnormal state is generated; otherwise, a hydraulic state flag representing the normal state is generated. Using the hydraulic state flag as the first-level judgment variable, the current state is classified into a hydraulic steady-state field or a hydraulic abnormal field. Using the valve body mechanical state flag and the valve stem mechanical state flag as the second-level judgment variables, fault location is performed in each field, and the corresponding fault judgment result is output.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: 1) By constructing a lumped parameter model of fluid dynamics, the flow rate signal is converted into a fluid excitation force signal. Then, by using a transfer function calibrated in intervals, the flow-induced vibration component is physically separated from the measured vibration signal, resulting in a pure mechanical vibration residual signal that reflects only the mechanical structure state. This method solves the problem of overlap between flow-induced vibration and mechanical fault vibration at the sensor measurement point from a physical perspective, avoiding the misjudgment of normal fluctuations in flow-induced vibration as mechanical faults and significantly reducing the false alarm rate.

[0007] 2) Triaxial vibration sensors are installed on both the valve body and valve stem. Signals from each axis are independently acquired, isolated, and their features extracted, preserving directional information. Vibration responses in different directions exhibit varying sensitivities to different fault modes. Split-axis processing ensures that anomalies in any direction can be independently captured, improving the detection sensitivity for faults in specific directions. Furthermore, a data-driven threshold construction strategy based on the full-condition discrete envelope is proposed. This strategy traverses the local discreteness of each feature index across all healthy operating conditions and extracts the maximum extremum as the global safety boundary. This strategy avoids the false alarms or missed alarms caused by fixed thresholds during variable operating conditions.

[0008] 3) A two-level hierarchical cross-verification decision structure based on Boolean logic was constructed. Hydraulic state indicators were used as the first-level decision variables to divide the hydraulic steady-state field and the hydraulic anomaly field. Mechanical state indicators of the valve body and valve stem were used as the second-level decision variables to locate faults within each field. When hydraulic anomalies occur while mechanical conditions are normal, the problem is clearly identified as upstream flow field disturbance or process fluctuation, indicating no fault in the valve body. False alarms are actively suppressed, effectively solving the engineering challenge of distinguishing between flow field disturbances and mechanical faults in variable-condition industrial environments.

[0009] 4) This invention organically combines flow-induced vibration stripping, interval transfer function calibration, triaxial channel-specific independent processing, full-condition discreteness envelope threshold strategy, and fluid-structure interaction characteristic cross-validation decision structure to form a complete online valve fault monitoring method. This method pre-calibrates the transfer function and fault characteristic frequencies through offline finite element analysis, which are then directly invoked during online monitoring, reducing the online computational load, meeting the real-time requirements of industrial edge computing devices, and can be widely applied to the online monitoring and fault diagnosis of various industrial valves, demonstrating good versatility and engineering practical value. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of a valve fault monitoring method based on the collaborative analysis of vibration and flow signals, provided as an embodiment of this application. Detailed Implementation

[0011] This application provides a valve fault monitoring method based on the collaborative analysis of vibration and flow signals. The technical solution of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0012] like Figure 1 As shown, this application provides a valve fault monitoring method based on the joint analysis of vibration and flow signals. The method specifically includes the following steps: S1: Install a flow meter on the upstream straight pipe section of the valve, install a first triaxial vibration sensor on the valve body, and install a second triaxial vibration sensor on the valve stem; obtain the valve opening signal from the valve actuator; synchronously collect the vibration signals output by the first and second triaxial vibration sensors, the flow signal output by the flow meter, and the valve opening signal under the valve operating state at a preset sampling frequency; perform low-pass filtering on the flow signal to remove high-frequency pulsation noise, and perform wavelet threshold noise reduction processing on the two vibration signals respectively to obtain preprocessed flow time series data, valve body triaxial vibration time series data, and valve stem triaxial vibration time series data.

[0013] Specifically, an electromagnetic or vortex flow meter is installed on the upstream straight pipe section of the valve, at a distance of three times the pipe diameter from the valve inlet. A first triaxial vibration sensor is installed on the outer wall of the valve body near the valve seat, and a second triaxial vibration sensor is installed on the top of the valve stem or on the side of the valve stem protruding from the stuffing box. Both the first and second triaxial vibration sensors are piezoelectric accelerometers. The valve opening signal is directly extracted from the valve actuator. For pneumatic or hydraulic actuators, the real-time displacement of the valve stem is obtained from the actuator's displacement sensor, and the valve stem displacement is converted into a valve opening percentage; for electric actuators, the valve opening percentage signal is directly read from the servo motor encoder or position feedback module. An industrial data acquisition card is used to synchronously acquire the three vibration signals output by the first and second triaxial vibration sensors, the flow signal output by the flow meter, and the valve opening signal acquired from the valve actuator at a preset sampling frequency. The preset sampling frequency is set according to the highest frequency of interest of the valve's mechanical structure, typically five to ten times the highest frequency of interest. The preset sampling frequency is set based on the highest frequency of interest in the valve's mechanical structure, typically five to ten times the highest frequency of interest. The acquired flow signal is input to a Butterworth low-pass filter with a preset cutoff frequency of 5 Hz to 50 Hz to remove high-frequency pulsation noise caused by pump pulsation or local eddy currents in the pipeline. The cutoff frequency is selected based on the characteristic frequency of fluid pulsation in the pipeline and the frequency band of valve throttling noise to ensure that high-frequency noise is filtered out while retaining the flow trend component. To eliminate the nonlinear phase delay introduced by the filter and ensure that the filtered flow time series data and vibration time series data are strictly aligned on the time axis, the low-pass filter adopts a forward-backward filtering strategy, i.e., zero-phase filtering. The three vibration signals output from the first triaxial vibration sensor and the three vibration signals output from the second triaxial vibration sensor are respectively subjected to wavelet thresholding noise reduction. The wavelet thresholding noise reduction uses Daubechies wavelets as the wavelet basis function, employing a soft thresholding function and a general thresholding rule. For each vibration signal, wavelet decomposition is first performed, the wavelet coefficients are thresholded, and then wavelet reconstruction is performed to obtain the denoised vibration signal. The three denoised vibration signals from the first triaxial vibration sensor are used as the preprocessed X-axis, Y-axis, and Z-axis vibration time series data of the valve body, denoted as V. body,x (t), V body,y (t) and V body,z (t). The three vibration signals after noise reduction from the second and third axis vibration sensors are used as the preprocessed valve stem X-axis, Y-axis, and Z-axis vibration time series data, respectively, denoted as V. stem,x (t), V stem,y (t) and V stem,z (t). The low-pass filtered flow signal is retained as preprocessed flow time-series data. The valve opening signal is retained as preprocessed valve opening data.

[0014] This embodiment focuses on the large-diameter electric butterfly valves of the ballast water system of an ocean-going cargo ship. The ballast water system is used to regulate the ship's draft and navigational stability. The system consists of multiple large-diameter electric butterfly valves controlling the inflow and outflow of water into each ballast tank. The nominal diameter of the valves ranges from DN400 to DN800, and the medium is seawater. Seawater has a high salt content and contains silt particles, which have a corrosive and erosive effect on the valve sealing surfaces and moving parts. During ship navigation, the ballast water system frequently operates the valves according to the loading and unloading status and weather and sea conditions; the number of valve operations during a single voyage can reach dozens to hundreds. After long-term operation, valves mainly exhibit two types of faults: Valve body-side faults mainly include: wear and aging of the valve plate sealing ring due to seawater immersion and silt erosion, resulting in incomplete valve closure and water leakage between compartments; increased clearance of the valve shaft bearing due to seawater corrosion, causing jamming during opening and closing or triggering abnormal valve body vibration; Valve stem-side faults mainly include: loosening of the transmission components of the electric actuator due to long-term vibration, resulting in inconsistency between the valve opening feedback signal and the actual opening, and in severe cases, loosening at the connection between the valve stem and the valve plate, causing impact vibration in the valve stem transmission system, making it impossible for the valve to open and close normally.

[0015] In this embodiment, an electromagnetic flowmeter is installed on the upstream straight pipe section of the ballast water main valve to acquire real-time flow signals. A first triaxial vibration sensor is installed at the bearing seat on top of the valve body to monitor the vibration state of the valve body. A second triaxial vibration sensor is installed at the end of the valve shaft near the actuator to monitor the vibration state of the valve stem. Both sensors are piezoelectric accelerometers resistant to salt spray corrosion to adapt to the high salt spray environment at sea. The valve opening signal is provided by the rotary transformer built into the electric actuator and acquired through the ship's centralized monitoring system. The installation of the above sensors does not change the original structure of the valve and does not affect the normal operation of the ballast water system. During normal ship navigation, online monitoring is implemented according to the following steps: an industrial data acquisition card synchronously acquires various signals at a preset sampling frequency of 1 kHz, including: three vibration signals output by the first triaxial vibration sensor, three vibration signals output by the second triaxial vibration sensor, the flow signal output by the electromagnetic flowmeter, and the valve opening signal acquired from the electric actuator. Considering the electromagnetic interference in the ship's electrical network and the low-frequency fluctuations in seawater pressure within the pipeline caused by changes in the ship's draft, a 5 Hz low-pass filter is applied to the flow signal to remove wave-induced pulsating noise. Wavelet thresholding is performed on both vibration signals. Daubechies wavelets are selected as the wavelet basis function, and a soft thresholding function and general thresholding rules are used. Wavelet decomposition, wavelet coefficient thresholding, and wavelet reconstruction are performed on each vibration signal to obtain the denoised vibration signal. The three denoised vibration signals from the first triaxial vibration sensor are vector-synthesized to obtain the preprocessed valve body vibration time-series data. The three denoised vibration signals from the first triaxial vibration sensor are used as the preprocessed valve body X-axis, Y-axis, and Z-axis vibration time-series data, respectively. The three denoised vibration signals from the second triaxial vibration sensor are used as the preprocessed valve stem X-axis, Y-axis, and Z-axis vibration time-series data, respectively. The low-pass filtered flow signal is used as the preprocessed flow time-series data. At this point, four sets of time-series data that are strictly synchronized in time and have effectively suppressed noise are output: flow rate time-series data, valve body vibration time-series data, valve stem vibration time-series data, and valve opening data. These preprocessing operations ensure the quality of the input signals for subsequent feature extraction and fault determination, providing a fundamental guarantee for the accuracy of the entire diagnostic method.

[0016] S2: Using preprocessed flow time-series data and valve opening signal as input, the force exerted by the fluid on the valve plate under the current operating condition is calculated through a pre-constructed fluid dynamics lumped parameter model to obtain the fluid excitation force signal; based on the current valve opening signal, the valve body vibration transfer function and valve stem vibration transfer function matching the valve opening signal are determined; based on the fluid excitation force signal and the two transfer functions, the first flow-induced vibration response component at the valve body position and the second flow-induced vibration response component at the valve stem position are calculated; the preprocessed valve body triaxial vibration time-series data is synchronously subtracted from the first flow-induced vibration response component to obtain the valve body triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component; the preprocessed valve stem triaxial vibration time-series data is synchronously subtracted from the second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component.

[0017] Furthermore, the fluid excitation force signal is obtained, including: Obtain the instantaneous flow rate value corresponding to the preprocessed flow time series data, and the instantaneous valve opening value corresponding to the valve opening signal; Determine the equivalent flow area of ​​the valve under the current operating conditions based on the instantaneous opening value; The instantaneous flow rate, instantaneous opening value, and equivalent flow area of ​​the valve are substituted into the fluid dynamics lumped parameter model. The static pressure difference force is calculated based on Bernoulli's equation, and the steady-state hydrodynamic force is calculated based on the momentum theorem. The static pressure difference force and the steady-state hydrodynamic force are vector-synthesized, and the resultant force is used as the fluid excitation force signal of the fluid acting on the valve plate under the current operating condition.

