Intelligent fault prediction method for pneumatic valve island based on multi-protocol integration

Abstract

This invention discloses an intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration, belonging to the field of pneumatic valve island fault prediction technology. It includes constructing a protocol sensing layer data cube by integrating multi-protocol edge gateways to synchronously collect multi-source sensor data under the same timestamp reference, based on the heterogeneous characteristics of the valve island's physical topology and protocols. This invention achieves a paradigm shift from macroscopic pressure judgment to microscopic feedforward early warning by constructing the protocol sensing layer data cube, fusing protocol-topology joint feature maps, quantifying microscopic flow stability, predicting flow reconstruction probability, generating dynamic multimodal intervention commands, verifying protocol adaptive closed-loop, and performing digital twin health simulation. This forms a complete intelligent prediction-intervention-optimization closed loop, enabling the system to intelligently replenish air and automatically adjust throughout the process, achieving zero-blockage operation of the valve island and significantly improving system reliability and maintenance efficiency.

Description

Technical Field

[0001] This invention belongs to the field of pneumatic valve island fault prediction technology, specifically referring to a pneumatic valve island intelligent fault prediction method based on multi-protocol integration. Background Technology

[0002] As a key material conveying system in industrial fields such as thermal power generation and cement production, the pneumatic ash conveying system's core component, the pneumatic valve island, directly affects the stability and efficiency of the entire system. The valve island is responsible for controlling the conveying process of ash in the pipeline, achieving continuous material conveying by precisely adjusting the airflow and valve opening.

[0003] However, existing intelligent fault prediction methods for pneumatic valve islands still have certain shortcomings. These include data asynchrony and fragmentation caused by heterogeneous protocols, lack of deep integration between physical topology and protocol features, reliance on macroscopic pressure for fault warnings that lag behind microscopic flow state changes, static and fixed intervention commands that cannot adapt to real-time protocol requirements, lack of a closed-loop verification mechanism for intervention effects to optimize protocol weights, and missing component wear trend prediction capabilities. These core defects prevent the system from transitioning from manual unblocking after pipe blockage to microscopic feedforward early warning. Fault response is lagging, intervention efficiency is low, and maintenance relies on post-event repairs, failing to achieve the goal of fully intelligent air replenishment and zero pipe blockage operation. Therefore, an intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a pneumatic valve island intelligent fault prediction method based on multi-protocol integration, comprising the following steps: S1. Based on the physical topology and protocol heterogeneity of the valve island, multi-protocol edge gateways are integrated to synchronously collect multi-source sensor data under the same timestamp reference, and a protocol perception layer data cube is constructed. S2. Based on the protocol perception layer data cube, map the physical topology of the valve island into a graph structure and associate the node attributes with the edge weights to construct a protocol-topology joint feature graph. S3. Based on the protocol-topology joint feature map, calculate the pressure pulsation singular spectrum entropy under the fractal dimension by extracting the high-frequency pressure pulsation components along the pipeline, and quantify the microscopic flow stability. S4. Based on the quantified micro-fluidic characteristics and protocol-topology joint feature map, construct a flow regime reconfiguration probability prediction model and output the flow regime reconfiguration probability distribution; S5. Based on the fact that the probability of local aggregation in the flow regime reconstruction probability distribution exceeds a preset threshold, multimodal feedforward intervention commands are dynamically generated through the valve island topology location and protocol priority. S6. Based on the trend of micro-fluidic feature changes after the execution of multimodal feedforward intervention commands, the intervention effect of protocol adaptation is verified by reverse correction of the correlation strength weights of different protocol data in the protocol-topology joint feature graph. S7. Based on the protocol, the closed-loop verification of the intervention effect throughout the entire process data is carried out. The wear trend of key components of the valve island is simulated by mapping to the digital twin model to simulate the health status of the valve island.

[0006] Preferably, in step S1, based on the process design drawings of the pneumatic ash conveying system, the spatial positions and connection relationships of the equipment controlled by the valve island, such as the silo pumps, pipelines, valves, pressure transmitters, flow meters, intelligent air supply valves, and intelligent gradient valves, are sorted out to form a physical topology diagram that includes equipment nodes, communication links, and bus connection methods. Multi-protocol edge gateways are installed in the control cabinet or valve island box of each furnace, and the clocks of each gateway are synchronized to the same master clock source through optical fiber or dedicated synchronization cable. In the gateway, an independent data acquisition task is assigned to each protocol port. The acquisition cycle is set according to the minimum sampling frequency requirement of each sensor. The gateway adds a globally unified time tag to each data packet through a hardware timestamp unit and stores the multi-source data in a circular buffer in chronological order to form the original multi-protocol heterogeneous data stream. The data in the buffer is organized into a data cube structure according to four dimensions, including time dimension, protocol dimension, physical node ID dimension, and sensor type dimension.

