Shuttle valve sealing failure early warning method fusing vibration analysis and flow characteristics

Abstract

The invention provides a shuttle valve sealing failure early warning method fusing vibration analysis and flow characteristics, and belongs to the technical field of industrial fluid monitoring based on deep learning. The method comprises the following steps: firstly, collecting shuttle valve sealing failure characteristic data, including shuttle valve three-axis vibration signals, pipeline fluid, material physical parameters and environment working condition data; secondly, a multi-physical field data completion model is constructed, unsupervised learning is carried out on the collected pipeline fluid and environment working condition data, and completion pipeline flow data and environment working condition data with spatial distribution characteristics are generated; thirdly, constructing a shuttle valve sealing health index prediction model based on multi-physical field information fusion, and inputting the collected data and the complemented data into the model to obtain a shuttle valve sealing health index; and finally, shuttle valve sealing failure mode classification is carried out based on the obtained shuttle valve sealing health index. According to the method, the problem that a failure process cannot be comprehensively captured by a single feature is solved, the spatial-temporal distribution feature of an unmonitored area is effectively deduced, and dynamic quantitative evaluation is realized.

Description

Technical Field

[0001] This invention belongs to the field of industrial fluid monitoring technology based on deep learning, and particularly relates to a method for early warning of shuttle valve seal failure that integrates vibration analysis and flow characteristics. Background Technology

[0002] As a core component of industrial fluid control systems, the shuttling valve's sealing performance directly affects system safety and operational efficiency. Traditional monitoring methods often rely on single-modal data such as vibration signals or flow parameters, making it difficult to comprehensively capture the early characteristics of seal failure. Seal failure is often accompanied by the coupled evolution of multiple physical fields, including abnormal mechanical vibration, fluid pressure fluctuations, and material wear. Existing methods suffer from problems such as delayed early warning and low classification accuracy due to data silos and a lack of physical mechanism modeling. Therefore, there is an urgent need for an intelligent early warning method that integrates multi-physical field information.

[0003] Currently, the main methods for early warning of shuttle valve seal failure include the following: Vibration analysis: This method detects seal failure by analyzing the vibration signals of the shuttle valve. It primarily relies on changes in the frequency and amplitude of the vibration signal to determine the sealing condition. However, vibration analysis alone may not fully capture the complex mechanisms of seal failure.

[0004] Flow monitoring method: This method assesses the sealing performance of a shuttle valve by monitoring changes in pipeline flow and pressure. Abnormal fluctuations in flow data may indicate seal failure, but this method also has limitations, as it cannot comprehensively consider other influencing factors.

[0005] Material testing method: This method assesses wear and aging by testing the physical parameters of the valve's sealing surface material. While it provides information on the material's condition, it struggles to dynamically reflect the process of seal failure.

[0006] These methods typically focus only on single or local features, neglecting the impact of multiphysics coupling effects on seal failure, resulting in low early warning accuracy. Furthermore, most existing methods lack the ability to dynamically characterize the seal failure process, making early warning and timely maintenance difficult. Summary of the Invention

[0007] To address the above problems, this invention proposes a shuttle valve seal failure early warning method that integrates vibration analysis and flow characteristics, comprising the following steps: S1, Real-time acquisition of shuttle valve seal failure characteristic data, including shuttle valve triaxial vibration signal data, pipeline fluid data, material physical parameter data, and environmental condition data; the pipeline fluid data includes pipeline flow rate data and pipeline pressure data; the material physical parameter data includes shuttle valve sealing surface hardness data and shuttle valve sealing surface thickness attenuation rate data; the environmental condition data includes pipeline temperature data, pipeline humidity data, and pipeline chloride ion concentration data; S2, input the pipeline fluid data collected in S1 and the environmental condition data without pipeline chloride ion concentration data into the constructed multiphysics data completion model, perform real-time unsupervised training and learning, and generate completed pipeline flow data and environmental condition data with spatial distribution characteristics based on M pipeline position coordinates; the multiphysics data completion model defines boundary conditions for each physical field. S3 inputs the triaxial vibration signal data, material physical parameter data, pipeline chloride ion concentration data, and pipeline fluid data and other environmental condition data obtained in S1 into the constructed multi-physics information fusion shuttle valve sealing health index prediction model, and outputs the shuttle valve sealing health index prediction result. S4. Compare the predicted shuttle valve seal health index obtained in S3 with the preset threshold to obtain the shuttle valve seal failure mode.

[0008] Preferably, the method for collecting shuttle valve seal failure characteristic data in S1 includes: The shuttle valve triaxial vibration signal data is obtained by using a magnetically attached triaxial accelerometer fixed to the top surface and two sides of the shuttle valve body in the orthogonal direction to form the XYZ axis, ensuring that the sensor is coupled with the valve body vibration mode, and capturing the dynamic response of the high-frequency component of the valve core opening and closing impact and the mid-frequency noise of the sealing ring friction. The material physical parameter data are selected uniformly from the shuttle valve sealing surface. Microhardness testers were deployed at several points to measure the hardness of each point. The hardness value at each point is used to obtain a comprehensive range of hardness values. The hardness data of the shuttle valve sealing surface at each sampling point were obtained; and the 3D profilometer was used to scan the shuttle valve sealing surface to obtain the surface roughness and thickness attenuation, and the thickness attenuation rate was calculated. The chloride ion concentration data of the pipeline is obtained by deploying a corrosion potentiometer in the shuttle valve body. The chloride ion concentration data characterizes the corrosion intensity inside the pipeline.

[0009] Preferably, the multiphysics data completion model is used to complete four types of physical field data: pipeline flow rate data, pipeline pressure data, temperature data, and humidity data. The model consists of an input layer, a hidden layer, Gaussian kernel functions for pipeline flow rate, pipeline pressure, temperature, and humidity, and an output layer. The defined physical field boundary conditions include: Pipeline flow data boundary conditions include inlet boundary, wall boundary, and outlet boundary; Pipeline pressure data boundary conditions include inlet boundary, outlet boundary, and wall boundary; Temperature data boundary conditions include isothermal walls, adiabatic walls, convective heat transfer walls, and inlet fluid temperature boundary conditions. Humidity data boundary conditions include adsorption wall boundary conditions and adiabatic wall boundary conditions.

