Abstract: Currently, there is no complete system for automatic locomotive detection and fault diagnosis technologies. This paper designs a fault detection system based on an expert system and considering the actual conditions of a testing center. The theoretical basis and design methodology of the system are presented, and a diagnostic method based on an expert knowledge base, diagnostic reasoning, and knowledge acquisition mechanisms is proposed, which can effectively meet the requirements for the detection and diagnosis of railway locomotives.
Keywords: fault diagnosis, knowledge base, expert system
1. Introduction
Currently, diesel locomotives are one of the main traction power sources in my country's railway transportation, and their quality directly affects the efficiency of railway transportation organization. Therefore, improving locomotive quality and operational efficiency while reducing maintenance costs has been a crucial part of locomotive maintenance work for decades. This system is a Railway Locomotive Fault Diagnose Expert System (RLFDES). Based on expert systems, this system incorporates research and discussion based on the actual conditions of the testing center, closely integrating fault diagnosis technology with expert system development principles, and combining fault diagnosis practice with computer and network technologies. By linking theory with practice, it has essentially formed a complete and feasible design scheme.
2. Basic characteristics of locomotive malfunctions
Internal combustion locomotives are highly complex power systems, with tens of thousands of components coupled together during operation, resulting in complex external output signals. The locomotive operation process is a complex dynamic and random process, and any observation data at different times is not repeatable. From the perspective of system theoretical characteristics, locomotive faults have the following characteristics.
(1) Hierarchical nature: Tens of thousands of parts of a locomotive work together and the components are coupled with each other, which determines the hierarchical nature of locomotive failures. A failure is composed of multiple levels of causes.
(2) Sudden and gradual: Sudden failures occur instantly, without obvious symptoms, and are difficult to predict, while gradual failures are characterized by gradual progression and partial functional failure, and can be predicted.
(3) Ambiguity: Locomotive malfunctions and symptoms are not one-to-one correspondences and are random. The boundaries between symptoms are ambiguous, and some information is uncertain.
(4) Trend: Most locomotive malfunctions have a trend that changes over time, that is, they develop from minor symptoms to significant symptoms.
The above points are only based on analysis from one aspect. In practical applications, we should take stochastic processes as the starting point and use various modern scientific analysis tools to comprehensively judge the attributes, composition and development of locomotive malfunctions.
3. Railway locomotive fault diagnosis system
3.1 System Introduction
The railway locomotive fault diagnosis system consists of two main modules: a diesel locomotive fault diagnosis system and an electric locomotive fault diagnosis system. A brief explanation of the diesel locomotive fault diagnosis system will follow.
1) Data loading.
This function can transmit data from all the current testing instruments for internal combustion locomotives to the fault diagnosis system via communication.
2) Parameter settings.
The main task was to determine the threshold parameters for diesel locomotive equipment.
3) Fault diagnosis.
By using data detected by testing instruments and expert systems, the performance of locomotive equipment is determined, diagnostic results are obtained, and decisions are made on whether the equipment needs maintenance or repair.
4) Query and print.
Complete the query and printing of test data and fault diagnosis results.
5) Trend analysis.
By setting the conditions for trend analysis, we can gain a comprehensive understanding of the locomotive's operation.
3.2 Composition of the Expert System for Railway Locomotive Fault Diagnosis
Faults in railway locomotive equipment directly affect the safe and economical operation of locomotives and even the entire railway system. Fault diagnosis is a complex and highly experience-based technical task. Locomotive equipment failures have many causes, requiring rapid, effective, and accurate fault identification and timely troubleshooting. Utilizing expert systems for fault diagnosis and providing solutions to assist maintenance personnel in handling incidents and improve the safe and economical operation of locomotives is a specific application of expert systems in railway locomotive fault diagnosis systems.
Figure 1 System structure block diagram
The expert system mainly consists of a knowledge base, reasoning execution mechanism, interpretation mechanism, knowledge acquisition mechanism, and human-computer interface. The system structure is shown in Figure 1.
