Thesis Defence: Minimizing road construction costs from corridor selection to earthwork
August 1 at 10:00 am - 2:00 pm
Paavai Manimaran Vanjeenathammal, supervised by Dr. Yves Lucet, will defend their thesis titled “Minimizing road construction costs from corridor selection to earthwork” in partial fulfillment of the requirements for the degree of Master of Science in Computer Science.
An abstract for Paavai Manimaran Vanjeenathammal’s thesis is included below.
Defences are open to all members of the campus community as well as the general public. Please email yves.lucet@ubc.ca to receive the Zoom link for this defence.
Abstract
The optimization of road networks for wind farms presents unique challenges due to complex terrain and construction cost constraints. Traditional approaches often rely on manual alignment selection, which is time-consuming and may result in suboptimal solutions. This research extends road design optimization from a single alignment to a full network connecting multiple turbines to a common access point.
We propose a three-phase optimization framework, called TriPhase model, to generate a cost-efficient and terrain-feasible road network. The first phase involves corridor selection, where the terrain is discretized into a graph. A Steiner minimum tree formulation identifies a minimum-cost network while enforcing gradient and curvature constraints. This structured approach ensures an optimal corridor layout across the terrain.
In the second phase, we loop over all road segments and optimize each horizontal alignment using a bilevel model that integrates both horizontal and vertical alignment components for each road segment. This formulation is adapted from an existing bilevel optimization model where the upper level refines road geometry and the lower level solves a mixed-integer linear program to minimize earthwork cost. To ensure compatibility with the Gurobi solver, the model is reformulated explicitly. Although this approach allows efficient segment-level optimization, it does not guarantee global optimality across the entire network.
In the final phase, the vertical alignment is further refined across the entire network using a convex quadratically constrained quadratic program.
The output of the TriPhase model is validated using real-world terrain data and compared against Softree’s RoadEng® Terrain and Location module, a widely used industry standard. By explicitly modelling cost functions, geometric constraints, and earthwork requirements across all phases, our numerical experiments show that our method finds solutions that are up to 14% cheaper than manually computed alignments.