[0010] A preferred unsupervised alternating training strategy is used to train the multiphysics data completion model, and includes the following processes: S21, uniform sampling is performed within the length sampling interval of the shuttle valve pipe to obtain... Pipe location coordinates ; S22, use the boundary conditions of pipeline flow data as the loss function of pipeline flow Gaussian kernel function, use the boundary conditions of pipeline pressure data as the loss function of pipeline pressure Gaussian kernel function, use the boundary conditions of pipeline pressure data as the loss function of pipeline pressure Gaussian kernel function, use the boundary conditions of temperature data as the loss function of temperature Gaussian kernel function, and use the boundary conditions of humidity data as the loss function of humidity Gaussian kernel function. S23, will The model is completed using multiphysics data. First, the input layer, hidden layer, pipeline flow rate Gaussian kernel function, and output layer are trained, with the parameters of the pipeline pressure, temperature, and humidity Gaussian kernel functions fixed. Once the boundary loss of the pipeline flow rate data decreases to a preset threshold, the parameters of the pipeline flow rate Gaussian kernel function are frozen. Second, the input layer, hidden layer, pipeline pressure Gaussian kernel function, and output layer are trained, with the parameters of the pipeline flow rate, temperature, and humidity Gaussian kernel functions fixed. Once the boundary loss of the pipeline pressure data decreases to a preset threshold, the parameters of the pipeline pressure Gaussian kernel function are frozen. The parameters of the Gaussian kernel function for pressure are determined. The input layer, hidden layer, temperature Gaussian kernel function, and output layer are trained again, with the parameters of the pipeline flow rate Gaussian kernel function, pipeline pressure Gaussian kernel function, and humidity Gaussian kernel function remaining unchanged. Once the temperature data boundary loss decreases to a preset threshold, the temperature Gaussian kernel function parameters are frozen. Finally, the input layer, hidden layer, humidity Gaussian kernel function, and output layer are trained again, with the parameters of the pipeline flow rate Gaussian kernel function, pipeline pressure Gaussian kernel function, and temperature Gaussian kernel function remaining unchanged. Once the humidity data boundary loss decreases to a preset threshold, the humidity Gaussian kernel function parameters are frozen. S24, Gaussian distribution sampling is performed on the length sampling interval of the shuttle valve pipe to obtain... Pipe location coordinates ;Will Input the trained multiphysics data completion model to obtain completed pipeline flow rate data, pipeline pressure data, temperature data, and humidity data.

[0011] Preferably, the shuttle valve sealing health index prediction model includes a multi-physics feature extraction module, a multi-physics feature enhancement module, and a prediction output module; The multiphysics feature extraction module includes a vibration feature extraction network, a pipeline feature extraction network, a material physical parameter feature extraction network, and an environmental condition feature extraction network; it extracts valve body vibration signal features, pipeline fluid time sequence features, shuttle valve sealing surface hardness data features and thickness attenuation rate features, environmental spatiotemporal features, and time-dependent features, respectively. The multiphysics feature enhancement module includes a vibration signal feature enhancement network and a pipeline fluid feature enhancement network, which are used to enhance the extracted vibration signal features and pipeline fluid temporal features, respectively. The prediction output module is used to perform feature scale alignment, feature fusion, and feature activation transformation on the spatiotemporal features of the completed temperature data, the spatiotemporal features of the completed humidity data, the bidirectional time dependence features of the chloride ion concentration data, the aggregated features of the shuttle valve sealing surface hardness data, the time dependence features of the thickness attenuation rate, and the multi-scale enhancement features of the shuttle valve triaxial vibration signal and the pipeline flow enhancement features obtained by the multiphysics feature enhancement module to obtain the shuttle valve sealing health index.

[0012] Preferably, the vibration feature extraction network consists of 3×3 convolutional layers, 5×5 convolutional layers, and 7×7 convolutional layers, which are used to extract small-scale, medium-scale, and large-scale features of the triaxial vibration signal of the shuttle valve, respectively, to capture the short-term, medium-term, and long-term changes in the valve body vibration. The pipeline feature extraction network includes a long short-term memory network layer and a graph convolutional layer; wherein the long short-term memory network layer is used to capture pipeline flow time-series dependency features and pipeline pressure time-series dependency features, and the graph convolutional layer is used to extract spatial features from the pipeline flow time-series dependency features and pipeline pressure time-series dependency features to obtain high-dimensional pipeline flow features and high-dimensional pipeline pressure features containing spatiotemporal information. The material physical parameter feature extraction network consists of a graph attention layer and a long short-term memory network layer. The hardness data of the shuttle valve sealing surface is input into the graph attention convolution layer to perform spatial aggregation of the hardness values ​​of M sampling points to obtain the aggregated features of the hardness data of the shuttle valve sealing surface. The thickness attenuation rate is input into the long short-term memory network layer to capture the relationship between the thickness attenuation rate and time to obtain the time-dependent features of the thickness attenuation rate. The environmental feature extraction network comprises a forward long short-term memory (LSTM) network layer, an inverse LSM network layer, and a spatial attention layer. The forward LSM network layer extracts the forward time dependence features of the completed temperature, humidity, and chloride ion concentration data. The inverse LSM network layer extracts the inverse time dependence features of the completed temperature, humidity, and chloride ion concentration data. The spatial attention layer performs weighted summation on the forward and inverse time dependence features of the completed temperature and humidity data, respectively, to deeply extract the variation trends of the three features in the pipeline space, thereby obtaining the spatiotemporal features of the completed temperature and humidity data. The forward and inverse time dependence features of the chloride ion concentration data are concatenated to obtain the bidirectional time dependence features of the chloride ion concentration data.

