Combined evidence model to enhance pavement condition prediction from highly uncertain sensor data

Despite the significant advancements in data collection technology and the increased use of automated data collection systems, pavement condition assessment mainly relies on ground-based monitoring. The increasing availability of low-cost sensors and remote sensing technologies provide opportunities to explore more efficient approaches. These data, however, are often characterized by high uncertainties. This study explores the application of Evidence Theory to incorporate highly uncertain sensor data. The capabilities of the proposed approach are assessed in a case study comparing the estimated pavement condition derived from the proposed Evidence Theory approach and the traditional approaches of Markov Deterioration Process and the Bayesian approach proposed in the Partially Observable Markov Decision Process. This paper also explores how to incorporate aspects such as reliability and conflict in the combination of multiple sensors data and discusses the impacts on the results of various parameters and methods in the application of Evidence Theory. The results show that the proposed Evidence Theory approach produces errors that are 44% lower than any of the other methods. The sensitivity analysis exploring different treatments of conflict and the introduction of reliability measures show the importance of an adequate calibration of Evidence Theory parameters in real life applications.

Automated summary

This study explores the application of Evidence Theory to incorporate highly uncertain sensor data, showing that this method produces errors that are 44% lower than any of the other methods. Adequate calibration of parameters is crucial in real-life applications.