Power equipment metal material failure analysis intelligent auxiliary system and use method

Abstract

The invention discloses a power equipment metal material failure analysis intelligent auxiliary system and a use method. The system adopts a four-layer architecture of a knowledge source layer, a knowledge storage layer, a knowledge processing layer and a man-machine interaction layer. The knowledge source layer stores information such as macroscopic images, standard specifications and failure cases in a classified manner according to failure modes; the knowledge storage layer guarantees data security and query efficiency through a backup mechanism and multivariate retrieval; the knowledge processing layer identifies a failure mode based on a convolutional neural network and dynamically optimizes an inference model; the man-machine interaction layer provides diversified input and output modes. According to the use method, efficient and collaborative failure analysis is achieved through the processes of information uploading, system analysis, supplementation and perfection, result application and feedback optimization. The system and the method have the advantages that the problems of scene splitting, low efficiency and the like in a traditional mode are solved, and the accuracy and the working efficiency of failure analysis of the metal material of the power equipment are improved.

Description

Technical Field

[0001] This invention relates to the field of failure analysis technology for metallic materials in power equipment, and specifically to an intelligent auxiliary system for failure analysis of metallic materials in power equipment and a method for using the system. Background Technology

[0002] With the continuous development of power grids, the failure of metal components in power equipment has gradually become a prominent hidden danger threatening the safe and stable operation of power grids. The failure of metal components not only leads to an increase in the proportion of unplanned power outages, significantly increasing emergency repair costs, but may also trigger cascading failures, causing serious economic losses and social impacts.

[0003] Accurate and efficient failure analysis of metal components can promptly locate the root cause of the fault, clarify the failure mechanism, and thus provide a scientific basis for the maintenance, repair and upgrading of power grid equipment, effectively reducing the failure rate and improving the reliability and economy of power grid operation.

[0004] However, when conducting failure analysis on metal components, power companies currently mostly adopt a linear model of "sample delivery by grassroots teams - testing by technicians," which has many shortcomings: First, the analysis scenario is fragmented. Experimenters are isolated from the fault site, and the fault information transmitted by grassroots teams is limited. Many key elements cannot be reproduced, which leads to analysis conclusions that are often one-sided and fail to reveal the complete chain of action.

[0005] Second, there is a bottleneck in grassroots capabilities. Grassroots team members have shortcomings in professional knowledge and experience, and face difficulties in describing fault phenomena, selecting sampling locations, and controlling preservation conditions, all of which affect the accuracy of analytical conclusions.

[0006] Third, the process is redundant and inefficient. Traditional processes involve many steps, and cross-institutional coordination and sample transportation consume a lot of time. Moreover, due to the lack of on-site information, technicians often need to communicate repeatedly with grassroots teams to supplement information, which leads to longer analysis cycles, increased resource consumption, and may also result in a "broad-based" experimental strategy, causing resource misallocation.

[0007] Fourth, the standards system is complex. The supervision system for metal components is intricate, and the relevant standards are updated rapidly. Technical personnel face a triple dilemma: information acquisition, clause interpretation, and standard implementation. This leads to an increase in the invalidation rate of test reports, creating a negative cycle.

[0008] Therefore, traditional failure analysis models are insufficient to meet the needs of the rapid development of new power systems, and there is an urgent need to transform to a digital analysis paradigm of "field-laboratory collaboration" to achieve collaborative analysis between the field and the laboratory, thereby improving the efficiency and accuracy of failure analysis. Summary of the Invention

[0009] To achieve the above objectives, this invention provides an intelligent auxiliary system for failure analysis of metallic materials in power equipment, and also proposes a method for using the system.