[0013] Preferably, the multiphysics feature enhancement module specifically comprises: The vibration signal feature enhancement network comprises a gated unit layer and a feature splicing layer. First, the small-scale, mesoscale, and large-scale features of the obtained triaxial vibration signal of the shuttle valve are input into the gated unit layer. The gated unit layer weights the vibration signal features along the time dimension, making the model focus more on the vibration signal features at the current moment and less on the vibration signal features at historical moments, thus obtaining enhanced small-scale, mesoscale, and large-scale features of the vibration signal. Second, the enhanced small-scale, mesoscale, and large-scale features of the vibration signal are input into the feature splicing layer to obtain multi-scale enhanced features of the triaxial vibration signal of the shuttle valve. The pipeline flow feature enhancement network includes a spatial attention layer, a gating unit layer, and a feature splicing layer. First, the obtained high-dimensional pipeline flow features and high-dimensional pipeline pressure features are input into the spatial attention layer. The spatial attention layer aggregates the spatial distributions of the high-dimensional pipeline flow features and high-dimensional pipeline pressure features to obtain spatially weighted pipeline flow features and pipeline pressure features. Then, the spatially weighted pipeline flow features and pipeline pressure features are input into the gating unit for time-dimensional weighting and input into the feature splicing layer to obtain pipeline flow enhancement features that characterize pipeline flow changes.

[0014] Preferably, the prediction output module includes a feature concatenation layer, a pooling layer, a cross-modal attention fusion layer, and a softmax activation function layer; the specific data processing includes: The spatiotemporal features of the completed temperature data, the spatiotemporal features of the completed humidity data, and the bidirectional time-dependent features of the chloride ion concentration data are input into the pooling layer for feature alignment. The aligned features are then input into the feature splicing layer to obtain the environmental condition features. The hardness data of the valve sealing surface is aggregated and the thickness decay rate time-dependent feature is input into the pooling layer for feature alignment. The aligned features are then input into the feature splicing layer to obtain the material physical parameter features. Multi-scale enhancement features of the shuttle valve triaxial vibration signal, pipeline flow enhancement features, environmental operating condition features, and material physical parameter features are input into a pooling layer for feature alignment. The aligned features are then input into a cross-modal attention fusion layer. The cross-modal feature fusion layer contains four dual attention heads, which respectively represent the interaction relationships between the four feature sequences to obtain fused features. The fused features are input into the softmax activation function layer to obtain the prediction results of the valve sealing health index.

[0015] Preferably, the health index prediction result is a continuous value in the range of 0 to 1, where 0 is the complete failure boundary, representing the complete failure of the shuttle valve sealing performance, and 1 is the complete sealing boundary, representing the healthy sealing performance of the shuttle valve.

[0016] Preferably, the specific process of S4 includes: The predicted shuttle valve seal health index obtained by S3 is compared with a preset threshold to obtain the shuttle valve seal failure mode, including three modes: severe failure of shuttle valve seal performance, deterioration of shuttle valve seal performance, and healthy shuttle valve seal performance. When the predicted shuttle valve seal health index is less than 0.3, it indicates severe failure of shuttle valve seal performance, which requires immediate repair. When the predicted shuttle valve seal health index is greater than 0.3 and less than 0.7, it indicates deterioration of shuttle valve seal performance, which requires early warning and enhanced monitoring of the shuttle valve sealing surface. When the predicted shuttle valve seal health index is greater than 0.7, it indicates a healthy state of shuttle valve seal performance, which does not require manual intervention.

[0017] Compared with the prior art, the beneficial effects brought about by the innovation of this invention include: (1) Multiphysics Data Fusion and Dynamic Characterization: For the first time, the shuttle valve sealing failure problem is regarded as a multiphysics coupling problem. Four types of heterogeneous data, namely vibration signal, flow time series data, material physical parameters and environmental conditions, are fused. Through spatiotemporal feature extraction and enhancement, the complex mechanism of sealing failure is dynamically characterized. This solves the problem that a single feature cannot fully capture the failure process, and realizes the inference of the spatiotemporal distribution characteristics of unmonitored areas using limited sensor data, thus making up for the deficiency of insufficient sensor deployment.

[0018] (2) Dynamic health index prediction model: Construct a sealing health index prediction model based on multi-physics information fusion. Through feature extraction, enhancement and cross-modal fusion, generate a health index in the continuous range of 0-1 to realize dynamic quantitative evaluation of sealing performance from "completely healthy" to "completely failed". Attached Figure Description

[0019] Figure 1This is a flowchart illustrating the overall technical route of the present invention.

[0020] Figure 2 This is a schematic diagram of the alternating training of the multiphysics data completion model of the present invention.

[0021] Figure 3 This is a schematic diagram of the model structure for predicting the sealing health index of the shuttle valve according to the present invention.

[0022] Figure 4 This is a graph showing the degradation trend and early warning threshold of the shuttle valve sealing health index in an embodiment of the present invention.

[0023] Figure 5 This is a confusion matrix diagram for classifying the failure modes of the shuttle valve seal in an embodiment of the present invention.

[0024] Figure 6 This is a performance comparison chart of different early warning methods in embodiments of the present invention. Detailed Implementation

[0025] This invention proposes a method for early warning of shuttle valve seal failure that integrates vibration analysis and flow characteristics, such as... Figure 1 As shown, the overall process includes: First, collecting shuttle valve seal failure characteristic data, including shuttle valve triaxial vibration signals, pipeline fluid data, material physical parameters, and environmental condition data; second, constructing a multi-physics data completion model, performing unsupervised learning on the collected pipeline fluid data and environmental condition data to generate completed pipeline flow data and environmental condition data with spatial distribution characteristics; third, constructing a shuttle valve seal health index prediction model based on multi-physics information fusion, inputting the collected data and completed data into the model to obtain the shuttle valve seal health index; finally, classifying shuttle valve seal failure modes based on the obtained shuttle valve seal health index.

[0026] The invention will be further described below with reference to specific embodiments.

[0027] I. Data Collection on Characteristics of Shuttle Valve Seal Failure To achieve early warning of shuttle valve seal failure, this invention collects multimodal data related to shuttle valve seal failure, including vibration signal data, pipeline data, material parameter data, and environmental condition data, as detailed below: This system collects triaxial vibration signal data from a shuttle valve to comprehensively capture the mechanical characteristics when the valve seal fails. A PCB 352C65 ICP triaxial accelerometer is rigidly fixed to the top surface and two sides (X / Y / Z axes) of the shuttle valve body using a magnetic base, ensuring modal coupling between the sensor and the valve body vibration. This effectively captures the dynamic response of the high-frequency components of the valve core's opening and closing impact and the mid-frequency noise from the sealing ring friction. The sampling frequency is 20kHz, and the duration of a single acquisition is [duration missing]. .

