Prediction of Selective Laser Sintering Part Quality Using Deep Learning

Abstract

Additive manufacturing (AM) is increasingly gaining attention from academicians, researchers and industry due to its significant advantages relative to the conventional subtractive Manufacturing. But it is very difficult to tune AM processing parameters due to their greater impact on the printed part microstructure and on the performance of the printed parts. It is extremely difficult to derive a structure–property–performance relationship of an AM process for additive manufacturing employing conventional analytical and numerical models. Over the past decade, machine learning has been growing its applications to transform the manufacturing towards a new paradigm—smart manufacturing. At present, the machine learning (ML) technique has been found to be a promising technique for performing regression analysis and complex pattern recognition without any use of framing and solving the physical models. The neural network (NN) is the most popular machine learning model among ML techniques owing to strong computational power, data set availability and sophisticated architecture. This chapter aims at helping academicians and industry to understand the fundamentals about deep learning for additive manufacturing. In this chapter, the authors have included a review about additive manufacturing and the deep neural network learning benefits to solve additive manufacturing problems like product quality prediction. An illustration in selective laser sintering additive manufacturing to predict the quality in terms of minimum shrinkage ratio through a deep neural learning is demonstrated. In this chapter, the quality of a product in terms of minimum shrinkage ratio produced by selective laser sintering (SLS) was predicted from important parameters of the SLS process—surrounding working temperature, laser scanning speed, layer thickness, scanning mode, hatch distance, laser power and interval time. The relationship among SLS parameters with the shrinkage ratio is modelled using the deep neural network approach because the SLS parameters are considered to be multitudinous and a nonlinear system. This developed machine learning system employed supervised deep learning, including process parameters as input characteristics and the quality of a product in terms of minimum shrinkage ratio as the output characteristic. This employed weight decay and the dropout technique to overcome the overfitting problem is found in the deep neural network technique. The predicted output characteristic was compared to the actual characteristic. The shrinkage ratio found here by the deep neural network can be employed to determine information for shrinkage compensation in the SLS process.