Abstract Araújo

Fretting fatigue: our recent developments on the use of Artificial Neural Networks and on the conduct of complex tests

The aim of this plenary lecture is to present our recent developments on the use of Artificial Neural Networks (ANNs) to fretting fatigue, as well as to highlight their strengths and limitations in estimating life under such challenging mechanical problems. Fretting fatigue occurs when there is reverse frictional micro-slip in mechanical joints. It is a key issue in the fatigue design of overhead conductors, blade-disc connection in fans/turbines of aeroengines, riveted lap joints of aircraft fuselages, nub-groove connections in flexible risers, etc. Although ANNs have existed since the 1940s, their application to fretting fatigue, or even to fatigue as a whole, is recent and innovative. This study demonstrates the feasibility of developing a hybrid predictive model with high generalisation capabilities that surpass traditional methods used to compute fatigue life under fretting. This life estimation methodology is so-called hybrid as it incorporates critical plane-based equivalent stresses as input parameters for the ANN, thus introducing a physical relevance to the estimation of the crack initiation process. Hundreds of tests under different contact geometries, loading conditions, and materials have been considered to train, validate and test such an ANN. The capacity of extending the methodology to estimate fretting life under quite different geometries and materials, never seen by the ANN, is explored and discussed. To further challenge the accuracy and generality of the proposed ANN, a complex experimental test program is conducted. It involves out-of-phase and asynchronous combination of contact and bulk fatigue loads in high/low (H-L) cycle fatigue regime. This loading condition tries to mimic, as closely as possible, the loads faced by contact interfaces in aircraft during a complete flight. Results proved that the Hybrid ANN provided better life estimates than classical multiaxial fatigue methodologies (critical plane-based) considered to estimate life under fretting fatigue. However, the existence of long crack propagation phases in the mixed H-L fretting fatigue tests proved difficult to capture by the ANN.

Keywords: fretting fatigue, machine learning, ANN, tests