A method and system for fault diagnosis and early warning of distributed control systems

Abstract

This application relates to the field of distributed control system technology, and provides a method and system for fault diagnosis and early warning of distributed control systems. The method involves deploying multiple sensor arrays at key nodes of the distributed control system to collect mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data in real time, and comparing these data with preset safety thresholds and fault modes in a historical fault case library. When data anomalies occur, the system uses graph structure modeling to trace the vibration propagation path of adjacent nodes, locates the fault source by the direction of amplitude difference, and matches the real-time generated topology network of undetermined faults with a predefined reference fault topology network. The propagation characteristics of the vibration waveform are used to infer the fault origin, and intelligent diagnosis is achieved by combining this with a historical fault mode library. Dynamic analysis of fault trends is realized, and the accuracy of fault identification is improved through real-time data comparison and topology matching.

Description

Technical Field

[0001] This application relates to the field of distributed control systems, and more specifically, to a method and system for fault diagnosis and early warning of distributed control systems. Background Technology

[0002] The content in this section provides only background information related to this application and may not constitute prior art.

[0003] Fault diagnosis and early warning in distributed control systems involves analyzing equipment operating parameters acquired by the data acquisition and monitoring control system using industrial monitoring and scheduling software to identify abnormal patterns and pinpoint fault sources. Utilizing multi-node collaborative diagnostic technology, it rapidly diagnoses distributed faults such as sensor failures and communication interruptions. Simultaneously, it achieves early warning through threshold analysis and trend prediction, pushing risk information to the dispatch center. This mechanism significantly improves the reliability and operational efficiency of industrial equipment.

[0004] In the prior art, Chinese Patent CN119937502A discloses a distributed fault diagnosis method and device for a large-scale industrial closed-loop control process. This patented distributed fault diagnosis method for closed-loop control processes mainly relies on historical fault-free data, which suffers from drawbacks such as data uniformity, lack of real-time and dynamic adaptability, and insufficient fault early warning mechanisms. Specifically, this method fails to fully utilize multi-source, multi-dimensional environmental data in the distributed control system, limiting the accuracy and comprehensiveness of fault diagnosis. Furthermore, this method focuses on building models using historical data, making it difficult to quickly respond to sudden fault situations in the system, and its static modeling method cannot capture the dynamic propagation characteristics of faults in a timely manner.

[0005] Therefore, there is an urgent need for a fault diagnosis and early warning method and system for distributed control systems to improve the accuracy and comprehensiveness of fault diagnosis, while also being able to respond quickly to sudden fault situations in the system. Summary of the Invention

[0006] To address the aforementioned technical problems, the purpose of this application is to provide a fault diagnosis and early warning method and system for a distributed control system. By analyzing vibration to determine the development trend of faults, the method enables real-time monitoring and alarm of system faults, thereby improving the accuracy of fault diagnosis and early warning capabilities.

[0007] The objective of this application is achieved through the following technical solution:

[0008] In a first aspect, the present invention provides a method for fault diagnosis and early warning of a distributed control system, comprising:

[0009] Multiple sensor arrays are deployed at each key node of the distributed control system to collect various environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data; and a safety threshold is preset for each type of environmental data at each key node.

[0010] Read the predefined fault modes and fault propagation paths from the historical fault case library; use the graph structure modeling method to represent the physical components or functional modules that may fail in the distributed control system as network nodes, and represent the fault propagation relationship of the physical components or functional modules as directed connection edges between network nodes, forming a reference fault topology network corresponding to each fault mode;

[0011] Read the complete topology network of the distributed control system. When any type of environmental data of any critical node exceeds the corresponding safety threshold, find an adjacent critical node on each different path of the critical node and compare the real-time mechanical vibration data of the adjacent critical node with the data of the same time point in different periods of historical mechanical vibration data.

[0012] If the difference in amplitude in the mechanical vibration data is greater than a preset range, then the adjacent key nodes are identified as potential propagation nodes. The search continues to find an adjacent key node on each different path of the potential propagation node, and the real-time mechanical vibration data of the adjacent key nodes of the potential propagation node are compared with the data of the same time point in different periods of the historical mechanical vibration data until the comparison result shows that the difference in amplitude in the mechanical vibration data is within the preset range, and the search operation ends. At the same time, all potential propagation nodes are output and a topology network of pending faults is generated.

[0013] Using the amplitude difference in mechanical vibration data from large to small or from small to large as the transmission direction, the vibration propagation path is determined on the undetermined fault topology network; the key nodes corresponding to the fault source are determined according to the opposite direction of the transmission direction.

[0014] The vibration propagation path is matched with the reference fault topology network. When the matching degree is greater than the preset value, it is determined to be a fault and the corresponding fault mode is output. The location information of the vibration propagation path and the fault mode are sent to the terminal for early warning.

[0015] Furthermore, the step of presetting security thresholds for each type of environmental data at each critical node specifically includes:

[0016] Establish a multimodal training dataset for key nodes, which includes samples of mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data collected under historical normal operating conditions, and label the equipment load level and operating condition labels corresponding to the samples.

