Multi-objective optimisation of crush properties, production costs and CO2 footprint of 6xxx series aluminium extrusionsFriday (10.11.2017) 10:35 - 10:55 Part of:
In the present paper, a new modelling concept is used to optimise the crush performance of square hollow section 6xxx aluminium profiles. The crush performance is optimised at the same time as production costs and CO2 emissions throughout the process chain are attempted kept at minimum levels. This is obtained by using a novel holistic modelling concept which is closely related to various aspects of Industry 4.0 including “big data based simulation and optimisation”. The concept implies that each operation along the process chain is represented by different types of predictive models that are sequentially coupled. The models represent a wide spectre of disciplines and include physically based material models, different types of cost models as well as models for estimation of environmental impact during production. A multi-objective optimisation program is used to couple these models into one common software platform in a way which enables the outputs from one model to be input to the next model in the production chain. The models are run sequentially in several iterations until user defined acceptance levels on properties, costs and sustainability indices are obtained. In the present work, the predicted crush behaviour was first compared with previously conducted quasi-static, axial crushing experiments for one specific alloy composition. These comparisons revealed a very good agreement between predicted and measured force-displacement curves. The positive results for the verification of the crush behaviour are a pre-requisite for the second part of this work, i.e. the optimisation procedure. Different optimisation strategies for multi-objective optimisation were explored including MOGA-II (Multi-Objective Genetic Algorithm for fast Pareto converges). Finally, Response Surface Models (RSMs) or metamodels were used to approximate the relationship between the inputs and outputs, and both interpolating RSMs as well as approximating RSMs including Neural Networks were tested out. The results from this work show that excellent crush properties can be obtained using relatively cheap post consumed scrap metal as a basis, which gives low material costs and low CO2 emissions.