A virtual-real mapping digital twin facility fault prediction and health management system

Abstract

The present application relates to the technical field of industrial automation control and facility health management, and particularly relates to a virtual-real mapping digital twin facility fault prediction and health management system, comprising: a state mirror module that obtains target facility operation data and maps to a virtual-real mapping digital twin model to obtain a current facility state mirror; a boundary construction module that determines a critical fault boundary containing a boundary state; a facility topology construction module that constructs a facility control topology graph; a reverse deduction module that reversely deduces a candidate reachable path from the boundary state to the current facility state mirror; a forward rechecking module that generates a virtual disturbance sequence and screens an effective reachable path; a path determination module that determines a minimum fault triggering path; and a health management output module that generates a defense shaping strategy. The present application can improve the reliability of fault path prediction and the pertinence of health management strategies.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control and facility health management technology, and in particular to a facility fault prediction and health management system based on a virtual-real mapping digital twin. Background Technology

[0002] Fault prediction and health management of industrial facilities typically rely on sensor data acquisition, operational status monitoring, threshold alarms, historical fault sample training, and digital twin simulation. For example, one existing digital twin-based equipment monitoring solution first collects equipment operating status, condition data, and environmental condition data, then constructs a digital twin of the equipment, and judges fault risk based on monitoring data or adaptive thresholds. Another digital twin-based health status prediction solution typically uses a digital twin model to simulate different operating conditions or fault modes, generates simulation data to expand the training sample, and then combines health indices or risk indicators to output the equipment health status. These solutions can achieve equipment status monitoring and fault early warning to a certain extent, but their judgment logic mostly starts from the current operating status to make positive predictions for the future, or identifies the current state based on the similarity of historical fault samples.

[0003] In the existing solutions described above, the forward prediction process typically requires enumerating a large number of possible evolution paths starting from the current facility state, and the causal path between the fault boundary and the current state is unclear. When the target facility includes multiple devices, sensors, actuators, control variables, and process connections, the same abnormal boundary may be caused by multiple paths. For example, actuator lag, equipment degradation, control variable offset, sensor drift, or load fluctuations may all lead to similar boundary states. Relying solely on forward simulation, health indices, or fault probabilities can easily lead to two types of problems: one is that low-probability but low-cost dangerous paths are missed, and the other is that paths that may appear in the model but are difficult to reproduce in real facilities are misclassified as high-risk paths. As a result, although the fault prediction results provide a risk value, it is difficult to explain from which facility node the risk originates, along which state influence relationship it propagates, or which control variable amplifies it.

[0004] Therefore, the core problem of existing technologies is that in the virtual-real mapping digital twin scenario, there is a lack of a path screening mechanism that can first trace candidate paths backward from the critical fault boundary, and then verify the real reachability of candidate paths from the current facility status mirror. This makes it easy for fault prediction results to remain at the level of risk scoring or status alarms, making it difficult to determine the minimum fault triggering path that is truly likely to push the target facility to the critical fault boundary, and also difficult to generate health management strategies that can improve the triggering cost for this path. Summary of the Invention

[0005] This invention relates to the field of industrial automation control and facility health management technology, and proposes a facility fault prediction and health management system based on a virtual-real mapping digital twin. The system aims to solve the problems of existing facility fault prediction processes that mainly rely on positive prediction of the current state, which easily leads to omission of fault paths, false path misjudgment, and lack of path-specificity in health management strategies.

[0006] To address the aforementioned technical problems, this invention provides a facility fault prediction and health management system based on a virtual-real mapping digital twin, comprising a state mirroring module, a boundary construction module, a facility topology construction module, a reverse inference module, a forward verification module, a path determination module, and a health management output module.

[0007] The state mirroring module is used to collect operational data from the target facility and map this data to a virtual-physical mapping digital twin model to obtain a current facility state mirror. The operational data may include sensor measurements, control variables, actuator feedback values, process parameters, alarm status, and maintenance status within the target facility. By mapping this operational data according to facility nodes, the virtual-physical mapping digital twin model can obtain a digital state foundation corresponding to the current operational state of the target facility.

[0008] The boundary construction module is used to determine critical fault boundaries, including boundary states, based on preset safety constraints, process constraints, control constraints, and maintenance constraints. These boundary states can include hard boundary states, soft boundary states, or a combination of both. Hard boundary states can be formed when pressure, temperature, flow rate, current, or vibration reaches corresponding preset critical conditions; soft boundary states can be formed when control margins are insufficient, maintenance windows are unavailable, backup paths are unavailable, or production cycle time is limited. Therefore, critical fault boundaries reflect both safety and process limitations during facility operation, as well as the impact of maintenance accessibility and control margins on fault risk.

[0009] The facility topology construction module is used to construct a facility control topology diagram based on the equipment connection relationships, control logic relationships, and measurement point layout relationships of the target facility. The facility control topology diagram includes facility nodes and topology edges. Facility nodes include equipment nodes, sensor nodes, actuator nodes, control variable nodes, and process connection nodes. Topology edges are used to characterize the state influence relationships between facility nodes. The topology edges can be configured with forward propagation direction, backward deduction direction, propagation delay range, response gain, and topology edge confidence level, enabling the facility control topology diagram to be used to constrain the backward deduction and forward verification processes of fault paths.

[0010] The reverse deduction module is used to reverse-deduct from the boundary state to the current facility state under the constraints of the facility control topology graph, obtaining candidate reachable paths. Specifically, the reverse deduction module can start from the facility node that forms the boundary state, search for upstream facility nodes along the reverse deduction direction of the topology edge, and filter the searched paths by combining the propagation delay range, response gain, and topology edge confidence. Paths whose propagation delay falls within the propagation delay range, whose node response direction is consistent with the response direction corresponding to the response gain, whose node state difference falls within a preset difference range, and whose topology edge confidence meets the preset confidence condition can be used as candidate reachable paths.