[0017] The time-series feature extraction module processes the three types of environmental data to obtain multiple target features, including frequency domain energy distribution features extracted from mechanical vibration data, regional temperature difference gradient features extracted from temperature field distribution data, and harmonic distortion rate features extracted from power supply current waveform data.

[0018] The extracted target features and operating condition labels are input into a parallel machine learning model. The model includes a vibration analysis subnetwork, a temperature field regression subnetwork, and a current evaluation subnetwork. Each subnetwork uses an attention mechanism to dynamically allocate feature weights.

[0019] During model training, the dual optimization objectives are to minimize prediction error and maximize fault sensitivity. The final output is a dynamic safety threshold curve under different operating conditions. This threshold curve is automatically adjusted according to the real-time load rate of the equipment and a buffer is provided to avoid false triggering.

[0020] Furthermore, the step of determining the fault source based on the opposite direction of transmission specifically includes:

[0021] The main nodes are selected from the undetermined fault topology network. The selection criteria for main nodes include: when the main node has multiple connection paths, the amplitude increases / decreases radially from the main node to the critical nodes on each path; when the main node has only one path, it is selected directly.

[0022] Connect all major nodes, using the amplitude difference in the mechanical vibration data from largest to smallest or smallest to largest as the transmission direction.

[0023] Generate vibration propagation paths; if there are multiple vibration propagation paths, the intersection of all vibration propagation paths is taken as the fault source location; if there is only one vibration propagation path, the amplitude difference of each node on the vibration propagation path is fitted into a curve, and the node closest to the extreme value of the amplitude difference is selected as the fault source.

[0024] Furthermore, after determining the critical node corresponding to the fault source based on the opposite direction of transmission, the process also includes:

[0025] In the undetermined fault topology network, mark the turning nodes on the vibration propagation path. The turning nodes satisfy the condition that their amplitude difference increases or decreases by a preset value within a preset time.

[0026] Prioritize matching the reference fault topology network for the path where the turning point is located;

[0027] When a match is successful, the source of the fault is traced directly from that turning point.

[0028] Furthermore, the steps for finding an adjacent critical node on each different path of the critical node specifically include:

[0029] If any adjacent critical node does not respond when retrieving its data, the critical node is marked as a corrupted node, the corrupted node is determined to be a potential propagation node, and an adjacent critical node is found on each different path of the potential propagation node.

[0030] Furthermore, the formula for matching the vibration propagation path with the reference fault topology is:

[0031]

[0032] In the formula, The score represents the matching degree, ranging from [0,1], with a higher value indicating a higher matching degree. This represents the total number of nodes along the vibration propagation path. In the undetermined fault topology network, the first... The difference in mechanical vibration amplitude at each node; This is to reference the historical vibration amplitude difference of the corresponding node in the faulty topology network; The absolute difference in node data; For the first The weight coefficient of each node is determined by the node's hierarchical depth in the network topology; This represents the hop distance from the node to the fault source.

[0033] Furthermore, after sending the location information of the vibration propagation path and the fault mode to the terminal for early warning, it also includes:

[0034] Once a fault is identified and the corresponding fault mode is output, the fault mode is matched with the solutions in the preset solution library. The solutions include the recommended isolation measures, the order of switching backup equipment, and the list of node equipment components to be checked first.

[0035] The solution is attached to the warning message and sent to the terminal.

[0036] Secondly, the present invention provides a fault diagnosis and early warning system for a distributed control system, comprising:

[0037] The data acquisition module is used to deploy multiple sensor arrays at each key node of the distributed control system to collect various environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data; and to preset safety thresholds for each type of environmental data at each key node.

[0038] The reference fault topology network module is used to read predefined fault modes and fault propagation paths from the historical fault case library. Using a graph structure modeling method, physical components or functional modules that may fail in the distributed control system are represented as network nodes, and the fault propagation relationship of physical components or functional modules is represented as directed connection edges between network nodes, forming a reference fault topology network corresponding to each fault mode.

[0039] The comparison module is used to read the complete topology network of the distributed control system. When any type of environmental data of any critical node exceeds the corresponding safety threshold, it finds an adjacent critical node on each different path of the critical node and compares the real-time mechanical vibration data of the adjacent critical node with the data of the same time point in different periods of the historical mechanical vibration data.

[0040] The pending fault topology network module is used to determine that if the difference in amplitude in the mechanical vibration data is greater than a preset range, the adjacent key nodes are identified as potential propagation nodes. The module continues to search for an adjacent key node on each different path of the potential propagation node, and compares the real-time mechanical vibration data of the adjacent key nodes of the potential propagation node with the data of the same time point in different periods of the historical mechanical vibration data, until the comparison result shows that the difference in amplitude in the mechanical vibration data is within the preset range, at which point the search operation ends. At the same time, all potential propagation nodes are output and a pending fault topology network is generated.