[0018] Specifically, firstly, the instantaneous flow rate value corresponding to the current sampling moment is extracted in real time from the preprocessed flow time-series data stream, with the unit being cubic meters per second. Simultaneously, the instantaneous valve opening value at the same moment is extracted, expressed as a percentage. Based on this instantaneous opening value, the valve's equivalent flow area under the current operating condition is obtained by consulting a pre-calibrated valve opening-equivalent flow area lookup table. In this embodiment, the valve's equivalent flow area refers to the actual physical geometric flow cross-sectional area that the fluid can pass through when flowing through the valve at the current opening. This area is the actual flow cross-section calculated based on the geometric projection relationship between the valve plate and the flow channel, mainly used for subsequent physical mechanism modeling of flow velocity and pressure difference in Bernoulli's equation. The valve opening-equivalent flow area lookup table used to obtain the valve's equivalent flow area is pre-established in this example as follows: the theoretical equivalent flow area is calculated based on a geometric analytical method, and the butterfly valve's valve plate has a circular disc structure. When the valve is fully closed, the valve plate is perpendicular to the flow channel cross-section, completely obstructing the flow channel, and the equivalent flow area is zero. As the valve rotates from the fully closed to the fully open position, the valve plate gradually tilts, and its projected area on the plane perpendicular to the flow direction gradually decreases. The change in this projected area is determined by the geometric relationship between the nominal diameter of the valve plate and the opening angle; the projected area is the area of ​​the valve plate circle multiplied by the cosine of the current opening angle. When fully closed, the opening angle is zero degrees, the cosine value is one, and the projected area equals the area of ​​the valve plate circle; when fully open, the opening angle is ninety degrees, the cosine value is zero, and the projected area is zero; the projected area for any intermediate opening angle is calculated using the same cosine relationship. The cross-sectional area of ​​the flow channel is directly converted from the valve body's flow channel diameter and is a fixed value. The valve's equivalent flow area equals the cross-sectional area of ​​the flow channel minus the projected area of ​​the valve plate at the current opening. The projected area is largest when fully closed, and the equivalent flow area is smallest; when fully open, the projected area is zero, and the equivalent flow area equals the cross-sectional area of ​​the flow channel. Utilizing the aforementioned geometric projection principles, the opening degree is calculated sequentially from fully closed to fully open in preset steps of 1° or 5° to obtain discrete equivalent flow area data covering the entire travel distance. An initial lookup table is then constructed and stored in the monitoring system database. Considering that butterfly valves in the ballast water system of ocean-going cargo ships are subjected to long-term seawater erosion, slight wear may occur on the valve plates or sealing rings. During regular ship maintenance, the static leakage of the valves in the fully closed state is measured to calculate the basic deviation value of the equivalent flow area caused by wear, and the initial lookup table is then adjusted by overall translation. During daily online monitoring, the system directly queries the corrected lookup table based on the current instantaneous opening value. If the instantaneous opening value falls between two discrete points in the table, linear interpolation is used to obtain the corresponding real-time equivalent flow area.

[0019] The pre-processed instantaneous flow rate, instantaneous opening value, and the valve equivalent flow area obtained from the table are input into the pre-constructed fluid dynamics lumped parameter model. This model is based on Bernoulli's equation and momentum theorem. The static pressure difference force and steady-state hydrodynamic force are calculated and then summed to obtain the fluid excitation force signal.

[0020] First, calculate the static pressure difference force F before and after the valve core. pressure Based on Bernoulli's principle of energy conservation in fluid mechanics, under a simplified model that neglects fluid viscous friction and local resistance losses, the theoretical pressure difference ΔP before and after the valve core is approximately equal to the kinetic energy increment of the fluid flowing through the throttling orifice. The theoretical pressure difference ΔP before and after the valve core is simplified to a kinetic energy difference term. Therefore, the formula for the theoretical pressure difference ΔP(t) at the current sampling time is as follows: ; Where ρ is the fluid density, and in this example, the seawater density is preferably taken as a standard value or a measured temperature correction value, v th (t) represents the throttling section. Real-time flow rate at the location, v in (t) represents the real-time flow velocity at the valve inlet; Calculate the real-time flow velocity v at the throttling section at the current moment. th (t), the formula is as follows: ; Where Q(t) is the instantaneous flow rate at the current moment, and A eq (t) represents the real-time equivalent flow area at the current moment, where the throttling section refers to the minimum flow channel cross-section formed when the fluid flows through the valve plate, A. eq (t) represents the real-time area value of the minimum flow channel cross-section under the current valve opening; Calculate the real-time flow velocity v at the valve inlet at the current moment. in (t), the formula is as follows: ; Where Q(t) is the instantaneous flow rate at the current moment, and A in The cross-sectional area of ​​the valve inlet pipe is a constant determined by the nominal diameter of the pipe; Multiply the calculated pressure difference ΔP(t) by the effective working area A of the valve seat sealing surface. seat This refers to the effective force-bearing projected area of ​​the valve seat sealing surface. This area is the valve seat sealing surface area already indicated in the valve's factory drawings. During system installation and commissioning, it is entered into the monitoring system as a known structural parameter to obtain the static pressure difference force F at the current moment. pressure (t), the formula is as follows: ; Since there is inevitably a pressure loss when the fluid flows through the valve, and the upstream pressure is higher than the downstream pressure, the direction of this static pressure difference force is along the valve stem axis and points to the back of the valve plate.

[0021] Next, the steady-state hydrodynamic force F generated by the change in fluid flow direction is calculated. momentum According to the momentum theorem in fluid mechanics, when a fluid flows through the throttling edge of a valve plate, its flow direction changes drastically, resulting in a significant change in its axial momentum. The valve core exerts a force on the flow field that alters the fluid's momentum. Based on the principle of action and reaction, the fluid will inevitably generate an axial reaction force of equal magnitude and opposite direction on the valve plate, i.e., the steady-state hydrodynamic force. Calculate the steady-state hydrodynamic force F at the current moment. momentum (t), the formula is as follows: ; Wherein, θ is the angle between the direction of the fluid outflow velocity at the throttling section and the valve stem axis. In a preferred embodiment of this example, this angle θ is only related to the valve plate geometry and the current opening degree of the butterfly valve. In this embodiment, during the offline calibration stage, a visualization experimental platform is built for sample valves of the same model. Under normal temperature water medium and different discrete opening degrees, the streamline morphology downstream of the throttling section is captured using tracer and high-speed photography technology. The angle between the main flow direction and the valve stem axis corresponding to each opening degree is measured and recorded, and a feature mapping table of valve opening degree-jet angle θ is constructed and stored in the database. Since the jet angle θ is only determined by the inherent geometry of the valve plate and valve body and is not affected by later usage factors such as operating time and medium corrosion, offline calibration is only required once, and no subsequent correction or update is needed. During real-time monitoring, the system queries the feature mapping table based on the current instantaneous opening value; if the current opening value is between two discrete calibration opening degrees, linear interpolation is performed on the adjacent calibration values ​​to obtain the real-time jet angle θ under the current operating condition and substitute it into the steady-state hydrodynamic formula to complete the calculation.

[0022] Finally, in the standard centerline butterfly valve structure, the static pressure differential force is generated by the pressure difference across the valve core, and its direction is along the valve stem axis towards the back of the valve plate; the steady-state hydrodynamic force is generated by the change in fluid momentum, and its axial component also acts along the valve stem axis. Since the two forces are collinear in space and have the same direction, their vector sum can be directly simplified to an algebraic sum, and the fluid excitation force signal F at the current moment is... fluid (t) is calculated using the following formula: ; By calculating each sampling moment individually, a complete fluid excitation force time-series signal can be obtained. This fluid excitation force signal is the axial resultant force acting on the valve plate, and its direction is along the valve stem axis. For a standard centerline butterfly valve structure, the pressure decreases when the fluid flows through the throttling section, and the static pressure difference force points towards the back of the valve plate along the valve stem axis. The steady-state hydrodynamic force is calculated according to the momentum theorem, and the resultant force of the two usually points towards the valve plate closing direction. In summary, the above fluid dynamics lumped parameter model is constructed based on the ideal fluid assumption and is preferably applicable to single-phase incompressible fluids without significant cavitation. The model ignores fluid viscous friction and flow contraction effects, so the calculated real-time fluid excitation force is a theoretical approximation. In the fault diagnosis logic of this scheme, the core criterion is the degree of deviation of the real-time excitation force from the healthy baseline excitation force. Due to the systematic deviations introduced by the model simplification, such as viscous friction and flow contraction, which have high stability and consistency throughout the valve's entire life cycle, or can be corrected by introducing an offline calibrated flow coefficient, this deviation will exist simultaneously in the healthy baseline value and the real-time monitoring value. When performing difference calculations or trend comparisons, such systematic deviations are effectively offset or significantly reduced, thus not affecting the extraction and judgment of fault characteristics. Furthermore, the offline calibration parameter θ of the jet angle introduced into the model accurately corrects the vector direction of the steady-state hydrodynamic force, further ensuring the model's engineering reliability in reflecting the changing trends of fluid dynamic behavior. Therefore, this simplified model fully meets the engineering needs of real-time monitoring and fault early warning of valve hydraulic status in industrial settings.

[0023] Furthermore, the preprocessed triaxial vibration time series data of the valve body is synchronously subtracted from the first flow-induced vibration response component to obtain the valve body triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component; the preprocessed triaxial vibration time series data of the valve stem is synchronously subtracted from the second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component, including: The system monitors the current valve opening signal in real time and determines whether the current valve opening signal matches the opening node in the pre-calibrated transfer function database. The transfer function database stores time-domain sequence parameters that characterize the valve body vibration transfer function and valve stem vibration transfer function under several calibrated openings. The time-domain sequence parameters include components in three orthogonal directions: X, Y, and Z. If no matching opening node is found, the time-domain sequence parameters of the vibration transfer functions of the valve body and valve stem corresponding to the adjacent calibrated opening nodes are extracted according to the position of the current valve opening signal between the adjacent calibrated opening nodes. The time-domain sequence parameters are then interpolated in the direction to obtain the three-axis time-domain sequence parameters of the real-time valve body vibration transfer function and the real-time valve stem vibration transfer function under the current opening. The fluid excitation force signal is convolved with the triaxial time-domain sequence parameters of the real-time valve body vibration transfer function and the triaxial time-domain sequence parameters of the real-time valve stem vibration transfer function in the corresponding directions to obtain the first flow-induced vibration response components of the valve body in the X, Y, and Z axes, which characterize the vibration response generated by the fluid excitation at the valve body, and the second flow-induced vibration response components of the valve stem in the X, Y, and Z axes, which characterize the vibration response generated by the fluid excitation at the valve stem. On the time axis, the preprocessed valve body triaxial vibration time series data is subtracted from the corresponding first flow-induced vibration response component point by point in the direction to obtain the valve body triaxial pure mechanical vibration residual signal. The preprocessed valve stem triaxial vibration time series data is subtracted point by point in each direction from the corresponding second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal.

[0024] Specifically, after obtaining the fluid excitation force signal, it is necessary to use this signal to separate the flow-induced vibration component from the measured vibration signal, obtaining a pure mechanical vibration residual signal that reflects only the mechanical structure state. The measured vibration signal contains two vibration components from different sources. The first is flow-induced vibration caused by throttling, eddies, and pressure pulsations when the fluid flows through the valve plate. This vibration always exists during normal valve operation, and its intensity varies with flow rate and opening degree, belonging to normal fluid dynamics response. The second is abnormal vibration caused by mechanical structural faults such as valve plate wear, valve stem loosening, and increased bearing clearance. This is the target signal that needs to be extracted for fault diagnosis. The two vibrations overlap at the sensor measurement point. If feature extraction and fault determination are performed directly on the measured vibration signal, the fluctuations of flow-induced vibration will be misjudged as mechanical faults, leading to an increased false alarm rate. This solution calculates the fluid excitation force signal using a fluid dynamics lumped parameter model, then calculates the flow-induced vibration response component using a transfer function, and subtracts this component from the measured vibration signal, thereby solving the problem of overlap between flow-induced vibration and mechanical fault vibration at the physical level. The core of the stripping process is to obtain a transfer function that matches the current butterfly valve opening, input the fluid excitation force signal into the transfer function to calculate the flow-induced vibration response component, and then subtract this component from the measured vibration signal.