[0011] The forward verification module generates a virtual disturbance sequence based on the candidate reachable paths and performs forward inference starting from the current facility state mirror image to select effective reachable paths from the candidate reachable paths that can reach the critical fault boundary. The forward verification module can extract the arrangement order of facility nodes, the node state differences between adjacent facility nodes, the propagation timing of topological edges, and the changes and directions of control variables from the candidate reachable paths, and generate a virtual disturbance sequence based on the extraction results. The virtual disturbance sequence may include at least one of sensor drift disturbance, actuator hysteresis disturbance, control bias disturbance, load fluctuation disturbance, equipment degradation disturbance, and topological edge weakening disturbance. After the virtual disturbance sequence is applied to the virtual-real mapping digital twin model according to the propagation timing corresponding to the candidate reachable paths, the forward verification module determines whether the forward inference result reaches the critical fault boundary within a preset prediction time window, and whether the facility node response order, control variable change direction, and reached boundary state during the forward inference process are consistent with the candidate reachable paths; when the above conditions are met, the corresponding candidate reachable path is determined as an effective reachable path.

[0012] The path determination module is used to calculate the path evaluation value of each effective reachable path based on the triggering cost, topology propagation complexity, and actual facility reachability deviation, and to determine the effective reachable path with the smallest path evaluation value as the minimum fault triggering path. Preferably, the path evaluation value can be calculated in the following manner: ,in, Indicates a valid reachable path Path evaluation value, Indicates a valid reachable path The trigger cost required to reach the critical failure boundary from the current facility state mirror image Indicates a valid reachable path Topology propagation complexity in facility control topology diagrams Indicates a valid reachable path Positive reproducibility in positive retesting The value range is from 0 to 1. Indicates a valid reachable path Actual facility accessibility deviation, , , These represent the weighting coefficients corresponding to the triggering cost, topology propagation complexity, and actual facility reachability deviation, respectively. The path determination module determines the effective reachable path with the smallest path evaluation value as the minimum fault triggering path.

[0013] The health management output module is used to determine the fault entry node, propagation relay node, and key control variables based on the minimum fault triggering path, and to generate a defense shaping strategy to improve the trigger cost corresponding to the minimum fault triggering path. The health management output module can perform trigger cost sensitivity analysis on the facility nodes in the minimum fault triggering path, and adjust the state of the facility nodes in the minimum fault triggering path in the virtual-real mapping digital twin model, recalculating the trigger cost corresponding to the minimum fault triggering path. The facility node located at the beginning of the minimum fault triggering path and causing the largest increase in trigger cost can be identified as the fault entry node; the facility node located between the fault entry node and the facility node forming the boundary state can be identified as the propagation relay node; the control variable corresponding to the control variable node causing the largest increase in trigger cost can be identified as the key control variable.

[0014] Furthermore, the health management output module can generate candidate defense shaping strategies based on the fault entry node, propagation relay node, and key control variables, and input each candidate defense shaping strategy into the virtual-real mapping digital twin model for re-verification and path evaluation value calculation. When a candidate defense shaping strategy increases the trigger cost corresponding to the minimum fault triggering path and does not generate a new effective reachable path with a smaller path evaluation value, the candidate defense shaping strategy is determined as the defense shaping strategy. The defense shaping strategy may include at least one of the following: adjusting control setpoints, limiting load, increasing sampling frequency, calibrating sensors, performing early maintenance, enabling backup paths, or adjusting the maintenance window.

[0015] Furthermore, the facility fault prediction and health management system based on a virtual-real mapping digital twin can also include a closed-loop update module. This module records the actual anomaly path based on the facility node response sequence, control variable change direction, and reached boundary state when an anomaly occurs during the target facility's operation, and compares the actual anomaly path with the minimum fault triggering path. When the actual anomaly path matches the minimum fault triggering path, the positive reproducibility of the effective reachable path forming the minimum fault triggering path is improved, or the topological edge reliability of the corresponding topological edge in the minimum fault triggering path is improved. When the actual anomaly path does not match the minimum fault triggering path, the propagation delay range, response gain, and topological edge reliability in the facility control topology diagram are corrected, or the generation parameters of the virtual disturbance sequence are corrected.

[0016] The technical effects and advantages of this invention are as follows: This invention maps the target facility's operational data to a current facility state mirror using a state mirroring module, determines critical fault boundaries including boundary states using a boundary construction module, and constructs a facility control topology graph using a facility topology construction module. This ensures that the current state of the target facility, the fault boundary, and the state influence relationships between facility nodes all have a predictable basis. Based on this, a reverse deduction module, starting from the boundary state, deduces candidate reachable paths from the current facility state mirror under the constraints of the facility control topology graph. A forward verification module then generates a virtual disturbance sequence based on the candidate reachable paths and performs forward deduction from the current facility state mirror to select effective reachable paths that can reach the critical fault boundary. Therefore, this invention transforms fault prediction from a one-way forward prediction to a two-way path confirmation process of "boundary reverse deduction + forward verification," reducing false paths that only exist in the digital twin model but are difficult to reproduce in the real facility, and lowering the probability of missing low-trigger-cost dangerous paths. This invention further utilizes a path determination module to calculate the path evaluation value of each effective reachable path based on trigger cost, topology propagation complexity, and actual facility reachability deviation. The effective reachable path with the lowest evaluation value is then identified as the minimum fault-triggered path, transforming fault prediction results from a single risk value into interpretable path outcomes. The health management output module determines the fault entry node, propagation relay node, and key control variables based on the minimum fault-triggered path and generates a defensive shaping strategy to increase the trigger cost corresponding to the minimum fault-triggered path. Thus, the health management strategy can directly target the path segments most prone to triggering faults, such as adjusting key control variables, proactively maintaining fault entry nodes, increasing the sampling frequency of relevant nodes, or activating backup paths. This increases the trigger cost required for the target facility to reach the critical fault boundary, improving fault prediction reliability, the targeted nature of maintenance decisions, and the facility's resilience. Attached Figure Description

