Unsupervised clustering for condition monitoring of electric drives

Thumbnail Image

Files

master_Srinivasa_Perumal_Byri_Manoj_2025.pdf (15.27 MB)

URL

Journal Title

Journal ISSN

Volume Title

School of Electrical Engineering | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2025-09-27

Department

Major/Subject

Electrical Power and Energy Engineering

Mcode

Degree programme

Master's Programme in Automation and Electrical Engineering

Language

en

Pages

69

Series

Abstract

Electric drives are critical to modern industry, where unscheduled downtime translates directly into high financial losses and safety risks. Conventional condition monitoring relies on generalized rules/thresholds and analytical models, which are efficient but brittle under operational changes and ineffective with incomplete failure information. Data-driven approaches address these limitations and provides better adaptability. However, most are trained offline, rely on labeled fault data or assume computational resources and connectivity that only cloud or external edge systems could afford. This thesis addresses the gap by developing an embedded-safe, unsupervised clustering framework for the condition monitoring of electric drives and demonstrates its on-drive deployment using ABB Crealizer. After studying numerous clustering methods, a density-based clustering algorithm suitable for real-time embedded constraints is selected. The chosen algorithm is re-engineered for embedded systems with several optimizations to guarantee cyclic runtime and fixed memory execution within the Crealizer environment. The developed embedded-safe clustering algorithm is coupled with efficient anomaly detection framework which encode both anomaly severity and their spatial context in a single continuous metric. The developed intelligent condition monitoring (iCM) solution is validated on two use cases, powertrain process monitoring and DC-link capacitor monitoring. In both, the system collects healthy input features, trains the clustering algorithm deterministically and achieves high performance in identifying anomalies without additional sensors and cloud dependence. The results show that the developed on-drive unsupervised clustering method is a practical and interpretable solution for real-time condition monitoring of electric drives.

Description

Supervisor

Belahcen, Anouar

Keywords

electric drives, condition monitoring, unsupervised machine learning, clustering, embedded-ML, process monitoring, Drive-as-a-Sensor (DaaS)

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202510208295

Collections

[dipl] Sähkötekniikan korkeakoulu / ELEC