| In recent years, with the rapid development of science and technology, social progress, growth of national economy, improvement of people's awareness of traffic and advances of railway technology, there is no doubt that more and higher requirements are going to be with the construction and maintenance of railway lines, and this has become a difficult problem to solve. The successful introduce of Tamping Machine in our country made these difficult problems into reality and opened a new era of railway automation maintenance. Tamping Machine, as the heart of railway maintenance, a long-term operation in harsh environments will definitely increase the probability of the failure of equipment parts. Only Fault Diagnosis Expert System could solve these problems comprehensively. The most important part of Fault Diagnosis Expert System is the implementation of fault diagnosis. By the early stage of manual analysis, lately into single neural network machine learning and Fault Diagnosis Expert System after that, predecessors accumulated a lot of experience. Our system uses Tamping Machine Fault Diagnosis Expert System based on the improved BP neural network, neural network and expert system are combined to solve fault diagnosis problems, the improved BP neural network uses Steepness Factor to reduce the error of function value, so that convergence rate raises and improves diagnosis rates. Fault The design of Diagnosis Expert System has a very active significance for Tamping Machine fault diagnosis domestic and foreign.Firstly, we illustrate domestic and foreign present situation of Tamping Machine, propose the main type of Tamping Machine such as GD08-32, GD09-32 and GWD-320, analysis the main failure types, point out its deficiency according to traditional fault diagnosis program, and put forward the design of remote fault diagnosis Expert System based on Web, in which we propose Inference Enginen Machine Learning?Interpreter and Expert System with neural network knowledge base as the core. The design of Fault Diagnosis Expert System based on Wed ensures timeliness and effectiveness of fault diagnosis, greatly improve the efficiency and offers a lot of convenience to the front-line staff.Secondly, on the basis of BP neural network's principle and characteristics and analysis the importance of knowledge representation and reasoning in fault diagnosis, do a further analysis of BP neural network algorithm, point out the shortcomings of traditional algorithm. The method of Gradient descent and momentum factor both speed up convergence rate but will result in local minimum problem. Therefore, we propose method of improved activation function?method of self-adaption of study speed?method of optimizing network structure?method of improved error function, the four algorithm improvements according to present experience. And at the same time, we design Web fault diagnosis Expert System in detail on the aspect of Server?Database Server?Database?Client technology. Security Certificate technology?Inference Engine technology. Knowledge Base technology.Finally, by introducing the concept of steepness factor, we improve BP neural network algorithm. The introducing of steepness factor could effectively improve the convergence rate according to algorithm. Compared with the traditional 1200 times iterations algorithm, the improved algorithm stops training when it reaches 40 times iterations, the error value turns to 0, meets the convergence requirements, greatly enhance diagnosis rate. Also, through system simulation, we input 11 sets of values and 13 sets of values to neural network training and analysis the results, proving the correctness and practicality of improved BP neural network algorithm design.Various test data show that the improved BP neural network algorithm and Tamping Machine Fault Diagnosis Expert System based on Web meet the requirement of practicability, instantaneity, accuracy and security, which is the best choice in Tamping Machine Fault Diagnosis field. |
PDF Full Text Request