Industrial electrical system event chain diagnosis method and system based on EdgeBrain engine driving

Abstract

The invention discloses an industrial electrical system event chain diagnosis method and system based on an EdgeBrain engine. The method comprises the following steps: firstly, constructing a multi-level fault library model comprising a representation layer, a middle layer and a root cause layer, and forming an event chain structure for describing a potential fault logic and a causal relationship; data are collected from equipment such as a PLC in real time and input into a streaming rule EdgeBrain engine to generate multi-stage event streams, step-by-step reasoning is performed in a multi-stage fault library model based on the event streams, and finally visual diagnosis results including link paths, node states and trigger parameters are output. According to the method, specific elements can be quickly tracked from top layer alarm within millisecond-level time, automatic generation of fault paths and accurate positioning of root causes are achieved, the average fault positioning time is shortened to be within 5 minutes from traditional 30 minutes, and the diagnosis efficiency and accuracy are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and edge intelligence diagnostic technology, and in particular to a method and system for diagnosing event chains in industrial electrical systems based on the EdgeBrain engine. Background Technology

[0002] In existing industrial equipment fault diagnosis systems, basic signal acquisition and alarm prompts are typically achieved through PLC logic, SCADA monitoring software, or host computer programs. Common solutions include: PLC internal logic alarm: An alarm is triggered after detecting a change in signal state using a ladder diagram or logic expression; SCADA / Host Computer Alarm System: After collecting PLC messages, it outputs alarm information through fixed threshold and condition combinations, with an actual false trigger rate of up to 50%. Cloud-based intelligent diagnostics: Some systems upload on-site data to cloud servers and then use AI or rule engines for analysis.

[0003] While the above methods can detect that a device malfunction has occurred, the following problems still exist: (1) Static fault rules: In existing SCADA or PLC systems, fault judgment is mostly achieved through fixed logic expressions, which cannot be dynamically expanded; (2) Message-level diagnostic fragmentation: Most diagnostic logic is single-level triggering, that is, an alarm is triggered when a certain point exceeds the limit. There is a lack of multi-level correlation analysis of the internal logic structure of the equipment (such as the causal relationship between relays, contactors and circuit breakers), which makes it impossible to directly locate the root cause. (3) High latency in the cloud: Some AI diagnostics rely on cloud inference, which cannot meet the millisecond-level response requirements of the crane control system; (4) Lack of visual rule management: On-site maintenance personnel cannot easily view and modify triggering conditions, duration, suppression logic, etc.; (5) High false trigger rate: Taking the fixed logic expression in SCADA as an example, there will be a high false trigger rate of more than 50% in actual use. Summary of the Invention

[0004] This invention provides a method and system for diagnosing event chains in industrial electrical systems based on the EdgeBrain engine, in order to overcome the aforementioned technical problems.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: A diagnostic method for event chains in industrial electrical systems based on the EdgeBrain engine, comprising the following steps: S1. Establish a fault-signal mapping list containing all potential fault modes, triggering conditions and their logical relationships; S2. Based on the fault-signal mapping list, establish a multi-level fault library model. The multi-level fault library model includes: a manifestation layer, an intermediate layer, a root cause layer, and several message event nodes, which are divided according to root cause logic. The manifestation layer, intermediate layer, and root cause layer form a multi-level event chain structure through several message event nodes. S3. Collect real-time data streams from industrial control equipment; S4. Establish a streaming rule EdgeBrain engine and input the real-time data stream into the streaming rule EdgeBrain engine to generate several levels of event streams for fault diagnosis. S5. Using the aforementioned event streams as input, perform step-by-step causal reasoning in the multi-level fault database model until the final event chain diagnosis result is generated and visualized. The final event chain diagnosis result includes the fault propagation link path, the status of each level of node, and triggering parameters.

[0006] Furthermore, the streaming rule EdgeBrain engine includes: a data preprocessing module, a filter module, a streaming rule calculation module, a time window module, and a trigger module, wherein: The data preprocessing module is used to filter, de-jitter, remove outliers, interpolate and align with timestamps on the real-time data stream, and output the preprocessed data stream. The filter module is used to perform validity screening and debouncing on the preprocessed data stream, and also to perform alarm suppression and deduplication on the trigger event objects fed back by the streaming rule calculation module. The time window module is used to perform event detection on the data processed by the filter module based on the time window and transmit it to the streaming rule calculation module; The streaming rule calculation module is used to perform real-time judgment and generate trigger event objects based on the duration and frequency of data transmitted by the time window module. The trigger module is used to generate an event stream with event identifiers, i.e., valid event objects, based on the trigger event objects processed by the filter module.