[0025] The transfer function describes the dynamic relationship between the excitation force on the valve plate and the vibration response generated at the valve body or valve stem. Although the butterfly valve exhibits nonlinear characteristics throughout the entire adjustment stroke, the stripping algorithm in this embodiment employs linear time-invariant system theory. This is because the nonlinearity of the butterfly valve mainly stems from the changes in the flow area and flow field boundary conditions with the opening degree. Under a single steady-state opening degree, the contact stiffness between the valve plate and the valve seat, as well as the physical parameters of the mechanical structure such as mass, damping, and stiffness, remain constant. At this point, the valve structure dynamically degenerates into a linear time-invariant system. Based on this, this embodiment decomposes the complex nonlinear time-varying dynamic characteristics of the butterfly valve into a series of locally linear time-invariant subsystems indexed by the opening degree. Therefore, the linear time-domain sequence parameters, i.e., the unit impulse response function, can be used to describe the transmission characteristics from the force on the valve plate to the acceleration at the sensor measurement point under a specific opening degree. For the same butterfly valve, when the opening degree changes, the contact state between the valve plate and the valve seat, the constraint stiffness of the valve stem, and the boundary conditions within the flow channel all change accordingly, causing the dynamic characteristics such as the natural frequency and damping ratio of the structure to drift. Therefore, the transfer functions from the force on the valve plate to the vibration of the valve body, and from the force on the valve plate to the vibration of the valve stem, are different at different valve opening degrees. This scheme characterizes this nonlinear time-varying characteristic that changes with the opening degree by establishing a transfer function database covering the entire stroke. This embodiment adopts a segmented calibration strategy to characterize this nonlinear time-varying characteristic that changes with the opening degree. The transfer function database is pre-calibrated during the valve offline stage. The database stores the valve body vibration transfer function and valve stem vibration transfer function corresponding to several calibrated opening degrees, all stored in the form of time-domain sequence parameters of unit impulse response functions. The calibrated opening degree covers the entire stroke of the butterfly valve from fully closed to fully open. The time-domain sequence parameters of the valve body and valve stem corresponding to all calibrated opening degrees together constitute the transfer function database, which is stored in the monitoring system.

[0026] During online monitoring, the system monitors the current butterfly valve opening signal in real time. First, it determines whether the current opening signal falls within a specific opening interval node in the transfer function database. If the current opening value is exactly equal to a calibrated opening, the system directly retrieves the valve body time-domain sequence parameters and valve stem time-domain sequence parameters corresponding to that node from the database, using them as the real-time transfer function for the current opening. If the current opening value does not fall within any opening interval node, the system determines the position of the current opening value between two adjacent opening interval nodes. For example, if the current opening value is 35%, located between the calibrated nodes 30% and 40%, the system extracts the valve body time-domain sequence parameters corresponding to 30% opening and the valve stem time-domain sequence parameters corresponding to 40% opening from the database, respectively. To ensure the causality and stability of the interpolated sequence, this embodiment preferably uses a conformal piecewise cubic interpolation or cubic spline interpolation algorithm. In the time-domain sequence of two adjacent nodes, interpolation is performed point by point at the same sampling time to obtain the real-time valve body time-domain sequence parameters and the real-time valve stem time-domain sequence parameters at the current 35% opening.

[0027] The flow-induced vibration stripping method in this embodiment is preferably used for steady-state conditions of single-phase incompressible fluids without severe cavitation. Regarding single-phase incompressible fluids, this condition refers to the fluid maintaining a single phase and a relatively constant density during flow. In the ballast water system of ocean-going cargo ships, the medium is seawater, a typical single-phase liquid fluid. Under normal pressure and temperature conditions within the ballast water pipeline, seawater maintains a single-phase liquid flow without vaporization or phase change, satisfying the physical assumptions of incompressible fluids. Regarding the absence of severe cavitation, cavitation typically occurs when the fluid flows through the throttling section of a valve, where the local pressure is lower than the fluid's saturated vapor pressure at that temperature, resulting in bubble cavitation. However, in ballast water systems, seawater temperatures are typically between 0°C and 30°C, corresponding to extremely low saturated vapor pressures; simultaneously, the pipeline operating pressure is usually maintained at a few tenths to a few megapascals. Even if the pressure decreases at the valve throttling point, its absolute pressure is still much higher than the saturated vapor pressure of seawater, lacking the physical conditions for severe cavitation. Regarding steady-state conditions, these require valve opening, flow rate, and pressure to be relatively constant. During normal navigation, once the ballast tank's water intake and discharge operations are adjusted, the large-diameter electric butterfly valve will remain fixed at a specific opening to maintain a stable draft and hull balance. At this time, the system is in a long-term stable operating state, fully conforming to the theoretical premise of a linear time-invariant system upon which this scheme is based. Therefore, the operating state of the large-diameter electric butterfly valve in the ballast water system of the ocean-going cargo ship, as selected in this scheme, during normal navigation, meets the three conditions of single-phase incompressible fluid, absence of severe cavitation, and steady-state operation.

[0028] After obtaining the real-time transfer function at the current opening degree, the resulting fluid excitation force signal F is... fluid(t) is used as input. To reduce the real-time computational load, this embodiment employs a piecewise convolution algorithm based on Fast Fourier Transform (FFT) to efficiently implement FFT in the frequency domain. fluid (t) are respectively compared with the real-time valve body time-domain sequence parameter h body (t) and real-time valve stem time-domain sequence parameters h stem The time-domain convolution operation of (t) is performed. A piecewise convolution algorithm based on Fast Fourier Transform is used to perform convolution on the long sequence of fluid excitation force signals F. fluid (t) The data is divided into multiple shorter segments according to a preset frame length; after necessary zero-padding is performed on each segment and the unit impulse response sequence of the valve body or valve stem, a Fast Fourier Transform is performed on each segment, and the spectra of the two segments are multiplied in the frequency domain. Then, the time domain signal is restored by an Inverse Fast Fourier Transform; finally, the output segments are overlapped and seamlessly spliced ​​using either the overlap-preservation method or the overlap-addition method to obtain the complete flow-induced vibration response components. Due to the fluid excitation force signal F... fluid (t) represents the axial resultant force along the valve stem axis. Since both the valve body transfer function and the valve stem transfer function contain components in three orthogonal directions (X, Y, Z), the convolution operation is performed independently for each direction, i.e., the triaxial flow-induced vibration response components of the valve body and the valve stem are calculated separately for each direction. Based on the above convolution operation relationship, the first flow-induced vibration response component V at the current valve body position is... fluid_body (t) is calculated using the following formula: ; Among them, symbols This represents the temporal convolution operation. F fluid (t) represents the fluid excitation force signal, in N and h. body (t) represents the real-time valve body time-domain sequence parameters at the current opening degree, in m / (s²·N), characterizing the acceleration response caused by a unit excitation force. V fluid_body (t) represents the flow-induced vibration response component at the valve body position at the current moment, calculated in m / s².

[0029] The second flow-induced vibration response component V at the valve stem position at the current moment fluid_stem (t) is calculated using the following formula: ; Among them, h stem (t) represents the real-time valve stem time-domain sequence parameters at the current valve opening, in m / (s²·N). V fluid_stem (t) represents the calculated flow-induced vibration response component at the valve stem position at the current moment, in m / s². The preprocessed valve body X-axis vibration time-series data V... body,x (t) and the first flow-induced vibration response component V fluid_body,x(t) Subtracting the samples at each sampling point along the time axis yields the valve body X-axis pure mechanical vibration residual signal V. res_body,x (t). Similarly, the Y-axis and Z-axis pure mechanical vibration residual signals V of the valve body are obtained respectively. res_body,y (t) and V res_body,z (t). The preprocessed valve stem X-axis vibration time series data V stem,x (t) and the second flow-induced vibration response component V fluid_stem,x Subtracting the values ​​point by point (t) yields the valve stem X-axis pure mechanical vibration residual signal V. res_stem,x (t). Similarly, the Y-axis and Z-axis pure mechanical vibration residual signals V of the valve stem are obtained respectively. res_stem,y (t) and V res_stem,z (t). Since all channels are synchronously acquired under the same sampling clock, and the low-pass filter has adopted a forward-backward filtering strategy to eliminate phase distortion, the vibration time series data and the flow-induced vibration response components in each corresponding direction are strictly aligned on the time axis, and the phase error of point-by-point subtraction can be ignored. In the above pure mechanical vibration residual signal, the vibration component caused by normal fluid excitation has been stripped away, and the remaining components mainly come from the dynamic response of the butterfly valve's mechanical structure itself. At the same time, there may also be a small amount of random flow field disturbances or model mismatch errors that were not considered in the model. The above residual signal will be used as the input for step three to extract fault characteristic indicators and generate mechanical condition indicators.

[0030] Furthermore, the transfer function database stores several time-domain sequence parameters characterizing the valve body vibration transfer function and the valve stem vibration transfer function at various calibrated opening degrees, including: Establish a finite element model of the valve, which includes the geometric and material property parameters of the valve body, valve core, valve stem, and stuffing box. In the finite element model, the valve opening is set to multiple preset calibration openings. Under each calibration opening, a dynamic excitation force covering the actual working frequency band of the valve is applied to the valve core for harmonic response analysis. The frequency domain vibration response in the three orthogonal directions of X, Y, and Z at the installation positions of the first and second triaxial vibration sensors is obtained. The frequency domain vibration response at the installation position of the first triaxial vibration sensor, obtained from simulation at each calibration opening degree, is divided by the applied dynamic excitation force to obtain the frequency domain transfer function of the valve body vibration at that calibration opening degree; the frequency domain vibration response at the installation position of the second triaxial vibration sensor, obtained from simulation at each calibration opening degree, is divided by the applied dynamic excitation force to obtain the frequency domain transfer function of the valve stem vibration at that calibration opening degree. Inverse Fourier transforms were performed on the frequency domain transfer functions of the valve body vibration and the valve stem vibration to obtain the triaxial time-domain sequence parameters of the valve body vibration transfer function and the valve stem vibration transfer function under each calibrated opening degree, and then stored in the transfer function database.

[0031] Specifically, the transfer function is a mathematical expression describing the dynamic relationship between input and output in a linear time-invariant system. In this scheme, the transfer function specifically describes the causal relationship between the fluid excitation force acting on the valve plate and the vibration acceleration at the measuring point of the valve body or valve stem under a certain fixed butterfly valve opening. This relationship is expressed as a unit impulse response function h(t) in the time domain, with the input force signal and the output acceleration signal correlated through convolution. In the frequency domain, the transfer function H(f) is defined as the complex ratio of the spectrum of the output signal to the spectrum of the input signal, i.e., H(f) = Y(f) / X(f), where X(f) is the spectrum of the input force and Y(f) is the spectrum of the output acceleration. In this scheme, the transfer function plays a crucial role in converting the fluid excitation force signal into a flow-induced vibration response component. Although the structural dynamic characteristics of the butterfly valve exhibit nonlinear time-varying characteristics with the opening degree, within the vibration amplitude range of normal valve operation, the structural dynamic characteristics of the valve at each fixed opening degree can be regarded as a linear time-invariant system. Based on this, this scheme proposes a segmented calibration strategy, decomposing the full-stroke nonlinear system into a series of locally linear subsystems parameterized by the opening degree, and establishing a transfer function database covering all closed to all open states. This strategy transforms the complex nonlinear time-varying problem into a linear time-invariant problem at discrete opening degree nodes, and achieves continuous acquisition of the transfer function at arbitrary opening degrees through online interpolation. In this embodiment, the transfer function obtained by using a 1-Newton unit amplitude excitation force as the unit impulse response is used in the calibration stage. During online monitoring, it is compared with the calculated actual fluid excitation force signal F. fluid (t) performs convolution operation, and utilizes the homogeneity of the linear system to automatically adapt to real-time excitation forces of different amplitudes, thereby accurately restoring the flow-induced vibration response components under the current working conditions.