[0007] Preferably, in step S2, based on the device list in the valve island physical topology, each independently addressable physical entity is defined as a graph node, a unique identifier is assigned to each node, and a mapping relationship is established with the physical node ID field in the protocol sensing layer data cube. For each node, the associated multi-protocol sensor time series data is extracted from the protocol sensing layer data cube, and the multi-dimensional features are spliced ​​to form an initial node attribute vector. Based on the valve island physical topology, the connection relationship between nodes is determined, and finally an undirected or directed graph edge set is formed, and each edge is assigned a weight. Finally, determine the edge weight of each edge. The implementation is as follows: ; In the formula, This represents the edge weight, the weight between node i and node j, with a value range of [0, 1]. It also represents the strength of the association between node i and node j. This represents the physical distance between node i and node j, with a value range of [0, ..., 0]. ], This indicates the system's maximum pipe length setting. This represents the physical distance attenuation coefficient, with a value range of [0, 2]. This represents the communication delay between node i and node j. This indicates the maximum allowable communication latency threshold of the system. Indicates the protocol coupling coefficient. This represents the correlation decay coefficient of the operating conditions. This represents the absolute value of the mutual information coefficient of the sensor time-series data between node i and node j; The node set, node attribute vector, edge set, and edge weights are combined into graph structure data, represented as follows: Where V is the set of nodes, E is the set of edges, X is the node attribute matrix, and W is the edge weight matrix, with edge weights... Let be the element value in the i-th row and j-th column of the edge weight matrix W. The initial graph structure is processed by a graph attention network for message passing and feature updating. At each layer, the node updates its own feature representation by aggregating the feature vectors of its neighbors and performing weighted fusion based on the edge weights and attention coefficients. After multiple iterations, the final feature of each node not only includes its own multi-protocol sensor information, but also integrates the features of physically adjacent nodes and protocol-related nodes, deeply fusing protocol heterogeneous data with physical topology. The updated node feature matrix is ​​used as the protocol-topology joint feature map.

[0008] Preferably, in step S3, based on the topological connection relationship of the protocol-topology joint feature map, a start point and an end point are specified for each pipeline segment to form an ordered node sequence, and the relative positions of the nodes in the pipeline are marked. Pressure time series data of the corresponding pipeline nodes are extracted from the protocol-topology joint feature map. Adaptive bandpass filtering is applied to the pressure time series to separate high-frequency pressure pulsation signals. The fractal dimension D of the extracted high-frequency pressure pulsation signals is calculated, resulting in: , In the formula, D represents the fractal dimension. For regular flow state, It is a chaotic flow state. Representing scale The minimum number of boxes required to cover the signal below. represents the scaling parameter, and log represents the logarithmic function; Based on fractal dimension, multi-scale singular spectrum analysis is performed on high-frequency pressure pulsation signals to calculate the singular spectral entropy S, which is achieved as follows: , In the formula, S represents the singular spectral entropy, and k represents the total number of singular index categories. Let p denote the singular exponent probability distribution, and p represent the probability distribution. , This represents the singularity index.

[0009] Preferably, in step S3, the singular spectral entropy is mapped to a stability index to calculate the microscopic flow stability, which is implemented as follows: , In the formula, Represents the stability of the microscopic flow regime, with values ​​ranging from [0, 1]. S represents the maximum value of the singular spectral entropy in historical data, and S represents the current singular spectral entropy.

[0010] Preferably, in step S4, a flow regime reconfiguration probability prediction model is constructed based on the quantized micro-flow regime characteristics and the protocol-topology joint feature map, and the output flow regime reconfiguration probability distribution is implemented as follows: , In the formula, Represents the probability distribution of flow regime reconfiguration. This represents the activation function. Including discrete phase stability probability Probability of local cluster formation and the probability of irreversible reconfiguration of the flow state , Represents the activation function, when S5 intervention is triggered at the time. This represents the i-th component in the flow regime reconfiguration probability distribution. This represents the weights and biases of the fully connected layer. This represents the global valve island state output by the Transformer encoder. X represents the global state feature vector of the valve island generated by global average pooling, and X represents the node feature matrix of the protocol-topology joint feature graph.

[0011] Preferably, in step S5, the probability of local aggregation in the flow regime reconstruction probability distribution is... With preset threshold In comparison, if If the risk of local aggregation is high, the intervention process is triggered; otherwise, the process is terminated. Microfluidic stability less than a preset threshold is selected from the protocol-topology joint feature map. Based on the edge weights between nodes, a high-risk propagation path is constructed: Starting from high-risk nodes, the path is extended along the edge weights that exceed a preset threshold to determine the range of influence of clustering and generate a list of high-risk topological regions. For each node in a high-risk topology region, obtain the associated protocol type and sort the high-risk nodes from high to low according to protocol priority; Generate a protocol priority ranking list, and generate a multimodal intervention instruction set based on the high-risk topology area list and the protocol priority ranking list.

[0012] Preferably, in step S6, after the intervention command is executed, new sensor data is synchronously collected through the multi-protocol edge gateway of step S1, the microfluidic stability index of the key node is recalculated, the change rate of the microfluidic stability index before and after the intervention is generated, and the data is corrected by grouping according to protocol type. The corrected protocol-topology joint feature map is input into the prediction model to generate a new flow regime reconstruction probability distribution. If the effect does not meet the expected threshold, the weight recalibration is triggered.

[0013] Preferably, in step S7, the entire process data related to the closed-loop verification of the intervention effect is continuously collected and stored, including snapshots of the protocol-topology joint feature map at each intervention trigger, time-series sequences of microfluidic features, generated intervention instruction sets, instruction execution feedback status, microfluidic recovery curves after intervention, intervention effect evaluation scores, and correction records of the association strength weight of protocol data. Key components of the valve island are recorded. For key execution components in the valve island, including intelligent air replenishment valves, intelligent gradient valves, solenoid valve coils, valve core assemblies, position feedback switches, etc., multi-physics field coupled digital twin models are established respectively. Geometric models are established based on component design parameters, wear mechanism models are embedded, and input / output interfaces corresponding to physical entities are configured to simulate the cumulative wear state under a given action sequence.

