Supervised load identification of 18 fixed-speed motors based on their turn-on transient current

TL;DR

This study demonstrates that supervised learning can accurately identify 18 different fixed-speed motors from their turn-on transient current in real environments, despite normalization and variability, but cannot classify their mechanical output type.

Contribution

It provides a comprehensive empirical validation of load identification using transient current for multiple motors in real-world conditions, addressing previous limitations.

Findings

Achieved 97.7% f1-score in motor identification

Normalization to 1A steady-state current does not hinder classification

Mechanical output type cannot be reliably inferred from transient current

Abstract

Several studies have already shown that the transient current can be successfully used to identify electric loads. However, most of the proposed methods were validated using only a handful of loads whose electric properties often differed significantly from each other. Therefore, a much more challenging empirical study was carried out here using 18 different fixed-speed motors in a real work environment. A further difficulty was introduced by normalizing the current amplitude to a value of 1~A during steady state. It is shown that a classifier can distinguish these 18 motors even under those conditions with an f1-score of 97.7% after a supervised training period. In addition to that it was tested whether the mechanical output type (pump, fan or compressor) of the fixed-speed motors could be inferred using only the transient current. However, the classification results indicate that this is not possible.