Anomaly Identification in A Liquid-Coffee Vending Machine Using Electrical Current Waveforms

Abstract

This paper proposes an anomaly identification method for a liquid-coffee vending machine using electrical current waveforms. The method consists of preprocessing of a series of current values collected from the machine, training of multiple classifiers corresponding to individual target anomalous operations, and anomaly detection by means of the classifiers. Preprocessing improves detection accuracy by excluding current values that represent non-target operations. Multiple classifiers corresponding to individual target operations are trained using pre-processed data and the ground truth. An operation with the maximum likelihood normalized by the total number of individual operations is identified as the current anomaly. Evaluations using electrical current values obtained from an actual coffee vending machine shows a false positive rate and a false negative rate of, respectively, 0% and 6.7%, for lack of beans and 2% and 0% for water leakage, both of which are major reasons for degraded coffee quality.