Abstract
Powder Metallurgy (PM) is the branch of Metallurgy that deals with the design/production of near net-shaped sintered workpieces with different shapes and characteristics. The produced sintered workpieces are used in automotive, aviation, and aerospace industries, just to name a few. The quality of the produced sintered workpieces largely depends on powder compaction techniques and the accurate adjustments of process parameters. Currently, adjustments of these process parameters are done manually and thus resulting in laborious and time-intensive effort. To this end, this paper explores the use of Machine Learning (ML) in the compaction process and proposes an accurate and light-weight ML-based pipeline to estimate the quality characteristics of the produced workpieces in the powder metallurgy domain. More specifically, it presents a pipeline for workpiece’s mass and lengths estimation by exploiting some novel hand-crafted features and comparing well selected ML prediction models, namely, Random Forest (RF), Adaboost (ADA), and Gradient Boosting (GB). The chosen models are trained on a combination of features extracted from environmental and sensory raw data to estimate the mass and lengths of the next produced workpiece. We have implemented and evaluated our scheme on a dataset collected in a real production environment and we have found that GB is the most consistent and accurate one with the lowest Root Mean Squared Error (≈ 0.0886%). The results of an extensive experimentation have proven the relevance of the selected features and the accuracy of GB.