Quantifying Uncertainties in Performance Deterioration Modeling with Practical Considerations in Transportation Asset Management

Performance modeling plays a major role in transportation asset management systems. It is employed to forecast asset deterioration and hence support decision making processes involving future conditions of assets. However, current performance modeling does not provide a measure of reliability associated with the prediction. This research proposes the use of multilevel Bayesian regression modeling for mixing prior knowledge with experimental observations in order to develop deterioration modeling with the ability to quantify the associated reliability. It is demonstrated that such an approach, not only provides measures of variability for the model parameters and the predicted performance but, it is also capable of addressing a range of practical limitations. The methodology can explicitly consider materials properties, structural capacity (or strength), external loading and environmental effects in the model by adapting classical mechanistic models.;Three case studies were used to demonstrate the applicability of the methodology and present varied approaches for dealing with some of the practical limitations. In the first case study, the goal was to demonstrate the validity of the methodology. It was shown that the Bayesian regression modeling can reproduce the results of Uzan's rut depth progression model based on the AASHO Road Test data.;The second case study focused on developing a reliable performance prediction model in the presence of very limited time series data. The use of apparent age is proposed to develop an initial performance model from limited data for the Costa Rica national road network.Two separate multilevel models were analyzed: (a) one for estimating the AASHTO pavement layer strength coefficients and (b) another one for estimating the probabilistic performance deterioration model.;Another case study using the New Brunswick road network adapts Paterson's pavement roughness model to address practical limitations such as missing data, incorporating expert criteria and predictors belonging to different data structures. A sensitivity analysis found that the environmental factor m plays a very significant role in the performance of roads in New Brunswick.;The multi-level Bayesian regression modeling can be used in trade-off and optimization decision making tools when allocating limited investment resources to produce planning decisions with the associated reliability information.