TWI

Neural-fuzzy modelling techniques were employed to develop a transparent model-based approach that creates linguistic indices directly from spectral-temporal FSW (friction stir welding) data in order to analyse the relationship between process parameters and weld quality. ARTEMIS, an online data capture system, was used to record a high-resolution dataset during FSW trials on AA 5083 aluminium alloy plates (thickness 6 mm) welded with a "Tri-flute" tool at rotational speeds of 280-580 rpm and traverse speeds of 168-812 mm/min. Weldment quality was characterised by bend tests, surface finish examination and cross-sectional microstructure observation. A spectral-temporal methodology based on a fast Fourier transform was applied to analysis of the bending forces recorded by the ARTEMIS tool. The ability of the model to predict welding performance and monitor performance deterioration under tool wear conditions is discussed.

9th International Symposium, 15-17 May 2012, Session 4B: Modelling I, Paper 03