Degradation Modelling by Markov Chain Monte Carlo Simulation with Application to Railway Bridge Condition Prediction 【2016.11.10 10:30am, S509】 |
【大 中 小】【打印】【关闭】 |
2016-11-8
Colloquia & Seminars
Speaker |
Tieling Zhang, University of Wollongong |
Title |
Degradation Modelling by Markov Chain Monte Carlo Simulation with Application to Railway Bridge Condition Prediction |
Time |
11月10日星期四10:30 - 11:30AM |
Venue |
数学院思源楼S509 |
Abstract |
In a large city, or a railway or highway network, there are a great number of bridges suffering from degradation process. To maintain these bridges in good operation status, it costs a huge amount of money in every year. Therefore, it is a need to predict the future condition of bridges in order to make an optimised plan for maintenance service. Given the situation, this present topic is concerned with degradation modelling for predicting the future conditions of railway bridge elements. In this study, three existing approaches to deterioration modelling namely Regression based Nonlinear Optimization (RNO), Bayesian Maximum Likelihood (BML) and Markov Chain Monte Carlo (MCMC) simulation with Metropolis-Hasting Algorithm (MHA) were examined to account for missing inspection data in history. In order to group the bridges, contribution factors to rail bridge deterioration were identified from expert opinions and previous studies. Bridge inspection data over 15 years of 1000 Australian railway bridges were collected from a main industrial partner in Australia and reviewed. The transition probability matrixes (TPMs) of Markov chain model were estimated using the above mentioned three approaches for bridge deck transoms with similar characteristics. Network level condition state prediction results are tested by using statistical hypothesis tests to validate the suitability of the developed deterioration models and to compare the performance of different methods. Although the estimated TPMs are slightly different by these three approaches, the MCMC simulation with MHA shows the best performance with lowest Chi-square value. Therefore, it is verified that MCMC method is capable of generating more accurate TPMs compared to the other two methods for condition prediction under the condition that only limited historical inspection data are available. |
Affiliation |
Dr. Tieling Zhang is serving as deputy director of Engineering Asset Management Group in the Faculty of Engineering at the University of Wollongong (UOW). He is a technical leader to drive several research projects funded by Energy Pipeline CRC and CRC for Railway Innovation, Australia. Before he joined UOW, he was a reliability specialist working in Vestas Technology R&D Singapore where he was responsible for system reliability in new wind turbine system development. Before that, he also held a few other research positions in university and other industry. He received a PhD degree in system engineering from Tokyo University of Marine Science and Technology in 2001. He has very strong expertise in data processing and modelling, system reliability engineering, condition based prediction, process simulation and optimisation. He published over 80 research articles in the related fields and completed more than 15 major research projects. He is an invited referee serving for over 20 international journals and, reviewer/technical team member for many international conferences. He holds 5 patents and 7 others published and pending for grant. He is teaching subjects in systems engineering and supervising PhD students who are doing research in the related fields. He received two best paper awards. |
|
|