[0007] Furthermore, the specific steps of the streaming rule calculation module in performing real-time judgment and generating trigger event objects based on the duration and frequency of data transmitted by the time window module include: The data stream is filtered in real time based on the conditional expressions defined in the multi-level fault database model. When the data transmitted by the time window module meets one or more of the following conditions: condition expression, set duration threshold, or occurrence frequency threshold, the data stream is determined to meet the triggering condition, and the data stream that meets the triggering condition is transmitted to the trigger module as a trigger event object.

[0008] Furthermore, the multi-level fault library model is defined as a three-layer fault diagnosis logic framework, and a multi-level event chain structure is constructed through the parent-child relationship of message event nodes: The message event nodes of the root cause layer are used to describe the specific component failures at the lowest level. The message event node of the intermediate layer is the parent node of the root cause layer event node, and it is used to describe the subsystem or functional module problem that leads to the phenomenon in the upper layer. The message event node of the appearance layer is the parent node of the intermediate layer event node, and it is used to describe the system-level fault symptoms that are ultimately manifested. Each message event node contains the following structured fields: Identification information: Node ID, fault name, and level; Association relationships refer to node topology information: parent node ID and a list of child node IDs; Triggering condition: conditional expression; Criterion parameters: time window type, window length, number of attempts threshold, and suppression conditions; Auxiliary information: Repair recommendations, fault severity level, and handling tools.

[0009] Furthermore, the specific steps of the data preprocessing module in filtering, de-jittering, outlier removal, interpolation, and timestamp alignment of the real-time data stream include: The null and invalid values ​​in the real-time data stream are filtered. Perform debounce and outlier removal on the filtered data; Interpolation processing is performed on data with sampling gaps; Timestamp alignment is performed on data from different acquisition channels.

[0010] Furthermore, the specific steps of the filter module in performing validity screening and debouncing on the preprocessed data stream, i.e., the data associated with the triggering event object, include: 1) Perform a validity check on the preprocessed data stream. The check should include at least the following: Determine whether the preprocessed data stream is within the preset numerical range; Determine whether the preprocessed data stream contains missing, abnormal, or invalid identifiers; Determine whether the timestamps of the preprocessed data stream are continuous and whether they exceed the allowable time deviation range; If the preprocessed data stream meets all the above validity requirements (i.e., all judgment results are negative), then the preprocessed data stream is a valid data stream; otherwise, it is discarded or marked as abnormal data. 2) Perform debouncing on the preprocessed data stream that passes the validity check. The debouncing methods include, but are not limited to: The state of the preprocessed data stream is compared within multiple consecutive sampling periods. Only when the state of the preprocessed data stream remains consistent for a preset number of consecutive times or a preset time period is the change in the state of the preprocessed data stream considered valid.

[0011] An event chain diagnostic system for industrial electrical systems driven by the EdgeBrain engine, comprising: The fault database management module is used to build a multi-level fault database model based on the fault-signal mapping list, and to update and manage the multi-level fault database model. The data acquisition module is used to acquire real-time data streams from industrial control equipment; The streaming rules EdgeBrain engine is used to generate several levels of event streams for fault diagnosis based on the real-time data stream. The event chain reasoning module is used to perform step-by-step causal reasoning in the multi-level fault library model based on the several levels of event flow until the final event chain diagnosis result is generated. The real-time fault diagnosis and recording module is used to record the event results output in real time by the EdgeBrain engine. The event results output in real time by the EdgeBrain engine include: fault name, level, triggering condition, duration, associated nodes, timestamp, event chain path, and manual handling status.

[0012] Beneficial Effects: This invention, through its constructed streaming rule EdgeBrain engine, can extract several levels of event streams from real-time data streams for fault diagnosis. Based on these event streams, it performs step-by-step reasoning within a multi-level fault database model until the final event chain diagnostic result is generated. This invention can rapidly trace from top-level alarms to specific components within milliseconds. Through top-down cascading and bottom-up backtracking logic of the event chain, it can automatically generate fault paths and accurately locate root causes, with a location error of less than one electrical node. This invention can quickly and accurately locate the root cause of equipment faults, significantly reducing manual troubleshooting steps and lowering the average fault location time from the traditional 30 minutes to less than 5 minutes. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of the event chain diagnosis method for industrial electrical systems based on the EdgeBrain engine in this invention; Figure 2 This is a flowchart illustrating the diagnostic process in an embodiment of the present invention; Figure 3 This is a diagram of the multi-level event chain diagnostic structure in an embodiment of the present invention; Figure 4 This is a timing diagram of the diagnostic process in an embodiment of the present invention; Figure 5 This is a structural diagram of the diagnostic system in an embodiment of the present invention; Figure 6-8 This is a performance graph for scenario three in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] This embodiment provides a method and system for diagnosing event chains in industrial electrical systems based on the EdgeBrain engine, such as... Figure 1 As shown, the specific steps include: S1. Establish a fault-signal mapping list containing all potential fault modes, triggering conditions and their logical relationships. The fault-signal mapping list defines what signal combinations characterize what faults under what conditions. Specifically, the PLC ladder diagram, electrical schematic diagram, I / O table, relay and contactor contact table are obtained. Based on human experience and fault tree analysis (FTA), the above information is systematically sorted out to identify and extract all signal points related to potential faults, the logical conditions between signal points and the timing relationship of state changes, thereby forming a fault-signal mapping list to prepare for the construction of a digital fault database.