[0010] The technical solution of this invention: an intelligent auxiliary system for failure analysis of metallic materials in power equipment, comprising a four-layer architecture: a knowledge source layer, a knowledge storage layer, a knowledge processing layer, and a human-computer interaction layer; The knowledge source layer stores the corresponding knowledge source data according to various failure modes. The knowledge source data includes macroscopic photographs, scanning electron microscope images, standards for the composition of material grades, standards for material selection of equipment components, typical failure cases, and standards for experimental testing corresponding to each failure mode. The knowledge storage layer adopts a mechanism that combines regular full backups and real-time incremental backups to back up all data of the intelligent auxiliary system to local disks and cloud storage. It has full-text search, keyword search, and semantic search functions, and improves response speed by optimizing database statements. The knowledge processing layer includes a failure mode recognition module based on a convolutional neural network. This module receives input information and calls knowledge source data from the knowledge source layer. It compares and infers the input information with the knowledge source data to derive the same type of failure mode and generate failure causes and solutions. A knowledge evaluation mechanism is also included to dynamically optimize the inference model through comparison with actual cases and expert review feedback. The human-computer interaction layer features a clear menu and intuitive icon design, supports manual input, voice input, and image recognition input, and displays analysis results in charts and graphs.

[0011] The typical failure cases of the knowledge source layer include fields such as equipment model, material grade, macroscopic image, failure phenomenon, analysis conclusion, and solution.

[0012] The convolutional neural network of the knowledge processing layer is trained with no less than 1,000 sets of failure image samples to form a failure image comparison set.

[0013] The failure modes include corrosion, fracture, and wear.

[0014] Ideally, a full backup should be performed every seven days.

[0015] Furthermore, the voice input supports Mandarin recognition, and the image recognition input has an accuracy rate of ≥85% for identifying invalid areas.

[0016] Preferably, the chart data includes bar charts, line charts, and pie charts.

[0017] A method of using the above-mentioned intelligent auxiliary system for failure analysis of metallic materials in power equipment includes the following steps: S1: Operators upload images and basic information of failed metal materials through the human-machine interface layer. The images are macroscopic photos, and the basic information includes equipment name, years of operation, and material grade. S2: The knowledge processing layer calls the failure mode recognition module based on convolutional neural network, combines the information from the knowledge source layer to generate a preliminary judgment, and feeds back the sampling requirements through the human-computer interaction layer, including the sampling location, quantity, and data to be supplemented, including the operating environment temperature, humidity, and recent load conditions. S3: The operator completes the sampling and supplements the data according to the feedback sampling requirements, and re-uploads it to the intelligent auxiliary system; S4: The knowledge processing layer performs reasoning based on the information uploaded in steps S1 and S3, and generates analysis results that include failure causes, solutions, verification experiment priorities (such as prioritizing component analysis) and reference standards (such as corresponding standard numbers). S5: Technicians review the results and conduct experimental verification, then upload the verification conclusions to the intelligent assistance system through the human-computer interaction layer; S6: The knowledge processing layer updates the knowledge source data and reasoning model based on feedback, improving the accuracy of subsequent failure analysis of metal materials in power equipment.

[0018] Preferably, the response time of the failure mode identification module in step S2 is ≤5 seconds.

[0019] Preferably, in step S4, the priority of the verification experiment is ranked according to the degree of failure impact and the detection efficiency.

[0020] The advantages of this invention compared to existing technologies are as follows: Solving the problem of fragmented scenarios: Through images and real-time data uploaded by operators, technicians can intuitively understand the failure situation on site, and combined with system reasoning, the analysis conclusions are more comprehensive.

[0021] Improved analytical accuracy: Through the system's rich knowledge resources and efficient reasoning and judgment, combined with the accurate identification of failure modes by convolutional neural networks, the failure modes and causes of metallic materials can be determined more accurately, providing a comprehensive and reliable reference for failure analysis and improving the accuracy of failure analysis results.

[0022] Improved analysis efficiency: The diverse input methods and intuitive result display of the human-computer interaction layer, as well as the system's automatic recommendation of verification experiment priority order and related experimental testing standards, reduce the workload of operators and technicians and greatly improve the efficiency of failure analysis.

