Application of Influence Diagram and Sequential Hypothesis Testing Method in Bridge Maintenance Decision Making

Bridge inspection is important in making maintenance or reconstruction decisions. Sometimes a single inspection fails to accurately reflect the actual condition, and additional inspections are needed. The paper addresses this issue from the perspective of minimizing the potential loss, that is, balancing the financial consequences of inspecting to obtain additional information and the financial consequences of making a decision with inadequate information. An influence diagram is a tool used for decision support. The research described uses the framework of an influence diagram to build a bridge inspection and maintenance model. The sequential hypothesis testing method is a statistical approach used to estimate the expected cost of the whole decision process. In this paper, it is used to calculate the potential loss for the influence diagram model. INTRODUCTION There are more than 575,000 bridges over 20 feet in length in the United States recorded in the existing National Bridge Inventory (NBI) (1). Inspecting, maintaining, and repairing these bridges can be time -consuming and costly. More importantly, if an agency cannot detect an unsafe condition or does not make good decisions, the resulting price is perhaps not merely a waste of funds, but the loss of life. The detection of construction deficiencies, deterioration, and subsequent damage depends on the inspection devices and methods used. The emergence of new technologies and the invention of better detection implements will narrow the difference between the detected condition and the actual condition. Yet, making good decisions on the time intervals of maintenance and rehabilitation (M&R) activities and the distribution of funds, which is much more complex, needs both expert experience and good data. Errors during data collection are inevitable. Some errors are caused by inexperienced inspectors, while others are caused by inspection devices. Although researchers and engineers have developed some methods for dealing with the effects of uncertainty in data collection, one hundred percent accuracy of the data cannot be