[0032] The specific calibration process for the transfer function database is as follows: First, a finite element model of the butterfly valve is established. Based on the geometric parameters provided in the butterfly valve design drawings, a three-dimensional solid model including the valve body, valve plate, valve stem, and stuffing box is constructed in the finite element software. The valve body is modeled according to the actual flow channel shape, the valve plate is modeled as a circular disc structure, the valve stem is modeled as a solid shaft-like structure, and the stuffing box is modeled as a ring-shaped constrained region. Each component is assigned corresponding material property parameters. The valve body and valve plate are made of duplex stainless steel with excellent seawater corrosion resistance, a density of approximately 7800 kg / m³, an elastic modulus of approximately 195 GPa, and a Poisson's ratio of approximately 0.3. The valve stem is made of precipitation-hardening stainless steel with an elastic modulus of approximately 195 GPa, a density of approximately 7750 kg / m³, and a Poisson's ratio of approximately 0.3. The constraint stiffness of the stuffing box region is simplified by treating it as a radially distributed linear spring element based on the packing type and actual preload, to simulate the radial clamping effect of the packing on the valve stem, while not restricting the rotational freedom of the valve stem around its own axis. The equivalent spring stiffness value is determined by the gland preload, packing transfer coefficient, packing-stem contact area, and packing friction coefficient, and can be calculated through gland preload calibration experiments or packing mechanical parameter manuals. Bearing constraints are applied to both ends of the valve stem to limit its radial displacement while retaining its rotational freedom around its own axis. Axial displacement constraints are applied to the top of the valve stem to simulate the axial limiting of the valve stem by an actuator or thrust bearing. Damping parameters are provided for subsequent harmonic response analysis; the modal damping ratio is preset to 2% to 5% based on the butterfly valve structure type.

[0033] Then, multi-opening harmonic response analysis is performed in the finite element model. The valve opening is set as the calibration opening in 5% increments from fully closed to fully open, up to 100% as the calibration node. At each calibration opening, the contact state and constraint relationship between the valve plate and the valve seat are updated by rotating the valve plate to the corresponding angular position. A dynamic excitation force along the valve stem axis is applied at the center position of the valve plate. The amplitude of the excitation force is 1N, and the frequency range covers the actual operating frequency band of the butterfly valve, from 0.5Hz to 2000Hz, covering the main excitation components such as low-frequency wave pulsation, pump pressure pulsation, and throttling noise. This axial excitation force is the equivalent axial resultant force of the fluid on the distributed surface pressure of the valve plate. This embodiment adopts this equivalent treatment based on the engineering fact that the fluid excitation under the steady-state operating condition of the butterfly valve is mainly axial. Since the finite element model is a three-dimensional solid model, the vibration response generated by this axial excitation force on the valve body and valve stem includes components in three orthogonal directions, corresponding to the three measurement axes of the first and second triaxial vibration sensors, respectively. Harmonic response analysis was performed using the modal superposition method, with the modal damping ratio ranging from 2% to 5% depending on the butterfly valve structure type. Considering the valve flow channel is filled with seawater, additional mass elements were added to the valve plate and the inner wall of the valve body flow channel to simulate the inertial effect of the surrounding seawater moving during structural vibration. The density of the added mass was taken as the seawater density of 1025 kg / m³. The vibration acceleration responses at the installation locations of the first and second triaxial vibration sensors were extracted in three orthogonal directions, and each direction was denoted as the valve body frequency domain vibration response Y. body,x (f), Y body,y (f), Y body,z (f) and the frequency domain vibration response of the valve stem Y stem,x (f), Y stem,y (f), Y stem,z (f).

[0034] Next, the frequency domain transfer function is calculated according to the definition of the transfer function for each calibrated opening degree. The frequency domain transfer function of the valve body in each direction is the ratio of the vibration response in that direction to the input excitation force, i.e. Where i takes three directions: x, y, and z. Similarly, the frequency domain transfer function of the valve stem is given by... Since the applied excitation force has an amplitude of 1 Newton, the frequency domain transfer function at each frequency point is numerically equal to the vibration response value at the corresponding frequency point. The frequency domain transfer functions of the valve body and valve stem are both complex values ​​at a set of discrete frequency points. Each frequency point contains amplitude and phase information, and its physical meaning is the frequency response of the anisotropic acceleration caused by a unit axial excitation force. The amplitude unit is meters per second squared (m / (s²·N)). When converting the above frequency domain transfer functions into time domain impulse responses, negative frequency domain data is supplemented based on the conjugate symmetry property of the real signal spectrum to construct a complete frequency domain sequence for inverse Fourier transform.

[0035] Finally, the frequency domain transfer function is converted into the time domain sequence parameters required for online calculation. Inverse Fourier transforms are then performed on the frequency domain transfer functions of the valve body in each direction to obtain the unit impulse response sequences h of the valve body in each direction. body,i (t), which represents the time-domain sequence parameters of the valve body vibration transfer function. Performing inverse Fourier transforms on the frequency domain transfer functions of the valve stem in each direction yields the unit impulse response sequences h of the valve stem in each direction. stem,i (t) represents the time-domain sequence parameters of the valve stem vibration transfer function. These parameters, composed of real values ​​at discrete-time sampling points, describe the complete decay of the valve body or valve stem's acceleration response over time under a unit axial pulse force on the valve plate. The time-domain sequence parameters of the valve body and valve stem at each calibrated opening degree are stored in the transfer function database. Each calibrated opening degree in the database corresponds to a set of three-dimensional time-domain sequence parameters for the valve body and a set of three-dimensional time-domain sequence parameters for the valve stem. The transfer function database calibration is complete. This database covers the entire stroke of the butterfly valve from fully closed to fully open, providing a complete offline data foundation for transfer function matching and interpolation during the online monitoring phase.

[0036] S3: For each axis component of the valve body pure mechanical vibration residual signal and the valve stem pure mechanical vibration residual signal, extract the time-domain kurtosis value and the frequency-domain amplitude corresponding to the preset fault characteristic frequency as feature values. Compare each feature value with the corresponding preset normal range. When any feature value exceeds its corresponding preset normal range, it is determined that there is an abnormality in the corresponding mechanical structure, and a mechanical state flag representing the abnormal state is generated. Otherwise, a mechanical state flag representing the normal state is generated. At the same time, calculate the root mean square value of the fluid excitation force signal and compare the root mean square value with the preset hydraulic steady-state threshold range. When the root mean square value exceeds the hydraulic steady-state threshold range, it is determined that there is instability in the flow field, and a hydraulic state flag representing the abnormal state is generated. Otherwise, a hydraulic state flag representing the normal state is generated.

[0037] Furthermore, when any feature value exceeds its corresponding preset normal range, an anomaly is determined in the corresponding mechanical structure, and a mechanical state flag representing the abnormal state is generated; otherwise, a mechanical state flag representing the normal state is generated. Simultaneously, the root mean square (RMS) value of the fluid excitation force signal is calculated and compared with a preset hydraulic steady-state threshold range. When the RMS value exceeds the hydraulic steady-state threshold range, instability in the flow field is determined, and a hydraulic state flag representing the abnormal state is generated; otherwise, a hydraulic state flag representing the normal state is generated, including: For each axis component of the valve body's pure mechanical vibration residual signal, its time-domain kurtosis value is extracted, and the frequency-domain amplitude corresponding to a preset first fault characteristic frequency is also extracted. The time-domain kurtosis value of each axis component is compared with the preset normal range of valve body kurtosis, and the frequency-domain amplitude corresponding to the first fault characteristic frequency of each axis component is compared with the preset normal range of valve body amplitude. When the time-domain kurtosis value of any component is not within the normal range of valve body kurtosis, or when the frequency domain amplitude corresponding to the first fault characteristic frequency of any component is not within the normal range of valve body amplitude, it is determined that there is an abnormality in the valve body mechanical structure, and a valve body mechanical state flag representing the abnormal state is generated; otherwise, a valve body mechanical state flag representing the normal state is generated. For each axis component of the valve stem's pure mechanical vibration residual signal, its time-domain kurtosis value is extracted, and the frequency-domain amplitude corresponding to a preset second fault characteristic frequency is also extracted. The time-domain kurtosis value of each axis component is compared with the preset normal range of valve stem kurtosis, and the frequency-domain amplitude corresponding to the second fault characteristic frequency of each axis component is compared with the preset normal range of valve stem amplitude. When the time-domain kurtosis value of any component is not within the normal range of valve stem kurtosis, or the frequency-domain amplitude corresponding to the second fault characteristic frequency of any component is not within the normal range of valve stem amplitude, it is determined that there is an abnormality in the valve stem mechanical structure, and a valve stem mechanical state flag representing the abnormal state is generated; otherwise, a valve stem mechanical state flag representing the normal state is generated. The root mean square (RMS) value of the fluid excitation force signal is calculated and compared with a preset hydraulic steady-state threshold range. If the RMS value is not within the hydraulic steady-state threshold range, the flow field is determined to be unstable, and a hydraulic state flag representing the abnormal state is generated; otherwise, a hydraulic state flag representing the normal state is generated.

[0038] Specifically, feature indicators reflecting the health status of the mechanical structure are extracted from the residual signals of the valve body's triaxial pure mechanical vibration and the valve stem's triaxial pure mechanical vibration. Corresponding mechanical state indicators are generated by comparing these indicators with preset normal ranges. Simultaneously, the root mean square value of the fluid excitation force signal is calculated to generate a hydraulic state indicator. These three indicators serve as input variables for a cross-validation decision structure. In this embodiment, kurtosis is selected as the time-domain feature indicator, and the frequency domain amplitude corresponding to the fault characteristic frequency is selected as the frequency-domain feature indicator, based on the vibration signal characteristics of common valve faults. Valve mechanical faults such as wear, loosening, and cracks typically exhibit two types of vibration characteristic changes: first, an increase in the impact component in the vibration signal; kurtosis is sensitive to impact signals and can effectively capture the weak impact characteristics in the early stages of the fault; second, the vibration energy caused by the fault is concentrated near the structure's natural frequency, where the frequency domain amplitude increases significantly. The two indicators characterize the state of the vibration signal in the time and frequency domains, respectively, complementing each other and constituting a multi-dimensional description of the fault characteristics. The root mean square value of the fluid excitation force signal reflects the effective strength of the force exerted by the fluid on the valve plate. Under healthy valve conditions, the fluid excitation force remains within a certain range during steady-state operation. When the upstream pump speed changes, pipeline valves are switched, or abnormal events such as water hammer occur, the effective intensity of the fluid excitation force will change significantly. By monitoring whether this root mean square value deviates from the healthy baseline, it is possible to independently determine whether the flow field is in a steady state. This indicator depends only on the flow signal and the lumped parameter model of fluid dynamics, and is physically independent of the vibration signal. Therefore, it can serve as an independent criterion for fluid-side judgment in subsequent fluid-structure interaction verification.