[0017] S2. Based on the fault-signal mapping list, establish a multi-level fault library model. The multi-level fault library model includes: a manifestation layer, an intermediate layer, a root cause layer, and several message event nodes, which are divided according to root cause logic. The manifestation layer, intermediate layer, and root cause layer form a multi-level event chain structure through several message event nodes. Specifically, the multi-level fault library model is defined as a three-layer fault diagnosis logic framework, and a multi-level event chain structure is constructed through the parent-child relationship of message event nodes: The message event nodes in the root cause layer are used to describe the specific component failures at the lowest level (such as a circuit breaker tripping or a relay failure). This is the endpoint of diagnosis, i.e., the root cause. The message event nodes of the intermediate layer are the parent nodes of the root cause layer event nodes. They are used to describe the problems of the subsystems or functional modules that cause the phenomena in the upper layer. They are both the result of the upper layer failure and the cause of the lower layer failure. The message event node of the surface layer is the parent node of the intermediate layer event node. It is used to describe the final system-level fault symptoms (such as "protection panel power-on failure"), which is also the starting point for subsequent diagnosis. Each message event node is a structured object with rich attributes, including a set of structured fields, including: Core identifiers: Node ID, fault name, and fault level; Relationships: Parent node ID (defining upstream) and list of child nodes (defining downstream) form an event chain; Triggering condition: Conditional expression (e.g., I54.0=1 AND I50.7=0), which is the logical relationship extracted from S1; Criterion parameters: time window type, window length, number of attempts threshold, and suppression conditions, used for anti-shake and confirmation of real faults; Auxiliary information: maintenance recommendations, fault severity levels, and handling tools, used to guide on-site operations and maintenance.

[0018] Specifically, this embodiment performs hierarchical modeling based on the fault-signal mapping list, dividing the fault information in the fault-signal mapping list into multiple levels according to causality and granularity, constructing a digital multi-level fault model containing complete topological relationships and criterion parameters, and providing real-time query services to the EdgeBrain engine and event chain reasoning module.

[0019] Specifically, fault database management includes the following: maintaining multi-level message trees and diagnostic knowledge, and supporting version control, canary releases, and online incremental updates.

[0020] S3. Acquire real-time data streams from industrial control equipment, such as real-time data streams from PLCs or other industrial equipment; Specifically, data is acquired from PLCs or other industrial devices via various industrial protocols (such as Modbus RTU / TCP, Siemens S7, OPC UA, etc.), including digital inputs, analog inputs, status signals, and coil outputs. The sampling period can be configured to a minimum of 50ms, and the sampling accuracy error is [not specified]. 5ms; achieves timestamp alignment and signal standardization; automatically caches data when the network connection is lost and replays it in chronological order after recovery to ensure data integrity; and transmits the collected data to the EdgeBrain engine in real time for preprocessing and judgment.

[0021] S4. Establish a streaming rule EdgeBrain engine and input the real-time data stream into the streaming rule EdgeBrain engine to generate several levels of event streams for fault diagnosis. S5. Using the aforementioned event streams as input, perform step-by-step causal reasoning in the multi-level fault database model until the final event chain diagnosis result is generated and visualized. The final event chain diagnosis result includes the fault propagation link path, the status of each level of node, and triggering parameters.

[0022] Specifically, this embodiment takes the EdgeBrain edge computing engine as the core and integrates industrial equipment signal acquisition, streaming rule engine judgment, multi-level event chain reasoning, fault database management and visualization display in a unified manner. It realizes a closed loop of autonomous diagnosis from data acquisition to fault root cause location, and can be applied to the judgment of large batches of rules in a single node, especially suitable for fault diagnosis in the field of industrial cranes.

[0023] Specifically, such as Figure 2 As shown, this embodiment first judges the first-level message. If it is valid, it immediately proceeds to the condition verification of the second-level message, and so on, cascading to the third and fourth levels. Each level of judgment supports time windows / number thresholds. After debouncing, it is confirmed as a valid event. After the lower-level event is confirmed as abnormal, the upper-level state is written back in reverse, forming a chain path of top-level fault-root cause. After any node recovers, it drives the upper-level node to automatically backtrack. When the network or device is abnormal, local caching and delayed playback are triggered to ensure link integrity. In practice, suppression strategies can also be set: specified events can be blocked in maintenance / debugging mode and power failure / emergency stop mode. And multiple first-level messages can be tracked in parallel according to actual needs.