[0023] Reduced resource consumption: By avoiding the traditional "broad-based" experimental testing strategy, and through precise failure mode identification and experimental testing recommendations, unnecessary experimental testing work is reduced, the consumption of related resources is decreased, and the efficiency of resource utilization is improved. Attached Figure Description

[0024] Figure 1 This is an architecture diagram of the system of the present invention; Figure 2 This is a flowchart illustrating the usage of the system of the present invention. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0026] Example 1 A smart auxiliary system for failure analysis of metallic materials in power equipment, comprising a four-layer architecture: Knowledge Source Layer: Knowledge source data is categorized and stored according to common failure modes of metallic materials in power equipment, such as corrosion, fracture, and wear, forming a structured catalog. The knowledge source data includes: ① Macroscopic photographs (resolution ≥ 1024×768) and scanning electron microscope images (magnification 100-2000x) of failure types; ② Standard numbers and text corresponding to material grades (e.g., Q235 steel) and material components; ③ Standards corresponding to material selection requirements for equipment components such as transformers and circuit breakers; ④ Typical cases (no fewer than 500) including equipment model, material grade, macroscopic images, failure phenomena, analysis conclusions, and solutions; ⑤ Standards corresponding to experiments such as chemical composition analysis and mechanical performance testing.

[0027] Knowledge Storage Layer: A hybrid storage approach is adopted, with structured data (such as standard clauses) stored in a relational database and unstructured data (such as images) stored on a file server. A mechanism combining regular full backups (every Sunday morning) and real-time incremental backups is employed to prevent data loss. Full backups are performed to the local disk, while real-time incremental backups are performed to cloud storage providers (such as Alibaba Cloud). Search functions support: ① Full-text search (e.g., searching for "copper corrosion cases"); ② Keyword search (e.g., entering "GB / T228" to search for tensile testing standards); ③ Semantic search (e.g., understanding "stress corrosion resistance of a certain material" and linking it to relevant standards). By optimizing SQL query statements (e.g., adding indexes), the response time for a single query is controlled to within 2 seconds.

[0028] The knowledge processing layer includes a failure mode recognition module based on convolutional neural networks (such as the ResNet50 architecture). Its internal failure image comparison set is trained from no fewer than 1000 sets of failure image samples. This module can identify failure modes such as corrosion (pitting corrosion, uniform corrosion) and fatigue fracture (fatigue streaks), achieving an accuracy rate of over 90%. The inference engine receives input information and calls upon knowledge source data from the knowledge source layer. Combining this with standards and cases in the knowledge source database, it compares and infers the input information with the knowledge source data to derive the same type of failure mode, generating failure causes (e.g., "stress concentration leads to fatigue fracture") and solutions (e.g., "optimize the structure to reduce stress"). An evaluation and optimization module is included. Through comparison with actual cases and expert review feedback, the inference results are compared with the verification conclusions of technical personnel to calculate the matching degree (e.g., ≥80% is considered acceptable). For unacceptable inference items, the corresponding case sample size is increased, the model is retrained, and the inference model is dynamically optimized. Human-Computer Interaction Layer: The interface features a clear menu and intuitive icons, employing a modular design divided into an information input area and a results display area. Supported input methods include: ① Manual input (device model, operating parameters, etc.); ② Voice input (implemented via Baidu Voice API, supports Mandarin, recognition accuracy ≥95%); ③ Image input (calls OpenCV library to identify failure areas, automatically extracting features such as size and color, failure area recognition accuracy ≥85%). Results display includes: ① Text report (including failure mode, cause, and recommendations); ② Bar chart (showing the probability percentage of various failure causes); ③ Pie chart (showing the distribution of experimental priorities). The usage method of this intelligent assistance system includes the following steps: S1: The operator uploads a macroscopic photo and basic information of the failure of the cable trench fixing hardware in a substation through the human-machine interaction layer, and manually enters: cable trench fixing hardware, operating time 3 years, hardware material is Q235, installed in underground cable well, and there is water accumulation nearby.

[0029] S2: Within 4 seconds, the system calls the convolutional neural network to identify the failure mode as "corrosion failure". Combined with the information from the knowledge source layer, it generates a preliminary judgment and feeds back the sampling requirements (take samples from the severely corroded area, take 3 samples of 10mm×10mm) and the data that needs to be supplemented (the humidity of the recent operating environment and the pH value of the water).