[0017] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts: Figure 1 A flowchart for fault prediction and health management of digital twin facilities that map virtual and real data; Figure 2 A flowchart for generating candidate reachable paths and performing positive verification. Detailed Implementation

[0018] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0019] Reference Figure 1-2 As shown, this embodiment provides a facility fault prediction and health management system based on a virtual-real mapping digital twin, applicable to target facilities with equipment connection relationships, control logic relationships, and measurement point layout relationships. For ease of explanation, this embodiment uses an industrial circulating cooling facility as the target facility. The industrial circulating cooling facility includes a circulating pump, regulating valve, heat exchanger, cooling tower, circulating pipeline, pressure sensor, flow sensor, temperature sensor, current sensor, vibration sensor, programmable controller, and upper-level monitoring equipment. The circulating pump drives the circulation of the cooling medium, the regulating valve regulates the flow rate of the cooling medium in the circulating pipeline, the heat exchanger facilitates heat exchange between the cooling medium and the cooled medium, the cooling tower reduces the temperature of the cooling medium, and the programmable controller outputs pump frequency setpoints, regulating valve opening setpoints, and cooling tower fan start / stop commands. The above-described industrial circulating cooling facility is only one specific implementation scenario; compressed air stations, chemical reaction facilities, packaging production lines, and CNC machining units with similar control links, equipment connection links, and measurement point layout links can also be configured according to this embodiment.

[0020] The virtual-real mapping digital twin-based facility fault prediction and health management system includes a state mirroring module, a boundary construction module, a facility topology construction module, a reverse inference module, a forward verification module, a path determination module, a health management output module, and a closed-loop update module. The state mirroring module, boundary construction module, and facility topology construction module are all connected to the virtual-real mapping digital twin model. The reverse inference module calls the boundary states output by the boundary construction module and the facility control topology diagram output by the facility topology construction module. The forward verification module calls the current facility state mirror output by the state mirroring module and the candidate reachable paths output by the reverse inference module. The path determination module calls the effective reachable paths output by the forward verification module. The health management output module calls the minimum fault triggering path output by the path determination module. The closed-loop update module calls the actual operational feedback of the target facility and updates the parameters used by the path determination module, facility topology construction module, and forward verification module.

[0021] The virtual-real mapping digital twin model includes a node state layer, a topology relationship layer, a control logic layer, a virtual disturbance execution layer, a state inference layer, and a feedback update layer. The node state layer stores the real-time state, historical state, and boundary state markers of each facility node in the target facility; the topology relationship layer stores the facility control topology diagram and the forward propagation direction, backward inference direction, propagation delay range, response gain, and topology edge reliability of the topology edges; the control logic layer stores the control correspondence between control variable nodes and equipment nodes, and actuator nodes, such as the control correspondence between pump frequency setpoints and circulating pump nodes, and the control correspondence between regulating valve opening setpoints and regulating valve nodes; the virtual disturbance execution layer receives virtual disturbance sequences and changes the corresponding facility node states, control variable states, or topology edge parameters according to the propagation timing in the virtual disturbance sequence; the state inference layer calculates the state changes of each facility node within a preset prediction time window based on the current facility state mirror, facility control topology diagram, control logic relationship, virtual disturbance sequence, and topology edge parameters; and the feedback update layer updates the forward reproducibility, topology edge reliability, propagation delay range, response gain, and virtual disturbance sequence generation parameters based on the comparison results between the actual abnormal path and the minimum fault triggering path.

[0022] The status mirroring module collects operational data from the target facility. This data includes pump frequency setpoint, actual pump frequency, pump current, pump vibration value, pump outlet pressure, regulating valve opening setpoint, regulating valve opening feedback value, circulating pipeline flow rate, heat exchanger inlet temperature, heat exchanger outlet temperature, cooling tower fan status, alarm status, maintenance window status, standby path status, and production cycle time status. The status mirroring module performs unified time alignment on data from programmable logic controllers (PLCs), sensors, actuators, and maintenance management equipment. Control variable data is recorded according to the control command output time, sensor measurement data is recorded according to the sampling time, actuator feedback data is recorded according to the feedback arrival time, and maintenance status data is recorded according to the maintenance status update time. The status mirroring module aggregates operational data within the same time window according to facility nodes and maps it to the node status layer of the virtual-physical mapping digital twin model to obtain the current facility status mirror.

[0023] The current facility state mirror represents a set of node-based operational states of the target facility at the current moment or within the current time window. Taking an industrial circulating cooling facility as an example, the current facility state mirror includes the states of the circulating pump node, regulating valve node, heat exchanger node, cooling tower node, pressure sensor node, flow sensor node, temperature sensor node, current sensor node, vibration sensor node, pump frequency control variable node, and regulating valve opening control variable node. The circulating pump node state includes the actual pump frequency, pump current, pump vibration value, and pump outlet pressure; the regulating valve node state includes the regulating valve opening setpoint, regulating valve opening feedback value, and opening response time; the heat exchanger node state includes the inlet temperature, outlet temperature, and heat exchange temperature difference; and the cooling tower node state includes the fan start / stop status and cooling medium return water temperature. After the state mirror module completes the above mapping, the virtual-real mapping digital twin model can use the current facility state mirror as the current state basis for subsequent reverse inference and forward verification.

