Abstract

The motivation of this work is the modeling of the hardening precipitate and hardness evolutions of fully hardened heat treatable aluminium alloys during friction stir welding (FSW) and/or heat treatment processes. The model used is based on the kinetics of dissolution of precipitates model for hardened aluminium alloys given by Myhr and Grong (1991). This model contains a single independent variable, the time, and a single state variable, the volume fraction of hardening precipitates. A key point of this model is the identification of the effective activation energy for precipitates dissolution and the master curve defining the model, which was given by a look-up table. The goal of this work is to find an estimation of the effective activation energy and to model the dissolution rate of hardening precipitate in aluminium alloys using neural networks, avoiding the use of look-up tables. For this purpose a new and more convenient parametrization of the master curve is defined, a neural networks class is proposed, an objective functional is defined and a variational problem including independent parameters is solved. The novel methodology has been applied to different aluminium alloys, including the AA 6005A T6, AA 7449 T79 and AA 2198 T8. Experimental tests have been carried out for those aluminium alloys in order to get the HV1 hardness after isothermal heat treatments at different temperatures and for different treatment time durations. The effective activation energy for hardening precipitates dissolution and the master curve of the model have been obtained, using different network architectures, for the aluminium alloys considered in this work.