[0030] S3: The operator completes the sampling and supplements the data according to the feedback sampling requirements (the humidity of the recent operating environment is above 72%, and the pH of the water is 6.0), and then re-uploads it to the system.

[0031] S4: The system generates the following analysis results: The cause of failure may be "acidic water accumulation and excessive humidity in the operating environment leading to electrochemical corrosion of the fittings"; the solution is "to remove the accumulated water, reduce the humidity in the operating environment, and replace the fittings with epoxy anti-corrosion coating"; it is recommended to prioritize "energy dispersive spectroscopy analysis (refer to GB / T 17359-2023)".

[0032] S5: Technicians review the analysis results generated by the system and conduct relevant experimental verification on the failed samples to confirm that the failure was indeed caused by corrosion due to excessive humidity in the operating environment. The verification conclusions are then uploaded to the system through the human-machine interface layer.

[0033] S6: The system updates the knowledge source database and reasoning model based on feedback.

[0034] Example 2 An intelligent auxiliary system for failure analysis of metallic materials in power equipment, with the same system architecture as in Example 1.

[0035] The usage method of this intelligent assistance system includes the following steps: S1: The operator uploads a macroscopic photo and basic information about the failure of the capacitor unit connection line in a substation through the human-machine interface layer. The operator manually enters: capacitor, operating time 5 years, connection line material is T2 copper, and the fastening nut is obviously loose.

[0036] S2: Within 3 seconds, the system calls the convolutional neural network to identify the failure mode as "high temperature fuse failure". Combined with the information from the knowledge source layer, it generates a preliminary judgment and feeds back the sampling requirements (sampling at the fuse point and adjacent conductor segments, taking 3 samples) and the data to be supplemented (load conditions in the 30 minutes before the fuse failure) through the human-computer interaction layer.

[0037] S3: The operator completes the sampling and supplements the data according to the feedback sampling requirements (the load increases to 1.5 times the rated value 30 minutes before the circuit breaker is triggered), and re-uploads it to the system.

[0038] S4: System-generated analysis results: The cause of failure may be "loose fastening nuts leading to excessive contact resistance, causing overheating and melting under excessive load"; the solution is "replace the capacitor unit connection wire of the same model and tighten the new connection wire strictly according to the standard torque"; it is recommended to prioritize "resistance test (refer to GB / T 3048.4-2007)" and secondarily "composition analysis".

[0039] S5: Technicians review the analysis results generated by the system and conduct relevant experimental verification on the failed sample. They confirm that the failure is indeed a high-temperature melting failure caused by excessive contact resistance and upload the verification conclusion to the system through the human-computer interaction layer.

[0040] S6: The system updates the knowledge source database and reasoning model based on feedback.

[0041] Example 3 A method for using the aforementioned intelligent auxiliary system for failure analysis of metallic materials in power equipment includes the following steps: S1: The operator opens the system at the fault site, takes a macroscopic photo of the metal material failure and the fracture surface of the circuit breaker contacts with a mobile phone, and uploads it to the system; at the same time, the operator voice inputs "Equipment model: LW36-126, service life: 8 years", and the system automatically converts it into text.

[0042] S2: The system completes the analysis within 5 seconds and initially judges it to be a fatigue fracture failure. The feedback through the human-machine interface layer is: "The sampling locations are 3 points on the edge of the fracture; vibration frequency data of the equipment operation in the past 3 years need to be supplemented."

[0043] S3: The operator takes samples as required, records the ambient temperature with a thermometer, adds the vibration frequency, and then uploads the data again.

[0044] S4: The system inference results show: "Failure mode: fatigue fracture; Possible cause: external vibration frequency is close to the natural frequency of the circuit breaker contact system (probability 70%); Solution: optimize the installation structure to avoid resonance; Recommended experiment: prioritize fatigue life test (refer to GB / T 3075-2021), then perform metallographic analysis", with a bar chart showing the percentage of causes.

[0045] S5: Technicians conduct relevant experiments based on system recommendations, confirm the cause of failure matches, and mark "verification passed" in the system.