[0014] Preferably, in step S7, the collected full-process operation data is associated and mapped with the established digital twin model, and the collected intervention effect closed-loop verification full-process data is injected into the corresponding digital twin model segment by segment in chronological order for simulation.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention realizes a paradigm shift from macroscopic pressure judgment to microscopic feedforward early warning by constructing an intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration. Through the construction of a protocol perception layer data cube, fusion of protocol-topology joint feature maps, quantification of microscopic flow stability, prediction of flow reconstruction probability, generation of dynamic multimodal intervention instructions, protocol adaptive closed-loop verification, and digital twin health simulation, a complete intelligent prediction-intervention-optimization closed loop is formed, enabling the system to intelligently replenish air and automatically adjust throughout the process, achieving zero-blockage operation of the valve island, and significantly improving system reliability and operation and maintenance efficiency. 2. This invention extracts high-frequency pressure pulsation components along the pipeline to calculate the singular spectral entropy of pressure pulsation under fractal dimension, quantifies the stability of the microscopic flow regime, reflects the global irregularity of pressure pulsation through fractal dimension, and characterizes the microscopic dynamic complexity through singular spectral entropy, forming a stability index. This allows for early detection of abnormal flow regime changes, identification of weak signals before local agglomeration, and early warning of pipe blockage before it occurs. This advances the fault prediction time to the microscopic flow regime change stage, significantly improving the accuracy of prediction. 3. This invention dynamically generates multimodal feedforward intervention commands based on the probability of local agglomeration formation exceeding a threshold. Based on the probability of local agglomeration formation in the flow reconfiguration probability distribution, combined with the valve island topology and protocol priority, commands for different protocol types are dynamically generated. High-priority protocol commands are executed first to ensure timely response of key equipment. The command type is customized according to the equipment characteristics. The multimodal feedforward intervention mechanism realizes the precision and real-time nature of intervention commands, enabling the system to trigger precise intervention before pipe blockage occurs, significantly improving the timeliness and effectiveness of intervention. 4. This invention performs closed-loop verification of the intervention effect by reversing the protocol association weights based on the trend of changes in microfluidic characteristics after intervention. The microfluidic changes after the execution of the intervention command are used as feedback, and the edge weights in the joint feature graph are dynamically adjusted according to the protocol type: the weights of high-priority protocols are increased or decreased according to the effect, while the adjustment range of medium and low-priority weights is reduced accordingly. The protocol adaptive mechanism significantly improves the intervention efficiency. Attached Figure Description

[0016] Figure 1 The following is the operation flow of the intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration of the present invention. Figure 1 ; Figure 2 The following is the operation flow of the intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration of the present invention. Figure 2 ; Figure 3 The following is the operation flow of the intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration of the present invention. Figure 3 ; Figure 4 The following is the operation flow of the intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration of the present invention. Figure 4 ; Figure 5 The following is the operation flow of the intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration of the present invention. Figure 5 . Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0018] Please see Figures 1-5 As shown, the present invention provides a technical solution comprising the following steps: S1. Based on the physical topology and protocol heterogeneity of the valve island, multi-protocol edge gateways are integrated to synchronously collect multi-source sensor data under the same timestamp reference, and a protocol perception layer data cube is constructed. S2. Based on the protocol perception layer data cube, map the physical topology of the valve island into a graph structure and associate the node attributes with the edge weights to construct a protocol-topology joint feature graph. S3. Based on the protocol-topology joint feature map, calculate the pressure pulsation singular spectrum entropy under the fractal dimension by extracting the high-frequency pressure pulsation components along the pipeline, and quantify the microscopic flow stability. S4. Based on the quantified micro-fluidic characteristics and protocol-topology joint feature map, construct a flow regime reconfiguration probability prediction model and output the flow regime reconfiguration probability distribution; S5. Based on the fact that the probability of local aggregation in the flow regime reconstruction probability distribution exceeds a preset threshold, multimodal feedforward intervention commands are dynamically generated through the valve island topology location and protocol priority. S6. Based on the trend of micro-fluidic feature changes after the execution of multimodal feedforward intervention commands, the intervention effect of protocol adaptation is verified by reverse correction of the correlation strength weights of different protocol data in the protocol-topology joint feature graph. S7. Based on the protocol, the closed-loop verification of the intervention effect throughout the entire process data is carried out. The wear trend of key components of the valve island is simulated by mapping to the digital twin model to simulate the health status of the valve island.

[0019] In this embodiment, in step S1, based on the process design drawings of the pneumatic ash conveying system, the spatial positions and connection relationships of the equipment controlled by the valve island, such as the silo pump, pipeline, valve, pressure transmitter, flow meter, intelligent air supply valve, and intelligent gradient valve, are sorted out to form a physical topology diagram that includes equipment nodes, communication links, and bus connection methods, and the protocol type of each sensor and its address mapping in the bus network segment are clarified.