[0024] In a specific embodiment, the streaming rule EdgeBrain engine includes: a data preprocessing module, a filter module, a streaming rule calculation module, a time window module, a trigger module, an action executor module, and a state machine management module, wherein: The data preprocessing module is used to filter, de-jitter, remove outliers, interpolate and align with timestamps on the real-time data stream, and output the preprocessed data stream. Specifically, the data preprocessing module performs filtering, dejittering, outlier removal, interpolation, and timestamp alignment on the real-time data stream, including the following steps: The null and invalid values ​​in the real-time data stream are filtered. Perform debounce and outlier removal on the filtered data; Interpolation processing is performed on data with sampling gaps; Timestamp alignment is performed on data from different acquisition channels.

[0025] Specifically, in this embodiment, the preprocessed data stream is used as the input for streaming rule judgment, thereby ensuring the stability and accuracy of subsequent time window judgment.

[0026] The filter module is used to perform validity screening and debouncing on the preprocessed data stream, and also to perform alarm suppression and deduplication on the trigger event objects fed back by the streaming rule calculation module. In a specific embodiment, the filter module performs validity screening and debouncing on the preprocessed data stream, i.e., the data associated with the triggering event object, including the following steps: 1) Before the preprocessed data stream enters the streaming rule calculation module, the filter module first performs a validity check on the preprocessed data stream. The check includes at least the following: Determine whether the preprocessed data stream is within the preset numerical range; Determine whether the preprocessed data stream contains missing, abnormal, or invalid identifiers; Determine whether the timestamps of the preprocessed data stream are continuous and whether they exceed the allowable time deviation range; When the preprocessed data stream meets the above validity requirements, i.e., all the judgment results are negative, it indicates that the preprocessed data stream is a valid data stream, and it is allowed to enter the subsequent time window and rule calculation process; otherwise, it is directly discarded or marked as abnormal data. 2) Perform debouncing on the preprocessed data stream that has passed the validity check to avoid false triggering caused by momentary jitter or short-term interference. The debouncing methods include, but are not limited to: The state of the preprocessed data stream is compared within multiple consecutive sampling periods. Only when the state of the preprocessed data stream remains consistent for a preset number of consecutive times or a preset time period is the state change of the preprocessed data stream considered valid. Specifically, through the above anti-shake processing, it can be ensured that the data entering the rule calculation stage reflects the actual working condition changes, and the data state changes that are shorter than the anti-shake threshold cannot trigger the rule judgment.

[0027] Specifically, the steps of the filter module in performing alarm suppression and deduplication on the output data of the streaming rule calculation module include: Determine if the current event is a duplicate of a similar event that is already in the Active state; Determine whether the same event has already occurred within the set suppression time window; Determine whether the current event is overridden or masked by a higher-priority event. If any of the above conditions are met, i.e., any judgment result is yes, then no new alarm event will be generated, and only the persistence status or count information of existing events will be updated.

[0028] Specifically, in this embodiment, the filter module can also be used for priority processing in multi-point parallel scenarios. That is, when multiple points and multiple rules are running in parallel, the filter module sorts and processes multiple events triggered simultaneously according to a preset priority. The processing rules include: Set high-priority events to be output first, and set low-priority events to be temporarily suppressed when high-priority events are in the active state; Within the same device or event chain, only the highest priority event is allowed to output alarm information. Other events are allowed to output only after the highest priority event has been processed. Through the priority mechanism, duplicate alarms or alarm storms are avoided in multi-point concurrent scenarios.

[0029] Specifically, the processing by the filter module ensures the accuracy and stability of subsequent event chain diagnostic results.

[0030] The time window module is used to perform event detection on the data processed by the filter module based on the time window and transmit it to the streaming rule calculation module; Specifically, the time window module supports three modes: sliding window, fixed window, and continuous window, to calculate latency. 50ms, supports custom window length (100ms-60s) and sliding step size.

[0031] The streaming rule calculation module is used to perform real-time judgment and generate trigger event objects based on the duration and frequency of data transmitted by the time window module. In a specific embodiment, the specific steps of the streaming rule calculation module in performing real-time judgment and generating trigger event objects based on the duration and frequency of data transmitted by the time window module include: The streaming rule calculation module is based on a streaming SQL rule mechanism, and performs real-time judgment and processing on the data stream according to the condition expressions defined in the multi-level fault database model: When the data transmitted by the time window module meets one or more of the following conditions: condition expression, set duration threshold, or occurrence frequency threshold, the data stream is determined to meet the triggering condition, and the data stream that meets the triggering condition is transmitted to the trigger module as a trigger event object.