[0041] The fault source identification module is used to determine the vibration propagation path on the undetermined fault topology network by taking the amplitude difference in mechanical vibration data from large to small or from small to large as the propagation direction; and to determine the key node corresponding to the fault source according to the opposite direction of the propagation direction.

[0042] The diagnosis and early warning module is used to match the vibration propagation path with the reference fault topology network. When the matching degree is greater than the preset value, it is determined to be a fault and the corresponding fault mode is output. The location information of the vibration propagation path and the fault mode are sent to the terminal for early warning.

[0043] Thirdly, the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps corresponding to the method in the first aspect.

[0044] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps corresponding to the method in the first aspect.

[0045] In summary, the technical solutions of this application have at least the following advantages and beneficial effects:

[0046] This invention utilizes multiple sensor arrays deployed at key nodes of a distributed control system to collect multi-dimensional environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data. This data is then combined with preset safety thresholds to achieve real-time monitoring. A reference fault topology network is constructed based on a historical fault case library, with physical components as nodes and fault propagation relationships as edges. When monitored data exceeds the threshold, the system compares the amplitude differences between real-time and historical vibration data from adjacent nodes, tracing the abnormal vibration propagation path step by step to pinpoint the fault source node. The system performs pattern matching between dynamically generated vibration propagation paths and the pre-stored fault topology network to achieve accurate fault type identification. Finally, fault pattern recognition and early warning are achieved by matching the actual propagation path with the reference fault network. This enables precise fault source location and propagation path prediction, improving the accuracy and timeliness of system fault diagnosis. Attached Figure Description

[0047] Figure 1 A flowchart of a fault diagnosis and early warning method for a distributed control system provided by the present invention;

[0048] Figure 2 A schematic diagram of the structure of a fault diagnosis and early warning system for a distributed control system provided by the present invention;

[0049] Figure 3 This is a schematic diagram of an electronic device provided by the present invention.

[0050] Reference numerals: 201, Data acquisition module; 202, Reference fault topology network module; 203, Comparison module; 204, Pending fault topology network module; 205, Fault source identification module; 206, Diagnosis and early warning module; 301, Processor; 302, Memory. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] like Figure 1 As shown in the embodiment of this application, a fault diagnosis and early warning method for a distributed control system includes:

[0053] S101. Deploy multiple sensor arrays at each key node of the distributed control system to collect various environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data; and preset safety thresholds for each type of environmental data at each key node.

[0054] Specifically, multiple sensor arrays (such as vibration accelerometers, infrared thermal imagers, and current transformers installed in the joint drive units of industrial robots) are deployed at each key node of the distributed control system. A multi-dimensional monitoring system is constructed by parallel acquisition of three types of heterogeneous environmental data: mechanical vibration waveforms (i.e., mechanical vibration data), temperature field infrared thermal imaging sequences (temperature field distribution data on equipment surfaces), and power supply current waveforms. Specifically, mechanical vibration data, through frequency domain energy distribution feature extraction (e.g., wavelet packet decomposition of the 0-10kHz frequency band and calculation of the energy proportion of each sub-band), can quantify internal bearing wear or gear meshing abnormalities. Temperature field distribution data, through regional temperature difference gradient feature extraction (e.g., dividing the motor casing into 32 grid regions and calculating the maximum temperature difference between adjacent grids), can identify local overheating or heat dissipation failure. Power supply current waveform data, through harmonic distortion rate feature extraction (e.g., analyzing the ratio of the fundamental wave amplitude to the 3rd / 5th / 7th harmonic amplitudes), can reflect power fluctuations or winding insulation degradation.

[0055] To establish a dynamic safety threshold, this invention first constructs a multimodal training dataset for key nodes: collecting three types of environmental data samples from equipment under historical normal operating conditions, and labeling them with corresponding equipment load levels and operating condition tags. Multiple target features and operating condition tags are processed through a parallel machine learning model—this model employs a parallel architecture of a vibration analysis sub-network (1D-CNN structure, i.e., One-Dimensional Convolutional Neural Network), a temperature field regression sub-network (spatiotemporal graph convolutional structure), and a current evaluation sub-network (LSTM structure, i.e., Long Short-Term Memory Network). An attention mechanism is used to dynamically allocate feature weights (e.g., increasing the vibration feature weight to 0.7 under heavy load conditions and increasing the temperature feature weight to 0.6 under light load conditions). Model training adopts a dual optimization objective of minimizing prediction error and maximizing fault sensitivity (e.g., introducing an F1-score penalty term, i.e., a model performance evaluation index that combines precision and recall). The final output is a dynamic safety threshold curve that automatically adjusts with the load rate, and a ±5% buffer is set to suppress false triggering caused by transient interference.

[0056] The formula for extracting the frequency domain energy distribution characteristics of mechanical vibration data through wavelet packet decomposition is as follows:

[0057]

[0058] In the formula, For the first The proportion of energy in the individual's body For the first The energy of each subband (calculated in the 0-10 kHz frequency band by wavelet packet decomposition). The total number of sub-bands, For total energy, This is the loop variable for summation.

