06/02/2022
By Sokny Long

The Francis College of Engineering, Department of Mechanical Engineering, invites you to attend a doctoral proposal defense by Behlol Nawaz on "Detection of Anomalies in Energy Systems with Machine Learning."

Ph.D. Candidate: Behlol Nawaz
Defense Date: Wednesday, June 8, 2022
Time: 10 a.m. to noon
Location: Perry 415

Committee Chair (Advisor): J. Hunter Mack, Associate Professor, Department of Mechanical Engineering, UMass Lowell

Committee Members:

  • Juan Pablo Trelles, Associate Professor, Department of Mechanical Engineering, UMass Lowell
  • Noah Van Dam, Assistant Professor, Department of Mechanical Engineering, UMass Lowell
  • Ruizhe Ma, Assistant Professor, Department of Computer Science, UMass Lowell

Brief Abstract:
Energy systems are a core component of most facets of today’s civilization and have a bearing on key global issues. However, the process of energy production creates environmental pollution, contributes to climate change and consumes finite natural resources. This has led to an increase in the diversification of the global energy mix as cleaner and more sustainable systems are being used in increasing proportions. Not only are the newer sources more sustainable, but the diversification of the energy mix also makes the overall energy production infrastructure more robust and resilient to certain shock events. However, one of the challenges that comes with increasingly heterogeneous energy sources is the maintenance. A larger variety of generating systems will have different malfunctions that are mostly technology specific. This requires domain specific expertise to detect, diagnose and repair. The maintenance can be time consuming and costly. Unexpected and unscheduled maintenance of energy systems already costs billions globally due to inconvenient and costly down times. Timely detection of such problems can help reduce the costs.

A single system capable of automatically detecting problems with multiple generation systems (such as photovoltaics (PV) panels, wind turbines, dams, micro-turbines etc.) would be more cost-effective at scale. Using visible and infrared images for the detection of problems with such varied technologies would make the detection system more generalizable across them, detect numerous faults and replace the time-consuming visual inspections required from valuable human resources. This proposal focuses on a machine learning based anomaly detection system for two basic cases of faults in fundamentally different areas of application using images. In addition to the direct contribution towards improving the cost and reliability of energy systems, research in the proposed direction will also make contributions to combustion science by enhancing the way combustion instabilities may be studied. It will also play a part in advancing the detection of faulty solar panels from thermographic images and finally it could add to the use of machine learning in computer vision.

All interested are invited to attend.