Maximum relative entropy-based probabilistic inference in fatigue crack damage prognostics

A general probabilistic inference procedure is proposed in this paper based on the Maximum relative Entropy (MrE) approach which generalizes both Bayesian and Maximum Entropy (MaxEnt) inference methodologies. The construction of the conditional probability (likelihood function) for general model-based inference problems is discussed in detail to systematically manage uncertainties from mechanism modeling, model parameters, and measurements. Analytical and numerical examples are used to investigate the sequence effect in the probabilistic inference using point observations and moment constraints. The developed methodology is applied to the engineering fatigue crack growth problem with experimental data for demonstration and validation. Following this, a detailed comparison between the classical Bayesian inference and the MrE inference is given.

► An inference procedure based on the maximum relative entropy method is developed.
► Both observation data and moment data can be processed in the method.
► Results of the fatigue crack growth application show advantages of the method.