[0059] The formula for calculating the regional temperature gradient characteristics of the temperature field distribution data using grid temperature difference is as follows:

[0060]

[0061] In the formula, This represents the maximum temperature difference between adjacent grid regions. For the first Temperature of each grid region It is the set of all adjacent grid pairs (e.g., adjacent grids are defined as spatially adjacent regions). , These are network indices for different sequences.

[0062] The formula for calculating the harmonic distortion rate characteristic of the power supply current waveform data is as follows:

[0063]

[0064] In the formula, Total harmonic distortion (THD) The fundamental amplitude, For the first The amplitude of the second harmonic. For the set of harmonics to be considered, For indexing.

[0065] Then, the dynamic security threshold is calculated, as follows:

[0066] Input feature vector :

[0067]

[0068] in, For vibration eigenvectors (e.g.) sequence), For temperature feature vectors (e.g.) Or the original temperature field distribution data). For current eigenvectors (e.g.) Or the original waveform data).

[0069] Set the operating condition labels: Load Level L and Operating Condition Label C. The output formula of the parallel machine learning model is:

[0070]

[0071] In the formula, To be under load and working conditions The predicted normal feature values, For the vibration analysis subnetwork output (1D-CNN structure), input ; Temperature field regression subnetwork output (spatiotemporal graph convolutional structure), input ; For the current evaluation subnetwork output (LSTM structure), input ; , , These are the corresponding attention weights, satisfying... The load is dynamically allocated based on the operating conditions.

[0072] Dynamic security threshold definition (for each feature):

[0073]

[0074] In the formula, The upper safety threshold (used for upper limit features such as vibration amplitude and temperature); This is the lower safety threshold (used for lower limit features, adjusted according to feature type). These are the predicted normal feature values. ×1.05 and ×0.95 are within a ±5% buffer range.

[0075] The model training optimization objectives (dual objectives) are:

[0076]

[0077] In the formula, For mean square error loss, The number of training samples, For the first The actual feature values ​​of each sample (historical normal operation data). , The first The load level and operating condition label of each sample.

[0078] Fault sensitivity maximization (F1-score penalty):

[0079]

[0080] In the formula, The F1-score penalty loss is used for classification tasks, defining a binary classification problem for anomaly detection. Precision and recall are the fault detection results based on threshold triggering, respectively.

[0081] The total loss function is:

[0082]

[0083] In the formula, , These are the weighting coefficients (balancing prediction error and fault sensitivity).

[0084] S102: Read the predefined fault modes and fault propagation paths from the historical fault case library; use the graph structure modeling method to represent the physical components or functional modules that may fail in the distributed control system as network nodes, and represent the fault propagation relationship of the physical components or functional modules as directed connection edges between network nodes, forming a reference fault topology network corresponding to each fault mode.

[0085] Specifically, the system reads predefined fault modes and propagation paths from a historical fault case library. These fault modes cover typical equipment failure types (such as bearing jamming, coil short circuits, etc.), while the propagation paths record the sequence of abnormal signals transmitted between physical components in historical faults. A graph structure modeling method is used to abstract potentially faulty physical components or functional modules in the system as network nodes (e.g., servo motors, reducers, and PLC control units in industrial robots). Simultaneously, the fault transmission relationships between components caused by physical connections or functional coupling are represented as directed edges between nodes (e.g., a gear tooth breakage fault in a reducer will cause an abnormal increase in the load torque of the servo motor, forming a directed edge from the reducer node to the motor node). Thus, each independent fault mode can be mapped to a directional reference fault topology network.

[0086] The modeling principle is based on the spatiotemporal correlation of fault propagation—when a component fails, the abnormal state it causes will spread along physical links or control signal flow paths in a specific direction. For example, in a servo drive unit, if the reducer lubrication fails (node ​​A), its abnormal vibration will first be transmitted to the connected motor bearing (node ​​B), and further cause end effector positioning drift through the drive shaft (node ​​C), thus forming a directed propagation chain of "A→B→C". Through graph structure modeling, the system can quantify the logical dependencies of the fault propagation path (e.g., the direction of the arrows represents the propagation sequence), enabling the subsequent diagnostic stage to quickly locate the source based on the network topology. Compared to traditional isolated node monitoring, this design significantly improves the systematic nature of fault tracing.

[0087] S103: Read the complete topology network of the distributed control system. When any type of environmental data of any critical node exceeds the corresponding safety threshold, find an adjacent critical node on each different path of the critical node, and compare the real-time mechanical vibration data of the adjacent critical node with the data of the same time point in different periods of the historical mechanical vibration data.