The knowledge base is a fundamental component of an expert system. The quantity and quality of knowledge it possesses are crucial factors in evaluating the system's performance and problem-solving capabilities. The inference execution mechanism is another essential component. Under certain control strategies, it searches the knowledge base for current fault phenomena based on the information in the search panel, identifying or selecting the corresponding causes and solutions. Users input a set of fault phenomena through a human-machine interface. The inference execution mechanism determines the fault and its cause based on rules in the rule base and information in the real-time database. If the rule is not found in the rule base, the knowledge acquisition mechanism is invoked to generate a new rule from the base.
3.2.1 Knowledge Base
The knowledge base is the core component of an expert system, and its completeness determines the system's capabilities and efficiency.
1. Real-time database
A real-time database is required for online system diagnostics. This database stores real-time monitoring data from the computer system, including operating parameters and switching data of the locomotive equipment. The method for establishing a real-time database is to use PowerBuilder's timers with trigger mechanisms to transform PowerBuilder's built-in Sybase SQL Anywhere database into an active database, periodically refreshing the data within it.
2. Forms of knowledge representation
This system uses production rules to represent knowledge. The general form of a production rule is: P→Q, which means that if P satisfies the condition, then Q may be derived. P represents a set of phenomena that occur when a fault occurs, and Q represents the cause of the fault and the solution derived from the set of phenomena.
For example, the main equipment of a diesel locomotive includes control circuits, main circuits, engines, generators, bearings, transformers, and electric motors. Common generator set faults include bearing overheating, severe generator vibration, generator fire, generator overspeed, abnormal generator operating noise, generator demagnetization, exciter polarity reversal during generator operation, exciter fire, electro-hydraulic converter control failure, and frequency measurement circuit faults. Common transformer faults include insulation damage between core laminations, partial core melting, poor contact between the core and grounding plates, loose core, coil breakage, winding-to-ground breakdown, winding phase-to-phase short circuit, tap changer failure, three-phase voltage imbalance, oil deterioration, oil level rise, oil level fall, overload, no-load, core vibration, and internal poor contact or breakdown. The relationship between faults and phenomena is complex; each fault is accompanied by one or more fault phenomena, and a single fault phenomenon may be caused by multiple faults. All possible phenomena and faults, along with their corresponding rules, are stored in a knowledge base for use by the expert system.
This system summarizes hundreds of rules. Here is a simple example to illustrate how to represent the correspondence between phenomena and faults (see Tables 1 and 2).
The correspondence between phenomena and faults can be expressed in the form of P→Q as follows:
R1:[（X1）→（H1）]
R2:[（X2）→（H1）]
R3:[（X3）→（H1）]
R4:[(X1)(X2)(X3)→(H1)]
R5:[(X1)(X3)(X4)(X5)→(H1)]
R6:[（Y1）→（H3）]
R7:[（Y1）→（H4）]
3. Structure of the knowledge base
As the analysis above shows, using a relational database is suitable for implementing the knowledge base. Large-scale relational database management systems (RDBMS) offer advantages such as large data storage capacity, fast query speed, ease of modification and expansion, and high reliability. This system's knowledge base is implemented using Sybase SQL Anywhere, a database built into PB7.0.
3.2.2 Reasoning Execution Mechanism
1 Fault Reasoning and Algorithms
The fault search algorithm directly affects the speed and accuracy of fault identification. This system proposes functional-level search and signal source transmission path search based on the characteristics of the test object. There is no fundamental difference between the two; the former uses functional modules as the search object, establishing the relationship between module feature data vectors and transmission functions and higher-level modules; the latter uses signal transmission paths as the search and tracing object, optimizing and segmenting faulty paths, and searching for faulty signal sources level by level, thereby locating the fault. This is essentially an object-oriented strategy.
2. Reasoning and transmission of credibility
Because fault diagnosis involves uncertainty, the testing and verification process is essentially a process of revising the reliability of the reasoning conclusion. When positive evidence appears, the reliability of the conclusion increases; when negative evidence appears, its reliability decreases. When the reasoning is not the final conclusion, it becomes an intermediate result and a condition for other reasoning, taking the minimum value.
