| PhD Viva

Name of the Speaker: Mr. Mannam Veera Narayana (EE18D302)
Guide: Prof. Devendra Jalihal
Co-Guide: Prof. Shiva Nagendra SM
Venue: BSB-104
Online meeting link: https://meet.google.com/ghz-josy-yyn
Date/Time: 5th April 2024 (Friday), 9:00 AM
Title: Developing Machine Learning calibration models for low-cost sensors

Abstract

Advancements in low-cost sensors (LCS) have enabled large-scale air pollution monitoring efforts, providing comprehensive data with high spatiotemporal resolution. Despite these advancements, concerns persist regarding data quality due to various error sources inherent in sensor-based monitoring. Calibration is a well-known technique that involves developing mathematical algorithms (calibration models) which transform sensor measurements into accurate values.

In the first part of the thesis, we discuss various error sources associated with low-cost sensor-based air quality monitoring. Subsequently, we present the calibration of low-cost sensors as a machine learning-based optimisation problem. In the latter part of the thesis, we introduce two novel calibration models: Estimated Error Augmented Two-phase Calibration (EEATC) and Sens-BERT, a BERT-based approach. These models have been implemented and demonstrated to outperform existing calibration models, offering the advantage of transferability to uncalibrated sensors. Results from our implementation are showcased using monitored datasets, as well as datasets available on USEPA data websites.