Identifying biases in deterioration models using synthetic sewer data.

Area of Science:

• Environmental Engineering
• Asset Management
• Infrastructure Modeling

Background:

• Assessing sewer deterioration models is hindered by a lack of reliable data, complicating model selection.
• Existing models require robust validation, which is difficult without comprehensive real-world datasets.

Purpose of the Study:

• To develop and utilize a network condition simulator (NetCoS) for generating synthetic sewer data to evaluate deterioration models.
• To investigate the impact of condition classification errors and sewer system age on model performance.

Main Methods:

• Generation of synthetic sewer network data using NetCoS, incorporating pipe deterioration, replacement, network expansion, and classification errors.
• Calibration of two sewer deterioration models (Markov and cohort) using the synthetic data.
• Comparative analysis of model performance and parameter estimation under varying conditions.

Main Results:

• Classification errors significantly bias parameter estimation in the Markov model, whereas the cohort model demonstrates greater robustness.
• Young sewer systems with few pipes in poor condition states strongly influence the parameter uncertainties of the cohort model.
• The Markov model shows less sensitivity to the age and condition state of sewer systems compared to the cohort model.

Conclusions:

• Synthetic data generation via NetCoS provides a valuable method for understanding sewer deterioration model strengths and weaknesses.
• Model selection should consider the potential impact of data quality (classification errors) and system age.
• The cohort model is more sensitive to system age and condition distribution, while the Markov model is more susceptible to classification errors.