Prediction of Variation from Heat Treatment

Abstract: This study is carried out at Scania’s production centre in Södertälje. The heat treatment process fora crown wheel is subject to a significant amount of parameters which are complex to model. Theaim of this study is to investigate the feasibility of improving process predictability for the casehardening depth of crown wheels by using advanced analytics.The project is initiated by studying in depth the steps involved in the case hardening process and theproperties of the raw material for the crown wheel. The field of machine learning is also studied andexplored to know its applications and the resources it requires. For the prediction of the output, dataanalysis is conducted to find the parameters affecting the process after which data is gathered fromdifferent databases and is cleaned and synthesised to be used for machine learning in Python. Aregression and classification analysis is made by using pre-existing algorithms for prediction fromthe Scikit-learn library. The models obtained are evaluated by using metrics and a classificationmodel is found to have the greatest prediction ability. A technical and business value evaluation ismade to judge its performance.This study resulted in the development of a basic tool which can make predictions on the casehardening depth, with certain limitations. Additionally, conclusions in regards to the differentparameters affecting the output were derived.