[0024] The boundary construction module determines the critical fault boundary, including boundary states, based on preset safety constraints, process constraints, control constraints, and maintenance constraints. Preset safety constraints include upper limits for pump current, pipeline pressure, vibration alarm threshold, and safe shutdown conditions. Preset process constraints include upper limits for heat exchanger outlet temperature, lower limits for circulating pipeline flow rate, and allowable range for heat exchanger temperature difference. Preset control constraints include upper limits for adjustable pump frequency, upper limits for adjustable regulating valve opening, lower limits for control margin, and duration of control deviation. Preset maintenance constraints include the availability of maintenance windows, the availability of backup paths, the availability of spare parts, and the allowable adjustment of production cycle time.

[0025] The boundary construction module determines hard boundary states based on preset safety and process constraints, and soft boundary states based on preset control and maintenance constraints. Hard boundary states include facility states that occur when pressure, temperature, flow rate, current, or vibration reaches corresponding preset critical conditions. Soft boundary states include facility operation or maintenance states that occur when control margins are insufficient, maintenance windows are unavailable, backup paths are unavailable, or production cycle time is limited. The boundary construction module determines at least one hard boundary state, at least one soft boundary state, or a combination of both as a boundary state. Critical fault boundaries consist of multiple boundary states, representing a set of states that require key control before the target facility enters an unacceptable operating state.

[0026] In one specific setup, the boundary states determined by the boundary construction module include: the heat exchanger outlet temperature reaches 80°C and the circulation pipeline flow rate is less than 70% of the set flow rate; the pump current reaches 120% of the rated current and the pump vibration value reaches the preset vibration alarm threshold; the regulating valve opening reaches more than 95%, the circulation pipeline flow rate is still lower than the required value, and the maintenance window is unavailable for the next two hours. The first boundary state is formed by temperature and flow rate, the second boundary state by current and vibration, and the third boundary state by insufficient control margin and unavailable maintenance window. The boundary construction module writes these boundary states into the node state layer and boundary state record table of the virtual-real mapping digital twin model, serving as the backtracking starting point for the reverse inference module.

[0027] The facility topology construction module constructs a facility control topology diagram based on the equipment connection relationships, control logic relationships, and measurement point layout relationships of the target facility. The module identifies equipment nodes, sensor nodes, actuator nodes, control variable nodes, and process connection nodes within the target facility as facility nodes. Equipment nodes include circulating pump nodes, heat exchanger nodes, and cooling tower nodes; sensor nodes include pressure sensor nodes, flow sensor nodes, temperature sensor nodes, current sensor nodes, and vibration sensor nodes; actuator nodes include regulating valve nodes and cooling tower fan nodes; control variable nodes include pump frequency control variable nodes, regulating valve opening control variable nodes, and fan start / stop control variable nodes; process connection nodes include circulating pipeline nodes and heat exchanger inlet / outlet connection nodes.

[0028] The facility topology construction module establishes topological edges between facility nodes with state-related influence relationships. Pump frequency control variable nodes affect circulating pump nodes; circulating pump nodes affect pressure sensor nodes, flow sensor nodes, current sensor nodes, and vibration sensor nodes; regulating valve opening control variable nodes affect regulating valve nodes; regulating valve nodes affect flow sensor nodes; flow sensor nodes and heat exchanger inlet temperature nodes jointly affect heat exchanger outlet temperature sensor nodes; and cooling tower fan nodes affect heat exchanger inlet temperature nodes. The facility topology construction module writes these state-related influence relationships into the topology relationship layer of the virtual-real mapping digital twin model, forming the facility control topology diagram.

[0029] The facility topology construction module configures the forward propagation direction, backward propagation direction, propagation delay range, response gain, and topology edge confidence for each topology edge. The forward propagation direction represents the direction in which the state influence is transmitted in the real target facility; for example, a pump frequency control variable node points to a circulating pump node, and a circulating pump node points to a flow sensor node. The backward propagation direction constrains the direction in which the backward propagation module mirrors the current facility state. The backward propagation direction is set according to the reverse direction of the forward propagation direction or a traceable causal direction. The propagation delay range represents the allowable time range between a change in the state of an upstream facility node and a response from a downstream facility node; for example, a change in the opening of a control valve may result in a flow response within 1 to 8 seconds, and a change in the frequency of a circulating pump may result in a pump current response within 1 to 5 seconds. The response gain represents the proportion and direction of the influence of a change in an upstream facility node on a change in a downstream facility node; for example, a decrease in the opening of a control valve usually results in a decrease in flow, and an increase in the pump frequency usually results in an increase in flow. Topological edge credibility represents the degree of credibility of the corresponding topological edge used for inference. Topological edge credibility can be determined by the amount of historical running data, historical response consistency rate and manual verification results, and the value range can be set to 0 to 1.

[0030] The backpropagation module starts from the boundary state and performs a mirrored backpropagation to the current facility state under the constraints of the facility control topology, obtaining candidate reachable paths. The backpropagation module first identifies the facility nodes that form the boundary state. For example, if the boundary state is that the heat exchanger outlet temperature reaches 80°C and the circulation pipeline flow rate is less than 70% of the set flow rate, the facility nodes forming this boundary state include the heat exchanger outlet temperature sensor node and the flow sensor node. Starting from the facility nodes that form the boundary state, the backpropagation module searches for upstream facility nodes along the backpropagation direction. For abnormal heat exchanger outlet temperatures, the backpropagation module searches along the backpropagation direction and finds the flow sensor node, heat exchanger inlet temperature node, cooling tower fan node, and regulating valve node. For insufficient circulation pipeline flow, the backpropagation module searches along the backpropagation direction and finds the regulating valve node, circulation pump node, and pump frequency control variable node.