[0028] This system collects time-series flow and pressure data from shuttle valve pipelines to comprehensively capture the internal flow and pressure characteristics of the pipeline when the shuttle valve seal fails. An E+H FTW330 turbine flow meter is installed in the straight pipe section at the shuttle valve inlet, achieving a leak-free connection via a magnetic coupling. It is used to collect pipeline flow data, with a single data acquisition duration of [duration missing]. The Keller PA-33X differential pressure transmitter was symmetrically installed at positions 5D upstream of the inlet and 3D downstream of the outlet to avoid valve flow field disturbance. Pressure time-series data was collected, with a single acquisition duration of [duration missing]. .

[0029] Physical parameter data of the shuttle valve sealing surface material were collected to quantify the shuttle valve seal failure state. Uniformly selected samples were taken from the shuttle valve sealing surface. A Zwick 3120 microhardness tester was deployed at each point to measure... The hardness value at each point is used to obtain a comprehensive range of hardness values. The hardness data of the valve sealing surface at each sampling point, with a single sampling time of [time value missing]. The sealing surface of the shuttle valve was scanned using a Keyence VK-X200 3D profilometer to obtain the surface roughness and thickness attenuation. The thickness attenuation rate was calculated, and the duration of a single acquisition was [duration missing]. .

[0030] Environmental condition data was collected to characterize the relationship between the shuttle valve's seal failure mechanism and environmental conditions. A SHT30 intelligent temperature and humidity sensor was installed within 50mm of the shuttle valve body using a metal-armored cable to collect pipeline temperature and humidity data. The duration of a single data collection session was... A Leici ZDJ-5B corrosion potentiometer was deployed in the shuttle valve body to obtain chloride ion concentration data in the pipeline. The duration of a single acquisition was [duration missing]. Chloride ion concentration data characterizes the corrosion intensity inside the pipeline.

[0031] The aforementioned sensors are triggered by a unified Siemens S7-1500 PLC controller, which transmits data in real time to the computer via the MODBUSRTU protocol. This data includes shuttle valve triaxial vibration signal, pipeline flow rate, pipeline pressure, pipeline temperature, pipeline humidity, shuttle valve sealing surface hardness, shuttle valve sealing surface thickness attenuation rate, and pipeline chloride ion concentration.

[0032] II. Constructing a Multiphysics Data Completion Model The four types of physical field data—flow rate, pressure, temperature, and humidity—in the shuttle valve pipeline exhibit spatial continuity, and the limited number of sensors deployed above cannot fully reflect the spatiotemporal distribution characteristics of these four types of physical field data. Therefore, to comprehensively characterize the spatial distribution characteristics of these four types of physical field data, this invention designs a multi-physics data completion model to supplement the data in unmonitored areas.

[0033] A multiphysics data completion model is constructed, which consists of an input layer, a hidden layer, a Gaussian kernel function for pipe flow, a Gaussian kernel function for pipe pressure, a Gaussian kernel function for temperature, a Gaussian kernel function for humidity, and an output layer.

[0034] 1. Define boundary conditions for the physics field: To ensure the accuracy of the simulation results of the multiphysics data completion model for the four physics fields, boundary conditions need to be defined for each of the four physics fields, as shown below: 1) Pipeline flow data boundary conditions include inlet boundary, wall boundary, and outlet boundary, as detailed below: in For flow velocity vector, For the location of the entrance boundary, For time, For the pipe inlet velocity that varies over time, For the normal vector boundary, For the coordinates of the pipe wall position, The speed is 0. For the pipe flow rate in S1, This indicates the location where the E+H FTW330 turbine flow meter is deployed in S1.

[0035] 2) Pipeline pressure data boundary conditions include inlet boundary, outlet boundary, and wall boundary, as detailed below: in For stress data, For the inlet pressure that changes over time, For export pressures that change over time, This is the pressure gradient vector; The pipe pressure at 5D upstream of the flow channel inlet in S1; The pipe pressure at 5D upstream of the flow channel inlet in S1. The pipeline pressure at the downstream 3D location of the outlet; and These are the positions of the Keller PA-33X differential pressure transmitter, 5D upstream of the inlet and 3D downstream of the outlet.

[0036] 3) Temperature data boundary conditions include isothermal wall, adiabatic wall, convective heat transfer wall, and inlet fluid temperature boundary conditions, as detailed below: in To maintain a constant wall temperature, For temperature data, The thermal conductivity of the fluid, The convective heat transfer coefficient is... The ambient temperature; The deployment location of the SHT30 intelligent temperature and humidity sensor in S1.

[0037] 4) Humidity data boundary conditions include adsorption wall boundary conditions and adiabatic wall boundary conditions, as detailed below: in The water vapor diffusion coefficient, Water vapor concentration, The saturated concentration of water vapor at the wall temperature; The humidity data is collected by the SHT30 intelligent temperature and humidity sensor in S1.

[0038] 2. An unsupervised alternating training strategy is used to train the multiphysics data completion model, such as... Figure 2 As shown: 1) Uniform sampling is performed along the length sampling interval of the shuttle valve pipeline to obtain... Pipe location coordinates ; 2) Use the boundary conditions of pipeline flow data as the loss function of pipeline flow Gaussian kernel function, the boundary conditions of pipeline pressure data as the loss function of pipeline pressure Gaussian kernel function, the boundary conditions of pipeline pressure data as the loss function of pipeline pressure Gaussian kernel function, the boundary conditions of temperature data as the loss function of temperature Gaussian kernel function, and the boundary conditions of humidity data as the loss function of humidity Gaussian kernel function. 3) The model is completed using multiphysics data. First, the input layer, hidden layer, pipeline flow rate Gaussian kernel function, and output layer are trained, with the parameters of the pipeline pressure, temperature, and humidity Gaussian kernel functions fixed. Once the boundary loss of the pipeline flow rate data decreases to a preset threshold, the parameters of the pipeline flow rate Gaussian kernel function are frozen. Second, the input layer, hidden layer, pipeline pressure Gaussian kernel function, and output layer are trained, with the parameters of the pipeline flow rate, temperature, and humidity Gaussian kernel functions fixed. Once the boundary loss of the pipeline pressure data decreases to a preset threshold, the parameters of the pipeline pressure Gaussian kernel function are frozen. The parameters of the Gaussian kernel function for pressure are determined. The input layer, hidden layer, temperature Gaussian kernel function, and output layer are trained again, with the parameters of the pipeline flow rate Gaussian kernel function, pipeline pressure Gaussian kernel function, and humidity Gaussian kernel function remaining unchanged. Once the temperature data boundary loss decreases to a preset threshold, the temperature Gaussian kernel function parameters are frozen. Finally, the input layer, hidden layer, humidity Gaussian kernel function, and output layer are trained again, with the parameters of the pipeline flow rate Gaussian kernel function, pipeline pressure Gaussian kernel function, and temperature Gaussian kernel function remaining unchanged. Once the humidity data boundary loss decreases to a preset threshold, the humidity Gaussian kernel function parameters are frozen. Multiphysics data completion: Gaussian distribution sampling is performed on the length sampling interval of the shuttle valve pipe to obtain... Pipe location coordinates ;Will Input the trained multiphysics data completion model to obtain completed pipeline flow rate data, completed pipeline pressure data, completed temperature data, and completed humidity data.