[0088] Specifically, when the environmental monitoring data (vibration, temperature, or current) of any critical node within the system (such as the main bearing unit of the coke oven gas compressor in a coking plant) exceeds the dynamic safety threshold, the system immediately initiates cross-equipment correlation diagnosis. First, it reads the plant's pre-set equipment topology network (e.g., the mechanical transmission chain of "main bearing → speed-increasing gearbox → high-pressure cylinder" in the gas compressor unit). Starting from the critical node that triggered the alarm, it traverses all its physical connection paths and selects the nearest downstream node on each path (e.g., selecting the coupling monitoring point when the main bearing alarms). For each adjacent node, its real-time mechanical vibration waveform is synchronously acquired, and historical benchmark data under the same operating conditions is retrieved through the operating condition matching engine (requiring load rate deviation ≤3% and speed fluctuation ≤1%). The temporal amplitude characteristics (e.g., the effective value RMS in the 0-10kHz frequency band) are then compared and aligned at the millisecond level. This design is based on the mechanical impedance matching principle—when a fault vibration wave propagates between rigidly connected equipment, the response signal of adjacent nodes should exhibit a linear attenuation relationship (the amplitude difference is stable within the range of 8-12dB under normal operating conditions). If a communication interruption is detected between adjacent nodes during a data request (e.g., a 500ms timeout in the Profibus-DP bus response), that node is marked as a hardware failure point (possibly due to a broken junction box caused by vibration), and the original alarm node is marked as a "potential propagation source." The system then initiates topology tracing, skipping the faulty node and continuing to detect the next nearest node (e.g., a speed-increasing gearbox). Taking a 2650m³ blast furnace blower unit in a steel plant as an example, when the vibration amplitude at the blower impeller end exceeds the standard (measured 4.8mm / s, threshold 2.5mm / s), the system simultaneously detects the connected hydraulic coupler and main shaft support nodes: it finds that the hydraulic coupler node communication is interrupted (CANOpen message lost), and the real-time vibration value of the main shaft support reaches 35μm (historical baseline value 22±2μm, deviation exceeding 59%). Based on this, it is determined that the fault propagates along the "impeller → main shaft support" path. This mechanism ensures accurate location of transmission chain faults through directional topology tracing and strong correlation verification of operating conditions. Simultaneously, by utilizing the communication-vibration dual-mode criterion, a combined fault of a broken gear tooth puncturing a signal cable was successfully identified in a case involving an induced draft fan in a thermal power plant.

[0089] S104: If the difference in amplitude in the mechanical vibration data is greater than the preset range, then the adjacent key node is determined to be a possible propagation node. Continue to search for an adjacent key node on each different path of the possible propagation node, and compare the real-time mechanical vibration data of the adjacent key node of the possible propagation node with the data of the same time point in different periods in the historical mechanical vibration data until the comparison result shows that the difference in amplitude in the mechanical vibration data is within the preset range, and end the search operation; at the same time, output all possible propagation nodes and generate a pending fault topology network.

[0090] Specifically, when the system detects that the amplitude difference between the real-time mechanical vibration data of an adjacent key node and the historical benchmark data (strictly filtered by the operating condition matching engine) exceeds a preset range (e.g., ±12dB), the system determines that the adjacent node is a potential propagation node. The principle behind this determination is that during vibration transmission, amplitudes can have a superposition effect, meaning the amplitude may increase or decrease. Therefore, comparing the difference allows for a direct assessment of whether the vibration of the same node has changed at different times. If the difference exceeds the limit, it indicates nonlinear distortion in the vibration transmission process (such as abnormal energy release due to structural cracks or sudden changes in mechanical impedance). The system then uses this node as a new starting point, selecting the next nearest node along each physical connection path, and repeatedly performs millisecond-level comparisons of real-time and historical vibration data (focusing on the RMS difference in the 0-10kHz frequency band). This iterative process continues to expand outwards until the amplitude difference of the newly detected node returns to the preset range (indicating that the vibration energy has been effectively absorbed at the healthy equipment node or that energy transmission has stopped at this point). Ultimately, the system aggregates all marked potentially propagating nodes and generates a pending fault topology network based on their physical connections (with directed edges representing the propagation direction).

[0091] A multi-level topology tracing mechanism addresses the ambiguity of propagation paths in complex fault scenarios. Traditional single-point monitoring is susceptible to propagation attenuation interference, making it difficult to distinguish between the fault source and propagation nodes. This step utilizes the attenuation characteristics of vibration wave propagation as a fault path identifier, avoiding a global scan of the system topology and reducing computational load.

[0092] S105: Using the amplitude difference in mechanical vibration data from large to small or from small to large as the transmission direction, determine the vibration propagation path on the undetermined fault topology network; determine the key node corresponding to the fault source based on the opposite direction of the transmission direction.

[0093] Specifically, the system, based on a topology network of undetermined faults, uses the changing trend of real-time amplitude differences between adjacent nodes in mechanical vibration data as the core criterion for fault propagation direction. According to the propagation characteristics of mechanical vibration waves between rigidly connected devices, the abnormal vibration energy generated by the fault source will exhibit gradient attenuation or superposition effects as it propagates along the physical link. Therefore, the system first analyzes the amplitude difference sequence of all possible propagating nodes in the undetermined fault topology network. If the sequence shows a monotonically increasing trend (e.g., the amplitude difference between nodes A→B→C is +8dB→+12dB→+15dB), the vibration propagation path is determined to be in the forward direction (i.e., spreading from low-amplitude nodes to high-amplitude nodes); otherwise, it is determined to be in the reverse direction (indicating that energy attenuates during propagation).