[0031] The reverse engineering module filters the searched paths based on propagation delay range, response gain, and topology edge reliability. Specifically, the module reads the current facility state mirror and historical time window data prior to the current facility state mirror to determine if the propagation delay falls within the propagation delay range. For example, if the valve opening feedback hysteresis occurs before the flow rate decreases, and the time difference between the two is within the propagation delay range of the topology edge corresponding to the valve node to the flow sensor node, then the propagation delay corresponding to that topology edge meets the requirements. The reverse engineering module determines whether the node response direction is consistent with the response gain. For example, if the response gain indicates that a decrease in valve opening leads to a decrease in flow rate, and the current facility state mirror and historical time window data show a decrease in valve opening and a decrease in flow rate, then the node response directions are consistent. The reverse engineering module determines whether the node state difference falls within a preset difference range. For example, if the difference between the flow rate decrease and the valve opening change does not exceed the preset difference range, then the node state difference meets the requirements. The reverse engineering module determines whether the topology edge reliability meets a preset reliability condition, such as a topology edge reliability of not less than 0.6. Paths that meet the requirements for propagation delay, node response direction, node state difference, and topological edge credibility are identified as candidate reachable paths.

[0032] Candidate reachable paths represent paths with interpretable relationships on the facility control topology when backtracking from the boundary state to the current facility state mirror. For example, the backtracking module can obtain candidate reachable path one: regulating valve node, flow sensor node, heat exchanger outlet temperature sensor node; candidate reachable path two: circulating pump node, flow sensor node, heat exchanger outlet temperature sensor node; candidate reachable path three: cooling tower fan node, heat exchanger inlet temperature node, heat exchanger outlet temperature sensor node. All of these candidate reachable paths originate from the backtracking of the facility control topology and are filtered based on propagation delay, response direction, node state differences, and topology edge reliability.

[0033] The forward verification module extracts the arrangement order of facility nodes, the differences in node states between adjacent facility nodes, the propagation timing of topological edges, and the changes and directions of control variables from the candidate reachable path, and generates a virtual perturbation sequence based on the extraction results. The arrangement order of facility nodes represents the sequence in which facility nodes evolve from their current state to the boundary state in the candidate reachable path. The differences in node states between adjacent facility nodes represent the corresponding differences in state values ​​between upstream and downstream facility nodes. The propagation timing of topological edges represents the time sequence and time interval of each node response. The changes and directions of control variables represent the changes that control variable nodes should undergo in the candidate reachable path; for example, the control variable node for regulating valve opening decreases by 5%, and the control variable node for pump frequency increases by 3Hz.

[0034] The forward verification module generates a virtual disturbance sequence based on the extracted results. The virtual disturbance sequence includes at least one of the following: sensor drift disturbance, actuator hysteresis disturbance, control bias disturbance, load fluctuation disturbance, equipment degradation disturbance, and topology edge weakening disturbance. Sensor drift disturbance is used to simulate the deviation of sensor output values ​​from the true state in the virtual-real mapping digital twin model. Actuator hysteresis disturbance is used to simulate the lag of actuator feedback relative to control commands. Control bias disturbance is used to simulate the deviation of control variables from the target control value. Load fluctuation disturbance is used to simulate changes in external operating load. Equipment degradation disturbance is used to simulate a decrease in equipment performance. Topology edge weakening disturbance is used to simulate a weakening of the state influence relationship between facility nodes. The virtual disturbance sequence is applied to the virtual disturbance execution layer of the virtual-real mapping digital twin model according to the propagation timing corresponding to the candidate reachable paths, and then the state inference layer calculates the state changes of the facility nodes within a preset prediction time window.

[0035] The forward verification module starts with the current facility state mirror image and performs forward inference, filtering out the effective reachable paths from the candidate reachable paths to the critical fault boundary. The preset prediction time window can be set according to the target facility's operational inertia, process response speed, and maintenance response time, for example, set to the next 30 minutes, the next 2 hours, or the next shift. Taking an industrial circulating cooling facility as an example, the preset prediction time window can be set to the next 2 hours. After applying the virtual disturbance sequence to the virtual-real mapping digital twin model, the forward verification module determines whether the forward inference result reaches the critical fault boundary within the preset prediction time window. If the state inference layer calculates that the heat exchanger outlet temperature reaches the boundary state, or the circulating pipeline flow rate is lower than the corresponding lower limit of the boundary state, then the forward inference result reaches the critical fault boundary.

[0036] The forward verification module also determines whether the response sequence of facility nodes, the direction of change of control variables, and the boundary states reached during the forward simulation are consistent with the arrangement sequence of facility nodes, the direction of change of control variables, and the boundary states in the candidate reachable path, respectively. For example, if the arrangement sequence of facility nodes in a candidate reachable path is a control valve node, a flow sensor node, and a heat exchanger outlet temperature sensor node, the direction of change of control variables is a decrease in the control valve opening, and the boundary states are that the heat exchanger outlet temperature reaches the upper limit and the flow rate is below the lower limit, the forward verification module applies a hysteresis disturbance to the control valve node. If the forward simulation shows that the control valve node first exhibits an abnormal response, followed by a decrease in flow rate at the flow sensor node, and finally the heat exchanger outlet temperature sensor node reaches the upper temperature limit, and the direction of change of the control valve opening is consistent with the direction of change of control variables in the candidate reachable path, then the forward verification module determines this candidate reachable path as a valid reachable path. If the forward simulation result reaches the critical fault boundary, but the response sequence of facility nodes, the direction of change of control variables, or the boundary states reached are inconsistent with the candidate reachable path, then the forward verification module does not determine this candidate reachable path as a valid reachable path.

