TY - JOUR
T1 - A data-driven probabilistic framework towards the in-situ prognostics of fatigue life of composites based on acoustic emission data
AU - Loutas, T.
AU - Eleftheroglou, Nikos
AU - Zarouchas, Dimitrios
PY - 2016
Y1 - 2016
N2 - An innovative prognostic data-driven framework is proposed to deal with the real-time estimation of the remaining useful life of composite materials under fatigue loading based on acoustic emission data and a sophisticated multi-state degradation Non Homogeneous Hidden Semi Markov Model (NHHSMM). The acoustic emission data pre-processing to extract damage sensitive health indicators and the maximum likelihood estimation of the model parameters from the training set are discussed in detail. In parallel, a Bayesian version of a well-established machine learning technique i.e. neural networks, is utilized to approach the remaining useful life estimation as a non-linear regression task. A comparison between the two algorithms training, operation, input-output and performance, highlights their ability to offer reliable remaining useful life estimates conditional on health monitoring data from composite structures under service loading. Both approaches result in very good estimations of the mean remaining useful life of unseen data. NHHSMM is concluded as the preferable option as it provides much less volatile predictions and more importantly is characterized by confidence intervals which shorten as more data come into play, an essential trait of a robust prognostic algorithm.
AB - An innovative prognostic data-driven framework is proposed to deal with the real-time estimation of the remaining useful life of composite materials under fatigue loading based on acoustic emission data and a sophisticated multi-state degradation Non Homogeneous Hidden Semi Markov Model (NHHSMM). The acoustic emission data pre-processing to extract damage sensitive health indicators and the maximum likelihood estimation of the model parameters from the training set are discussed in detail. In parallel, a Bayesian version of a well-established machine learning technique i.e. neural networks, is utilized to approach the remaining useful life estimation as a non-linear regression task. A comparison between the two algorithms training, operation, input-output and performance, highlights their ability to offer reliable remaining useful life estimates conditional on health monitoring data from composite structures under service loading. Both approaches result in very good estimations of the mean remaining useful life of unseen data. NHHSMM is concluded as the preferable option as it provides much less volatile predictions and more importantly is characterized by confidence intervals which shorten as more data come into play, an essential trait of a robust prognostic algorithm.
KW - SHM
KW - Remaining useful life
KW - Machine learning techniques
KW - Condition-based reliability
KW - Acoustic emission
KW - Composite materials
U2 - 10.1016/j.compstruct.2016.10.109
DO - 10.1016/j.compstruct.2016.10.109
M3 - Article
SN - 0263-8223
JO - Composite Structures
JF - Composite Structures
ER -