[0032] Specifically, the streaming rule calculation module continuously monitors the data stream. For each rule defined in the fault database (i.e., the conditional expression of a node), the streaming rule calculation module performs real-time calculations on the incoming data to determine whether the logical conditions set by the rule are met (e.g., I54.0=1 AND I50.7=0), and whether this met state has lasted for a set time threshold or reached a set number of occurrences. Once both the logical conditions and the duration / number of occurrences thresholds are met, the rule is determined to be valid. Subsequently, the trigger module generates a structured event object (e.g., "Level 1 Event: Protection panel power-on fault triggered").

[0033] Specifically, the triggering rules in the streaming rule calculation module include logical AND / OR operations, interval threshold judgments, and state change detection, among which: In logical AND / OR operations (triggered by combined conditions), a rule can contain multiple signal conditions simultaneously. The streaming computation submodule performs logical combination judgments on these conditions, such as: Logical AND: Signal A is true and signal B is true; Logical OR: Signal A is true or signal B is true; Triggering timing: The rule is valid when the combinational logic is valid as a whole within the time window.

[0034] Interval threshold judgment (analog signal trigger): For analog or continuous numerical signals, determine whether they fall within a specified interval, such as: Current value Temperature value is greater than the set upper limit threshold; Time window: The condition is considered to be met only when the value remains in this range for more than the set time, or when the number of occurrences within the window exceeds the threshold.

[0035] State change detection (edge ​​triggered): Used to determine whether the signal state has changed, rather than to determine the current value itself, such as: From 0 to 1 (rising edge); From 1 to 0 (falling edge); The time window module records the most recent state change, and the streaming rule calculation module determines whether the change meets the rule requirements; the rule is determined to be valid only when the state change meets the preset conditions.

[0036] Specifically, an example of the diagnostic process: Level 1 message fault diagnosis: such as "protection panel power-on fault", judge I54.0=1∧I50.7=0∧Q0.0=0. The judgment result is true when the three conditions of I54.0 state is 1, I50.7 state is 0 and Q0.0 state is 0 are met at the same time. At this time, "protection panel power-on fault" can be diagnosed. Level 2 message fault diagnosis: For the "Power-on relay_0K01 / 0K02 activation fault", Level 1 must be established, and Q0.x=1∧I0.y=0 must be verified. When Q0.x is 1 and I0.y is 0, "Power-on relay_0K01 / 0K02 activation fault" can be diagnosed. Level 3 message fault diagnosis: further subdivided into "contactor engagement abnormality", "control circuit abnormality", etc., corresponding to different downstream components respectively; Level 4 message fault diagnosis: Abnormality of the final stage component or specific contact (such as "Emergency circuit breaker_Q003 tripped"), confirmed as the root cause, output link.

[0037] Output the final fault diagnosis results: forming a top-down path and bottom-up root cause confirmation.

[0038] Specifically, this embodiment realizes dynamic cascading and backtracking of multi-level messages through event chain reasoning; it upgrades traditional point-triggered detection to chain-based diagnosis, forming a causal path.

[0039] The trigger module is used to generate an event stream with event identifiers, i.e., valid event objects, based on the trigger event objects processed by the filter module. An action executor module is used to publish the event stream to the event chain inference module; Specifically, the action executor module can publish the event stream to the event chain inference module, or execute internal system responses (such as reset, block, push message). If the setting is push message or reset, the internal system response will be executed after the rule is triggered.

[0040] The state machine management module is used to maintain the state machine of each message event node. The state machine's state set includes Normal, Active, Recovering, and Recovered. When the underlying event recovers, it automatically triggers the upper-layer node state backtracking to realize a self-healing mechanism.

[0041] Specifically, the state machine management module maintains the state of each event node. When a node is triggered, its state becomes Active; when the fault condition disappears and the data returns to normal, the node state reverts to Normal. When the root cause node recovers its state, the system automatically drives the state reversion of all its parent nodes, achieving automatic fault recovery confirmation without manual intervention. For example, after a fourth-level node recovers, the states of its corresponding third-level, second-level, and first-level nodes will also be updated accordingly, thus achieving automatic fault elimination confirmation without manual reset.

[0042] Specifically, the state machine management module adopts an asynchronous streaming architecture and supports... 3000 rules are executed in parallel, with an average CPU utilization rate of 60% response delay 100ms.

[0043] In a specific embodiment, S5, the specific steps of performing step-by-step reasoning in the multi-level fault database model based on the several levels of event flow until the final event chain diagnosis result is generated and visualized include: like Figure 2 and 3 As shown, top-down cascaded diagnostics include: Based on the primary event flow, query all its subordinate sub-message event nodes from the fault database; The rules corresponding to the sub-message event nodes are activated and judged level by level. For example, after a first-level event is triggered, the system will then judge whether the rules of its associated second-level event are valid, and then the third-level, the fourth-level, until the last-level node with no more subordinate nodes is found. Fault path formation: A fault chain is constructed based on the top-down cascaded diagnostic results, from the symptoms to the root cause. The generated structured fault chains can be published via message buses (such as NATS / MQTT) or displayed directly on a visual interface.

