Every year hundreds of millions of dollars are allocated towards the maintenance of pavements in the commonwealth. Therefore, even the slightest improvement in budgeting management can make a significant difference.
What it does
Optimized road maintenance planning seeks solutions that minimize the life-cycle cost of a road network while maximizing the pavement quality. We propose a solution to find the best pavement schedules for different budgets. The formulation of the optimization model is carried out in such a way that a cost-effective maintenance strategy is reached by preserving the performance level of the road network at a desirable level.
How we built it
For demonstration purposes, the tailored multi-objective meta-heuristic algorithm is applied to 1000 road segments in the commonwealth. Data is obtained from SmarterRoads  collected yearly by VDOT. Four variously priced maintenance projects are defined based on the level of repair needed for the segment. Results are shown for an 8-year pavement maintenance schedule. The final timetable can be chosen based on what amount of budgeting is available and is optimized to maintain the highest pavement quality level within the given 8-year timespan. One instance is shown in TABLE 1. Pavement condition for each year is presented in TABLE 2 if this maintenance schedule is chosen. Moreover, the proposed model is robust in transferring learnings from various resources to replace the current data logging system.
Challenges we ran into
Image dataset for the classification of pavement quality is not currently available. Adding these data sources will increase the model's accuracy.
Accomplishments that we're proud of
Development of a robust multi-objective meta-heuristic algorithm that automatically schedules pavement maintenance. Moreover, a transfer learning model for pavement image classification is developed.
What we learned
The great potential of Machine learning algorithms in traffic applications to save tremendous amounts of money.
What's next for SmartPave (Smart systematic pavement management)
The model is robust and can be applied to a crowdsourced database for a more up to date and recurrent scheduling. Moreover, the proposed model is robust in transferring learnings from various resources to replace the current data logging system. The data collection section of the model will be strengthen using other data sources such as Mendeley data  for pavement conditions of different materials, or pictures and sensor data  collected from smartphones. This will greatly reduce and with enough participation eliminate the need for yearly assessments performed by VDOT.
 VDOT, "SmarterRoads dataset," [Online]. Available: http://smarterroads.org/dataset. [Accessed 27 04 2018].  "Mendeley Datasets," [Online]. Available: https://data.mendeley.com/. [Accessed 27 04 2018].  K. Gopalakrishnan, S. K. Khaitan, A. Choudhary and A. Agrawal, "Deep Convolutional Neural Networks with transfer learning for computer vision-based data-driven pavement distress detection," Construction and Building Materials, vol. 157, pp. 322-330, 2017.