This thesis proposes an integrated approach for the analysis and optimization of an industrial HVAC system based on high-induction fabric air dispersion. The first phase of the research is based on an extensive empirical monitoring campaign within the production area. Through the acquisition and processing of environmental data (temperature, relative humidity, and dew point), thermographic and hygrometric maps were developed. These maps enabled the identification of thermal hotspots related to production cycles, simultaneously leading to the update and verification of the actual air distribution layout (Current State). To overcome the myopic nature of standard controllers, the second phase of the work introduces the development of a predictive model in a Python environment for the smart management of the system. Based on the Digital Twin paradigm with lumped-parameter modeling (RC network) and implementing advanced machine learning architectures (such as Long Short-Term Memory - LSTM recurrent neural networks), the system was calibrated using the historical thermohygrometric data collected in the field. The predictive algorithm is capable of learning the temporal dependencies between the activation of production lines, external weather forcings, and internal climatic variations. The simulated results demonstrate how the transition from a purely reactive to a proactive control allows for the anticipation of thermal transients, timely modulating the airflows within the fabric diffusers. This approach ensures the mitigation of localized heat peaks, a clear improvement in occupational comfort, and a significant optimization of overall energy consumption.
Il presente lavoro di tesi propone un approccio integrato per l'analisi e l'ottimizzazione di un impianto di condizionamento industriale basato su diffusione tessile ad alta induzione. La prima fase della ricerca si fonda su un'estesa campagna empirica di monitoraggio all'interno dell'area produttiva. Tramite l'acquisizione e l'elaborazione di dati ambientali (temperatura, umidità relativa e temperatura di rugiada), sono state elaborate mappe termografiche e igrometriche che hanno permesso di identificare le criticità termiche (hotspot) correlate ai cicli produttivi, portando contestualmente all'aggiornamento e alla verifica del layout aeraulico reale (Current State). Per superare la "miopia" dei controllori standard, la seconda fase del lavoro introduce lo sviluppo in ambiente Python di un modello predittivo per la gestione intelligente dell'impianto. Basandosi sul paradigma del Gemello Digitale (Digital Twin) con modellazione a parametri concentrati (rete RC) e implementando architetture di apprendimento automatico avanzate (come le reti neurali ricorrenti LSTM - Long Short-Term Memory), il sistema è stato calibrato utilizzando le storicità termoigrometriche raccolte sul campo. L'algoritmo predittivo è in grado di apprendere le dipendenze temporali tra l'attivazione delle linee produttive, le forzanti meteorologiche esterne e le variazioni climatiche interne. I risultati simulati dimostrano come la transizione da un controllo puramente reattivo a uno proattivo consenta di anticipare i transitori termici, modulando tempestivamente le portate d'aria nei diffusori tessili. Tale approccio garantisce l'abbattimento dei picchi di calore localizzati, un netto miglioramento del comfort lavorativo e una significativa ottimizzazione dei consumi energetici complessivi.
ANALISI E OTTIMIZZAZIONE DI UN IMPIANTO DI CLIMATIZZAZIONE INDUSTRIALE
D'APRILE, DANIELE
2025/2026
Abstract
This thesis proposes an integrated approach for the analysis and optimization of an industrial HVAC system based on high-induction fabric air dispersion. The first phase of the research is based on an extensive empirical monitoring campaign within the production area. Through the acquisition and processing of environmental data (temperature, relative humidity, and dew point), thermographic and hygrometric maps were developed. These maps enabled the identification of thermal hotspots related to production cycles, simultaneously leading to the update and verification of the actual air distribution layout (Current State). To overcome the myopic nature of standard controllers, the second phase of the work introduces the development of a predictive model in a Python environment for the smart management of the system. Based on the Digital Twin paradigm with lumped-parameter modeling (RC network) and implementing advanced machine learning architectures (such as Long Short-Term Memory - LSTM recurrent neural networks), the system was calibrated using the historical thermohygrometric data collected in the field. The predictive algorithm is capable of learning the temporal dependencies between the activation of production lines, external weather forcings, and internal climatic variations. The simulated results demonstrate how the transition from a purely reactive to a proactive control allows for the anticipation of thermal transients, timely modulating the airflows within the fabric diffusers. This approach ensures the mitigation of localized heat peaks, a clear improvement in occupational comfort, and a significant optimization of overall energy consumption.| File | Dimensione | Formato | |
|---|---|---|---|
|
tesi_DAprile_Daniele.pdf
File consultabile da postazioni biblioteca
Dimensione
4.92 MB
Formato
Adobe PDF
|
4.92 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/110896