Bayesian updating of a prediction model for sewer degradation

Abstract

Sewer degradation is mainly a stochastic process. The future condition of sewers can be predicted using models based on condition states. In The Netherlands, the SPIRIT model is being developed combining expert opinion and visual inspections. In this model the likelihood function of condition states is updated with inspections. A Dirichlet distribution is used to describe ‘subjective’ prior knowledge, i.e. expert knowledge. The results show that the model can be solved analytically reducing calculation time. In addition, the weight of experts and inspections is determined on the basis of prior information and data instead of estimated by subjective expert knowledge.

Keywords:

• asset management
• Bayesian statistics
• Dirichlet distribution
• expert knowledge
• informative prior
• prediction model
• sewer condition
• sewer rehabilitation