[0039] First, feature extraction is performed on each axis component of the valve body's pure mechanical vibration residual signal. The valve body's pure mechanical vibration residual signal includes the X-axis component V. res_body,x (t), Y-axis component V res_body,y (t) and Z-axis component V res_body,z(t), each axis component is independent. The purpose of retaining the three-axis directional information is to make full use of the vibration characteristics in each direction. During the operation of the butterfly valve, the vibration response in different directions has different sensitivities to different faults. For example, the impact vibration caused by the loosening of the connection between the valve body and the valve seat is mainly manifested in the radial direction of the valve body, while the vibration caused by the wear of the valve body bearing seat may be reflected in the axial direction. If the three-axis signals are combined into a scalar in advance, the directional information will be lost, reducing the detection sensitivity of faults in specific directions. Separate axis processing allows any abnormality in any direction to be captured independently, avoiding the problem of the abnormal features being diluted by the normal signals in other directions after synthesis. The valve body pure mechanical vibration residual signal mainly reflects the mechanical state of the valve seat, valve body bearing seat and valve body flow channel wall, while the valve stem pure mechanical vibration residual signal mainly reflects the mechanical state of the valve stem itself and the connection between the valve stem and the valve plate. The two are physically independent and have different emphases in the vibration transmission path. For each axis component, its time-domain kurtosis value is extracted. The kurtosis value is the fourth-order normalized moment of the signal probability distribution, reflecting the significance of the impact component in the vibration signal. When the valve body experiences wear, loosening, or cracking, the impact component in the vibration signal increases, and the kurtosis value increases accordingly. Kurtosis value K body,i The calculation formula is the average of the fourth power of the difference between the signal sampling point and its mean, divided by the fourth power of the standard deviation, where i represents the x, y, and z axes. Simultaneously, a Fast Fourier Transform (FFT) is performed on each axis component to extract the frequency domain amplitude corresponding to the preset first fault characteristic frequency from the spectrum. The FFT converts the time-domain vibration signal into a frequency-domain representation, with each frequency point in the spectrum corresponding to an amplitude. Extracting the amplitude at a specific frequency from the spectrum involves locating the frequency index position of the first fault characteristic frequency in the spectrum and reading the amplitude corresponding to that position as A. body,i The first fault characteristic frequency is the first-order natural frequency of the valve body or the characteristic frequency corresponding to the valve seat loosening fault, which is pre-extracted through finite element modal analysis of the valve. This frequency is chosen as the monitoring target because when a mechanical fault occurs in the valve body, the additional vibration energy caused by the fault is mainly concentrated near its natural frequency, and the change in vibration amplitude at this frequency can most sensitively reflect the change in the mechanical structure state. The specific determination method is detailed in subsequent embodiments. The kurtosis value K of each axis component is... body,i The value is compared with the preset normal range of valve body kurtosis, and the frequency domain amplitude A corresponding to the first fault characteristic frequency of each axis component is calculated. body,i The value is compared with the preset normal range of valve body amplitude. The normal range of valve body kurtosis and normal range of valve body amplitude are distribution intervals obtained by statistically analyzing multiple sets of operating data within a specific steady-state operating range under valve health conditions. When the kurtosis value K of any component... body,i Less than the lower limit of the normal range of valve body kurtosis or greater than the upper limit of the normal range of valve body kurtosis, or the frequency domain amplitude A of any component. body,iWhen the amplitude value is less than the lower limit of the normal range or greater than the upper limit of the normal range, it indicates that the vibration characteristics in at least one direction have deviated abnormally, indicating an anomaly in the valve body's mechanical structure and generating a valve body mechanical status flag representing the abnormal state. When the index value is exactly equal to the boundary value of the normal range, it is considered to be within the normal range and no anomaly judgment is triggered. Only when the kurtosis value and frequency domain amplitude of all axis components are within their respective normal ranges are a valve body mechanical status flag representing the normal state generated. This axis-by-axis, index-by-index full-coverage judgment method ensures that no valve body faults are missed.

[0040] Next, the same processing procedure is applied to each axis component of the valve stem's pure mechanical vibration residual signal. The valve stem's pure mechanical vibration residual signal includes the X-axis component V. res_stem,x (t), Y-axis component V res_stem,y (t) and Z-axis component V res_stem,z (t). The principle of split-axis processing is the same as that of valve body; vibration responses in different directions correspond to different valve stem failure modes. For example, vibrations caused by valve stem cracks are more sensitive in the axial direction, while vibrations caused by packing gland loosening are more prominent in the radial direction. The time-domain kurtosis value K is extracted for each axial component. stem,i The frequency domain amplitude A corresponding to the preset second fault characteristic frequency stem,i The second fault characteristic frequency is either the first-order natural frequency of the valve stem's longitudinal vibration or the characteristic frequency corresponding to the valve stem crack fault, which is also pre-extracted through finite element modal analysis. The kurtosis values ​​K of each axis component are then... stem,i The frequency domain amplitude A of each axis component is compared with the preset normal range of valve stem kurtosis. stem,i Compare with the preset normal range of valve stem amplitude. When the kurtosis value K of any component... stem,i Less than the lower limit of the normal range of valve stem kurtosis or greater than the upper limit of the normal range of valve stem kurtosis, or the frequency domain amplitude A of any component. stem,i If the value is less than the lower limit of the normal range for valve stem amplitude or greater than the upper limit of the normal range for valve stem amplitude, an abnormality in the valve stem mechanical structure is determined, and a valve stem mechanical status flag representing the abnormal state is generated; otherwise, a valve stem mechanical status flag representing the normal state is generated. The normal ranges for valve body kurtosis and amplitude are independent of the corresponding ranges for valve stem. This is because the valve body and valve stem differ in material, geometry, constraints, and stress states, resulting in numerically different normal vibration levels. Setting separate normal ranges avoids misjudgments caused by using a uniform threshold.

[0041] Finally, the root mean square value of the fluid excitation force signal is calculated to generate a hydraulic state indicator. The obtained fluid excitation force signal F... fluid (t), calculate its root mean square value F within a preset time window. rmsThe root mean square (RMS) value is the square root of the average of the squares of the amplitudes of the signal at each sampling point within the time window, reflecting the effective strength of the fluid excitation force. Let F... rms The value is compared with the preset hydraulic steady-state threshold range. The hydraulic steady-state threshold range is a distribution range obtained by statistically analyzing the root mean square values ​​of fluid excitation force under multiple steady-state conditions in a healthy valve state. When F rms When the value is less than the lower limit of the hydraulic steady-state threshold range or greater than the upper limit of the hydraulic steady-state threshold range, it indicates abnormal fluctuations in the flow field. This may be caused by changes in upstream pump operating conditions, valve switching in pipelines, or flow pulsations caused by waves. The flow field is determined to be unstable, and a hydraulic state flag representing the abnormal state is generated. Otherwise, a hydraulic state flag representing the normal state is generated. At this point, three Boolean state flags are output: valve body mechanical state flag, valve stem mechanical state flag, and hydraulic state flag. Each flag's value represents either an abnormal or normal state. These three flags independently reflect the valve's operating state from both the mechanical and fluid perspectives. They will serve as input variables for the two-level hierarchical cross-validation decision structure in step four. Through cross-validation of fluid-structure characteristics, the valve's fault type and location are ultimately determined.

[0042] Furthermore, the preset normal ranges for valve body kurtosis, valve body amplitude, valve stem kurtosis, valve stem amplitude, and hydraulic steady-state threshold intervals are all threshold intervals determined based on a pre-constructed threshold model. This threshold model is pre-constructed using the following method: Acquire historical operating data of the valve under known health conditions. The historical operating data includes operating parameters and corresponding characteristic index values. The operating parameters include at least the valve opening degree and flow signal. The characteristic index values ​​include the kurtosis value of each axis of the valve body, the amplitude value of each axis of the valve body, the kurtosis value of each axis of the valve stem, the amplitude value of each axis of the valve stem, and the root mean square value of the fluid excitation force signal. The historical operating data is fitted and analyzed to construct a benchmark mapping relationship between the values ​​of each characteristic index and the changes in operating parameters, and a benchmark surface equation for each characteristic index is established. For each characteristic indicator, the fluctuation and dispersion of historical operating data under healthy conditions are analyzed, and the maximum dispersion extreme value within the operating condition range covered by the historical operating data is extracted. The maximum dispersion extreme value is used to construct the full operating condition dispersion envelope, and the full operating condition dispersion envelope is used as the global safety boundary of the characteristic indicator. During online monitoring, the currently collected operating parameters are substituted into the reference surface equation to calculate the theoretical reference value of each characteristic index under the current operating conditions. Combined with the corresponding global safety boundary, the normal range of the dynamic threshold of each characteristic index under the current operating conditions is determined.

[0043] Specifically, the normal ranges and threshold intervals of the various feature indicators used for comparison in this example are not fixed static values, but are dynamically generated based on the real-time operating parameters of the valve through a pre-built threshold model. This threshold model extends the health baseline data of discrete operating points into a baseline surface in a continuous operating space through regression fitting, and then combines it with a data-driven threshold construction strategy based on the full operating condition dispersion envelope to generate dynamic threshold intervals that adaptively adjust with the operating conditions. This method is automatically generated by the system based on the actual operating data of the valve, reflecting the online adaptive technical characteristics of intelligent manufacturing. The threshold model is pre-built during the valve's offline stage and directly called during online monitoring. The following details the construction method and online application process of the threshold model.

[0044] The threshold model was constructed under the condition that the valve was in a known healthy state. A healthy state refers to the initial operating phase after the valve has passed factory acceptance or major maintenance, at which point the valve's mechanical structure is in a known good condition and has not yet experienced long-term wear and fatigue. Training data comes only from the health period confirmed by maintenance or acceptance records. During data acquisition, the system monitors the rate of change of valve opening α and flow rate Q in real time. Only when the absolute value of the rate of change of opening α / Δt is less than the preset opening change threshold, and the flow fluctuation amplitude is lower than the preset flow fluctuation threshold within a preset time window, is the valve considered to be in a steady-state condition, and only then is the collected data included in the training dataset. Taking the DN400 to DN800 large-diameter electric butterfly valve of the ship's ballast water system in this embodiment as an example, the valve actuator's full stroke action time is typically between 30 and 60 seconds, and the opening change rate does not exceed 2% per second. A preset opening change threshold of 1% per second can effectively distinguish between steady-state conditions and valve regulation processes. The flow pulsation in the ballast water system pipeline is mainly caused by draft changes due to waves. The pulsation period ranges from several minutes to tens of minutes. A preset time window of 300 seconds can cover multiple typical pulsation periods, ensuring the statistical reliability of the judgment. Under steady-state conditions, the pipeline flow fluctuation amplitude is usually within 3% of the current flow average. A preset flow fluctuation threshold of 3% can accommodate normal fluctuations and filter out abnormal transient events. Simultaneously, an improved Laida criterion is used to remove outliers from the collected raw characteristic index values. Data points deviating from the local mean of the operating point by more than 2.5 times the standard deviation are considered abnormal interference data and removed. Only data within the normal statistical distribution range are retained as healthy baseline values ​​to ensure the purity and representativeness of the training dataset. Under healthy steady-state operating conditions that meet the above screening conditions, multiple operating data covering different valve openings and different flow combinations are obtained. The valve opening range and flow range covered above together constitute the full operating condition, that is, the set of all steady-state operating conditions that the valve may experience under healthy conditions. The selection of operating points should cover the common opening and flow ranges encountered in actual valve operation. For example, an operating point should be selected every 10% of the opening, from 10% to 100%. Within each opening, three flow levels—low, medium, and high—should be selected, totaling approximately thirty operating points. At each operating point, the valve should operate stably for five to ten minutes, and the number of valid frames collected at each point should be no less than fifty. Following the processing flow of steps S1 and S2, the kurtosis values ​​K of each axis of the valve body should be extracted from the preprocessed three-axis pure mechanical vibration residual signal of the valve body. body,i The frequency domain amplitude A corresponding to the preset fault characteristic frequency body,i Where i represents the x, y, and z directions. The kurtosis values ​​K of each axis of the valve stem are extracted from the residual signal of the triaxial pure mechanical vibration of the valve stem. stem,i And frequency domain amplitude A stem,i Simultaneously, the fluid excitation force signal F fluid(t) Calculate the root mean square value F under this working condition.rms The average value of each characteristic index from multiple consecutive frames of data collected at a given operating point is used as the health baseline value for that characteristic index at that operating point. Simultaneously, the local standard deviation σ is calculated for each characteristic index across multiple frames of data at each operating point. local This local standard deviation reflects the inherent dispersion of the valve's healthy vibration under specific steady-state conditions. Data from all collected operating points are aggregated to form a training dataset. Each data set in the training dataset contains input and output variables. The input variables are two operating parameters: valve opening α and flow rate Q. There are thirteen output variables, including: the valve body X-axis kurtosis reference value K. body,x Valve body Y-axis kurtosis reference value K body,y Valve body Z-axis kurtosis reference value K body,z Valve body X-axis amplitude reference value A body,x Valve body Y-axis amplitude reference value A body,y Valve body Z-axis amplitude reference value A body,z Valve stem X-axis kurtosis reference value K stem,x Valve stem Y-axis kurtosis reference value K stem,y Valve stem Z-axis kurtosis reference value K stem,z Valve stem X-axis amplitude reference value A stem,x Valve stem Y-axis amplitude reference value A stem,y Valve stem Z-axis amplitude reference value A stem,z and hydraulic reference value F rms .