[0044] Specifically, the fault diagnosis results obtained in this embodiment not only include the final level fault information that is finally triggered, but more importantly, they include the complete event chain path (i.e., the fault propagation path), the triggering status of each level of node, the duration, the associated specific point values, and the maintenance suggestions brought out from the fault database.

[0045] This embodiment also proposes an event chain diagnostic system for industrial electrical systems, used to implement diagnostic methods, such as... Figure 4 and 5 As shown, it includes: The fault database management module is used to build a multi-level fault database model based on the fault-signal mapping list, and to update and manage the multi-level fault database model. The data acquisition module is used to acquire real-time data streams from industrial control equipment; The streaming rules EdgeBrain engine is used to generate several levels of event streams for fault diagnosis based on the real-time data stream. The event chain reasoning module is used to perform step-by-step causal reasoning in the multi-level fault library model based on the several levels of event flow until the final event chain diagnosis result is generated. The real-time fault diagnosis and recording module is used to record the event results output in real time by the EdgeBrain engine. The event results output in real time by the EdgeBrain engine include: fault name, level, triggering condition, duration, associated nodes, timestamp, event chain path, and manual handling status.

[0046] Specifically, the EdgeBrain engine enables millisecond-level writes and real-time synchronization, allowing operations and maintenance personnel to view data instantly.

[0047] The historical fault diagnosis record module is used to store all event chain trajectories, diagnostic conclusions and manual intervention records for a long time. Reports can be exported to the upper-level system or the cloud to facilitate equipment health assessment and reliability research. The human-computer interaction interface module provides a visual display and operation entry point, and its main functions include: Real-time display of event chain structure and node status; Rule configuration and parameter adjustment; Historical queries, log export, and report generation; Supports access via PC HMI and Web client; Non-programmers can add or modify rules and trace faults through the interface.

[0048] Specifically, this embodiment manages event chain node information in a structured manner within a fault database and combines this with a visual interface to enable online rule editing, hot parameter updates, and visual display of node status. Maintenance personnel can directly view the triggering conditions, duration, suggested measures, and equipment location for each node level on the interface, achieving WYSIWYG intelligent maintenance. Through visual interaction and real-time link display, the system significantly reduces the reliance of maintenance personnel on programming skills, enabling on-site personnel to directly complete diagnostic analysis and maintenance decisions, improving maintenance efficiency by 3-5 times.

[0049] Specifically, the system proposed in this embodiment achieves top-down cascading judgment and bottom-up root cause confirmation at the edge using a multi-level event chain, streaming rule engine, and fault database. It directly outputs actionable root cause and location information (device number / position / terminal / circuit), with the entire process executed locally. The response time is less than 100ms, independent of the cloud. Simultaneously, maintenance personnel no longer see isolated alarms, but a clear diagnostic conclusion, such as: Fault symptom: Sub-carriage master command reversal fault → Root cause: Sub-carriage #1 device incoming circuit breaker fault → Recommendation: Check and replace the circuit breaker, thus achieving immediate repair upon arrival at the problem location.

