Técnicas de ciência de dados aplicadas à detecção de padrões de falha em processos de pintura automotiva

Abstract

Automotive paint automation is a delicate process, consisting of multiple steps that neatly apply sensitive layers of painting to vehicles. Each layer has a particular purpose that is initially linked to protection and, finally, defines the final visual aspects. Due to the interdependence among paint layers, errors in internal layers often compromise subsequent layers. Thus, at the time the vehicle passes though the quality inspection, at the end of the process, it may be rejected for not complying with the required standards, being sent for rework. This article presents a data science approach to identifying patterns in the automotive painting process that could result in poor final product formation. The approach is implemented using the Google Collaborative tool, with libraries for the Python language, and applied to a real database of a painting process. Results show that it is possible to anticipate the control system performance patterns that have less probability of failure in the inspection process.