[0037] When no valid reachable path is selected within the preset prediction time window, the forward verification module outputs a result indicating no valid reachable path. The system maintains routine status monitoring and re-executes the following processes in the next sampling cycle: data acquisition, current facility status mirror generation, critical fault boundary construction, candidate reachable path reverse deduction, and forward verification.

[0038] The path determination module calculates the path evaluation value of each effective reachable path based on trigger cost, topology propagation complexity, and actual facility reachability deviation, and determines the effective reachable path with the lowest path evaluation value as the minimum fault triggering path. Trigger cost represents the disturbance cost required for an effective reachable path to reach the critical fault boundary from the current facility state mirror image. Trigger cost can be obtained by weighting the normalized disturbance intensity, disturbance duration, and control variable change of each virtual disturbance in the virtual disturbance sequence. The lower the disturbance intensity, the shorter the duration, and the smaller the control variable change, the lower the trigger cost. Topology propagation complexity represents the propagation complexity of the effective reachable path in the facility control topology graph. Topology propagation complexity can be obtained by weighting the number of topology edges traversed by the effective reachable path, the total propagation delay, the number of low-confidence topology edges, and the inverse of the topology edge confidence. The fewer nodes the path traverses, the shorter the propagation delay, and the higher the topology edge confidence, the lower the topology propagation complexity. Forward reproducibility represents the degree of reproducibility of the effective reachable path in forward verification. Positive reproducibility can be determined by the consistency of facility node response sequence, the consistency of control variable change direction, the consistency of reached boundary states, and arrival time deviation. Actual facility accessibility deviation is determined by... This indicates the degree of deviation of the effective reachable path from the actual operational patterns of the facility. When the response sequence of the facility nodes, the direction of change of the control variables, and the boundary states reached are all consistent with the candidate reachable path, and the arrival time deviation is small, the positive reproducibility is high and the deviation of the actual facility reachability is low.

[0039] The path determination module calculates the path evaluation value for each valid reachable path as follows: ,in, Indicates a valid reachable path Path evaluation value, Indicates a valid reachable path The trigger cost required to reach the critical failure boundary from the current facility state mirror image Indicates a valid reachable path Topology propagation complexity in facility control topology diagrams Indicates a valid reachable path Positive reproducibility in positive retesting The value ranges from 0 to 1, with a larger value indicating a higher degree of positive reproducibility. Indicates a valid reachable path Actual facility accessibility deviation, , , Let α, β, and γ represent the weighting coefficients corresponding to the triggering cost, topology propagation complexity, and actual facility reachability deviation, respectively. α, β, and γ are non-negative weighting coefficients, and α + β + γ = 1. During implementation, , and Normalization to the range of 0 to 1 ensures that data with different dimensions can be assigned the same evaluation value. The path determination module identifies the effective reachable path with the lowest evaluation value as the minimum fault-triggered path.

[0040] The health management output module determines the fault entry node, propagation relay node, and key control variables based on the minimum fault triggering path, and generates a defensive shaping strategy to improve the triggering cost corresponding to the minimum fault triggering path. The health management output module performs triggering cost sensitivity analysis on the facility nodes in the minimum fault triggering path, including control variable nodes. In the virtual-real mapping digital twin model, the health management output module adjusts the state of the facility nodes in the minimum fault triggering path and recalculates the corresponding triggering cost. For example, the health management output module reduces the hysteresis time of the regulating valve node by half and recalculates the triggering cost; reduces the equipment degradation disturbance intensity of the circulating pump node and recalculates the triggering cost; lowers the upper limit of the pump frequency control variable node and recalculates the triggering cost; and adjusts the control strategy of the regulating valve opening control variable node and recalculates the triggering cost.

[0041] The health management output module designates the facility node located at the beginning of the minimum fault triggering path as a candidate entry node. Among the candidate entry nodes, the facility node that causes the largest increase in triggering cost corresponding to the minimum fault triggering path is identified as the fault entry node. The health management output module also designates the facility node located between the fault entry node and the facility node forming the boundary state as a propagation relay node. Furthermore, the health management output module identifies the control variable corresponding to the control variable node that causes the largest increase in triggering cost corresponding to the minimum fault triggering path as a critical control variable. Taking the minimum fault triggering path formed by the regulating valve node, flow sensor node, and heat exchanger outlet temperature sensor node as an example, the regulating valve node is located at the beginning of the path. Adjusting the regulating valve node's state results in the largest increase in triggering cost, therefore the regulating valve node is the fault entry node. The flow sensor node is located between the regulating valve node and the heat exchanger outlet temperature sensor node, therefore the flow sensor node is the propagation relay node. Finally, adjusting the regulating valve opening setpoint corresponding to the regulating valve opening control variable node results in the largest increase in triggering cost, therefore the regulating valve opening setpoint is the critical control variable.