[0020] Specifically, a multi-protocol edge gateway is installed in the control cabinet or valve island box of each furnace. The multi-protocol edge network integrates ProfiNet master station, EtherCAT master station, IO-Link master station and PA protocol interface, and has a built-in IEEE 1588 precision time protocol or IEEE 802.1AS time-sensitive network synchronization module.

[0021] The clocks of each gateway are synchronized to the same master clock source via fiber optic cable or dedicated synchronization cable to ensure that the data acquisition timestamp deviation of all network segments is less than 1 microsecond. Specifically, the gateway assigns an independent data acquisition task to each protocol port, sets the acquisition cycle based on the minimum sampling frequency requirements of each sensor, and adds a globally unified timestamp to each data packet using a hardware timestamp unit. Multi-source data is then stored in a circular buffer in chronological order, forming the original multi-protocol heterogeneous data stream. The buffered data is organized into a data cube structure along four dimensions, including: Time dimension: arranged in a time stamp sequence, such as consecutive time points at 10ms intervals; Protocol dimension: Classified by protocol type, such as Modbus, CAN, Ethernet / IP; Physical Node ID dimension: A unique identifier for the device in the corresponding valve island physical topology; Sensor type dimension: Subdivided by sensor type, such as pressure, temperature, flow.

[0022] In this embodiment, in step S2, based on the device list in the valve island physical topology, each independently addressable physical entity, including the silo pump, intelligent air supply valve, intelligent gradient valve, pressure transmitter, flow meter, solenoid valve box, control host, etc., is defined as a graph node. A unique identifier is assigned to each node, and a mapping relationship is established with the physical node ID field in the protocol sensing layer data cube. For each node, the associated multi-protocol sensor time-series data, such as high-frequency components of pressure pulsation, valve opening feedback, solenoid valve current, communication link status, etc., are extracted from the protocol sensing layer data cube. Statistical features and time-domain features, such as mean, variance, peak value, energy spectral density, slope, zero-crossing rate, etc., are extracted for each parameter type within a fixed time window.

[0023] Specifically, multi-dimensional features are concatenated to form an initial node attribute vector, which serves as the input feature of the graph neural network. Based on the physical topology of the valve island, the connection relationships between nodes are determined, including: Physical connection edge: An edge is established between two nodes connected by physical media such as pipes, cables, and buses; Logical association edges: Edges are established between nodes belonging to the same conveying unit, such as a silo pump on a ash pipe or multiple devices controlled by the same solenoid valve box; Protocol-shared edges: Edges are established between nodes that are on the same bus segment or share the same gateway; This ultimately forms a set of undirected or directed graph edges, ensuring that each node is connected to at least one related node. Assign weights to each edge, taking into account the following factors: Physical distance: Normalized based on the actual pipe length or cable length between equipment rooms and used as a weighted component; Communication delay: Based on bus type and network segment hop count, the protocol conversion delay is calculated as a weighted component; Protocol coupling: Edge weights are higher for nodes of the same protocol type, and edge weights are appropriately reduced between nodes of heterogeneous protocols based on protocol conversion overhead; Operating condition correlation: Adjust the edge weights based on the mutual information coefficients of node parameters in historical operating data to reflect the correlation strength in actual operation; Finally, determine the edge weight of each edge. The implementation is as follows: , In the formula, This represents the edge weight, the weight between node i and node j, with a value range of [0, 1]. It also represents the strength of the association between node i and node j. This represents the physical distance between node i and node j, with a value range of [0, ..., 0]. ], This indicates the system's maximum pipe length setting. This represents the physical distance attenuation coefficient, with a value range of [0, 2]. This represents the communication delay between node i and node j. This indicates the maximum allowable communication latency threshold of the system. This represents the protocol coupling coefficient. For the same protocol, the value ranges from 1.0; for different protocols, the value ranges from [0.5, 0.8]. This represents the correlation decay coefficient of the operating conditions. This represents the absolute value of the mutual information coefficient of the sensor time series data between node i and node j, with a value range of [0, 1]. Specifically, the node set, node attribute vector, edge set, and edge weights are combined into graph structure data, represented as follows: Where V is the set of nodes, E is the set of edges, X is the node attribute matrix, and W is the edge weight matrix, with edge weights... Let be the value of the element in the i-th row and j-th column of the edge weight matrix W.

[0024] The initial graph structure is message-passing and feature-updating through a graph attention network. At each layer, nodes update their own feature representation by aggregating the feature vectors of their neighbors and performing weighted fusion based on edge weights and attention coefficients. After multiple iterations, the final feature of each node not only includes its own multi-protocol sensor information but also incorporates the features of physically adjacent nodes and protocol-related nodes, thereby deeply fusing heterogeneous protocol data with physical topology. The updated node feature matrix is ​​then used as the protocol-topology joint feature map.

[0025] In this embodiment, in step S3, based on the topological connection relationship of the protocol-topology joint feature graph, the continuous node sequence of the key pipeline segment of the valve island is determined, such as the main conveying pipeline chain from the material inlet valve to the outlet valve. The start point and end point are specified for each pipeline segment to form an ordered node sequence, and the relative position of the nodes in the pipeline is marked.

