• home
• current students
• graduate
• events

Presenter: Maryam Ahang
Supervisor:

Date: Fri, August 28, 2026
Time: 07:00:00 - 00:00:00
Place: Zoom - see below.

ABSTRACT

Zoom Details:

Meeting link: https://uvic.zoom.us/j/86727261624?pwd=KrTN0wNZ2vrxrjdbbG6fRJYV7iBSep.1

Meeting ID: 86727261624

Password: 669808

Abstract: Reliable condition monitoring of industrial systems remains challenging because fault data are often scarce, operating conditions vary, previously unseen failure modes may emerge, and machine-learning models can produce overconfident diagnostic decisions. This seminar introduces solutions to these challenges to improve the accuracy, adaptability, and trustworthiness of intelligent fault detection and diagnosis methods through three approaches.

This seminar covers N2FGAN, a signal-to-signal generative framework that learns to synthesize fault signals under operating conditions where only normal data is available. Moreover, to address previously unseen severe faults, a variational autoencoder–based health index is introduced to model degradation and detect fault conditions absent from the training data.

Finally, a hybrid, uncertainty-aware condition-monitoring framework is proposed that integrates sensor measurements, temporal information, and physics-informed residuals derived from nominal process models. Both feature-level and decision-level fusion strategies are investigated, demonstrating improved fault-diagnosis performance on CSTR and Tennessee Eastman Process benchmarks. Conformal prediction is further incorporated to quantify diagnostic uncertainty through calibrated prediction sets, showing that high classification accuracy does not necessarily imply well-calibrated confidence.

Together, this seminar covers generating missing fault data, detecting unseen failure modes, and ultimately combining data-driven learning, physical knowledge, temporal information, and uncertainty quantification to support more reliable intelligent condition monitoring in industrial applications.