[0042] The health management output module generates candidate defense shaping strategies based on the fault entry node, propagation relay node, and key control variables. These candidate strategies include at least one of the following: adjusting control setpoints, limiting load, increasing sampling frequency, calibrating sensors, early maintenance, enabling alternative paths, or adjusting the maintenance window. The health management output module inputs each candidate defense shaping strategy into the virtual-physical mapping digital twin model for re-verification and path evaluation value calculation. If a candidate defense shaping strategy increases the trigger cost corresponding to the minimum fault triggering path but does not generate a new effective reachable path with a smaller path evaluation value, the health management output module determines that candidate defense shaping strategy as the defense shaping strategy. For example, after the health management output module inputs the early maintenance control valve as a candidate defense shaping strategy into the virtual-physical mapping digital twin model, if the forward verification shows that the trigger cost corresponding to the original minimum fault triggering path has increased, and the path determination module does not find a new effective reachable path with a smaller path evaluation value after recalculation, then the early maintenance control valve is determined as the defense shaping strategy. If lowering the upper limit of pump frequency increases the trigger cost corresponding to the original minimum fault triggering path, but generates a new effective reachable path caused by insufficient cooling tower fan capacity, and the path evaluation value of the new effective reachable path is smaller, then lowering the upper limit of pump frequency will not be determined as a defensive shaping strategy.

[0043] The closed-loop update module records the actual anomaly path based on the response sequence of facility nodes, the direction of control variable changes, and the boundary states reached when anomalies occur during the operation of the target facility. It then compares the actual anomaly path with the minimum fault triggering path. During actual operation, if the control valve opening feedback first shows hysteresis, followed by a decrease in the circulation pipeline flow, and then the heat exchanger outlet temperature reaches the boundary state, the closed-loop update module records the actual anomaly path as the control valve node, flow sensor node, and heat exchanger outlet temperature sensor node. The closed-loop update module then compares the actual anomaly path with the minimum fault triggering path determined by the path determination module.

[0044] When the actual anomaly path coincides with the minimum fault triggering path, the closed-loop update module increases the positive reproducibility of the effective reachable path that forms the minimum fault triggering path, or increases the topological edge confidence of the corresponding topological edge in the minimum fault triggering path. The coincidence of the actual anomaly path with the minimum fault triggering path indicates that the minimum fault triggering path obtained from the aforementioned back-dive, forward verification, and path evaluation matches the anomaly evolution process of the actual target facility. Therefore, the closed-loop update module increases the positive reproducibility of this effective reachable path, or increases the topological edge confidence of the relevant topological edge in this path. When the actual anomaly path does not coincide with the minimum fault triggering path, the closed-loop update module corrects the propagation delay range, response gain, and topological edge confidence in the facility control topology diagram, or corrects the generation parameters of the virtual disturbance sequence. For example, if the actual abnormal path shows that the delay from the hysteresis of the control valve to the decrease in flow is longer than the original propagation delay range, the closed-loop update module will expand the propagation delay range from the control valve node to the flow sensor node; if the actual abnormal path shows that a certain topological edge cannot explain the actual propagation relationship for a long time, the closed-loop update module will reduce the credibility of that topological edge; if the actual abnormal path shows that the intensity of the actuator hysteresis disturbance does not match the actual value, the closed-loop update module will correct the generation parameters of the actuator hysteresis disturbance.

[0045] Through the above implementation methods, the facility fault prediction and health management system based on a virtual-real mapping digital twin can, before a serious fault occurs in the target facility, use the boundary state within the critical fault boundary as the backtracking target, reverse-engineer candidate reachable paths along the facility control topology map, and then perform forward verification through virtual disturbance sequences to screen out effective reachable paths that can actually reach the critical fault boundary. The path determination module calculates path evaluation values ​​using trigger cost, topology propagation complexity, and deviation from real facility reachability, and can determine the minimum fault triggering path that most needs prevention and control under the current facility state mirror from the effective reachable paths. The health management output module determines the fault entry node, propagation relay node, and key control variables based on the minimum fault triggering path, and generates a defense shaping strategy that can improve the trigger cost corresponding to the minimum fault triggering path. The closed-loop update module corrects the forward reproducibility, topology edge credibility, propagation delay range, response gain, and virtual disturbance sequence generation parameters based on the real abnormal path, making subsequent reverse-engineering, forward verification, and path evaluation more consistent with the actual operating rules of the target facility. This implementation can reduce path misjudgment caused by unidirectional prediction, improve the interpretability of fault prediction results, and enable health management strategies to directly apply to key links in the least fault-triggered path.

[0046] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A facility fault prediction and health management system based on a virtual-real mapping digital twin, characterized in that, include: The status mirroring module is used to collect operational data of the target facility and map it to a virtual-physical mapping digital twin model to obtain the current status mirror of the facility; The boundary construction module is used to determine the critical fault boundary, which includes the boundary state, based on preset safety constraints, process constraints, control constraints, and maintenance constraints; the facility topology construction module is used to construct the facility control topology diagram based on the equipment connection relationship, control logic relationship, and measurement point layout relationship of the target facility. The reverse deduction module is used to mirror and reverse deduct from the current facility state, starting from the boundary state, under the constraints of the facility control topology graph, to obtain candidate reachable paths. The forward verification module is used to generate a virtual disturbance sequence based on the candidate reachable paths, and perform forward deduction starting from the current facility state image to select the effective reachable path that can reach the critical fault boundary from the candidate reachable paths; The path determination module is used to calculate the path evaluation value of each effective reachable path based on the trigger cost, topology propagation complexity and actual facility reachability deviation, and to determine the effective reachable path with the smallest path evaluation value as the minimum fault triggering path; The health management output module is used to determine the fault entry node, propagation relay node and key control variables based on the minimum fault triggering path, and to generate a defense shaping strategy to improve the triggering cost corresponding to the minimum fault triggering path.

