The challenges facing the WSDOT Highway Maintenance Program come in many forms, including aging equipment, personnel shortages, and aging assets, as well as underfunding for both preservation and maintenance. The primary objective of this project was to develop a way to predict the performance of important highway assets. This could help the WSDOT Maintenance Division set performance targets that balance available funds, acceptable performance expectations, and maintenance division priorities, potentially preventing the need for expensive reactive maintenance.
Since the mid-1990s, WSDOT Maintenance has been evaluating the effectiveness of its maintenance program through outcome-based performance measures, referred to as level of service (LOS). The objective of this project was to give WSDOT Maintenance the ability to forecast the LOS performance of its assets by creating algorithms that will be used as a basis to develop prediction models. The prediction models can be used to forecast the condition of highway assets based on different performance measures across different maintenance activities and funding levels.
This project focused on six important highway assets for the development of algorithms: culverts, barriers/guardrails, traffic signal systems, ditches, slopes, and shoulders.
To develop the algorithms, the researcher first acquired WSDOT data on the LOS conditions of and expenditures on the six assets. The project also included a two-phase survey of WSDOT professionals to gather their insights into the factors that influence the LOS conditions of the six assets. The factors determined to affect the LOS condition of each asset were identified and ranked. Each asset was found to have more than ten highly ranked factors. As an example, the five most highly ranked factors for culvert maintenance were determined to be hydrological/weather conditions, previous maintenance dates, current LOS, scouring around the culvert/pipe, and material type.
Once the data had been analyzed, the project developed prediction model algorithms for each of the six assets. The algorithms are the first step to developing prediction models. Once the models have been validated and tested, they can be used to predict LOS conditions and trends under various funding levels. To advance the project in developing prediction models, the author recommends applying machine learning techniques for a future study. Multi-dimensional datasets and non-linear variables can be processed by machine learning models, producing more accurate results.
The results from this project will provide WSDOT with a robust platform to improve decision-making for asset management. WSDOT and other states may use the algorithms in developing models to assist in predicting asset conditions, calculating the base funds required for individual assets, and efficiently allocating resources. The results have great promise for enhancing the overall condition of roadway assets.
Report: WA-RD 932.1
Author: Kishor Shrestha, WSU Department of Construction Engineering
Sponsor: WSDOT
WSDOT Technical Monitors:
Kelly Shields
Bruce Castillo
WSDOT Project Manager: Doug Brodin