A methodology for developing bridge condition rating models based on limited inspection records

Transportation infrastructure facilities are the essential components for the continuous development of the economic and community well-being of any country. Maintaining such facilities is a difficult task for service authorities and to keep their assets in optimal conditions at all times. To assist with conducting such a task, infrastructure management systems (IMSs) have been developed for effective asset management. The main function of these systems is to minimise total operation costs for service authorities while maximising the benefits for public users. To obtain the right decisions from an IMS, software packages must have high quality asset information for the system's various analytical processes. For an IMS to correctly predict a mixture of future maintenance and repair needs, periodic inspection records are the key resources amongst other information requirements. However, many infrastructure facilities were already in existence long before the IMS technology was developed. Thus, many years of past inspection records for those structures are always lacking. In particular, the lack of such historical records which are required as inputs to IMSs is a very common operational problem in their implementation. The present research was instigated by the prevailing lack of past inspection records available to service agencies for the effective use of IMSs. Among IMSs, a bridge management system (BMS) has been selected for this thesis project to solve the abovementioned problem of historical data record gaps. Similar to IMS, BMS software packages also require various types of datasets to operate. One of the most significant BMS data requirements is the historical bridge element condition ratings forming part of the past bridge inspection records. In cases of insufficient or non-existent of such datasets, BMS software packages are unable to accurately predict future bridge status. This is because historical bridge condition rating data can affect approximately 60% of the BMS analysis modules. In practice, the most common problem faced by most bridge management agencies during the early stages of BMS implementation arises from incongruence of those input information between the bridge agency's existing bridge data and the input requirements of commercial BMS software packages. Although most bridge agencies in the past conducted inspections and maintenance, the format of such bridge inspection records is different from what are required for BMS implementations. These data incompatibilities are the major barriers to form a BMS database and consequently its implementation. This is achieved by utilising the existing incompatible and limited bridge inspection records to produce sufficient amounts of historical condition rating datasets for a BMS database. The proposed condition rating model adopts an Artificial Neural Network (ANN) technique to back-predict the unavailable historical condition rating patterns using a limited amount of existing bridge condition ratings. The ANN-based bridge element condition rating models presented in this thesis are constructed utilising limited datasets, i.e. bridge inspection records, obtained from three different bridge agencies. This thesis report covers three key topics: feasibility study, refined methodology and case studies. The feasibility study given in Chapter 3 evaluates the condition rating prediction accuracy of an ANN technique to facilitate the identification of problems and limitations. The refined methodology given in Chapter 4 helps to enhance the ANN model to cater for the problems and limitations identified in the feasibility study, resulting in the Backward Prediction Model (BPM). The case studies presented in Chapter 5 are carried out to validate the Backward Prediction Model. The proposed BPM provides a holistic perspective of effective BMS implementations and operations and offers significant contributions to solving the current BMS operational problems on the lacking of past inspection records.