[0046] S6: The system adds the case to the knowledge source database, updates the correlation weight of "resonance and fatigue fracture failure" in the reasoning model, and improves the accuracy of subsequent analysis.

Claims

1. An intelligent auxiliary system for failure analysis of metallic materials in power equipment, characterized in that... It includes a four-layer architecture: knowledge source layer, knowledge storage layer, knowledge processing layer, and human-computer interaction layer; The knowledge source layer stores the corresponding knowledge source data according to various failure modes. The knowledge source data includes macroscopic photographs, scanning electron microscope images, standards for the composition of material grades, standards for material selection of equipment components, typical failure cases, and standards for experimental testing corresponding to each failure mode. The knowledge storage layer adopts a mechanism that combines regular full backups and real-time incremental backups to back up all data of the intelligent auxiliary system to local disks and cloud storage. It has full-text search, keyword search, and semantic search functions, and improves response speed by optimizing database statements. The knowledge processing layer includes a failure mode recognition module based on a convolutional neural network. The failure mode recognition module can receive input information and call the knowledge source data from the knowledge source layer. It compares and infers the input information with the knowledge source data to obtain the same type of failure mode and generate the failure cause and solution. A knowledge assessment mechanism is in place to dynamically optimize the reasoning model through comparison with real-world cases and expert review feedback. The human-computer interaction layer features a clear menu and intuitive icon design, supports manual input, voice input, and image recognition input, and displays analysis results in charts and graphs.

2. The intelligent auxiliary system for failure analysis of metallic materials in power equipment according to claim 1, characterized in that, The typical failure cases of the knowledge source layer include fields such as equipment model, material grade, macroscopic image, failure phenomenon, analysis conclusion, and solution.

3. The intelligent auxiliary system for failure analysis of metallic materials in power equipment according to claim 1, characterized in that, The convolutional neural network of the knowledge processing layer is trained with no less than 1,000 sets of failure image samples to form a failure image comparison set.

4. The intelligent auxiliary system for failure analysis of metallic materials in power equipment according to claim 1, characterized in that, The failure modes include corrosion, fracture, and wear.

5. The intelligent auxiliary system for failure analysis of metallic materials in power equipment according to claim 1, characterized in that, A full backup should be performed every seven days.

6. The intelligent auxiliary system for failure analysis of metallic materials in power equipment according to claim 1, characterized in that, Voice input supports Mandarin recognition, and image recognition input has an accuracy rate of ≥85% for identifying invalid areas.

7. The intelligent auxiliary system for failure analysis of metallic materials in power equipment according to claim 1, characterized in that, Chart data includes bar charts, line charts, and pie charts.

8. A method of using the intelligent auxiliary system for failure analysis of metallic materials in power equipment as described in any one of claims 1-7, characterized in that, Includes the following steps: S1: Operators upload images and basic information of failed metal materials through the human-machine interface layer. The images are macroscopic photos, and the basic information includes equipment name, years of operation, and material grade. S2: The knowledge processing layer calls the failure mode recognition module based on convolutional neural network, combines the information from the knowledge source layer to generate a preliminary judgment, and feeds back the sampling requirements through the human-computer interaction layer, including the sampling location, quantity, and data to be supplemented, including the operating environment temperature, humidity, and recent load conditions. S3: The operator completes the sampling and supplements the data according to the feedback sampling requirements, and re-uploads it to the intelligent auxiliary system; S4: The knowledge processing layer performs reasoning based on the information uploaded in steps S1 and S3, and generates analysis results including failure causes, solutions, verification experiment priorities and reference standards. S5: Technicians review the results and conduct experimental verification, then upload the verification conclusions to the intelligent assistance system through the human-computer interaction layer; S6: The knowledge processing layer updates the knowledge source data and reasoning model based on feedback, improving the accuracy of subsequent failure analysis of metal materials in power equipment.

9. The method of use according to claim 8, characterized in that, The response time of the failure mode identification module in step S2 is ≤5 seconds.

10. The method of use according to claim 8, characterized in that, In step S4, the priority of the verification experiment is ranked according to the degree of failure impact and detection efficiency.

Images