Elevator Door Diagnosis with Low-cost Current Sensor based on Robust One-Class Learning Time-series Shapelets

Abstract

Recently, as the number of elevators in operation and the period since their installation have grown, the importance of maintenance of elevators is increasing. However, due to a shortage of personnel responsible for maintenance, there is a demand for efficiency in these processes, while prioritizing safety. In this study, we developed a prototype for elevator door diagnosis system that can detect signs of anomalies using current waveforms collected by low-cost current sensors. To compensate for the performance degradation due to the low reproducibility of waveforms, we adopt Robust One-Class Learning Time-series Shapelets (ROCLTS) with low-pass filter and waveform extraction. In our experimental evaluation, sufficient detection performance could be achieved even with training on only 20 normal waveforms.