Deep learning for material synthesis and manufacturing systems: A review

Abstract

Deep learning (DL) techniques are the evolutionary methods of machine learning (ML) advancements in which current industrial operations are focusing and this method is way far efficient in handling big data in a rapid pace and with autonomy. DL techniques are the analyzing tools for interpreting and operating according to the big data and process the data for variety of applications. This study discusses the evolution of DL methods from conventional ML techniques and the potentiality of DL over ML methods. This is followed by the enunciation of various computational DL methods which enhances the performance of technologies used in material synthesis, characterization and manufacturing methods. Studies focused on the determination crystal structure of a material by using deep neural network (DNN) tools which are trained from a huge crystallographic big data were also touched upon. Such crystallographic studies was reported to be possible by using atomic fingerprints of the multiple crystal structures which provides data regarding the topology of crystallographic regions and these data could be used for the training of DLL models that detects the crystal structure of an element in any crystallographic environment. Degradation of materials is another task that was successfully reported to be carried out using DL models. Though the material degradation is dependent on many factors and modelling of such degradation by using deterministic algorithms is difficult owing to the measuring constraints and interdependence of the governing variables, usage of appropriate DL tools and algorithms would make it easier. DL models for the synthesis of materials, analysis of spectroscopic results and deployment of manufacturing systems were individually discussed. This study is also extended towards the upcoming areas in which DL could be extensively applied and the challenges in it is also discussed.