An AI approach for quality improvement in heat treatment processing

Abstract: Export of heat treated steel goods has an important impact on the Swedish economy which brings performance demands and expectations on production to keep a competitive market position. Sustainability and efficiency are two important aspects in meeting these demands. This thesis studies how a data driven approach can be used to increase efficiency in manufacturing of rods produced for the mining industry.  The purpose of this thesis is to use a machine learning model suitable for classifying quality results for heat treated steel rods. This is done by comparing nine algorithms with the objective to tune and deploy the model best fitted while gaining insights in variables that have an impact on the quality output.  This thesis outset is a heat treatment process at Epirocs facility in Fagersta. Interviews are conducted to gain domain knowledge about important features and an AI pipeline is implemented to demonstrate its suitability for predicting quality given production and weather data in the form of time series and product-unique data points.  The result of the study shows that the machine learning algorithm random forest is indicated as most suitable among the analyzed. The study also shows that an AI pipeline with streaming data can be designed and efficiently implemented for quality improvement. Through this work, the authors have proved that machine learning can be used to improve the heat treatment process of rods, but the model still has room for improvement in feature selection and availability of larger and more detailed data at the facility.