Application of deep learning techniques in the in-Situ monitoring of Electron beam melting process

With the huge amount of data being generated from additive manufacturing processes and the development of backscattered electron signal detection from the electron beam melting process, and also the ability of deep learning networks to learn features from big data, there is the possibility of real time detection of defects. The real time data from backscattered electrons is in the form of time series data. This project involves the use of an unsupervised learning algorithm, the model used consists of a one dimensional Convolutional neural network and long short term memory (LSTM) Auto-encoder to learn the spatial and temporal features of the data. This project incorporates the learned model into the electron beam monitoring system to allow for real time process monitoring. Performance evaluation was performed using receiving operating characteristics on the model, the area under curve is 0.96 which indicates the model accurately detects the true positives.