The ability to predict flow in pressurized pipes is of critical importance in fields where an unforeseen event may have serious consequences, such as healthcare and renewable energies. The traditional flow measurement methods typically involve intrusive steps, such as breaking into the pipe system, that are costly, risky and time consuming. One possible approach to tackling the problem are piezoresistive sensors, which can detect changes in pressure and deformation from outside the pipe. Our goal is to come up with a supervised statistical model able to accurately predict the flow based on the time-domain signals obtained from the sensor. Our methodology, based on Frequency Selection, Generalized Additive Models and Normalized Conformal prediction, enables to reduce the complexity of the problem to a simple one on one model that provides accurate predictions of the flow rate. Training, validation and testing of the procedure are performed on signals data from the test setup at Haptica s.r.l located in Busto Arsizio. In view of the results, an analysis of the Statistical Theory of Turbulence is performed and, given the complexity of the physical phenomenon, the good performance of the model on the test setup is confirmed. Further analysis is needed to extend the conclusions to all possible setups.
La capacità di prevedere il flusso in tubazioni pressurizzate è di molto importante in settori in cui un evento imprevisto può avere gravi conseguenze, come l’assistenza sanitaria e le energie rinnovabili. I metodi tradizionali di misurazione del flusso richiedono tipicamente interventi intrusivi, come l’apertura del sistema di tubazioni, che sono costosi, rischiosi e dispendiosi in termini di tempo. Un possibile approccio per affrontare il problema è l’uso di sensori piezoresistivi, in grado di rilevare variazioni di pressione e deformazione dall’esterno della tubazione. Il nostro obiettivo è sviluppare un modello statistico supervisionato in grado di prevedere con precisione il flusso basandosi sui segnali nel dominio temporale ottenuti dal sensore. La nostra metodologia, basata su Selezione di Frequenze significative, Generalized Additive Models e Normalized Conformal prediction, consente di ridurre la complessità del problema a un semplice modello uno ad uno, che fornisce previsioni accurate del flusso. Il training, la validazione e il testing della procedura sono effettuati su dati provenienti dal setup presso Haptica s.r.l., situata a Busto Arsizio. Alla luce dei risultati, viene effettuata un’analisi della Teoria Statistica della Turbolenza e, data la complessità del fenomeno fisico, si conferma la buona performance del modello sul setup di test. Ulteriori analisi sono necessarie per estendere le conclusioni a tutti i possibili setup.
Non-invasive flow prediction in pressurized pipelines using piezoresistive sensors and statistical modeling
MARCHIONI, SAMUELE
2023/2024
Abstract
The ability to predict flow in pressurized pipes is of critical importance in fields where an unforeseen event may have serious consequences, such as healthcare and renewable energies. The traditional flow measurement methods typically involve intrusive steps, such as breaking into the pipe system, that are costly, risky and time consuming. One possible approach to tackling the problem are piezoresistive sensors, which can detect changes in pressure and deformation from outside the pipe. Our goal is to come up with a supervised statistical model able to accurately predict the flow based on the time-domain signals obtained from the sensor. Our methodology, based on Frequency Selection, Generalized Additive Models and Normalized Conformal prediction, enables to reduce the complexity of the problem to a simple one on one model that provides accurate predictions of the flow rate. Training, validation and testing of the procedure are performed on signals data from the test setup at Haptica s.r.l located in Busto Arsizio. In view of the results, an analysis of the Statistical Theory of Turbulence is performed and, given the complexity of the physical phenomenon, the good performance of the model on the test setup is confirmed. Further analysis is needed to extend the conclusions to all possible setups.File | Dimensione | Formato | |
---|---|---|---|
2024_12_Marchioni_Tesi.pdf
non accessibile
Descrizione: testo della tesi
Dimensione
7.18 MB
Formato
Adobe PDF
|
7.18 MB | Adobe PDF | Visualizza/Apri |
2024_12_Marchioni_Executive_Summary.pdf
non accessibile
Descrizione: testo dell'executive summary
Dimensione
1.94 MB
Formato
Adobe PDF
|
1.94 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/10589/231392