On the development of error-trained BP-ANN technique with CDM model for the HCF life prediction of aluminum alloy

Abstract

A novel method is presented for the high cycle fatigue (HCF) life prediction of aluminum alloys, and the error-trained back propagation artificial neural network (BP-ANN) technique with the continuum damage mechanics (CDM) model is developed. First, the experimental data and numerically computed fatigue lives by the CDM model are combined to constitute a database, and the relative errors are taken as the training targets for the ANN model. Second, the relationship is established between the relative errors and external parameters (such as fatigue loads, stress concentration factors and so on). The predicted errors are then used as “gains” to adjust the numerical results, as the final predicted fatigue lives. At last, the HCF lives of the LC4 specimens are predicted by three different methods. It is found that there exists a relatively large error in the predicted results by the CDM finite element method. For the ANN model trained only with the experimental data, the accuracy of the predicted fatigue lives are not as good as the proposed technique, which also could maintain a better stability.