Road authorities make significant investments in the planning of maintenance, repair, and rehabilitation of concrete bridges. In order to extract the optimal output in the form of good management decisions with least resources, a bridge management system (BMS) is essential. In BMSs, decisions regarding frequency of maintenance, conducting repairs and rehabilitation are based on inspection data collected for bridges using a condition rating manual. At present, deterioration caused by service conditions and deferred maintenance of old bridges are diagnosed using a condition monitoring system where a condition rating is given to each and every component based on visual inspection. Evaluating these conditions to arrive at a meaningful decision criterion is a challenge faced by many road authorities in the world. Currently most decisions are made considering the data collected for a given year, which is essentially a reactive decision making process. Significant advantages will be available if the data can be used to forecast the future behaviour of the bridge components. In view of this, an attempt has been made to develop deterioration models for existing bridges using Markov chain and artificial neural network methods to compare the suitability of these two methods. Whilst there have been many different methods proposed in the literature to predict future condition from existing condition monitoring data, they have not been widely accepted. These methods include the Markov process, the Gamma process, and deterministic methods, where a condition curve is derived from a large amount of discrete condition data. In this research, an attempt has been made to use two of the most popular methods, the Markov process and artificial neural networks (ANNs), to forecast deterioration using condition data from level 2 inspections by VicRoads. Visual inspection data has been sourced from the road transport authority of Victoria, Australia (VicRoads) to derive the models. In the development of the ANN model, seven parameters were considered, namely age, AADT (Average Annual Daily Traffic), commercial vehicle ratio, environmental exposure, length, width and number of spans. Six parameters were established by evaluating the significance of the outputs. The seventh parameter, exposure category, was included based on engineering judgement. Back Propagation Multi-Layer Perceptron (BP MLP) was used to develop the forecasting model. The developed model was shown to offer reasonable accuracy through the process of validation. A Markov process model was initially developed for the full set of data for a given bridge component. To improve the accuracy three different methods of calibration of transition matrices were adopted: percentage prediction, non-linear optimisation and Bayesian Monte Carlo approach. Bayesian method offered the best accuracy. Combining the data clustering based on ANN to improve the Markov based models was the innovation in the presented work. Finally, a decision making method for optimised management of bridge structures using the outcome of the Markov model is developed. The method offers decision making considering a threshold for rehabilitation as well as optimised allocation of available funding.
展开▼