3.2.3 Knowledge Acquisition Institutions
Knowledge acquisition generally refers to the learning of explicit knowledge, which can be verified, modified, and interpreted. Expert systems typically express these knowledge using rule sets. Rule sets should be continuously collected, organized, and expanded to improve system performance. Therefore, knowledge acquisition must involve discovering new rules based on examples or acquiring new rules from experts, continuously adding them to the knowledge base, and gradually improving the knowledge base. Knowledge (especially personalized knowledge) is usually intuitive knowledge gained through long-term practice. It is difficult to describe and master clearly and accurately, and often lacks universality, certainty, and effectiveness. It requires continuous improvement during application, making it a challenging acquisition process.
There are several learning methods for knowledge acquisition, including rote memorization, question-guided learning, learning from practical examples, analogical learning, and inductive summarization. The first two are most commonly used, involving communication between experts and knowledge engineers, but this is time-consuming. Since advanced knowledge discovery systems with sophisticated learning capabilities have not yet been developed, creating a practical knowledge acquisition tool is the most effective means of knowledge acquisition for expert systems.
4. Conclusion
This system is applied to the North Locomotive Depot of Zhengzhou Railway Bureau, primarily performing four inspection tasks: electrical components, main circuits, pressure waves, and bearings. Taking the electrical components as an example, it includes 12 circuits, designated as LCQ, YC1, YC2, RBC, RD1, RD2, CF, LL, LLC, LC, FL, and RX. Threshold values for different circuits are set based on expert experience to detect faults. For instance, when inspecting the main contacts (LLC) of the excitation contactor for faults, if the detected value is 450, exceeding the maximum threshold of 400, the expert system infers a poor contact fault and takes appropriate measures. Locomotives undergo inspection at the inspection station upon return. Inspection personnel board the locomotive with their testing instruments and proceed to the corresponding equipment locations, performing inspections at pre-set points. The data is temporarily stored in the instrument's memory. Following the general principles of expert systems, the fault diagnosis system has independent knowledge bases and inference engines. The knowledge base can be continuously modified and improved during use, and while its content varies depending on the locomotive model, it maintains a completely identical structure.
The inference engine is completely universal across various vehicle models and can adapt to the expansion and modification of the knowledge base, thus exhibiting excellent adaptability and versatility. In its application at the Locomotive Inspection Center of the Zhengzhou Railway Bureau's North Locomotive and Locomotive Maintenance Section, the system demonstrates advantages such as high reliability and wide applicability.
The innovation of this paper lies in introducing an expert system into the fault diagnosis of diesel locomotives in railway locomotives, applying expert knowledge to improve the reliability of locomotive fault diagnosis, and having a wide range of applications.
References
[1] Sun Zengqi, Zhang Zaixing, Deng Zhidong. Intelligent Control Theory and Technology [M]. Beijing: Tsinghua University Press, 1997.
[2] Wu Quanyuan. Artificial Intelligence and Expert Systems [M]. Hunan: National University of Defense Technology Press, 1995.
[3] Patton PJ, Frank PM, Clark RN. Fault Diagnosis in Dynamics system, Theory and Application[M]. New York: Practice Hall,1989.
[4] Tian Hongping, Zhang Jiadong. Research on fault diagnosis system for locomotive electrical circuits [J]. Internal Combustion Locomotive, 2000, (10): 10-12.
[5] Basu A. A Knowledge Representation Model for Multiuser Knowledge-based System [J]. IEEE Trans Know Data Eng. 1993,5(20):177-189.
[6] He Xingui. Knowledge Processing and Expert Systems [M]. Beijing: National Defense Industry Press. 1990.
[7] Sang Qingbing, Yang Xing, Shi Hui. Development of a general-purpose fault diagnosis expert system based on COM components [J]. Microcomputer Information, 2006, 2-1: 190-192
