Machine learning aided design of smart, self-sensing fiber-reinforced plastics

Abstract

Numerous techniques have been developed for the non-destructive evaluation (NDE) of impact damage in fiber reinforced plastics (FRPs), following the increasing demands for their safety and maintenance. Considering the large-scale detection and the vast amount of data involved, machine learning (ML) can be utilized in NDE for damage type analysis and impact damage localization. Furthermore, self-sensing using carbon fiber in FRPs is an emerging technique for NDE that can be combined with ML. In this study, ML was used to design smart FRPs by selecting the fiber type and electrode distance considering the cost and electromechanical sensitivity. Furthermore, a novel algorithm for structural health self-sensing was suggested using an artificial neural network. The developed ML algorithms are advantageous since they do not require a theoretical model when all the factors and the variables of FRPs, such as the maximum absorbed impact energy, maximum impact force, initial electrical resistance, number of electrodes, fiber types, and electrode distance, are to be considered. The algorithm was trained using given input data and the target, and the output could be successfully obtained when new input data were provided. Therefore, the proposed ML algorithms hold great potential and applicability to FRP design and for NDE methods.