[0050] Example 1: Taking an industrial crane as an example, the PLC ladder diagram of the industrial crane is used to deduce the primary fault messages, such as: auxiliary trolley master command reversal fault (I40.3=1 or I40.4=1), I53.2=1 and M10.4=0; secondary faults that may cause the primary fault include: (1) The auxiliary trolley moves forward one gear; (2) Rear limit of the auxiliary trolley - 6K36; (3) Faulty electronic control block of the auxiliary trolley; In a level-two fault, a level-three fault that could lead to a "faulty auxiliary trolley electronic control block" might include: (1) The incoming circuit breaker of auxiliary trolley No. 1 is faulty and the incoming circuit breaker of auxiliary trolley No. 2 is faulty; (2) Faulty auxiliary trolley brake circuit breaker; (3) The main device relay of the auxiliary vehicle is faulty; (4) The backup device relay of the auxiliary trolley is faulty; After identifying the first, second, and third level faults, configure the corresponding rules in this system: In the fault database, with "Sub-vehicle Master Command Reverse Fault" as the first-level node, the following hierarchy is defined: Level 1 message event node (appearance layer): Node name: Sub-vehicle master command reverse failure; Triggering condition: (I40.3=1 OR I40.4=1) AND (I53.2=1) AND (M10.4=0); Note: The controller was detected to be in reverse gear, but the electronic control feedback signal did not activate. Second-level message event node (middle layer): 1. The auxiliary trolley is abnormally moving forward in one gear: I40.2=1 and Q0.3=0; 2. Sub-car rear limit _6K36 trigger: I53.6=1; 3. Sub-carriage electronic control block fault: M11.0=1 or Q0.5=0; Level 3 message event node (root cause layer): 1. Fault in the incoming circuit breaker of auxiliary trolley #1: I10.1 = 0; 2. Fault in the incoming circuit breaker of auxiliary trolley #2: I10.2 = 0; 3. Auxiliary trolley brake circuit breaker fault: I12.3=0; 4. Faulty main unit relay in the auxiliary vehicle: Q0.2=0 and M15.0=0; 5. Backup device relay fault in auxiliary trolley: Q0.3=0 and M15.1=0; Diagnostic process: 1. Data Acquisition: Acquire PLC signal streams and send them to the streaming rule engine; 2. First-level judgment: Check whether the conditions at the representational level are met; 3. Second-level matching: Query the multi-level fault database model and judge the conditions of the lower-level nodes level by level; 4. Level 3 Reasoning: Execute root cause node determination and output event chain; 5. Alarms and Visualization: Generate "Level 1 → Level 2 → Level 3" paths and suggestions; Output the corresponding fault information: { "Fault": "Fault caused by the secondary driver instructing the vehicle to reverse"; "Cause": "Faulty incoming circuit breaker for Units 1 and 2"; "Level": "Level 3"; "Time": "2025-11-07T15:30:21"; "Suggestion": "Check the auxiliary trolley incoming line circuit breaker"}.

[0051] The performance test results for this system application example are as follows: Instance verification environment:

[0052] Specifically, this embodiment employs the self-developed EdgeBrain streaming rule engine and NATS message bus to achieve fully autonomous computation within the edge terminal, from data sampling, rule determination, event reasoning to alarm issuance. All determinations are completed at the edge, enabling closed-loop processing without uploading to the cloud, achieving "local determination - local alarm - local recovery". In x86 platform testing, the system's average diagnostic latency did not exceed 100ms, and CPU utilization was low. 80% accuracy ensures the real-time performance and reliability of fault identification and feedback. It also supports network outage caching and automatic replay mechanisms to guarantee uninterrupted diagnostic chain during network anomalies. Real-time edge-end diagnostics and closed-loop processing are achieved.

[0053] The performance test data is shown in the table below:

[0054] For multi-point concurrent scenarios, such as Scenario 3, the actual operating performance of this system is verified using the Advantech MIC-770 industrial PC as an example. It supports real-time monitoring of over 1500 points, concurrent execution of over 3000 rules, and a data acquisition frequency of 50ms. Figure 6-8 As shown, CPU usage was 81.2%; memory usage was 40.4%; and network I / O bandwidth usage was 9.16 MB / s. This embodiment optimizes the multi-threaded streaming computing architecture and asynchronous event queue scheduling mechanism of the EdgeBrain engine, enabling a single node to stably handle more than 1500 collection points with a high frequency of 50ms collection and more than 3000 rules simultaneously. The system maintains stable CPU utilization even under high concurrency. 85%) and low latency ( (100ms). Simultaneously employing priority-based decision-making and repetitive alarm suppression strategies, this effectively avoids alarm storms and system congestion. This optimization scheme ensures long-term stable operation and high reliability in complex industrial environments (such as cranes, electrical control cabinet clusters, and conveyor systems). It supports stable operation of single-node high concurrency and large-scale rule processing.

[0055] The above application examples demonstrate the originality of the method of this invention and the stability of the system under large-scale concurrency rules, and effectively improve the accuracy of electrical system fault diagnosis.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing event chains in industrial electrical systems based on the EdgeBrain engine, characterized in that, The specific steps include: S1. Establish a fault-signal mapping list containing all potential fault modes, triggering conditions and their logical relationships; S2. Based on the fault-signal mapping list, establish a multi-level fault library model. The multi-level fault library model includes: a manifestation layer, an intermediate layer, a root cause layer, and several message event nodes, which are divided according to root cause logic. The manifestation layer, intermediate layer, and root cause layer form a multi-level event chain structure through several message event nodes. S3. Collect real-time data streams from industrial control equipment; S4. Establish a streaming rule EdgeBrain engine and input the real-time data stream into the streaming rule EdgeBrain engine to generate several levels of event streams for fault diagnosis. S5. Using the aforementioned event streams as input, perform step-by-step causal reasoning in the multi-level fault database model until the final event chain diagnosis result is generated and visualized. The final event chain diagnosis result includes the fault propagation link path, the status of each level of node, and triggering parameters.

