Classification of wear mechanisms in hot forming of aluminium alloy by machine and deep learning approaches

Sustainable manufacturing emphasises lubricant-free forming to reduce costs, chemical exposure and environmental impact. In high-temperature forming of aluminium alloys, this approach results in direct contact between the part and tools, leading to material transfer and surface defects. Amorphous carbon coatings (such as DLC) limit these problems, although they are prone to mechanical wear and degradation at high temperatures. In order to better understand the transfer mechanisms and to optimise these forming processes, recent studies have used experimental tests to analyse the effects of temperature, sliding velocity and distance covered. Based on this research, this article focuses on the classification of defects (Plowing grooves, Peeling off and co-occurring defects case) associated with transfer mechanisms (abrasion, adhesion) during high-temperature forming of 6082-T6 aluminium alloys. Two approaches are compared: an analysis of 2D surface profiles using interferometry with different supervised and unsupervised machine learning algorithms, and an analysis of SEM images using a convolutional neural network. The aim is to automate the identification of defects, even in the presence of multiple defects, and to determine the parameters (statistical or roughness) that characterise these defects. Despite the challenges posed by unbalanced and uncleaned databases, the results show that the models are highly accurate. • Hot forming aluminium alloys with DLC-coated tools leads to defects • Combination of wear damages give rise to complex classification • Scanning electron microscope images and 3D surface topographies form the database • Key features of defects can be identified using machine learning algorithms • An uncleaned database can be used in deep learning to classify defects