Remaining useful life prediction of flax fibre biocomposites under creep load by acoustic emission and deep learning

Abstract

Natural fibre composites are increasingly explored for structural applications due to improvements in mechanical performance. For this, damage prognostics are crucial. We integrate acoustic emission (AE) and deep learning techniques to predict the remaining useful life of a flax fibre composite under long-term creep load. Derivatives of cumulative AE features with respect to time, such as cumulative hit and count rates, are introduced to reflect the performance degradation rate of the materials. These proposed features seem more relevant for creep lifespan than traditional AE features. Long short-term memory networks and temporal convolutional networks are adopted to estimate the composite’s remaining useful life. The two models' normalized root mean square errors are below 0.11, less than 20% of the error of a statistical Weibull-distribution benchmark model. Our study demonstrates that AE-based data-driven models can predict the performance degradation of composite materials subject to sustained load.