20th IMEKO TC10 Conference on "Measurement for Diagnostics, Optimization and Control to Support Competitiveness and Innovation"

Anomaly detection of ventilation fan by training a variational auto-encoder with current waveforms of brushless DC motor

Not scheduled

20m

Lisbon

Oral presentation T3 - Condition monitoring and maintenance of industrial processes, plants and complex systems.

Speaker

Yuko Hiranaka

Description

Anomalies in motor-driven equipment can be detected potentially by measuring the current of the driving motor. We attempt to utilize unsupervised learning of motor current to detect load fluctuations and bearing anomalies in ventilation fans. By training a variational auto-encoder on the cyclic current waveforms of a motor under normal conditions, anomalies—such as deviations in cycle length, peak current, and waveform patterns—can be detected using anomaly scores derived from latent variables. This approach is applied to experimental data from a small brushless DC motor used in a ventilation fan to verify its effectiveness.

Presentation materials

26IMEKO_TC10_hiranaka_260430.pdf