[0045] Next, using the valve opening α and flow rate Q as independent variables and the baseline values ​​of each characteristic index as dependent variables, a baseline mapping relationship between the values ​​of each characteristic index and the operating conditions is constructed through regression fitting. The regression fitting employs a multinomial regression method, with each of the thirteen output variables independently establishing a general-form baseline surface equation: ; Among them, Y j This represents the baseline value of the j-th feature index, where j ranges from 1 to 13, and j=1 corresponds to the baseline value K of the valve body X-axis kurtosis. body,x j=2 corresponds to the valve body Y-axis kurtosis reference value K body,y j=3 corresponds to the valve body Z-axis kurtosis reference value K body,z j=4 corresponds to the valve body X-axis amplitude reference value A body,x j=5 corresponds to the reference value A of the valve body's Y-axis amplitude. body,y j=6 corresponds to the reference value A of the valve body's Z-axis amplitude. body,z j=7 corresponds to the valve stem X-axis kurtosis reference value K stem,x j=8 corresponds to the valve stem Y-axis kurtosis reference value K stem,y j=9 corresponds to the valve stem Z-axis kurtosis reference value K stem,zj=10 corresponds to the valve stem X-axis amplitude reference value A stem,x j=11 corresponds to the reference value A of the valve stem Y-axis amplitude. stem,y j=12 corresponds to the reference value A of the valve stem Z-axis amplitude. stem,z j=13 corresponds to the hydraulic reference value F rms c j,0 To c j,5 The undetermined coefficients of the reference surface equation for the j-th feature index are obtained by solving the operating point data of that feature index in the training dataset using the least squares method. Each of the thirteen feature indices independently solves for a set of undetermined coefficients. For intermediate operating points not directly covered in the training data, each reference surface equation can output the reference value under any combination of opening and flow rate in the continuous operating space.

[0046] During online monitoring, the system inputs the current real-time valve opening α(t) and current flow rate Q(t) into thirteen independent reference surface equations to calculate the thirteen characteristic index reference values ​​Y under the current operating condition. j,ref Y j,ref =Y j In the formula (α(t),Q(t)), Y j,ref This represents the real-time baseline value of the j-th feature index under the current operating condition. Each calculated baseline value represents the normal level that the corresponding feature index should have when the valve is in a fully healthy state under the current operating condition. Meanwhile, to determine the dynamic threshold range of each feature index, this embodiment proposes a data-driven threshold construction strategy based on the full-condition discrete envelope. The full-condition discrete envelope refers to finding the maximum fluctuation amplitude that the valve may experience at all known healthy operating points, using this maximum fluctuation amplitude as the safety redundancy boundary of the feature index, encompassing the normal fluctuation range of all healthy operating points. The core of this strategy is: during the offline modeling stage, for each feature index, iterate through its local standard deviations at all healthy operating points, extract the maximum dispersion extremum, and use the historical worst fluctuation boundary corresponding to this extremum to constrain the currently monitored dynamic threshold. For each feature index, after obtaining a training dataset covering the entire operating condition range, the calculated local standard deviation σ of the feature index at each independent operating point is used... local Subsequently, the local standard deviations of this characteristic index across all operating conditions are iterated, and the maximum dispersion extremum σ is extracted. max σm ax =max(σ local,1 ,σ local,2 ,...,σ local,m In the formula, m is the total number of healthy operating points, and σ local,1 To σ local,m This represents the local standard deviation of this characteristic index at each operating point. σ maxThis characterizes the maximum fluctuation boundary that this feature index may exhibit under all healthy operating conditions of the valve, i.e., the full-condition dispersion envelope. By extracting this maximum dispersion extremum, this strategy essentially constructs a global robustness boundary for each feature index based on historical health data. This boundary covers the worst-case fluctuation situation of the valve under all known healthy operating conditions, ensuring that in subsequent online monitoring, as long as the actual fluctuation amplitude of the valve does not exceed the maximum health fluctuation amplitude that has occurred historically, no false alarms will be triggered. Based on this maximum dispersion extremum σ... max Using a preset confidence factor k, where k takes the value of 3, corresponding to a 3-standard-deviation confidence interval in statistics, the half-width W = k·σ of the dynamic threshold interval is determined. max Each of the thirteen feature indices independently undergoes the maximum dispersion extremum extraction operation described above, yielding thirteen independent half-width values ​​W. j , where j ranges from 1 to 13, corresponding one-to-one with the subscript j of the reference surface equation. The upper limit of the dynamic threshold Y for each characteristic index. j,upper and lower limit value Y j,lower Calculate Y using the following formulas respectively: j,upper =Y j,ref +W j Y j,lower =Y j,ref -W j In the formula, Y j,ref W represents the real-time baseline value of the j-th characteristic index under the current operating conditions. j The maximum dispersion extreme value σ of the j-th feature index max,j The threshold half-width is obtained by multiplying by the confidence factor k. Taking the valve body X-axis kurtosis corresponding to j=1 as an example, Y... 1,ref That is, its benchmark value K body,x,ref W1 is its half-width W Kbx The upper limit is K body,x,ref +W Kbx The lower limit is K body,x,ref -W Kbx The dynamic threshold intervals for the remaining twelve feature indicators are generated in the same way. All dynamic threshold intervals for the current moment have been generated. These dynamic threshold intervals will serve as the basis for comparison with real-time feature indicator values.

[0047] The innovation of this solution lies in the fact that traditional dynamic thresholds often only focus on local fluctuations under the current operating conditions, using methods such as moving averages or local regression to set thresholds. This is prone to frequent threshold jumps due to sudden changes in operating conditions, leading to system instability. In contrast, the data-driven threshold construction strategy proposed in this solution, based on the entire operating condition dispersion envelope, extracts the maximum dispersion extremum to construct a global robust boundary for each feature index based on historical health data. This mechanism, which constrains the current monitoring boundary with the worst historical fluctuations, greatly reduces reliance on human experience and ensures that the monitoring thresholds have extremely high anti-interference capabilities and stability when the valve is running continuously across operating conditions, effectively avoiding false alarms caused by operating condition switching. Furthermore, as valve operating data accumulates, the system can periodically update the equations of each benchmark surface and the maximum dispersion extremum using newly added health condition data, allowing the threshold model to gradually approximate the health characteristics of the valve under real service conditions, demonstrating the self-learning and self-optimization technical characteristics of intelligent manufacturing.

[0048] Furthermore, the first fault characteristic frequency is the first natural frequency of the valve body or the characteristic frequency corresponding to the valve seat loosening fault, and the second fault characteristic frequency is the first natural frequency of the longitudinal vibration of the valve stem or the characteristic frequency corresponding to the valve stem crack fault. The first and second fault characteristic frequencies are pre-extracted by performing finite element modal analysis on the valve.

[0049] Specifically, the first and second fault characteristic frequencies are pre-extracted through finite element modal analysis during the valve's offline phase. First, a finite element model of the butterfly valve is established. This finite element model can be constructed using the same modeling method as the finite element model used for transfer function calibration, including the geometric and material property parameters of the valve body, valve plate, valve stem, and stuffing box. The valve body is modeled according to the actual flow channel shape, the valve plate is modeled as a circular disc structure, the valve stem is modeled as a solid shaft-like structure, and the stuffing box is modeled as a ring-shaped constrained region. Each component is assigned corresponding material property parameters. The valve body and valve plate are made of duplex stainless steel with a density of approximately 7800 kg / m³, an elastic modulus of approximately 195 GPa, and a Poisson's ratio of approximately 0.3. The valve stem is made of precipitation-hardening stainless steel with an elastic modulus of approximately 195 GPa, a density of approximately 7750 kg / m³, and a Poisson's ratio of approximately 0.3. The above material parameters are only examples of preferred embodiments; in actual applications, they should be set according to the physical properties corresponding to the specific material grade used for the valve. The constraint stiffness of the stuffing box region is simplified by treating it as a radially distributed linear spring element based on the packing type and actual preload, to simulate the radial clamping effect of the packing on the valve stem, while not restricting the rotational freedom of the valve stem about its own axis. Bearing constraints are applied to both ends of the valve stem to restrict its radial displacement, while retaining its rotational freedom about its own axis. Axial displacement constraints are applied to the top of the valve stem to simulate the axial limiting of the valve stem by the actuator or thrust bearing. Then, modal analysis is performed based on the above finite element model. The purpose of modal analysis is to solve for the natural frequencies and corresponding mode shapes of the valve structure. In the finite element software, after applying boundary constraints to the model, the eigenvalue equations of the structure are solved to obtain the natural frequencies and their corresponding mode shapes. The first natural frequency of the valve body is extracted, which is the lowest natural frequency of the valve body under boundary constraints. The corresponding mode shape is the bending vibration of the valve body flow channel wall in the radial direction. Simultaneously, the first natural frequency of the longitudinal vibration of the valve stem is extracted. This frequency is the natural frequency of the valve stem's reciprocating vibration along its own axis, and the corresponding mode shape is the extension and contraction vibration of the valve stem in the axial direction. The first natural frequency of the valve body is used as the first fault characteristic frequency because when the valve seat becomes loose or a gap appears at the connection between the valve body and the pipeline, the local constraint stiffness of the valve body decreases, and the vibration energy concentrates near the first natural frequency, resulting in a significant increase in the vibration amplitude at this frequency. The first natural frequency of the longitudinal vibration of the valve stem is used as the second fault characteristic frequency because when fatigue cracks appear in the valve stem or the connection between the valve stem and the valve plate becomes loose, the axial stiffness of the valve stem decreases, the longitudinal vibration natural frequency shifts, and the vibration amplitude at this frequency also increases significantly. In addition, in the early stages of cracking or loosening, the fault scale is small and may not be enough to cause a significant shift in the overall structural natural frequency, but the vibration energy at this fault characteristic frequency will increase significantly. Therefore, monitoring the frequency domain amplitude change at this characteristic frequency is often more sensitive and faster in capturing early fault characteristics than directly monitoring the frequency shift.Selecting these two natural frequencies as fault characteristic frequencies can most sensitively reflect changes in the mechanical structure of the valve body and stem, providing a physical basis for fault determination in S3. Finally, the first-order natural frequency of the valve body and the first-order natural frequency of longitudinal vibration of the valve stem, extracted from finite element modal analysis, are pre-set into the monitoring system as the first and second fault characteristic frequencies, respectively. These fault characteristic frequencies are mainly determined by the geometry and material properties of the valve body and stem. During service, they may drift slightly due to factors such as material corrosion, wear, or changes in packing preload. Therefore, it is preferable to pre-extract them offline and update them through periodic verification or online calibration when necessary.

