A Multi-objective Evolutionary Optimization Approach to Procedural Noise Mitigation for Near-ground Aircraft

Open Access
Author:
Christian, Andrew W
Graduate Program:
Acoustics
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
February 22, 2013
Committee Members:
  • Victor Ward Sparrow, Thesis Advisor
Keywords:
  • Acoustics
  • Aircraft noise
  • Community noise
  • Multi-objective optimization
  • Evolutionary optimization
Abstract:
This thesis demonstrates the viability and utility of using contemporary multi-objective evolutionary algorithms to minimize the noise impact of military aircraft on residential areas in the vicinity of airbases by suggesting novel flight procedures. That is, this thesis is searching for an algorithmic means of finding paths through the sky that aircraft can take which offer good performance in terms of noise as well as in terms of other possible flight metrics (e.g. fuel consumption). To do this, the current version of the US Navy noise-modelling program (NoiseMap 7) is combined with a recent multi-objective evolutionary algorithm (epsilon-MOEA). First, an overview of the pertinent mathematics of optimization is given, during which the uses of both an evolutionary algorithm and the multi-objective approach are defended. The basic acoustics of noise are discussed and a method for the aggregation of noise exposure across physically distributed populations is presented. The formulation of NoiseMap as a multi-objective optimization problem follows. A complication that makes the NoiseMap/epsilon-MOEA combination inefficient is discussed, and a solution is proposed and shown to be effective. Several other multi-objective optimization problems are presented which will be used for benchmarking and refining the optimization method. The population distribution of the Asheville, NC area is used as a hypothetical test case for this method. Several experimental optimizations are run serially so that the results of one inform the formulation of the next. The ultimate result is an optimization approach which, if appropriately parallelized on a modern desktop computer, can consistently produce subjectively accurate results within the span of 8 hours. The utility that is afforded to a decision maker by having these results is presented. Future directions for refinement and improvement of this method are discussed.