[0039] III. Constructing a Prediction Model for Shuttle Valve Sealing Health Index Based on Multi-Physics Field Information Fusion Model structure as follows Figure 3 As shown, it includes a multi-physics feature extraction module, a multi-physics feature enhancement module, and a prediction output module. The collected shuttle valve triaxial vibration signal data, shuttle valve sealing surface material physical parameter data, and chloride ion concentration data, as well as the completed pipeline flow data, completed pipeline pressure data, completed temperature data, and completed humidity data, are input into the multi-physics information fusion shuttle valve sealing health index prediction model to obtain the shuttle valve sealing health index prediction result.

[0040] 1. The multiphysics feature extraction module includes vibration feature extraction networks, pipeline feature extraction networks, material physical parameter feature extraction networks, and environmental condition feature extraction networks; details are as follows: 1) Input the obtained triaxial vibration signal data of the shuttle valve into the vibration feature extraction network to obtain multi-scale features of valve body vibration; the vibration feature extraction network consists of 3×3 convolutional layers, 5×5 convolutional layers and 7×7 convolutional layers, which are used to extract small-scale features, medium-scale features and large-scale features of the triaxial vibration signal of the shuttle valve, respectively, to capture the short-term, medium-term and long-term changes of valve body vibration.

[0041] 2) Input the obtained completed pipeline flow and pressure data into the pipeline feature extraction network to obtain pipeline flow time-series features and pipeline pressure time-series features. The pipeline feature extraction network includes a long short-term memory network layer and a graph convolutional layer. The long short-term memory network layer is used to capture pipeline flow time-series dependency features and pipeline pressure time-series dependency features, and the graph convolutional layer is used to extract spatial features from the pipeline flow time-series dependency features and pipeline pressure time-series dependency features to obtain high-dimensional pipeline flow features and high-dimensional pipeline pressure features containing spatiotemporal information.

[0042] 3) Input the obtained physical parameter data of the shuttle valve sealing surface material into the material physical parameter feature extraction network, which consists of a graph attention layer and a long short-term memory network layer; input the hardness data of the shuttle valve sealing surface into the graph attention convolutional layer to spatially aggregate the hardness values ​​of M sampling points to obtain the aggregated features of the hardness data of the shuttle valve sealing surface; input the thickness decay rate into the long short-term memory network layer to capture the relationship between the thickness decay rate and time to obtain the time-dependent features of the thickness decay rate.

[0043] 4) The obtained complete temperature data, complete humidity data, and chloride ion concentration data are input into the environmental condition feature extraction network to obtain environmental condition features. The environmental feature extraction network includes a forward long short-term memory network layer, an inverse long short-term memory network layer, and a spatial attention layer. The forward long short-term memory network layer is used to extract the forward time dependence features of the complete temperature data, complete humidity data, and chloride ion concentration data. The inverse long short-term memory network layer is used to extract the inverse time dependence features of the complete temperature data, complete humidity data, and chloride ion concentration data. The spatial attention mechanism layer is used to perform weighted summation on the forward and inverse time dependence features of the complete temperature data and the complete humidity data, respectively, to deeply extract the changing trends of the three features in the pipeline space and obtain the spatiotemporal features of the complete temperature data and the complete humidity data. The forward and inverse time dependence features of the chloride ion concentration data are concatenated to obtain the bidirectional time dependence features of the chloride ion concentration data.

[0044] 2. The multiphysics feature enhancement module includes a vibration signal feature enhancement network and a pipeline fluid feature enhancement network, which are used to enhance the obtained vibration signal features and pipeline flow characteristics, respectively, as follows: 1) The vibration signal feature enhancement network consists of a gated unit layer and a feature splicing layer. First, the small-scale, medium-scale, and large-scale features of the obtained triaxial vibration signal of the shuttle valve are input into the gated unit layer. The gated unit layer is used to weight the vibration signal features in the time dimension, so that the model pays more attention to the vibration signal features at the current moment and reduces the attention to the vibration signal features at historical moments, thereby obtaining enhanced small-scale, medium-scale, and large-scale features of the vibration signal. Second, the enhanced small-scale, medium-scale, and large-scale features of the vibration signal are input into the feature splicing layer to obtain multi-scale enhanced features of the triaxial vibration signal of the shuttle valve.

[0045] 2) The pipeline flow feature enhancement network consists of a spatial attention layer, a gating unit layer, and a feature splicing layer. First, the obtained high-dimensional pipeline flow features and high-dimensional pipeline pressure features are input into the spatial attention layer. The spatial attention layer aggregates the spatial distributions of the high-dimensional pipeline flow features and high-dimensional pipeline pressure features to obtain spatially weighted pipeline flow features and pipeline pressure features. Then, the spatially weighted pipeline flow features and pipeline pressure features are input into the gating unit for time-dimensional weighting and then input into the feature splicing layer to obtain pipeline flow enhancement features that characterize pipeline flow changes.

