Evaluating deep learning models for fault diagnosis of a rotating machinery with epistemic and aleatoric uncertainty