2. The facility failure prediction and health management system according to claim 1, characterized in that: The path determination module is used to calculate the path evaluation value of each of the effective reachable paths in the following manner: ,in, Indicates a valid reachable path Path evaluation value, Indicates a valid reachable path The trigger cost required to reach the critical failure boundary from the current facility state mirror image Indicates a valid reachable path Topology propagation complexity in facility control topology diagram Indicates a valid reachable path Positive reproducibility in positive retesting The value range is from 0 to 1. Indicates a valid reachable path Actual facility accessibility deviation, , , These represent the weighting coefficients corresponding to the triggering cost, topology propagation complexity, and actual facility reachability deviation, respectively. The path determination module determines the effective reachable path with the smallest path evaluation value as the minimum fault triggering path.

3. The facility failure prediction and health management system according to claim 1, characterized in that, The facility topology construction module is used to identify equipment nodes, sensor nodes, actuator nodes, control variable nodes, and process connection nodes in the target facility as facility nodes, and to establish topological edges between facility nodes with state influence relationships. The topological edges are configured with a forward propagation direction, a backward inference direction, a propagation delay range, a response gain, and a topological edge confidence level. The backward inference direction is used to constrain the backward inference module to trace back from the boundary state to the current facility state.

4. The facility failure prediction and health management system according to claim 3, characterized in that, The reverse deduction module is used to search for upstream facility nodes along the reverse deduction direction, starting from the facility node that forms the boundary state, and to filter the searched paths according to the propagation delay range, the response gain, and the topology edge confidence; wherein, the paths whose propagation delay falls within the propagation delay range, whose node response direction is consistent with the response direction corresponding to the response gain, and whose node state difference falls within a preset difference range are determined as candidate reachable paths.

5. The facility failure prediction and health management system according to claim 4, characterized in that, The forward verification module is used to extract the arrangement order of facility nodes, the node state differences between adjacent facility nodes, the propagation time sequence corresponding to the topology edge, and the change amount and direction of the control variable from the candidate reachable path, and generate a virtual disturbance sequence based on the extraction results; the virtual disturbance sequence includes at least one of sensor drift disturbance, actuator hysteresis disturbance, control bias disturbance, load fluctuation disturbance, equipment degradation disturbance, and topology edge weakening disturbance, and the virtual disturbance sequence is applied to the virtual-real mapping digital twin model according to the propagation time sequence corresponding to the candidate reachable path.

6. The facility failure prediction and health management system according to claim 5, characterized in that, The forward verification module is used to determine whether the forward simulation result reaches the critical fault boundary within a preset prediction time window after the virtual disturbance sequence is applied to the virtual-real mapping digital twin model, and to determine whether the facility node response order, control variable change direction, and reached boundary state during the forward simulation process are consistent with the facility node arrangement order, control variable change direction, and boundary state in the candidate reachable path, respectively; when the forward simulation result reaches the critical fault boundary and the above judgment results are consistent, the corresponding candidate reachable path is determined as a valid reachable path.

7. The facility failure prediction and health management system according to claim 1, characterized in that, The boundary construction module is used to determine hard boundary states based on preset safety constraints and process constraints, and to determine soft boundary states based on preset control constraints and maintenance constraints. It determines at least one hard boundary state, at least one soft boundary state, or a combination of both as the boundary state. The hard boundary state includes the facility state formed when pressure, temperature, flow rate, current, or vibration reaches the corresponding preset critical conditions. The soft boundary state includes the facility operation or maintenance state formed when control margin is insufficient, maintenance window is unavailable, backup path is unavailable, or production cycle is limited.

8. The facility failure prediction and health management system according to claim 6, characterized in that, The health management output module is used to perform trigger cost sensitivity analysis on facility nodes in the minimum fault triggering path, and the facility nodes include control variable nodes. In the virtual-real mapping digital twin model, the state of the facility nodes in the minimum fault triggering path is adjusted respectively, and the triggering cost corresponding to the minimum fault triggering path is recalculated; The facility node located at the beginning of the minimum fault triggering path along the propagation direction from the current facility state to the critical fault boundary is selected as a candidate entry node. Among the candidate entry nodes, the facility node that causes the largest increase in the triggering cost corresponding to the minimum fault triggering path is determined as the fault entry node. The facility node located between the fault entry node and the facility node that forms the boundary state is determined as the propagation relay node. The control variable corresponding to the control variable node that causes the largest increase in the triggering cost corresponding to the minimum fault triggering path is determined as the critical control variable.

9. The facility failure prediction and health management system according to claim 8, characterized in that, The health management output module is used to generate candidate defense shaping strategies based on the fault entry node, propagation relay node, and key control variables. Each candidate defense shaping strategy is input into the virtual-real mapping digital twin model to re-verify and recalculate the path evaluation value. When a candidate defense shaping strategy increases the trigger cost corresponding to the minimum fault triggering path and does not generate a new effective reachable path with a smaller path evaluation value, the candidate defense shaping strategy is determined as the defense shaping strategy.

10. The facility failure prediction and health management system according to claim 9, characterized in that, It also includes a closed-loop update module, which is used to record the actual abnormal path based on the response sequence of facility nodes, the direction of change of control variables and the boundary state reached when an anomaly occurs during the operation of the target facility, and compare the actual abnormal path with the minimum fault triggering path; when the actual abnormal path is consistent with the minimum fault triggering path, the positive reproducibility of the effective reachable path that forms the minimum fault triggering path is improved, or the topological edge credibility of the corresponding topological edge in the minimum fault triggering path is improved; When the actual abnormal path is inconsistent with the minimum fault triggering path, the propagation delay range, response gain, and topology edge reliability in the facility control topology diagram are corrected, or the generation parameters of the virtual disturbance sequence are corrected.

Images