[0026] Specifically, pressure time series data of corresponding pipeline nodes are extracted from the protocol-topology joint feature map. Adaptive bandpass filtering is applied to the pressure time series, such as a center frequency of 10Hz and a bandwidth of 5Hz-20Hz, dynamically adjusted based on the valve island system characteristics, to separate high-frequency pressure pulsation signals. The fractal dimension D of the extracted high-frequency pressure pulsation signals is calculated, resulting in: , In the formula, D represents the fractal dimension. For regular flow state, It is a chaotic flow state. Representing scale The minimum number of boxes required to cover the signal below. represents the scaling parameter, and log represents the logarithmic function; Based on fractal dimension, multi-scale singular spectrum analysis is performed on high-frequency pressure pulsation signals to calculate the singular spectral entropy S, which is achieved as follows: , In the formula, S represents the singular spectral entropy, and k represents the total number of singular index categories. Let p denote the singular exponent probability distribution, and p represent the probability distribution. , Represents the singularity index, local scaling behavior, and the verification condition is: If the verification fails, such as if the error is >5%, the singular spectrum is recalculated. The larger the S value, the more unstable the microfluidic state is. For example, S > 1.0 indicates high instability.

[0027] In this embodiment, in step S3, the singular spectral entropy is mapped to a stability index to calculate the microscopic flow stability, which is implemented as follows: , In the formula, Represents the stability of the microscopic flow regime, with values ​​ranging from [0, 1]. S represents the maximum value of the singular spectral entropy in historical data, and S represents the current singular spectral entropy.

[0028] In this embodiment, in step S4, a flow regime reconfiguration probability prediction model is constructed based on the quantized micro-flow regime characteristics and the protocol-topology joint feature map, and the output flow regime reconfiguration probability distribution is implemented as follows: , In the formula, Represents the probability distribution of flow regime reconfiguration. This represents the activation function. Including discrete phase stability probability Probability of local cluster formation and the probability of irreversible reconfiguration of the flow state , Represents the activation function, when S5 intervention is triggered at the time. This represents the i-th component in the flow regime reconfiguration probability distribution. This represents the weights and biases of the fully connected layer. This represents the global valve island state output by the Transformer encoder. Let X represent the global state feature vector of the valve island generated by global average pooling, and let X represent the node feature matrix of the protocol-topology joint feature graph. for N represents the total number of valve island nodes. This represents the feature vector of the i-th node in the Transformer output.

[0029] In this embodiment, in step S5, the probability of local aggregation in the flow regime reconstruction probability distribution is... With preset threshold In comparison, if If the risk of local aggregation is high, the intervention process is triggered; otherwise, the process is terminated, and the current valve island state is maintained.

[0030] Screening for microfluidic stability values ​​below a preset threshold from the protocol-topology joint feature map. Based on the edge weights between nodes, a high-risk propagation path is constructed: Starting from high-risk nodes, the influence range of cluster formation is determined by extending along pathways whose edge weights exceed a preset threshold. Generate a list of high-risk topology regions, such as pipe segment A, valve group B, and flow meter C; For each node in a high-risk topology region, obtain the associated protocol type and sort the high-risk nodes from highest to lowest protocol priority: High-priority protocols, such as Ethernet / IP: Device response latency <10ms, priority is given to allocating intervention resources.

[0031] Medium-priority protocols, such as CAN, have a response delay of 10–50 ms and are assigned as secondary protocols.

[0032] Low-priority protocols, such as Modbus RTU, have a response latency >50ms and only fill in the gaps after high / medium priority instructions are executed. Generate a protocol priority sorting list.

[0033] Specifically, the multimodal intervention instruction set generated based on the high-risk topology area list and protocol priority ranking list includes: instruction type, dynamic logic, and instruction structure; Instruction type definition: Valve adjustment command: For valve nodes, the parameters are based on the local agglomeration intensity. For example, the higher the probability of local agglomeration formation, the greater the opening adjustment range. Airflow pulse command: For air source equipment, such as frequency adjustment, the parameters are based on the pressure pulsation component of the pipeline section; Protocol adaptation instructions: These are customized for different protocol types. For example, Ethernet / IP instructions include precise timestamps, while Modbus instructions are simplified to addresses and values.

[0034] Dynamic generation logic: For each high-risk node, generate instructions according to protocol priority: High-priority nodes, such as Ethernet / IP, generate high-precision instructions, such as increasing the valve opening of pipeline segment A by 15%, with timestamp = current time + 5ms. Medium-priority nodes, such as CAN, generate medium-precision commands, such as increasing the airflow frequency of valve group B to 25Hz. Low-priority nodes, such as Modbus, generate low-precision commands, such as gas pressure compensation of pipeline segment C +3kPa. Multimodal integration: Instructions are grouped according to protocol priority to form an execution sequence, with high-priority instructions sent first to ensure real-time performance; Instruction structure output: Command ID: A unique identifier; Target location, such as pipeline segment A - valve group B; Protocol type; Specific parameters, such as opening percentage, frequency (Hz), and pressure (kPa); Time constraints, such as execution must be completed within 10ms.

[0035] In this embodiment, in step S6, after the intervention command is executed, new sensor data is synchronously collected through the multi-protocol edge gateway of S1, focusing on high-frequency pressure pulsation signals along the pipeline, recalculating the microfluidic stability index of key nodes, and generating the rate of change of the microfluidic stability index before and after the intervention. The rate of change of the microfluidic stability index is... ,in, This represents the microfluidic stability index after the intervention is implemented. This represents the microfluidic stability index before intervention, with trends marked and grouped and corrected according to protocol type: High-priority protocol: if Increase the weight of the associated edges in the protocol; if If the condition is ineffective or worsens, reduce its weight. Medium priority protocol: Correction range halved; Low priority protocols: only Adjust the weights only when necessary, otherwise maintain the original values; The corrected protocol-topology joint feature map is input into the prediction model to generate a new flow regime reconstruction probability distribution. If the effect does not meet the expected threshold, the weight recalibration is triggered, the protocol weight adjustment history is recorded, and the protocol priority rule table is updated.

