Machine learning for low-cost particulate matter monitoring
keywords ANDROID, API MANAGEMENT, FLASK, FLUTTER, INTERNET OF THINGS, IOS, MOBILE APP DEVELOPEMENT, SMART CITIES, WIRELESS SENSOR NETWORKS
Reference persons FILIPPO GANDINO
Research Groups DAUIN - GR-05 - ELECTRONIC CAD & RELIABILITY GROUP - CAD
Thesis type RESEARCH
Description A high concentration of particulate matter (PM) in the air we breathe can have a serious impact on our bodies, causing a series of health problems such as strokes, heart diseases, and lung cancer.
The monitoring of PM concentration levels is often performed by environmental agencies using a network of fixed stations that are spread on the territory. However, due to the high cost of the instrumentation, it is not possible to achieve high spatial granularity, especially in urban environments.
In the past few years, low-cost light-scattering sensors have been introduced in the market. They could enable the creation of much denser networks, but they still suffer from many problems such as low precision, low accuracy, and high frequency of faults.
The thesis work is divided into two parts. The first concerns the application and tuning of machine learning algorithms to tackle different problems:
- sensor calibration: better tuning of the currently used machine learning algorithms (linear regression, random forest) and testing of different ones to improve the reliability of measurements.
- outlier removal: detect and remove outliers during the calibration phase.
- fault detection: predict when sensors are going to break, to allow corrective actions (turn on a redundant one)
The second part is the re-design of the software library used by the research group to analyze sensor data:
- great focus on modularity
- decoupling from the data model
- integration of newly developed algorithms
Required skills Good programming skills
Some experience with machine learning
Deadline 10/02/2023 PROPONI LA TUA CANDIDATURA