[0094] After determining the propagation direction, the system performs a backtracking analysis of the topology network: nodes with ≥2 connected paths are selected as primary nodes, and their radial diffusion characteristics are verified. For example, if a node connects to three paths simultaneously, and the amplitude difference from its output to the downstream node decreases unidirectionally, then the node meets the radiation characteristics of a fault source. For linear topologies with only a single path, the amplitude difference of all nodes on the path is fitted into a curve, and the node closest to the extreme point (highest / lowest point) of the amplitude difference is selected as a candidate fault source. When multiple independent propagation paths exist (such as two paths, "bearing housing → coupling" and "gearbox → output shaft"), in a compressor system, the system automatically calculates the intersection node of each path, which is the physical location of the fault source.

[0095] To improve the diagnostic efficiency of complex faults, the system marks inflection nodes in the undetermined fault topology network where significant abrupt changes in amplitude difference occur (the abrupt change threshold is defined as an amplitude difference change rate ≥ 30% between adjacent nodes). For example, in the case of a power plant turbine unit, if a 42% abrupt change in amplitude difference is detected between the high-pressure cylinder node and the intermediate-pressure cylinder node (the difference suddenly increases from +3dB to +12dB), the system immediately marks it as an inflection node. Subsequently, the system prioritizes matching the path of this node with the reference fault topology network in the historical fault case database (e.g., matching the "rotor dynamic balance failure" propagation pattern), directly tracing the fault source backward from this node, thus shortening the diagnostic link.

[0096] S106: Match the vibration propagation path with the reference fault topology network. When the matching degree is greater than the preset value, it is determined to be a fault and the corresponding fault mode is output. The location information of the vibration propagation path and the fault mode are sent to the terminal for early warning.

[0097] Specifically, the system first calculates the similarity between the vibration propagation path determined in step S105 and the reference fault topology network constructed in step S102. This reference fault topology network is derived from a historical fault case library and predefines the propagation paths and node topologies for different fault modes. The calculation process uses a graph structure matching algorithm (such as a subgraph isomorphism or similarity measurement model) to generate a matching degree value by quantifying the consistency between the vibration propagation path and the node sequence, connection direction, and propagation timing in the reference network. When the matching degree value exceeds a preset threshold (e.g., set to 80% to balance sensitivity and anti-interference), the system determines that the current scenario conforms to a specific fault mode and outputs the corresponding fault mode identifier (e.g., bearing wear or gear cracking). The principle is that the fault propagation of a distributed control system has predictable regularity. The historical fault case library captures the spatiotemporal correlation of fault propagation between physical components through graph structure modeling, enabling the real-time detected vibration propagation path to be quickly mapped to a known fault mode through similarity comparison, thereby avoiding the risk of misjudgment based on isolated node monitoring.

[0098] Subsequently, the system transmits the location information of the vibration propagation path (including the key nodes of the fault source and all possible propagation nodes on the transmission path) and the determined fault mode to the terminal monitoring system in real time via industrial communication protocol for early warning. The location information accurately identifies the physical location of the fault (such as equipment unit coordinates or topology node number), while the fault mode describes the specific failure type (such as lubrication failure or insulation degradation).

[0099] After the warning information is sent, the system further performs a solution matching operation: it queries the output fault mode against a preset solution library, which pre-stores standardized response strategies for each fault mode. The solutions include three core elements: recommended isolation measures (such as cutting off power to the faulty component or physically isolating the drivetrain), the switching sequence of backup equipment (such as prioritizing the activation of redundant pump sets or switching to backup motors), and a list of node equipment components to be checked first (such as specifying specific inspection items for bearings, seals, or windings). The system quickly retrieves matching items using a pattern mapping algorithm (such as hash table retrieval based on fault mode encoding) and appends the solution to the original warning information before sending it to the terminal.

[0100] The matching degree between the vibration propagation path and the reference fault topology network is calculated using a dynamic time warping algorithm, and the calculation formula is as follows:

[0101]

[0102] In the formula, The matching score ranges from [0,1], with larger values ​​indicating higher matching scores; the total number of nodes along the vibration propagation path is also represented. In the undetermined fault topology network, the first... The difference in mechanical vibration amplitude at each node; This is to reference the historical vibration amplitude difference of the corresponding node in the faulty topology network; The absolute difference in node data; For the first The weight coefficient of each node is determined by the node's hierarchical depth in the network topology; This represents the hop distance from the node to the fault source.

[0103] Based on the same inventive concept, such as Figure 2 As shown, the present invention provides a fault diagnosis and early warning system for a distributed control system, comprising:

[0104] The data acquisition module 201 is used to deploy multiple sensor arrays at each key node of the distributed control system to collect various environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data; and to preset safety thresholds for each type of environmental data at each key node.