[0036] In this embodiment, in step S7, data related to the closed-loop verification of the intervention effect are continuously collected and stored, including snapshots of the protocol-topology joint feature map at each intervention trigger, time-series sequences of microfluidic features, generated intervention instruction sets, instruction execution feedback status, microfluidic recovery curves after intervention, correction records of intervention effect evaluation scores and protocol data association strength weights, and accumulated operational data of key components of the valve island, such as the number of times the intelligent air replenishment valve core moves, the adjustment frequency of the intelligent gradient valve, the energizing duration and current waveform of the solenoid valve coil, and the number of times the valve position switch moves. For key execution components in the valve island, including the intelligent air replenishment valve, intelligent gradient valve, solenoid valve coil, valve core assembly, and position feedback switch, multi-physics field coupled digital twin models are established respectively.

[0037] Specifically, a geometric model is established based on the component design parameters, such as valve core stroke, coil resistance, spring stiffness, and sealing material. Wear mechanism models, such as contact wear, fatigue accumulation, and thermal wear, are embedded in the model. Input and output interfaces corresponding to the physical entity are configured. Each digital twin model can receive the component's historical action records and operating parameters to simulate the cumulative wear state under a given action sequence.

[0038] In this embodiment, in step S7, the collected full-process operation data is associated and mapped with the established digital twin model. The mapping rules are as follows: each physical component's unique identifier corresponds to a digital twin instance; each action record of the component corresponds to an input event of the twin model, such as valve opening / closing, gas volume adjustment, and pulse output; the component's operating parameters correspond to the twin model's boundary conditions, such as gas source pressure, ambient temperature, and medium characteristics; and the component's real-time status feedback, such as valve position signals and current monitoring values, is used to compare and verify with the simulated status output by the twin model.

[0039] Specifically, the collected data from the closed-loop verification of the intervention effect are injected into the corresponding digital twin model in chronological order for simulation. The simulation process adopts an event-driven mechanism: whenever a component action record is injected, the twin model calls the wear mechanism model to calculate the incremental microscopic damage caused to the component by the action based on the working parameters of the action and the cumulative number of actions, and accumulates the damage into the current wear state variable. For cases of successful intervention and failed intervention, different load characteristics are marked, such as the pulse impact load borne by the valve core when the intervention is successful and the abnormal pressure continuously borne by the valve when the intervention fails, so that the twin model can distinguish the wear mode under different working conditions.

[0040] The digital twin model generates wear trend simulation curves for each key component based on the injected complete historical action sequence. The horizontal axis of the curve represents time, and the vertical axis represents wear quantification indicators, such as wear depth of valve core sealing surface, insulation aging index of electromagnetic coil, and fatigue accumulation coefficient of spring. The curve simultaneously displays the simulated wear accumulation path and the wear status at the current moment, and compares it with the component's design life threshold to indicate the proportion of life that has been consumed.

[0041] Based on the wear trend simulation curves of each component, a comprehensive simulation assessment report of the overall health status of the valve island is generated. The report includes: the current wear level classification of each key component, the remaining life simulation prediction, the wear rate analysis, and the contribution analysis of historical intervention actions to component wear. By comparing the wear simulation results under different intervention strategies, the impact of the intervention actions themselves on the component life is evaluated.