[0050] The first fault characteristic frequency can also be taken as the characteristic frequency corresponding to the valve seat loosening fault. When the valve seat becomes loose, the connection stiffness between the valve seat and the valve body decreases, causing a change in the local constraint conditions of the valve body, and the natural frequency of the structure shifts relative to the healthy state. To extract the characteristic frequency corresponding to this fault, the valve seat loosening condition is simulated in the finite element model. The connection constraint between the valve seat and the valve body is modified from a fixed constraint to an elastic constraint, reducing the connection stiffness to simulate the degree of loosening. Modal analysis is then performed again on the modified finite element model to extract the first natural frequency of the valve body, which is the characteristic frequency corresponding to the valve seat loosening fault. In actual monitoring, if the first natural frequency of the valve body in the healthy state is extracted in advance, the valve seat loosening is identified by monitoring the amplitude change at this frequency. If the characteristic frequency under the valve seat loosening condition is extracted in advance, this frequency is directly used as the monitoring object. Both methods can effectively capture the vibration characteristic changes caused by valve seat loosening. Similarly, the second fault characteristic frequency can also be taken as the characteristic frequency corresponding to the valve stem crack fault. When fatigue cracks appear on the valve stem, the local stiffness of the cracked area decreases, and the natural frequency of the valve stem's longitudinal vibration shifts to lower frequencies. To extract the characteristic frequency corresponding to this fault, the valve stem crack condition is simulated in the finite element model, and a local stiffness reduction region is introduced at the crack-prone location on the valve stem. Modal analysis is then performed again on the modified finite element model to extract the first-order natural frequency of the valve stem's longitudinal vibration. This frequency is the characteristic frequency corresponding to the valve stem crack fault. In actual monitoring, if the first-order natural frequency of the valve stem's longitudinal vibration under healthy conditions is extracted beforehand, the valve stem crack is identified by monitoring the amplitude change or frequency shift at this frequency. If the characteristic frequency under valve stem crack conditions is extracted beforehand, this frequency is directly used as the monitoring target. Both methods can effectively capture the vibration characteristic changes caused by valve stem cracks.

[0051] S4: Input the obtained valve body mechanical status flag, valve stem mechanical status flag, and hydraulic status flag into a two-level hierarchical cross-verification decision structure based on Boolean logic to determine the field anomaly. The hydraulic status flag is used as the first-level determination variable to classify the current state into a hydraulic steady-state field or a hydraulic anomaly field. The valve body mechanical status flag and valve stem mechanical status flag are used as the second-level determination variables to locate the fault in each field and output the corresponding fault determination result.

[0052] Furthermore, the decision-making rules for the two-level hierarchical cross-validation decision structure include: Level 1 determination: The hydraulic state indicator is used as the determination variable; when the hydraulic state indicator represents a normal state, it is determined that the current state is in a hydraulic steady-state field; when the hydraulic state indicator represents an abnormal state, it is determined that the current state is in a hydraulic abnormal field. Second-level judgment: The valve body mechanical status indicator and the valve stem mechanical status indicator are the judgment variables; In the hydraulic steady-state field: If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate a normal state, the valve system is considered to be operating normally. If the valve body mechanical status indicator indicates an abnormal state while the valve stem mechanical status indicator indicates a normal state, it is determined that there is a fault in the valve body mechanical structure. If the valve body mechanical status indicator indicates a normal state and the valve stem mechanical status indicator indicates an abnormal state, it is determined that there is a fault in the valve stem transmission assembly. If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate an abnormal state, it is determined that there is a serious fault in the overall mechanical structure of the valve. In hydraulically anomalous fields: If the valve body mechanical status indicator and the valve stem mechanical status indicator both indicate a normal state, it is determined that there is an upstream flow field disturbance or process fluctuation, and the valve body is not faulty. If the valve body mechanical status indicator indicates an abnormal state while the valve stem mechanical status indicator indicates a normal state, it is determined that a fluid-induced valve body failure has occurred. If the valve body mechanical status indicator indicates a normal state and the valve stem mechanical status indicator indicates an abnormal state, it is determined that a fluid-induced valve stem failure has occurred. If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate an abnormal state, it is determined that the valve has experienced a fluid-induced overall failure.

[0053] Specifically, the valve body mechanical status flag, valve stem mechanical status flag, and hydraulic status flag are used as inputs. A two-level hierarchical cross-validation decision structure is used to cross-validate the fluid-structure characteristics, ultimately determining the valve's fault type and location. The three status flags independently reflect the valve's operating state from both the mechanical and fluid perspectives. The valve body and valve stem mechanical status flags are generated based on the residual signals of pure mechanical vibration after removing the flow-induced vibration component, reflecting the state of the valve's mechanical structure itself. The hydraulic status flag is generated based on the root mean square value of the fluid excitation force signal, reflecting whether the flow field is in a steady state. Since the aforementioned steps have removed the flow-induced vibration component from the vibration signal, the mechanical status flags are unaffected by normal flow field fluctuations. The fluid excitation force signal depends only on the flow rate signal and the lumped parameter model of fluid dynamics; therefore, the hydraulic status flag is also unaffected by mechanical vibration. The two sets of flags are physically independent, which is the physical basis for the two-level hierarchical cross-validation to effectively distinguish between operating condition disturbances and actual mechanical faults. The two-level hierarchical judgment utilizes this independence to verify the true cause of mechanical anomalies through the flow field state, and to eliminate false alarms caused by flow field disturbances through the mechanical state. When the hydraulic status indicator is normal but the mechanical status indicator is abnormal, the flow field is in a steady state, and the mechanical anomaly cannot be caused by flow field fluctuations; it can only be attributed to a fault in the mechanical structure itself. When the hydraulic status indicator is abnormal but the mechanical status indicator is normal, flow field instability is isolated on the fluid side and has not been transmitted to the mechanical structure, thus ruling out the possibility of valve body failure and actively suppressing false alarms.

[0054] The first-level decision in the two-tiered hierarchical cross-verification decision structure uses hydraulic state indicators as the decision variables. The hydraulic state indicators are either indicative of a normal or abnormal state. When the hydraulic state indicator indicates a normal state, it means the current flow field is in a steady state, the fluid excitation force signal fluctuates within the normal range, and there are no abnormal events such as sudden changes in upstream pump operating conditions or water hammer in the pipeline. In this case, the system is determined to be in a hydraulically steady-state field. When the hydraulic state indicator indicates an abnormal state, it means the flow field is unstable, possibly due to changes in upstream pump speed, valve switching in the pipeline, or flow pulsations caused by waves. In this case, the system is determined to be in a hydraulically abnormal field. The purpose of the first-level decision is to divide the valve's operating environment into two fields: hydraulically steady-state and hydraulically abnormal, providing the environmental background for the second-level decision. The second-level decision uses the valve body mechanical state indicator and the valve stem mechanical state indicator as decision variables to locate the fault within the field defined by the first-level decision. It should be noted that the valve body mechanical structure includes the valve seat, valve body bearing housing, valve body flow channel wall, and valve plate sealing ring. Faults in the valve body mechanical structure include valve seat loosening, valve body bearing housing wear, valve body flow channel wall corrosion, and valve plate sealing ring wear and aging. The valve stem drive assembly includes the valve stem itself, the valve stem-valve plate connection, and the electric actuator drive components. Faults in the valve stem drive assembly include valve stem cracks, loosening of the valve stem-valve plate connection, packing gland loosening, and loosening of the electric actuator drive components. Fault location is performed within the area defined by the first-level judgment. The valve system, as the final judgment object in the fault diagnosis conclusion, refers to the entire valve system, including the valve body, valve stem, and its drive assembly.

[0055] In a hydraulically steady-state field, the flow field is in a normal state, ruling out the possibility of mechanical abnormalities caused by flow field fluctuations. If both the valve body and valve stem mechanical status indicators show a normal state, it indicates that the vibration characteristics of both the valve body and valve stem are within the normal range, and the valve system is considered to be operating normally. If the valve body mechanical status indicator shows an abnormal state while the valve stem mechanical status indicator shows a normal state, and the valve body exhibits abnormal vibration while the valve stem is normal, since the interference of flow field fluctuations has been eliminated, the abnormality can only be attributed to a fault in the valve body's mechanical structure itself, and a fault is determined to exist in the valve body's mechanical structure. If the valve body mechanical status indicator shows a normal state while the valve stem mechanical status indicator shows an abnormal state, a fault is similarly determined to exist in the valve stem transmission assembly. If both the valve body and valve stem mechanical status indicators show an abnormal state, and both the valve body and valve stem exhibit abnormal vibration simultaneously, a serious fault is determined to exist in the overall mechanical structure of the valve.

[0056] In hydraulically abnormal fields, flow field instability can lead to two possible sources of abnormal mechanical status indicators: one is vibration response caused by flow field fluctuations transmitted through the structure, and the other is vibration due to mechanical structure failure. If both the valve body and valve stem mechanical status indicators are normal, it indicates that although abnormal flow field fluctuations have occurred, the mechanical structure has not produced an abnormal vibration response. Flow field instability is isolated on the fluid side and has not been transmitted to the mechanical structure. Therefore, the abnormality is determined to originate from upstream flow field disturbances or process fluctuations, and the valve body itself is fault-free, effectively avoiding false alarms caused by flow field fluctuations. If the valve body mechanical status indicator is abnormal while the valve stem mechanical status indicator is normal, flow field instability and valve body mechanical abnormality coexist. Flow field instability has induced abnormal vibration in the valve body, indicating fluid-induced valve body failure. Similarly, if the valve body mechanical status indicator is normal while the valve stem mechanical status indicator is abnormal, fluid-induced valve stem failure is also determined. If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate an abnormal state, and the flow field instability has been transmitted to the entire valve mechanical structure, it is determined that the valve has experienced fluid-induced overall failure.

[0057] The eight state combinations and their corresponding judgment results of the above two-level hierarchical mutual verification decision structure can be constructed as the following Boolean logic truth table. In the table, H represents the hydraulic state flag, B represents the valve body mechanical state flag, and S represents the valve stem mechanical state flag. 0 represents the normal state, and 1 represents the abnormal state.

[0058] When H=0, B=0, S=0, it is determined that the valve system is operating normally. When H=0, B=1, S=0, it is determined that there is a fault in the mechanical structure of the valve body. When H=0, B=0, S=1, it is determined that there is a fault in the valve stem drive assembly. When H=0, B=1, S=1, it is determined that the overall mechanical structure of the valve has a serious fault. When H=1, B=0, S=0, it is determined that the upstream flow field is disturbed or the process fluctuates, but the valve body is not faulty. When H=1, B=1, S=0, it is determined that there is a fault in the fluid-induced valve body mechanical structure. When H=1, B=0, S=1, it is determined that the fluid-induced valve stem actuation assembly is faulty. When H=1, B=1, S=1, it is determined that the overall mechanical structure of the fluid-induced valve has a fault.

[0059] During online monitoring, at each sampling moment, the system executes the first-level field determination and the second-level fault location sequentially according to the three status flags output in step three, following the aforementioned two-level hierarchical mutual verification decision structure, and outputs the corresponding fault determination result. This decision structure, through hierarchical mutual verification of the hydraulic field and mechanical state, achieves accurate fault source location and effective elimination of environmental interference, effectively solving the engineering problem that flow field fluctuations are easily misjudged as mechanical faults in traditional multi-feature fusion methods.

[0060] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A valve fault monitoring method based on the collaborative analysis of vibration and flow signals, characterized in that, The method includes: A flow meter is installed upstream of the valve, a first triaxial vibration sensor is installed at the valve body, and a second triaxial vibration sensor is installed at the valve stem. Vibration signals output by the first and second triaxial vibration sensors, flow signals output by the flow meter, and valve opening signals are synchronously acquired at a preset sampling frequency during valve operation. The flow signals are low-pass filtered, and the two vibration signals are wavelet thresholded to obtain preprocessed flow time-series data, valve body triaxial vibration time-series data, and valve stem triaxial vibration time-series data. Using preprocessed flow time-series data and valve opening signal as input, the force exerted by the fluid on the valve plate under the current operating condition is calculated through a pre-constructed fluid dynamics lumped parameter model, yielding the fluid excitation force signal. Based on the current valve opening signal, the valve body vibration transfer function and valve stem vibration transfer function matching the valve opening signal are determined. Based on the fluid excitation force signal and the two transfer functions, the first flow-induced vibration response component at the valve body position and the second flow-induced vibration response component at the valve stem position are calculated. The preprocessed valve body triaxial vibration time-series data is synchronously subtracted from the first flow-induced vibration response component to obtain the valve body triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component. The preprocessed valve stem triaxial vibration time-series data is synchronously subtracted from the second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component. For each axis component of the valve body pure mechanical vibration residual signal and the valve stem pure mechanical vibration residual signal, the time-domain kurtosis value and the frequency-domain amplitude corresponding to the preset fault characteristic frequency are extracted as feature values. Each feature value is compared with the corresponding preset normal range. When any feature value exceeds its corresponding preset normal range, it is determined that there is an abnormality in the corresponding mechanical structure, and a mechanical state flag representing the abnormal state is generated; otherwise, a mechanical state flag representing the normal state is generated. At the same time, the root mean square value of the fluid excitation force signal is calculated, and the root mean square value is compared with the preset hydraulic steady-state threshold interval. When the root mean square value exceeds the hydraulic steady-state threshold interval, it is determined that there is instability in the flow field, and a hydraulic state flag representing the abnormal state is generated; otherwise, a hydraulic state flag representing the normal state is generated. The obtained valve body mechanical status flag, valve stem mechanical status flag, and hydraulic status flag are input into a two-level hierarchical cross-verification decision structure based on Boolean logic to determine field anomalies. The hydraulic status flag is used as the first-level decision variable to classify the current state into a hydraulic steady-state field or a hydraulic anomaly field. The valve body mechanical status flag and valve stem mechanical status flag are used as the second-level decision variables to locate faults in each field and output the corresponding fault determination results.

