Output details

Evolutionary optimisation of noisy multi-objective problems using confidence-based dynamic resampling

This journal paper describes international collaborative research between the University of Skovde (Sweden), De Montfort University, Volvo Aero and Volvo Car (Sweden). The OPTIMIST project was partially funded by the Knowledge Foundation (KK Stiftelsen) in Sweden.

Real world situations such as manufacturing plants often contain complexities making the effective use of conventional analytical modelling exceptionally difficult. Such complexities can be addressed by simulation-based optimisation, however, signal ‘noise’ can present further problems; this is also the case using evolutionary selection processes where ‘noise’ can derail the convergences of optimisation processes. This paper presents a new technique to efficiently deal with ‘noise’ in multi-objective optimisation.

The new technique was applied to good effect in two case study implementations of real situations; the first involved the optimisation of engine component manufacture in a new manufacturing cell at Volvo Aero; the second a camshaft manufacturing line producing fifteen different camshafts for both petrol and diesel engines at Volvo Car. Both manufacturing lines are highly complex, involving automated machines and manual stations, and signal ‘noise’ would induce both machine breakdowns and human error.

The technique formulated for dealing with ‘noise’ conditions was rigorously evaluated against four extant noise-reduction techniques. In comparison to the existing techniques, the new method was proved superior in all cases when tested against the same benchmarks.

The European Journal of Operations Research is very highly regarded in the Engineering Science and Computing Science domains.

(Volvo contact: leif.pehrsson@volvocars.com – Senior Engineer/Production Technology Manager)