Abstract

Surfaces, from structured surfaces [1] to additive surfaces [2], are the critical parts of components. We have developed a new method for surface characterisation combining light scattering and machine learning (ML). The method has been successfully applied to the binary classification (acceptable vs. defective) of micro-structured [3,4] and additive manufacturing (AM) surfaces [5-8], particularly laser-based powder bed fusion (PBF-LB) surfaces. The general idea is to impinge laser light onto a surface and process the scattered reflection pattern by a ML binary classifier. We have developed the following ML models to characterise PBF-LB surfaces: convolutional autoencoder (CNN) [5], autoencoder based on a multilayer, fully connected network (ANN) [6], classifier based on principal component analysis (PCA) [7], and classifier based on a one-class support vector machine (SVM) [8]. All these methods share a common approach where the ML system is internally used to learn an optimal strategy to encode/decode only the scattering patterns of acceptable surfaces. Defected surfaces are then detected during operation, when poor encoding/decoding performance is observed on a new pattern. The single-class approach is advantageous for AM because the ML model can be trained without the need of examples from defective surfaces. In these methods, light scattering patterns (digital images) are used as input data and the classification result is output from the ML model. Image resolution is a critical parameter, which can significantly influence classification performance in terms of accuracy and speed. A higher image resolution means the image contains more information, thus giving the ML classifier more data to operate upon, which typically should result in higher classification accuracy, whilst at the same time processing speed is lowered. On the other hand, a lower image resolution implies less information to process, which is typically detrimental to classification accuracy, but results in higher processing speeds. Finding a method to select image resolution that results in an optimal balance between classification accuracy and processing speed is the focus of this paper.