Long-term Prediction of Bridge Element Performance Using Time Delay Neural Networks (TDNNs)

摘要

A bridge is principally designed to have a long service life. However, due to number factors, itrncould fail prematurely, and could cause loss of human life. In order to ensure the optimum bridgernserviceability, systematic asset management is essential for effective decision-making ofrnmaintenance, repair and rehabilitation (MR&R). Systematic asset management can be achieved by arncomputer-based bridge management system (BMS). Successful BMS development requires arnreliable bridge deterioration model, which is the most crucial component in a BMS. Historicalrncondition ratings obtained from biennial bridge inspections are a major resource for predictingrnfuture bridge deterioration via BMSs. However, available historical condition ratings from mostrnbridge agencies are very limited, thus posing a major barrier for predicting reliable future bridgernperformance.rnThis paper presents the progressive research on the development of a reliable bridge deteriorationrnmodel using advanced Artificial Intelligence (AI) techniques. The development is organised in threernmajor steps: (1) generating unavailable past bridge element condition ratings using the BackwardrnPrediction Model (BPM) - this helps to provide sufficient historical deterioration patterns for eachrnelement; (2) predicting long-term condition ratings based on the outcome of Step 1 using TimernDelay Neural Networks (TDNNs); and (3) improving long-term prediction accuracy of Step 2 byrnemploying Case-based Reasoning (CBR). This paper mainly focuses on the first two steps of thernresearch. Promising results are reported for the reliable long-term prediction of bridge elementrnperformance.