The Research Of Railway Vehicles Wheel-Flat Real-time Detection Technology

As the important part of the project "The Research and Development of Wheel-flat Detection Platform" which comes from Heilongjiang Province Science and Technology Department, this dissertation demonstrated the overall scheme of wheel-flat detection system based on DSP and developed the corresponding algorithm of wheel-flat detection based on DSP.The wheel-flat is one of the hidden dangers of major train operation accidents, it has a great significance to detect the wheel-flat accurately and rapidly for guaranteeing the train's traffic safety. Thus there is a great practical significance to develop an on-line and real time wheel-flat detection system.Firstly the thesis choosed the vibrational acceleration detection method to detect wheel-flat through analyzing and comparing various existing dynamic detection methods at home and abroad. According to the principle of vibrational acceleration detection method and the requirements of the system design specifications, the thesis demonstrated the scheme of wheel-flat detection system based on DSP in detail. In accordance with the working flow of wheel-flat detection system based on DSP, the algorithm of wheel-flat detection based on DSP was given and expounded detailedly. In the part of algorithm development, the thesis firstly analysed the existing nonstationary time-frequency analysis methods which were applicable to wheel-flat detection, based on the nonstationary characteristics of short lasting time and fast mutation for wheel-flat signal collected by vibration accelerometer. Through analysis and comparation time-frequency analysis method based on S-transform was proposed for the first time to detect and locate vibration signal with wheel-flat, and the superiority of S-transform compared with the presented methods was verified by simulation. Afterward, the thesis systematically expounded the whole algorithm of wheel-flat detection including signal pretreatment, characteristics extraction and location detection. In the part of characteristics extraction algorithm, the thesis mainly used threshold judgement methods and S-transform detection algorithm, and the method of short-time energy estimation was proposed in the threshold judgement. Simulation results showed that by short-time energy estimation the algorithm could prevent undetection and misjudgment efficiently. Finally, the detection algorithm's realizations on DSP were also provided, mainly including algorithm for AD acquisition and communication between DSP and ARM. In the end, the thesis accomplished the debugging on the board of DSP in the CCS emulator environment and the whole detection platform's united debugging. The debugging results have manifested that the performance indexes in every respect of the system are all achieved the system design specifications, and the validity of the algorithm and the reliability and stability of the system are all verified through a great deal data experiments for a long time.