[0046] 3. The prediction output module includes a feature splicing layer, a pooling layer, a cross-modal attention fusion layer, and a softmax activation function layer. It is used to perform feature scale alignment, feature fusion, and feature activation transformation on the obtained spatiotemporal features of the completed temperature data, the completed humidity data, the bidirectional time-dependent features of the chloride ion concentration data, the aggregated features of the shuttle valve sealing surface hardness data, the time-dependent features of the thickness attenuation rate, and the obtained multi-scale enhancement features of the shuttle valve triaxial vibration signal and the pipeline flow enhancement features to obtain the shuttle valve sealing health index, as detailed below: 1) Input the spatiotemporal features of the completed temperature data, the spatiotemporal features of the completed humidity data, and the bidirectional time-dependent features of the chloride ion concentration data into the pooling layer for feature alignment, and input the aligned features into the feature splicing layer to obtain the environmental condition features; 2) The hardness data of the shuttle valve sealing surface is aggregated and the thickness decay rate time-dependent feature is input into the pooling layer for feature alignment. The aligned features are then input into the feature splicing layer to obtain the material physical parameter features. 3) Input the multi-scale enhancement features of the shuttle valve triaxial vibration signal, pipeline flow enhancement features, environmental operating condition features, and material physical parameter features into the pooling layer for feature alignment, and input the aligned features into the cross-modal attention fusion layer; the cross-modal feature fusion layer contains 4 dual attention heads, which respectively represent the interaction relationship between the four feature sequences to obtain fused features; 4) Input the fused features into the softmax activation function layer to obtain the predicted result of the shuttle valve sealing health index. The predicted result of the health index is a continuous value in the range of 0 to 1. 0 is the complete failure boundary, which means that the shuttle valve sealing performance is completely failed, and 1 is the complete sealing boundary, which means that the shuttle valve sealing performance is healthy.

[0047] IV. Obtaining the Early Warning Mode for Shuttle Valve Seal Failure The predicted results of the shuttle valve seal health index are compared with preset thresholds to obtain the shuttle valve seal failure modes, including three modes: severe failure of shuttle valve seal performance, deterioration of shuttle valve seal performance, and healthy shuttle valve seal performance. When the predicted result of the shuttle valve seal health index is less than 0.3, the shuttle valve seal performance is severely failed and requires immediate repair. When the predicted result of the shuttle valve seal health index is greater than 0.3 and less than 0.7, the shuttle valve seal performance is deteriorated and requires early warning and enhanced monitoring of the shuttle valve sealing surface. When the predicted result of the shuttle valve seal health index is greater than 0.7, the shuttle valve seal performance is healthy and no manual intervention is required.

[0048] V. Explanation of Experimental Results The experiment simulated the entire life cycle of a shuttle valve from 0 to 1000 hours on a hydraulic platform. Vibration, flow, material, and environmental data were simultaneously collected using accelerometers, flow meters, hardness testers, and temperature and humidity sensors. Physical information was input into the model, fusing multi-source features. After 1000 rounds of training, a health index degradation curve was generated. Verification at actual failure points showed that the model could accurately classify the three stages: normal, warning, and failure. Experimental results are as follows: Figure 4 As shown.

[0049] Figure 4 In the diagram, the horizontal axis represents operating time (hours), covering the entire lifespan of the shuttle valve (0-1000 hours). The vertical axis represents the sealing health index (0-1), where 1 represents complete health and 0 represents complete failure. The blue curve represents the sealing performance degradation curve under multi-field coupling (vibration, flow rate, material aging, environmental conditions), conforming to the coupling law of EPDM material fatigue and fluid erosion. The orange dashed line is the warning threshold line, triggering the yellow warning area (0.3≤health index<0.7), indicating deterioration of sealing performance. The red solid line is the failure threshold line, triggering the red failure area (health index<0.3), indicating complete loss of sealing function. The green background area indicates that the shuttle valve is completely normal, with a health index ≥0.7 and stable sealing (no abnormal vibration / flow). The yellow background area indicates a warning: 0.3≤health index<0.7, requiring enhanced monitoring (predictive maintenance initiated). The red background area indicates shuttle valve failure, with a health index <0.3, requiring maintenance. Experiments show that the model achieves three-stage management of "normal-warning-failure," and the actual failure points are all located in the warning and failure areas, verifying the effectiveness of the model.

[0050] The experiment simulated four failure scenarios in an industrial hydraulic system: seal wear, material aging, etc., collecting 50 samples for each scenario. After 5-fold cross-validation, the average accuracy rate for the four failure scenarios reached 92%. Misclassifications were concentrated in failure types with similar characteristics, validating the classification accuracy of the method. The results are as follows: Figure 5 As shown.

[0051] Figure 5 In the figure, the horizontal axis represents the predicted failure mode, and the vertical axis represents the actual failure mode, both including four typical failure types: seal wear, material aging, fluid leakage, and valve core malfunction. The thermodynamic values ​​in the figure represent the number of samples, and the diagonal values ​​(46, 48, 45, 47) show the number of correctly classified samples for the four failure types, with an average accuracy of 92%. False positives mainly occurred between seal wear and material aging (6 misclassifications) and between fluid leakage and valve core malfunction (8 misclassifications). Overall, this invention demonstrates that the method has high accuracy in classifying seal failure modes and can effectively distinguish failure types.

[0052] The experiment was conducted on the same hydraulic platform, comparing the proposed method, vibration-only analysis, and flow-only analysis. Accuracy, recall, and F1 score were evaluated through 100 failure scenario tests. The results are as follows: Figure 6 As shown.

[0053] Figure 6 In the diagram, the horizontal axis represents the three methods (this invention, vibration-only analysis, and flow-only analysis), and the vertical axis represents the three performance metrics (precision, recall, and F1 score). Experimental results show that the accuracy of this invention (0.92) > the vibration-only method (0.76) > the flow-only method (0.71), demonstrating the advantages of multi-source feature fusion. This invention also has a higher recall (0.90) and a lower false negative rate. The F1 score (overall performance) of this invention is 0.91, the best among the three methods, balancing accuracy and recall.

