Growing population and urbanization has led to an increase in municipal solid waste production and a subsequent increment in the involvement of landfills for waste disposal. These sites generate emissions under the form of methane and many odors, which are cause of a substantial increase in air pollution and of many complaints by citizens living in nearby areas. Methane detection is relevant as the gas, when detected, can be used as fuel or burned to reduce its environmental impact, and is currently detected via costly devices which are mainly efficient at high concentrations of CH4 in air. Landfill odors are instead monitored as reference for possible unwanted compound leaks in the landfill disposal infrastructure and are currently detected using bulky devices called electronic noses, placed statically in areas close to the site. This study relied on multiple laboratory tests at different concentrations of methane, conducted on a toolbox of off-the-shelf VOC sensors and a micro-controller, programmed ad-hoc to acquire their measurements. The same toolbox was then used to acquire odors collected directly on landfill odor emission sources and to train a deep learning model to distinguish between leachate, fresh waste, biogas and air odors consequently deployed on an embedded device via tinyML techniques. This work led to the development of MOLE (Methane and Odors Low-cost Electronic nose), a prototype of a portable, low-cost device pro vided with sensors sensitive to as low concentrations of methane in air as 5ppm and capable of classifying different landfill odors with a built in deep learning model. The results of this thesis laid the bases for cheap alternatives to detection of low concentrated methane and portable substitutes to electronic noses used for odors detection and classification.
La crescita dell’industrializzazione ha portato ad un incremento nella produzione di rifiuti urbani con un con seguente aumento di utilizzo di discariche per il loro smaltimento. Queste strutture generano emissioni sotto forma di metano e molti odori, i quali sono causa di un sostanziale aumento di inquinamento atmosferico e di molteplici lamentele da parte dei cittadini residenti in aree limitrofe. Il rilevamento del metano è importante in quanto il gas, se identificato, può essere riutilizzato come carburante o bruciato per ridurre il suo impatto ambientale. Attualmente il metano viene rilevato attraverso dispositivi costosi e particolarmente efficienti nella rilevazione di alte concentrazioni di CH4 in aria. L’odore di discarica è invece monitorato col fine di individuare possibili malfunzionamneti o perdite della discarica attraverso strumenti ingombranti chiamati nasi elettronici, i quali vengono installati staticamente in aree vicine la discarica. Questo studio si è basato su numerosi test in laboratorio a differenti concentrazioni di metano, utilizzando un toolbox di sensori VOC commerciali ed un microcontrollore, programmato su misura per acquisire i valori da loro rilevati. Lo stesso toolbox è stato poi impiegato per acquisire odori campionati direttamente da sorgenti odorifiche situate in discarica e per allenare un modello di deep learning atto a differenziare fra odori di percolato, rifiuto fresco, biogas e aria. Tale modello è stato poi distribuito direttamente su un dispositivo embedded tramite tecniche di tinyML. Questo lavoro ha portato allo sviluppo di MOLE (Methane and Odors Low-cost Electronic nose), un prototipo di un dispositivo portatile ed economico, presentante un set di sensori sensibili a basse concentrazioni di metano, fino a 5 ppm, e capace di classificare diversi odori di discarica grazie ad un modello di deep learning integrato. I risultati di questa tesi gettano le fondamenta per alternative economiche per la misura di basse concentrazioni di metano e a possibili sostituti portatili dei nasi elettronici impiegati per le rilevazioni di odori e la loro classificazione.
MOLE: a portable, low-cost electronic nose for landfills methane detection and odors classification using tinyML
Vernola, Dario
2023/2024
Abstract
Growing population and urbanization has led to an increase in municipal solid waste production and a subsequent increment in the involvement of landfills for waste disposal. These sites generate emissions under the form of methane and many odors, which are cause of a substantial increase in air pollution and of many complaints by citizens living in nearby areas. Methane detection is relevant as the gas, when detected, can be used as fuel or burned to reduce its environmental impact, and is currently detected via costly devices which are mainly efficient at high concentrations of CH4 in air. Landfill odors are instead monitored as reference for possible unwanted compound leaks in the landfill disposal infrastructure and are currently detected using bulky devices called electronic noses, placed statically in areas close to the site. This study relied on multiple laboratory tests at different concentrations of methane, conducted on a toolbox of off-the-shelf VOC sensors and a micro-controller, programmed ad-hoc to acquire their measurements. The same toolbox was then used to acquire odors collected directly on landfill odor emission sources and to train a deep learning model to distinguish between leachate, fresh waste, biogas and air odors consequently deployed on an embedded device via tinyML techniques. This work led to the development of MOLE (Methane and Odors Low-cost Electronic nose), a prototype of a portable, low-cost device pro vided with sensors sensitive to as low concentrations of methane in air as 5ppm and capable of classifying different landfill odors with a built in deep learning model. The results of this thesis laid the bases for cheap alternatives to detection of low concentrated methane and portable substitutes to electronic noses used for odors detection and classification.File | Dimensione | Formato | |
---|---|---|---|
2024_06_Vernola_Tesi_01.pdf
solo utenti autorizzati dal 29/06/2025
Descrizione: Tesi
Dimensione
4.29 MB
Formato
Adobe PDF
|
4.29 MB | Adobe PDF | Visualizza/Apri |
2024_06_Vernola_Executive_Summary_02.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Executive summary
Dimensione
1.02 MB
Formato
Adobe PDF
|
1.02 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/223118