[0105] The reference fault topology network module 202 is used to read the predefined fault modes and fault propagation paths in the historical fault case library; using the graph structure modeling method, the physical components or functional modules that may fail in the distributed control system are represented as network nodes, and the fault propagation relationship of the physical components or functional modules is represented as directed connection edges between network nodes, forming a reference fault topology network corresponding to each fault mode.

[0106] The comparison module 203 is used to read the complete topology network of the distributed control system. When any type of environmental data of any key node exceeds the corresponding safety threshold, it finds an adjacent key node on each different path of the key node and compares the real-time mechanical vibration data of the adjacent key node with the data of the same time point in different periods of the historical mechanical vibration data.

[0107] The pending fault topology network module 204 is used to determine that if the difference in amplitude in the mechanical vibration data is greater than a preset range, the adjacent key nodes are identified as possible propagation nodes. The module continues to search for an adjacent key node on each different path of the possible propagation node, and compares the real-time mechanical vibration data of the adjacent key nodes of the possible propagation node with the data of the same time point in different periods of the historical mechanical vibration data until the comparison result shows that the difference in amplitude in the mechanical vibration data is within the preset range, and then ends the search operation. At the same time, it outputs all possible propagation nodes and generates a pending fault topology network.

[0108] The fault source identification module 205 is used to determine the vibration propagation path on the undetermined fault topology network based on the direction of propagation of the amplitude difference in the mechanical vibration data from large to small or from small to large; and to determine the key node corresponding to the fault source based on the opposite direction of propagation.

[0109] The diagnosis and early warning module 206 is used to match the vibration propagation path with the reference fault topology network. When the matching degree is greater than the preset value, it is determined to be a fault and the corresponding fault mode is output. The location information of the vibration propagation path and the fault mode are sent to the terminal for early warning.

[0110] Based on the same inventive concept, such as Figure 3 As shown, the present invention provides an electronic device, including: a memory 302, a processor 301, and a computer program stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program, it implements a fault diagnosis and early warning method for a distributed control system.

[0111] Based on the same inventive concept, the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for fault diagnosis and early warning of a distributed control system.

[0112] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for fault diagnosis and early warning of a distributed control system, characterized in that, include: Multiple sensor arrays are deployed at each key node of the distributed control system to collect various environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data; and a safety threshold is preset for each type of environmental data at each key node. Read the predefined fault modes and fault propagation paths from the historical fault case library; use the graph structure modeling method to represent the physical components or functional modules that may fail in the distributed control system as network nodes, and represent the fault propagation relationship of the physical components or functional modules as directed connection edges between network nodes, forming a reference fault topology network corresponding to each fault mode; Read the complete topology network of the distributed control system. When any type of environmental data of any key node exceeds the corresponding safety threshold, find an adjacent key node on each different path of the key node and compare the real-time mechanical vibration data of the adjacent key node with the data of the same time point in different periods of historical mechanical vibration data. If the difference in amplitude in the mechanical vibration data is greater than a preset range, the adjacent key node is determined to be a possible propagation node. The search continues to find an adjacent key node on each different path of the possible propagation node, and the real-time mechanical vibration data of the adjacent key node of the possible propagation node is compared with the data of the same time point in different periods of the historical mechanical vibration data until the comparison result shows that the difference in amplitude in the mechanical vibration data is within the preset range, and the search operation ends. At the same time, all possible propagation nodes are output and an undetermined fault topology network is generated. Using the amplitude difference in mechanical vibration data from large to small or from small to large as the transmission direction, the vibration propagation path is determined on the topology network of the undetermined fault; the key node corresponding to the fault source is determined according to the opposite direction of the transmission direction. The vibration propagation path is matched with the reference fault topology network. When the matching degree is greater than a preset value, it is determined to be a fault and the corresponding fault mode is output. The location information of the vibration propagation path and the fault mode are sent to the terminal for early warning. The matching degree between the vibration propagation path and the reference fault topology network is calculated using a dynamic time warping algorithm.

2. The method for fault diagnosis and early warning of a distributed control system according to claim 1, characterized in that, The step of presetting a security threshold for each type of environmental data at each key node specifically includes: Establish a multimodal training dataset for key nodes, which includes samples of mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data collected under historical normal operating conditions, and label the equipment load level and operating condition labels corresponding to the samples. The time-series feature extraction module processes the three types of environmental data to obtain multiple target features, including extracting frequency domain energy distribution features from mechanical vibration data. The extracted target features and operating condition labels are input into a parallel machine learning model. The model includes a vibration analysis subnetwork, a temperature field regression subnetwork, and a current evaluation subnetwork. Each subnetwork uses an attention mechanism to dynamically allocate feature weights. The model is trained with the dual optimization objectives of minimizing prediction error and maximizing fault sensitivity. It ultimately outputs dynamic safety threshold curves under different operating conditions. These threshold curves are automatically adjusted according to the real-time load rate of the equipment and are equipped with buffer zones to avoid false triggering.