2. The valve fault monitoring method based on the joint analysis of vibration and flow signals as described in claim 1, characterized in that, Obtain the fluid excitation force signal, including: Obtain the instantaneous flow rate value corresponding to the preprocessed flow time series data, and the instantaneous valve opening value corresponding to the valve opening signal; Determine the equivalent flow area of ​​the valve under the current operating conditions based on the instantaneous opening value; The instantaneous flow rate, instantaneous opening value, and equivalent flow area of ​​the valve are substituted into the fluid dynamics lumped parameter model. The static pressure difference force is calculated based on Bernoulli's equation, and the steady-state hydrodynamic force is calculated based on the momentum theorem. The static pressure difference force and the steady-state hydrodynamic force are vector-synthesized, and the resultant force is used as the fluid excitation force signal of the fluid acting on the valve plate under the current operating condition.

3. The valve fault monitoring method based on the joint analysis of vibration and flow signals as described in claim 2, characterized in that, The preprocessed triaxial vibration time series data of the valve body is synchronously subtracted from the first flow-induced vibration response component to obtain the valve body triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component; the preprocessed triaxial vibration time series data of the valve stem is synchronously subtracted from the second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal stripped of the flow-induced vibration component, including: The system monitors the current valve opening signal in real time and determines whether the current valve opening signal matches the opening node in the pre-calibrated transfer function database. The transfer function database stores time-domain sequence parameters that characterize the valve body vibration transfer function and valve stem vibration transfer function under several calibrated openings. The time-domain sequence parameters include components in three orthogonal directions: X, Y, and Z. If no matching opening node is found, the time-domain sequence parameters of the vibration transfer functions of the valve body and valve stem corresponding to the adjacent calibrated opening nodes are extracted according to the position of the current valve opening signal between the adjacent calibrated opening nodes. The time-domain sequence parameters are then interpolated in the direction to obtain the three-axis time-domain sequence parameters of the real-time valve body vibration transfer function and the real-time valve stem vibration transfer function under the current opening. The fluid excitation force signal is convolved with the triaxial time-domain sequence parameters of the real-time valve body vibration transfer function and the triaxial time-domain sequence parameters of the real-time valve stem vibration transfer function in the corresponding directions to obtain the first flow-induced vibration response components of the valve body in the X, Y, and Z axes, which characterize the vibration response generated by the fluid excitation at the valve body, and the second flow-induced vibration response components of the valve stem in the X, Y, and Z axes, which characterize the vibration response generated by the fluid excitation at the valve stem. The valve body is the installation position of the first triaxial vibration sensor, and the valve stem is the installation position of the second triaxial vibration sensor. On the time axis, the preprocessed valve body triaxial vibration time series data is subtracted from the corresponding first flow-induced vibration response component point by point in the direction to obtain the valve body triaxial pure mechanical vibration residual signal. The preprocessed valve stem triaxial vibration time series data is subtracted point by point in each direction from the corresponding second flow-induced vibration response component to obtain the valve stem triaxial pure mechanical vibration residual signal.

4. A valve fault monitoring method based on the collaborative analysis of vibration and flow signals according to claim 3, characterized in that, The transfer function database stores time-domain sequence parameters characterizing the valve body vibration transfer function and valve stem vibration transfer function at several calibrated opening degrees, including: Establish a finite element model of the valve, which includes the geometric and material property parameters of the valve body, valve core, valve stem, and stuffing box. In the finite element model, the valve opening is set to multiple preset calibration openings. Under each calibration opening, a dynamic excitation force covering the actual working frequency band of the valve is applied to the valve core for harmonic response analysis. The frequency domain vibration response in the three orthogonal directions of X, Y, and Z at the installation positions of the first and second triaxial vibration sensors is obtained. The frequency domain vibration response at the installation position of the first triaxial vibration sensor, obtained from simulation at each calibration opening degree, is divided by the applied dynamic excitation force to obtain the frequency domain transfer function of the valve body vibration at that calibration opening degree; the frequency domain vibration response at the installation position of the second triaxial vibration sensor, obtained from simulation at each calibration opening degree, is divided by the applied dynamic excitation force to obtain the frequency domain transfer function of the valve stem vibration at that calibration opening degree. Inverse Fourier transforms were performed on the frequency domain transfer functions of the valve body vibration and the valve stem vibration to obtain the triaxial time-domain sequence parameters of the valve body vibration transfer function and the valve stem vibration transfer function under each calibrated opening degree, and then stored in the transfer function database.

5. A valve fault monitoring method based on the collaborative analysis of vibration and flow signals according to claim 3, characterized in that, When any feature value exceeds its corresponding preset normal range, an anomaly is determined in the corresponding mechanical structure, and a mechanical state flag representing the abnormal state is generated; otherwise, a mechanical state flag representing the normal state is generated. Simultaneously, the root mean square (RMS) value of the fluid excitation force signal is calculated and compared with a preset hydraulic steady-state threshold range. When the RMS value exceeds the hydraulic steady-state threshold range, instability in the flow field is determined, and a hydraulic state flag representing the abnormal state is generated; otherwise, a hydraulic state flag representing the normal state is generated, including: For each axis component of the valve body's pure mechanical vibration residual signal, its time-domain kurtosis value is extracted, and the frequency-domain amplitude corresponding to a preset first fault characteristic frequency is also extracted. The time-domain kurtosis value of each axis component is compared with the preset normal range of valve body kurtosis, and the frequency-domain amplitude corresponding to the first fault characteristic frequency of each axis component is compared with the preset normal range of valve body amplitude. When the time-domain kurtosis value of any component is not within the normal range of valve body kurtosis, or when the frequency domain amplitude corresponding to the first fault characteristic frequency of any component is not within the normal range of valve body amplitude, it is determined that there is an abnormality in the valve body mechanical structure, and a valve body mechanical state flag representing the abnormal state is generated; otherwise, a valve body mechanical state flag representing the normal state is generated. For each axis component of the valve stem's pure mechanical vibration residual signal, its time-domain kurtosis value is extracted, and the frequency-domain amplitude corresponding to a preset second fault characteristic frequency is also extracted. The time-domain kurtosis value of each axis component is compared with the preset normal range of valve stem kurtosis, and the frequency-domain amplitude corresponding to the second fault characteristic frequency of each axis component is compared with the preset normal range of valve stem amplitude. When the time-domain kurtosis value of any component is not within the normal range of valve stem kurtosis, or the frequency-domain amplitude corresponding to the second fault characteristic frequency of any component is not within the normal range of valve stem amplitude, it is determined that there is an abnormality in the valve stem mechanical structure, and a valve stem mechanical state flag representing the abnormal state is generated; otherwise, a valve stem mechanical state flag representing the normal state is generated. The root mean square (RMS) value of the fluid excitation force signal is calculated and compared with a preset hydraulic steady-state threshold range. If the RMS value is not within the hydraulic steady-state threshold range, the flow field is determined to be unstable, and a hydraulic state flag representing the abnormal state is generated; otherwise, a hydraulic state flag representing the normal state is generated.

6. A valve fault monitoring method based on the collaborative analysis of vibration and flow signals according to claim 5, characterized in that, The preset normal ranges for valve body kurtosis, valve body amplitude, valve stem kurtosis, valve stem amplitude, and hydraulic steady-state threshold intervals are all threshold intervals determined based on a pre-constructed threshold model. This threshold model is pre-constructed using the following method: Acquire historical operating data of the valve under known health conditions. The historical operating data includes operating parameters and corresponding characteristic index values. The operating parameters include at least the valve opening degree and flow signal. The characteristic index values ​​include the kurtosis value of each axis of the valve body, the amplitude value of each axis of the valve body, the kurtosis value of each axis of the valve stem, the amplitude value of each axis of the valve stem, and the root mean square value of the fluid excitation force signal. The historical operating data is fitted and analyzed to construct a benchmark mapping relationship between the values ​​of each characteristic index and the changes in operating parameters, and a benchmark surface equation for each characteristic index is established. For each characteristic indicator, the fluctuation and dispersion of historical operating data under healthy conditions are analyzed, and the maximum dispersion extreme value within the operating condition range covered by the historical operating data is extracted. The maximum dispersion extreme value is used to construct the full operating condition dispersion envelope, and the full operating condition dispersion envelope is used as the global safety boundary of the characteristic indicator. During online monitoring, the currently collected operating parameters are substituted into the reference surface equation to calculate the theoretical reference value of each characteristic index under the current operating conditions. Combined with the corresponding global safety boundary, the normal range of the dynamic threshold of each characteristic index under the current operating conditions is determined.

7. A valve fault monitoring method based on the collaborative analysis of vibration and flow signals according to claim 5, characterized in that, The first fault characteristic frequency is the first natural frequency of the valve body or the characteristic frequency corresponding to the valve seat loosening fault. The second fault characteristic frequency is the first natural frequency of the longitudinal vibration of the valve stem or the characteristic frequency corresponding to the valve stem crack fault. The first and second fault characteristic frequencies are pre-extracted by performing finite element modal analysis on the valve.

8. A valve fault monitoring method based on the collaborative analysis of vibration and flow signals according to claim 5, characterized in that, A two-level hierarchical cross-validation decision structure includes: Level 1 determination: The hydraulic state indicator is used as the determination variable; when the hydraulic state indicator represents a normal state, it is determined that the current state is in a hydraulic steady-state field; when the hydraulic state indicator represents an abnormal state, it is determined that the current state is in a hydraulic abnormal field. Second-level judgment: The valve body mechanical status indicator and the valve stem mechanical status indicator are the judgment variables; In the hydraulic steady-state field: If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate a normal state, the valve system is considered to be operating normally. If the valve body mechanical status indicator indicates an abnormal state and the valve stem mechanical status indicator indicates a normal state, the valve body mechanical structure is determined to be faulty. If the valve body mechanical status indicator indicates a normal state and the valve stem mechanical status indicator indicates an abnormal state, the valve stem transmission assembly is determined to be faulty. If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate an abnormal state, it is determined that there is a serious fault in the overall mechanical structure of the valve. In hydraulically anomalous fields: If the valve body mechanical status indicator and the valve stem mechanical status indicator both indicate a normal state, it is determined that there is an upstream flow field disturbance or process fluctuation, and the valve body is not faulty. If the valve body mechanical status indicator indicates an abnormal state and the valve stem mechanical status indicator indicates a normal state, it is determined that a fluid-induced valve body mechanical structure failure has occurred. If the valve body mechanical status indicator indicates a normal state and the valve stem mechanical status indicator indicates an abnormal state, it is determined that a fluid-induced valve stem transmission assembly failure has occurred. If both the valve body mechanical status indicator and the valve stem mechanical status indicator indicate an abnormal state, it is determined that a fluid-induced overall mechanical structure failure of the valve has occurred.

Images