2. The method for diagnosing the event chain of an industrial electrical system based on the EdgeBrain engine as described in claim 1, characterized in that, The streaming rule EdgeBrain engine includes: a data preprocessing module, a filter module, a streaming rule calculation module, a time window module, and a trigger module, wherein: The data preprocessing module is used to filter, de-jitter, remove outliers, interpolate and align with timestamps on the real-time data stream, and output the preprocessed data stream. The filter module is used to perform validity screening and debouncing on the preprocessed data stream, and also to perform alarm suppression and deduplication on the trigger event objects fed back by the streaming rule calculation module. The time window module is used to perform event detection on the data processed by the filter module based on the time window and transmit it to the streaming rule calculation module; The streaming rule calculation module is used to perform real-time judgment and generate trigger event objects based on the duration and frequency of data transmitted by the time window module. The trigger module is used to generate an event stream with event identifiers, i.e., valid event objects, based on the trigger event objects processed by the filter module.

3. The method for diagnosing the event chain of an industrial electrical system based on the EdgeBrain engine as described in claim 2, characterized in that, The specific steps by which the streaming rule calculation module performs real-time judgment and generates trigger event objects based on the duration and frequency of data transmitted by the time window module include: The data stream is filtered in real time based on the conditional expressions defined in the multi-level fault database model. When the data transmitted by the time window module meets one or more of the following conditions: condition expression, set duration threshold, or occurrence frequency threshold, the data stream is determined to meet the triggering condition, and the data stream that meets the triggering condition is transmitted to the trigger module as a trigger event object.

4. The method for diagnosing the event chain of an industrial electrical system based on the EdgeBrain engine as described in claim 3, characterized in that, The multi-level fault database model is defined as a three-layer fault diagnosis logic framework, and a multi-level event chain structure is constructed through the parent-child relationship of message event nodes: The message event nodes of the root cause layer are used to describe the specific component failures at the lowest level. The message event node of the intermediate layer is the parent node of the root cause layer event node, and it is used to describe the subsystem or functional module problem that leads to the phenomenon in the upper layer. The message event node of the appearance layer is the parent node of the intermediate layer event node, and it is used to describe the system-level fault symptoms that are ultimately manifested. Each message event node contains the following structured fields: Identification information: Node ID, fault name, and level; Association relationships refer to node topology information: parent node ID and a list of child node IDs; Triggering condition: conditional expression; Criterion parameters: time window type, window length, number of attempts threshold, and suppression conditions; Auxiliary information: Repair recommendations, fault severity level, and handling tools.

5. The event chain diagnosis method for industrial electrical systems based on the EdgeBrain engine as described in claim 4, characterized in that, The specific steps of the data preprocessing module in filtering, dejittering, outlier removal, interpolation, and timestamp alignment of the real-time data stream include: The null and invalid values ​​in the real-time data stream are filtered. Perform debounce and outlier removal on the filtered data; Interpolation processing is performed on data with sampling gaps; Timestamp alignment is performed on data from different acquisition channels.

6. The method for diagnosing the event chain of an industrial electrical system based on the EdgeBrain engine as described in claim 5, characterized in that, The specific steps of the filter module in performing validity screening and debouncing on the preprocessed data stream, i.e., the data associated with the triggering event object, include: 1) Perform a validity check on the preprocessed data stream. The check should include at least the following: Determine whether the preprocessed data stream is within the preset numerical range; Determine whether the preprocessed data stream contains missing, abnormal, or invalid identifiers; Determine whether the timestamps of the preprocessed data stream are continuous and whether they exceed the allowable time deviation range; If the preprocessed data stream meets all the above validity requirements (i.e., all judgment results are negative), then the preprocessed data stream is a valid data stream; otherwise, it is discarded or marked as abnormal data. 2) Perform debouncing on the preprocessed data stream that passes the validity check. The debouncing methods include, but are not limited to: The state of the preprocessed data stream is compared within multiple consecutive sampling periods. Only when the state of the preprocessed data stream remains consistent for a preset number of consecutive times or a preset time period is the change in the state of the preprocessed data stream considered valid.

7. An event chain diagnostic system for industrial electrical systems based on the EdgeBrain engine, used to implement the method of claim 1, characterized in that, include: The fault database management module is used to build a multi-level fault database model based on the fault-signal mapping list, and to update and manage the multi-level fault database model. The data acquisition module is used to acquire real-time data streams from industrial control equipment; The streaming rules EdgeBrain engine is used to generate several levels of event streams for fault diagnosis based on the real-time data stream. The event chain reasoning module is used to perform step-by-step causal reasoning in the multi-level fault library model based on the several levels of event flow until the final event chain diagnosis result is generated. The real-time fault diagnosis and recording module is used to record the event results output in real time by the EdgeBrain engine; The real-time event results output by the EdgeBrain engine include: fault name, level, triggering condition, duration, associated nodes, timestamp, event chain path, and manual handling status.

Images