The Damage Mode Automatic Identification of Al2TiO5 Flexible Ceramics Based on Machine Learning

Abstract

Machine learning is applied to study the damage evolution of Al2TiO5 (AT) flexible ceramics at different sintering temperatures to realize the automatic identification of damage modes. Three damage modes (i.e., friction of Al2TiO5 grains, damage of grain boundary glass phase, and fracture of single Al2TiO5 grain) can be successfully distinguished by processing feature values of acoustic emission (AE) collected from material damage through k‐means algorithm. Though the specimen sintered at 1600 °C possesses the largest grain, the mean feature value of AE in cluster‐3 (fracture of AT single grain) is not maximum since the decomposition of Al2TiO5 is serious at 1600 °C. Meanwhile, the amplitude‐centralized area of the corresponding cluster first expands and then decreases with the increase of sintering temperature. When the sintering temperature is 1500 °C, this centralized area reaches to the maximum. Correspondingly, the specimen sintered at 1500 °C has the maximum failure displacement and the most AE damage events. To realize the automatic identification of damage modes, random forests algorithm is applied to distinguish the damage modes of other samples in the same set. The almost same trend of damage modes evolution in new samples and training sample proves high efficiency of this method.