Development of vision based fault diagnosis system with real time fuzzy automata for railways

Railway vehicles are widely used in our country as well as in the world for passenger and cargo transportation. With the development of high-speed railway vehicles in recent years, the safety of rail lines has become very important. The components of the railway line must be inspected at regular intervals in order to ensure continued safety of transport. In this thesis study, image processing based methods using real time, fuzzy automata structure has been developed in order to diagnose failures in railway tracks. Within the scope of the thesis firstly complex fuzzy automata structure has been put forward and its applicability has been achieved. Image processing based contactless methods were then developed to diagnose failures such as possible wear, fracture, cracking or undulation in track surfaces, joints, transverses and other components on rail tracks, and experimental results are given. Scientific contribution and innovation are presented in three main points within the thesis. First, the real-time feasibility of fuzzy automata in the thesis has been researched and complex fuzzy automata have been developed. Then comparative results were obtained on image processing application. Advantages of complex fuzzy logic automata have been explored. Secondly, there are built-in methods that can be implemented on a train that can be applied in real time to diagnose faults in track surfaces and components in the thesis and run at a speed of about 100 km/h. Thirdly, techniques using thermal image for detection and condition monitoring of some errors on the rail in the thesis are presented and verified on experimental images. As a result, in this doctoral thesis, new methods for detecting most failures on the rail were suggested using fast imaging cameras and their performance was verified by experimental results. The studies carried out within the scope of the thesis were supported by the research project of TUBITAK 1001 numbered 114E202 and FUBAP PhD thesis research project of numbered MF.16.65.