[0054] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0055] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for early warning of shuttle valve seal failure integrating vibration analysis and flow characteristics, characterized in that, The process includes the following: S1, Real-time acquisition of shuttle valve seal failure characteristic data, including shuttle valve triaxial vibration signal data, pipeline fluid data, material physical parameter data, and environmental condition data; the pipeline fluid data includes pipeline flow rate data and pipeline pressure data; the material physical parameter data includes shuttle valve sealing surface hardness data and shuttle valve sealing surface thickness attenuation rate data; the environmental condition data includes pipeline temperature data, pipeline humidity data, and pipeline chloride ion concentration data; S2, input the pipeline fluid data collected in S1 and the environmental condition data without pipeline chloride ion concentration data into the constructed multiphysics data completion model, perform real-time unsupervised training and learning, and generate completed pipeline flow data and environmental condition data with spatial distribution characteristics based on M pipeline position coordinates; the multiphysics data completion model defines boundary conditions for each physical field. S3 inputs the triaxial vibration signal data, material physical parameter data, pipeline chloride ion concentration data, and pipeline fluid data and other environmental condition data obtained in S1 into the constructed multi-physics information fusion shuttle valve sealing health index prediction model, and outputs the shuttle valve sealing health index prediction result. S4. Compare the predicted shuttle valve seal health index obtained in S3 with the preset threshold to obtain the shuttle valve seal failure mode.

2. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 1, characterized in that, The methods for collecting shuttle valve seal failure characteristic data in S1 include: The shuttle valve triaxial vibration signal data is obtained by using a magnetically attached triaxial accelerometer fixed to the top surface and two sides of the shuttle valve body in the orthogonal direction to form the XYZ axis, ensuring that the sensor is coupled with the valve body vibration mode, and capturing the dynamic response of the high-frequency component of the valve core opening and closing impact and the mid-frequency noise of the sealing ring friction. The material physical parameter data are selected uniformly from the shuttle valve sealing surface. Microhardness testers were deployed at several points to measure the hardness of each point. The hardness value at each point is used to obtain a comprehensive range of hardness values. The hardness data of the shuttle valve sealing surface at each sampling point were obtained; and the 3D profilometer was used to scan the shuttle valve sealing surface to obtain the surface roughness and thickness attenuation, and the thickness attenuation rate was calculated. The chloride ion concentration data of the pipeline is obtained by deploying a corrosion potentiometer in the shuttle valve body. The chloride ion concentration data characterizes the corrosion intensity inside the pipeline.

3. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 1, characterized in that: The multiphysics data completion model is used to complete four types of physical field data: pipeline flow rate, pipeline pressure, temperature, and humidity. The model consists of an input layer, a hidden layer, Gaussian kernel functions for pipeline flow rate, pipeline pressure, temperature, and humidity, and an output layer. The defined physical field boundary conditions include: Pipeline flow data boundary conditions include inlet boundary, wall boundary, and outlet boundary; Pipeline pressure data boundary conditions include inlet boundary, outlet boundary, and wall boundary; Temperature data boundary conditions include isothermal walls, adiabatic walls, convective heat transfer walls, and inlet fluid temperature boundary conditions. Humidity data boundary conditions include adsorption wall boundary conditions and adiabatic wall boundary conditions.

4. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 3, characterized in that: An unsupervised alternating training strategy is used to train a multiphysics data completion model, and includes the following processes: S21, uniform sampling is performed within the length sampling interval of the shuttle valve pipe to obtain... Pipe location coordinates ; S22, use the boundary conditions of pipeline flow data as the loss function of pipeline flow Gaussian kernel function, use the boundary conditions of pipeline pressure data as the loss function of pipeline pressure Gaussian kernel function, use the boundary conditions of temperature data as the loss function of temperature Gaussian kernel function, and use the boundary conditions of humidity data as the loss function of humidity Gaussian kernel function. S23, will The model is completed using multiphysics data. First, the input layer, hidden layer, pipeline flow rate Gaussian kernel function, and output layer are trained, with the parameters of the pipeline pressure, temperature, and humidity Gaussian kernel functions fixed. Once the boundary loss of the pipeline flow rate data decreases to a preset threshold, the parameters of the pipeline flow rate Gaussian kernel function are frozen. Second, the input layer, hidden layer, pipeline pressure Gaussian kernel function, and output layer are trained, with the parameters of the pipeline flow rate, temperature, and humidity Gaussian kernel functions fixed. Once the boundary loss of the pipeline pressure data decreases to a preset threshold, the parameters of the pipeline pressure Gaussian kernel function are frozen. The parameters of the Gaussian kernel function for pressure are determined. The input layer, hidden layer, temperature Gaussian kernel function, and output layer are trained again, with the parameters of the pipeline flow rate Gaussian kernel function, pipeline pressure Gaussian kernel function, and humidity Gaussian kernel function remaining unchanged. Once the temperature data boundary loss decreases to a preset threshold, the temperature Gaussian kernel function parameters are frozen. Finally, the input layer, hidden layer, humidity Gaussian kernel function, and output layer are trained again, with the parameters of the pipeline flow rate Gaussian kernel function, pipeline pressure Gaussian kernel function, and temperature Gaussian kernel function remaining unchanged. Once the humidity data boundary loss decreases to a preset threshold, the humidity Gaussian kernel function parameters are frozen. S24, Gaussian distribution sampling is performed on the length sampling interval of the shuttle valve pipe to obtain... Pipe location coordinates ;Will Input the trained multiphysics data completion model to obtain completed pipeline flow rate data, pipeline pressure data, temperature data, and humidity data.

5. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 1, characterized in that: The shuttle valve sealing health index prediction model includes a multi-physics feature extraction module, a multi-physics feature enhancement module, and a prediction output module. The multiphysics feature extraction module includes a vibration feature extraction network, a pipeline feature extraction network, a material physical parameter feature extraction network, and an environmental condition feature extraction network; it extracts valve body vibration signal features, pipeline fluid time sequence features, shuttle valve sealing surface hardness data features and thickness attenuation rate features, environmental spatiotemporal features, and time-dependent features, respectively. The multiphysics feature enhancement module includes a vibration signal feature enhancement network and a pipeline fluid feature enhancement network, which are used to enhance the extracted vibration signal features and pipeline fluid temporal features, respectively. The prediction output module is used to perform feature scale alignment, feature fusion, and feature activation transformation on the spatiotemporal features of the completed temperature data, the spatiotemporal features of the completed humidity data, the bidirectional time dependence features of the chloride ion concentration data, the aggregated features of the shuttle valve sealing surface hardness data, the time dependence features of the thickness attenuation rate, and the multi-scale enhancement features of the shuttle valve triaxial vibration signal and the pipeline flow enhancement features obtained by the multiphysics feature enhancement module to obtain the shuttle valve sealing health index.

6. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 5, characterized in that: The vibration feature extraction network consists of 3×3 convolutional layers, 5×5 convolutional layers, and 7×7 convolutional layers, which are used to extract small-scale, medium-scale, and large-scale features of the triaxial vibration signal of the shuttle valve, respectively, to capture the short-term, medium-term, and long-term changes in the valve body vibration. The pipeline feature extraction network includes a long short-term memory network layer and a graph convolutional layer; wherein the long short-term memory network layer is used to capture pipeline flow time-series dependency features and pipeline pressure time-series dependency features, and the graph convolutional layer is used to extract spatial features from the pipeline flow time-series dependency features and pipeline pressure time-series dependency features to obtain high-dimensional pipeline flow features and high-dimensional pipeline pressure features containing spatiotemporal information. The material physical parameter feature extraction network consists of a graph attention layer and a long short-term memory network layer. The hardness data of the shuttle valve sealing surface is input into the graph attention convolution layer to perform spatial aggregation of the hardness values ​​of M sampling points to obtain the aggregated features of the hardness data of the shuttle valve sealing surface. The thickness attenuation rate is input into the long short-term memory network layer to capture the relationship between the thickness attenuation rate and time to obtain the time-dependent features of the thickness attenuation rate. The environmental condition feature extraction network comprises a forward long short-term memory (LSTM) network layer, an inverse LSM network layer, and a spatial attention layer. The forward LSM network layer extracts the forward time dependence features of the completed temperature, humidity, and chloride ion concentration data. The inverse LSM network layer extracts the inverse time dependence features of the completed temperature, humidity, and chloride ion concentration data. The spatial attention layer performs weighted summation on the forward and inverse time dependence features of the completed temperature and humidity data, respectively, to deeply extract the variation trends of the three features in the pipeline space, thereby obtaining the spatiotemporal features of the completed temperature and humidity data. The forward and inverse time dependence features of the chloride ion concentration data are concatenated to obtain the bidirectional time dependence features of the chloride ion concentration data.

7. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 6, characterized in that: The multiphysics feature enhancement module is specifically as follows: The vibration signal feature enhancement network includes a gated unit layer and a feature splicing layer. First, the small-scale, medium-scale, and large-scale features of the obtained triaxial vibration signal of the shuttle valve are input into the gated unit layer. The gated unit layer is used to weight the vibration signal features in the time dimension, so that the model pays more attention to the vibration signal features at the current moment and reduces the attention to the vibration signal features at historical moments, thereby obtaining enhanced small-scale features, enhanced medium-scale features, and enhanced large-scale features of the vibration signal. Secondly, the small-scale features, medium-scale features, and large-scale features of the enhanced vibration signal are input into the feature splicing layer to obtain the multi-scale enhancement features of the triaxial vibration signal of the shuttle valve. The pipeline flow feature enhancement network includes a spatial attention layer, a gating unit layer, and a feature splicing layer. First, the obtained high-dimensional pipeline flow features and high-dimensional pipeline pressure features are input into the spatial attention layer. The spatial attention layer aggregates the spatial distributions of the high-dimensional pipeline flow features and high-dimensional pipeline pressure features to obtain spatially weighted pipeline flow features and pipeline pressure features. Then, the spatially weighted pipeline flow features and pipeline pressure features are input into the gating unit for time-dimensional weighting and input into the feature splicing layer to obtain pipeline flow enhancement features that characterize pipeline flow changes.

8. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 7, characterized in that: The prediction output module includes a feature concatenation layer, a pooling layer, a cross-modal attention fusion layer, and a softmax activation function layer; the specific data processing includes: The spatiotemporal features of the completed temperature data, the spatiotemporal features of the completed humidity data, and the bidirectional time-dependent features of the chloride ion concentration data are input into the pooling layer for feature alignment. The aligned features are then input into the feature splicing layer to obtain the environmental condition features. The hardness data of the valve sealing surface is aggregated and the thickness decay rate time-dependent feature is input into the pooling layer for feature alignment. The aligned features are then input into the feature splicing layer to obtain the material physical parameter features. Multi-scale enhancement features of the shuttle valve triaxial vibration signal, pipeline flow enhancement features, environmental operating condition features, and material physical parameter features are input into a pooling layer for feature alignment. The aligned features are then input into a cross-modal attention fusion layer. The cross-modal attention fusion layer contains four dual attention heads, which respectively represent the interaction relationships between the four feature sequences to obtain fused features. The fused features are input into the softmax activation function layer to obtain the prediction results of the valve sealing health index.

9. A method for early warning of shuttle valve seal failure integrating vibration analysis and flow characteristics as described in any one of claims 1 to 8, characterized in that: The health index prediction result is a continuous value in the range of 0 to 1, where 0 is the complete failure boundary, representing that the shuttle valve sealing performance has completely failed, and 1 is the complete sealing boundary, representing that the shuttle valve sealing performance is healthy.

10. The shuttle valve seal failure early warning method integrating vibration analysis and flow characteristics as described in claim 9, characterized in that, The specific process of S4 includes: The predicted shuttle valve seal health index obtained by S3 is compared with a preset threshold to obtain the shuttle valve seal failure mode, including three modes: severe failure of shuttle valve seal performance, deterioration of shuttle valve seal performance, and healthy shuttle valve seal performance. When the predicted shuttle valve seal health index is less than 0.3, it indicates severe failure of shuttle valve seal performance, which requires immediate repair. When the predicted shuttle valve seal health index is greater than 0.3 and less than 0.7, it indicates deterioration of shuttle valve seal performance, which requires early warning and enhanced monitoring of the shuttle valve sealing surface. When the predicted shuttle valve seal health index is greater than 0.7, it indicates a healthy state of shuttle valve seal performance, which does not require manual intervention.

Images