3. The method for fault diagnosis and early warning of a distributed control system according to claim 1, characterized in that, The step of determining the fault source based on the opposite direction of the transmission direction specifically includes: Key nodes are selected from the undetermined fault topology network. The selection criteria for key nodes include: when a key node has multiple connection paths, the amplitude increases / decreases radially from the key node to the key nodes on each path; when a key node has only one path, it is directly selected. Connect all the main nodes, using the amplitude difference in the mechanical vibration data from largest to smallest or smallest to largest as the transmission direction. Generate vibration propagation paths; if there are multiple vibration propagation paths, the intersection of all vibration propagation paths is taken as the fault source location; if there is only one vibration propagation path, the amplitude difference of each node on the vibration propagation path is fitted into a curve, and the node closest to the extreme value of the amplitude difference is selected as the fault source.

4. The fault diagnosis and early warning method for a distributed control system according to claim 3, characterized in that, After determining the critical node corresponding to the fault source based on the opposite direction of the transmission direction, the method further includes: In the undetermined fault topology network, the turning nodes on the vibration propagation path are marked, and the turning nodes satisfy that their amplitude difference increases or decreases by a preset value within a preset time. Prioritize the matching of reference fault topology networks for the paths where the turning points are located; When a match is successful, the source of the fault is traced directly from that turning point.

5. The method for fault diagnosis and early warning of a distributed control system according to claim 1, characterized in that, The step of finding an adjacent key node on each different path of the key node specifically includes: If any of the adjacent critical nodes does not respond when acquiring its data, the critical node is marked as a corrupt node, the corrupt node is determined to be a potential propagation node, and an adjacent critical node is found on each different path of the potential propagation node.

6. The fault diagnosis and early warning method for a distributed control system according to claim 1, characterized in that, The formula for matching the vibration propagation path with the reference fault topology network is: In the formula, The score represents the matching degree, ranging from [0,1], with a higher value indicating a higher matching degree. This represents the total number of nodes along the vibration propagation path. In the undetermined fault topology network, the first... The difference in mechanical vibration amplitude at each node; This is to reference the historical vibration amplitude difference of the corresponding node in the faulty topology network; The absolute difference in node data; For the first The weight coefficient of each node is determined by the node's hierarchical depth in the network topology; This represents the hop distance from the node to the fault source.

7. The method for fault diagnosis and early warning of a distributed control system according to claim 1, characterized in that, After sending the location information of the vibration propagation path and the fault mode to the terminal for early warning, the method further includes: Once a fault is identified and the corresponding fault mode is output, the fault mode is matched with the solutions in the preset solution library. The solutions include suggested isolation measures, backup equipment switching sequence, and a list of node equipment components to be checked first. The solution is attached to the warning message and sent to the terminal.

8. A fault diagnosis and early warning system for a distributed control system, characterized in that, include: The data acquisition module is used to deploy multiple sensor arrays at each key node of the distributed control system to collect various environmental data, including mechanical vibration data, equipment surface temperature field distribution data, and power supply current waveform data; and to preset safety thresholds for each type of environmental data at each key node. The reference fault topology network module is used to read predefined fault modes and fault propagation paths from the historical fault case library; using a graph structure modeling method, physical components or functional modules that may fail in the distributed control system are represented as network nodes, and the fault propagation relationship of the physical components or functional modules is represented as directed connection edges between network nodes, forming a reference fault topology network corresponding to each fault mode. The comparison module is used to read the complete topology network of the distributed control system. When any type of environmental data of any key node exceeds the corresponding safety threshold, it finds an adjacent key node on each different path of the key node and compares the real-time mechanical vibration data of the adjacent key node with the data of the same time point in different periods of historical mechanical vibration data. The pending fault topology network module is used to determine that if the difference in amplitude in the mechanical vibration data is greater than a preset range, the adjacent key node is a possible propagation node. It continues to search for an adjacent key node on each different path of the possible propagation node, and compares the real-time mechanical vibration data of the adjacent key nodes of the possible propagation node with the data of the same time point in different periods of historical mechanical vibration data until the comparison result shows that the difference in amplitude in the mechanical vibration data is within the preset range, and then ends the search operation; at the same time, it outputs all possible propagation nodes and generates a pending fault topology network. The fault source identification module is used to determine the vibration propagation path on the undetermined fault topology network by taking the amplitude difference in mechanical vibration data from large to small or from small to large as the propagation direction; and to determine the key node corresponding to the fault source according to the opposite direction of the propagation direction. The diagnosis and early warning module is used to match the vibration propagation path with the reference fault topology network. When the matching degree is greater than a preset value, it is determined to be a fault and the corresponding fault mode is output. The location information of the vibration propagation path and the fault mode are sent to the terminal for early warning.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a fault diagnosis and early warning method for a distributed control system as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements a fault diagnosis and early warning method for a distributed control system as described in any one of claims 1 to 7.

Images