Study and Application of Machine Learning Methods in Modern Additive Manufacturing Processes

Introduction

Additive Manufacturing (AM), usually stated to as three dimensional printing or 3DP, is an advanced manufacturing technique that has the promising of disrupting the engineering industry on a balance not observed from the time when the manufacturing revolution (Barua et al.,2019). Paralleled to traditional manufacturing techniques, it has the benefits of manufacturing complicated parts with difficult geometric structure, exclusive microstructures and characteristics, in addition to concentrated lead period and budget. Consequently, in current existence time, additive manufacturing has involved an excessive arrangement of investigation awareness together industrial uses and academic research widespread. Additive manufacturing methods can be approximately categorized into seven classifications (Lee et al., 2001). Figure1 shows the different types of additive manufacturing process. The additive manufacturing methods connecting Machine Learning (ML) in this chapter generally classified into three groups, i.e. directed energy deposition (DED), powder bed fusion (PBF), and material extrusion process. Even though ML has also been used in additional additive manufacturing methods for example stereo-lithography (Wang et al., 2018) and materials jetting (Yuan et al., 2017). In powder bed fusion (PBF) process, an electron beam or laser is applied in place of the energy source to carefully melt powder bed which is homogeneously extent by repainting layer by layer (Barua et al., 2019). In the directed energy deposition (DED) method, a concentrated laser beam liquefies the incessant powder stream or wire which are deposited from nozzle obsessed by the melt pool so as to construct desire objects (Rahman et al., 2019). Fused deposition modelling (FDM) is an example of material extrusion procedure. The filament is melted by extruder heater and deposited layer by layer. Machine learning is an artificial intelligence (AI) method that permits a machine or system to study from data spontaneously and make conclusions or estimates without being plainly encoded (Datta et al., 2019). In the study, machine learning is in advance acceptance in health diagnostics (Ning et al.,2015), autonomous driving (Bruijne et al., 2016), smart manufacturing (Kourou et al., 2015), natural language processing (Datta et al., 2019) (Akerfeldt et al., 2016), object recognition (Liang et al., 2015) (LeCun et al., 2015), and material possessions expectation (Ward et al., 2016) (Pilania et al., 2013). ML algorithms are typically considered as supervised, unsupervised and reinforcement learning. Supervised learning allows a computer programme to study from a set of considered data in the teaching set so that it can classify unlabeled data from a trial set with the maximum probable exactness (Shi et al., 2016). The datasets can be in a diversity of forms counting forms of audio clips (O’Shea et al., 2016), images (Lempitsky et al., 2010), or typescript (Tong et al., 2001). There is a main function known as cost function, which analyses the mistake between the expected output standards and the real output standards. In the preparation method, the limitations (or weights) between neurons in neighboring layers are reorganized so as to decrease the cost function after every iteration (or epoch) (Daelemans dt al., 2003). In the testing procedure, the formerly hidden fresh data, specifically test set, is presented to deliver an impartial calculation of the simulation’s accurateness. Unsupervised learning concludes from unlabeled data (Weber et al., 2000). It is a data-driven machine learning method which can expose unseen configurations or cluster comparable data collected in an assumed haphazard dataset (Alabi et al., 2018). Unsupervised learning is extensively applied in irregularity revealing (Omar et al., 2013), market dissection, and references systems (Tanev et al., 2007).