MACHINE LEARNING IN LEGAL METROLOGY – DETECTING BREATHALYZERS’ FAILURES

Abstract

Metrological control of breathalyzers used at sobriety checkpoints is done by metrological institutes or police departments to ensure the accuracy of the results. Periodic checks carried out to ensure accurate measurements are not enough, as instruments can have errors between verifications that are not detected by traffic agents. In this article, we present a new proposal to evaluate instruments using machine learning algorithms capable of detecting failures before they occur. Historical instrument measurement data is used, with the application of classification techniques and thus labeling the instruments in order to indicate those that may previously fail before the next verification. Experiments are performed with fuel cells to identify which instruments have cells that can compromise measurement results during inspections. The study ends with the simulation of using the instrument to trace the wear curve over time. The results show that it is possible to apply machine learning to assist in the metrological control of breathalyzers and thus provide more security when these instruments are used in traffic inspections.