Refine
Document Type
- Doctoral thesis (1)
- Report (1)
- Working paper (1)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Language
- English (3)
Keywords
- Gemischt-ganzzahlige lineare Programmierung (3) (remove)
Institute
One of the standard approaches for solving discrete optimization problems which include the aspect of time, such as the traveling salesman problem with time windows, is to derive a so-called time-indexed formulation. If the problem has an underlying structure that can be described by a graph, the time-indexed formulation is usually based on a different, extended graph, commonly referred to as the time-expanded graph. The time-expanded graph can often be derived in such a way that all time constraints are incorporated in its topology, and therefore algorithms for the corresponding time-independent variant become applicable. The downside of this approach is that the sets of vertices and arcs of the time-expanded graph are much larger than the ones of the original graph. In recent works, however, it has been shown that for many practical applications a partial graph expansion that might contain time-infeasible paths, often suffices to find a proven optimal solution. These approaches, instead, iteratively refine the original graph and solve a relaxation of the time-expanded formulation in each iteration. When the solution of the current relaxation allows for a feasible schedule, an optimal solution can be derived from it and the algorithm terminates.
In this work, we first present new ideas that allow for the propagation of information about the optimal solution of a coarser graph to a more refined graph and show how these can be used in algorithms. More precisely, we present two general algorithms for solving Mixed Integer Linear Program formulations which we call iterative refinement and branch-and-refine. Iterative refinement basically is solving relaxations of the problem until a feasible solution to the original problem is found. Branch-and-refine is a kind of branch-and-bound algorithm that allows for the graph refinement to be carried out during the exploration of the branch-and-bound tree. For demonstrating the practical relevance of these algorithms, we not only study them in the context of academic examples but also apply them to two real-world problems. The first is a problem from the literature, where small passenger air-crafts have to be routed and scheduled to serve flight requests while fulfilling a variety of conditions on, for example, fuel consumption, weight, and detours. We show here that refinement algorithms can be used to improve the best known results from the literature. The second problem we consider is the task of optimally scheduling deliveries and charging times of delivery robots such that delays are minimized. In this case, we show that refinement algorithms perform better than a direct solution approach making use of state-of-the-art solvers.
In wire-arc additive manufacturing (WAAM), the desired workpiece is built layerwise by a moving heat source depositing droplets of molten wire on a substrate plate. To reduce material accumulations, the trajectory of the weld source should be continuous, but transit moves without welding, called deadheading, are possible. The enormous heat of the weld source causes large temperature gradients, leading to a strain distribution in the welded material which can lead even to cracks. In summary, it can be concluded that the temperature gradient reduce the quality of the workpiece. We consider the problem of finding a trajectory of the weld source with minimal temperature deviation from a given target temperature for one layer of a workpiece with welding segments broader than the width of the weld pool. The temperature distribution is modeled using the finite element method. We formulate this problem as a mixed-integer linear programming model and demonstrate its solvability by a standard mixed-integer solver.
We present a general numerical solution method for control problems with PDE-defined state variables over a finite set of binary or continuous control variables. We show empirically that a naive approach that applies a numerical discretization scheme to the PDEs (and if necessary a linearization scheme) to derive constraints for a mixed-integer linear program (MILP) leads to systems that are too large to be solved with state-of-the-art solvers for MILPs, especially if we desire an accurate approximation of the state variables. Our framework comprises two techniques to mitigate the rise of computation times with increasing discretization level parameters:
First, the linear system is solved for a basis of the control space in a preprocessing step. Second, certain constraints are just imposed on demand via the IBM ILOG CPLEX feature of a lazy constraint callback. These techniques are compared with an approach where the relations obtained by the discretization of the continuous constraints are directly included in the MILP. We demonstrate our approach on two examples: modeling of the spread of wildfire and the mitigation of water contamination. In both examples the computational results demonstrate that the solution time is significantly reduced by our methods. In particular, the dependence of the computation time on the size of the spatial discretization of the PDE is significantly reduced.