[0042] Working Principle: Real-time synchronous acquisition of cross-protocol data is achieved by deploying a multi-protocol edge gateway. This gateway integrates multiple industrial protocol interfaces and has a built-in high-precision time synchronization module. Based on the valve island physical topology design, the gateway is configured with an independent acquisition task for each protocol port, dynamically adjusts the acquisition cycle according to the sensor sampling frequency, and adds a global time tag to each data packet through a hardware timestamp unit. After the acquired data stream passes through a circular buffer, it is structured into a data cube according to four dimensions: time series, protocol type, physical node identifier, and sensor type. The valve island physical devices are mapped as graph structure nodes, and each node carries multi-dimensional characteristics: physical attributes include device spatial coordinates and topological location, protocol... The attributes cover communication type and address mapping, while dynamic attributes integrate sensor timing features such as pressure pulsation and valve opening. The connection relationship between nodes is constructed based on three types of topological logic: physical connection, logical association, and protocol sharing. The weight of each edge is determined by dynamic calculation: physical distance attenuation reflects the spatial relationship of equipment, communication delay quantifies protocol conversion overhead, protocol coupling degree distinguishes the interaction strength of homogeneous / heterogeneous protocols, and operating condition correlation is dynamically adjusted based on historical data mutual information. The graph neural network uses a multi-layer message passing mechanism to enable node features to be automatically weighted and fused when aggregating neighbor node information, so that the final feature of each node includes its own protocol characteristics and integrates the flow state information of physically adjacent and protocol-associated nodes. By focusing on the microscopic dynamic analysis of high-frequency pressure pulsation signals along the pipeline, the pressure time series of key nodes are extracted from the protocol-topology joint feature map. Adaptive bandpass filtering is used to separate high-frequency components related to flow anomalies. The fractal dimension is calculated using the box-counting method to quantify the global irregularity of pressure pulsations: regular flow states correspond to low fractal dimensions, while chaotic flow states correspond to high fractal dimensions. Based on this, multi-scale singular spectrum analysis is performed to extract the singularity index distribution and calculate the singular spectrum entropy. The entropy value directly characterizes the dynamic complexity of the microscopic flow state; the higher the entropy value, the more unstable the flow state. The verification process ensures that the fractal dimension and the weighted integral of the singular spectrum are consistent. The microscopic quantification method can capture weak signals formed by local aggregation in advance. Fault early warning has progressed from macroscopic pressure discrimination to microscopic dynamic feature analysis. A flow reconfiguration probability prediction model based on Transformer is constructed. The quantified microscopic flow stability features and protocol-topology joint feature map are input into the model. The global topological dependencies between nodes are captured by the Transformer encoder, and a global representation vector of the valve island state is generated by global average pooling. After concatenating the global representation vector with the microscopic stability features, a three-state probability distribution is output through a fully connected layer: discrete phase stability probability, local agglomeration formation probability, and irreversible flow reconfiguration probability. The output mechanism of the probability distribution enables the system to accurately quantify the flow state, and triggers intervention when the local agglomeration formation probability exceeds the threshold. A local clustering probability threshold-triggered intervention mechanism based on the probability distribution of flow regime reconstruction is used. High-risk nodes with low micro-stability are screened from the protocol-topology joint feature map. The clustering impact range is determined by extending along high-weighted paths, generating a list of high-risk topology regions. For high-risk region nodes, they are dynamically sorted according to protocol priority: high-priority protocol devices respond quickly and are assigned high-precision commands; medium-priority protocols are assigned medium-precision commands; and low-priority protocols are assigned low-precision commands. Command generation logic is customized according to device type: valve adjustment commands dynamically adjust opening based on clustering intensity; airflow pulse commands set frequency based on pressure pulsation components; and protocol adaptation commands optimize command format according to protocol characteristics. Multimodal commands are grouped by priority to form an execution sequence, ensuring high-priority commands are executed first. After the intervention commands are executed, new data is synchronously collected through multi-protocol edge gateways to calculate the rate of change in micro-flow regime stability, serving as a quantitative indicator of the intervention effect. Based on the rate of change, the edge weights in the protocol-topology joint feature map are dynamically adjusted according to protocol type: the weights of high-priority protocols increase based on the stability improvement, and the effect is further enhanced by adjusting the edge weights. If the effect is poor, the weight is reduced; the weight adjustment of medium-priority protocols is halved; low-priority protocols are only fine-tuned when the effect is significant; the corrected feature map is input into the prediction model to generate a new probability distribution to verify whether the intervention effect meets the standard; if the effect does not meet expectations, the weight recalibration is triggered and historical adjustment data is recorded, and the protocol priority rule table is updated; the data of the entire process of closed-loop verification of the intervention effect is mapped to the digital twin model of key components; a multi-physics coupling model is established for each component, a geometric model is constructed based on design parameters and wear mechanism is embedded; the data mapping follows an event-driven mechanism: component action records are used as input events, operating parameters are used as boundary conditions, and real-time status feedback is used for model verification; the model calculates the micro-damage increment based on the action sequence and operating parameters, and accumulates to generate a wear trend curve; the curve shows the change path of quantitative indicators such as wear depth and aging index over time, and compares it with the design life threshold to indicate the life consumption ratio; the health status report integrates the component wear degree, remaining life, wear rate and intervention contribution, realizing the transformation from post-blockage maintenance to wear trend prediction, ensuring that the health status of the valve island is controllable throughout the entire cycle.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

[0044] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for intelligent fault prediction of pneumatic valve islands based on multi-protocol integration, characterized in that, Includes the following steps: S1. Based on the physical topology and protocol heterogeneity of the valve island, multi-protocol edge gateways are integrated to synchronously collect multi-source sensor data under the same timestamp reference, and a protocol perception layer data cube is constructed. S2. Based on the protocol perception layer data cube, map the physical topology of the valve island into a graph structure and associate the node attributes with the edge weights to construct a protocol-topology joint feature graph. S3. Based on the protocol-topology joint feature map, calculate the pressure pulsation singular spectrum entropy under the fractal dimension by extracting the high-frequency pressure pulsation components along the pipeline, and quantify the microscopic flow stability. S4. Based on the quantified micro-fluidic characteristics and protocol-topology joint feature map, construct a flow regime reconfiguration probability prediction model and output the flow regime reconfiguration probability distribution; S5. Based on the fact that the probability of local aggregation in the flow regime reconstruction probability distribution exceeds a preset threshold, multimodal feedforward intervention commands are dynamically generated through the valve island topology location and protocol priority. S6. Based on the trend of micro-fluidic feature changes after the execution of multimodal feedforward intervention commands, the intervention effect of protocol adaptation is verified by reverse correction of the correlation strength weights of different protocol data in the protocol-topology joint feature graph. S7. Based on the protocol, the closed-loop verification of the intervention effect throughout the entire process data is carried out. The wear trend of key components of the valve island is simulated by mapping to the digital twin model to simulate the health status of the valve island.

2. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In S1, based on the process design drawings of the pneumatic ash conveying system, the spatial positions and connection relationships of the equipment controlled by the valve island are sorted out to form a physical topology diagram including equipment nodes, communication links and bus connection methods. Multi-protocol edge gateways are installed in the valve island box of each furnace, and the clocks of each gateway are synchronized to the same master clock source through optical fiber. In the gateway, an independent data acquisition task is assigned to each protocol port. The gateway uses a hardware timestamp unit to tag each data packet with a globally unified timestamp and stores the multi-source data in a circular buffer in chronological order to form the original multi-protocol heterogeneous data stream. The data in the buffer is organized into a data cube structure according to four dimensions, including time dimension, protocol dimension, physical node ID dimension, and sensor type dimension.

3. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In step S2, based on the device list in the valve island physical topology, each independently addressable physical entity is defined as a graph node. A unique identifier is assigned to each node, and a mapping relationship is established with the physical node ID field in the protocol sensing layer data cube. For each node, the associated multi-protocol sensor time series data is extracted from the protocol sensing layer data cube, and the multi-dimensional features are spliced ​​to form an initial node attribute vector. Based on the valve island physical topology, the connection relationship between nodes is determined, and finally an undirected graph edge set is formed. Each edge is assigned a weight, and finally the edge weight of each edge is determined. The node set, node attribute vector, edge set, and edge weights are combined into a graph structure data, represented as follows: Where V is the set of nodes, E is the set of edges, X is the node attribute matrix, and W is the edge weight matrix; the initial graph structure is processed by a graph attention network for message passing and feature updating, and the updated node feature matrix is ​​used as the protocol-topology joint feature graph.

4. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In step S3, based on the topological connection relationship of the protocol-topology joint feature map, a start point and an end point are specified for each pipeline segment to form an ordered node sequence, and the relative positions of the nodes in the pipeline are marked. Pressure time series data of the corresponding pipeline nodes are extracted from the protocol-topology joint feature map, and adaptive bandpass filtering is applied to the pressure time series to separate high-frequency pressure pulsation signals. The fractal dimension of the extracted high-frequency pressure pulsation signals is calculated. Based on fractal dimension, multi-scale singular spectrum analysis is performed on high-frequency pressure pulsation signals to calculate the singular spectral entropy S, which is achieved as follows: , In the formula, S represents the singular spectral entropy, and k represents the total number of singular index categories. Let p represent the singular exponent probability distribution, and p denote the probability distribution. , represents the singularity index, and log represents the logarithmic function.

5. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 4, characterized in that: In step S3, the singular spectral entropy is mapped to a stability index to calculate the microscopic flow stability, which is implemented as follows: , In the formula, Indicates the stability of the microscopic flow regime. S represents the maximum value of the singular spectral entropy in historical data, and S represents the current singular spectral entropy.

6. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In step S4, a flow regime reconfiguration probability prediction model is constructed based on the quantized microscopic flow regime characteristics and the protocol-topology joint feature map, and the output flow regime reconfiguration probability distribution is implemented as follows: , In the formula, Represents the probability distribution of flow regime reconfiguration. This represents the activation function. These represent the weights and biases of the fully connected layer, respectively. X represents the global valve island state output by the Transformer encoder, and X represents the node feature matrix of the protocol-topology joint feature map.

7. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In step S5, the probability of local aggregation formation in the flow regime reconstruction probability distribution is... With preset threshold In comparison, if If the risk of local clustering is high, the intervention process is triggered. Otherwise, the process will terminate; Nodes with microfluidic stability less than a preset threshold are selected from the protocol-topology joint feature graph, and high-risk propagation paths are constructed based on the edge weights between nodes; For each node in a high-risk topology region, obtain the associated protocol type, sort the high-risk nodes from high to low according to protocol priority, generate a protocol priority sorting list, and generate a multimodal intervention instruction set based on the high-risk topology region list and the protocol priority sorting list.

8. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In step S6, after the intervention command is executed, new sensor data is collected synchronously through the multi-protocol edge gateway, the microfluidic stability index of key nodes is recalculated, the change rate of microfluidic stability index before and after intervention is generated, and correction is performed by grouping according to protocol type; the corrected protocol-topology joint feature map is input into the prediction model to generate a new fluid state reconstruction probability distribution; if the effect does not reach the threshold expectation, weight recalibration is triggered.

9. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 1, characterized in that: In S7, data related to the closed-loop verification of the intervention effect are continuously collected and stored, including snapshots of the protocol-topology joint feature map at each intervention trigger, time-series sequences of microfluidic features, generated intervention instruction sets, instruction execution feedback status, microfluidic recovery curves after intervention, intervention effect evaluation scores, and correction records of protocol data association strength weights. Digital twin models of multi-physics coupling are established, geometric models are established based on component design parameters, wear mechanism models are embedded, and input / output interfaces corresponding to physical entities are configured to simulate the cumulative wear state under a given action sequence.

10. The intelligent fault prediction method for pneumatic valve islands based on multi-protocol integration according to claim 9, characterized in that: In step S7, the collected full-process operation data is associated and mapped with the established digital twin model, and the collected intervention effect closed-loop verification full-process data is injected into the corresponding digital twin model segment by segment in chronological order for simulation.

Images