Industrial undercounter dishwashers operate under demanding thermal conditions and must comply with strict hygiene standards that require sustained high water temperatures throughout every wash cycle. Over time, degradation phenomena such as limescale deposition on the heating element reduce the effective thermal power delivered to the water, progressively impairing the machine's ability to meet these requirements. Early and reliable detection of such faults is therefore of both practical and safety relevance. This thesis develops a model-based fault detection system for the boiler of an industrial undercounter dishwasher, designed and validated in collaboration with Electrolux Professional. The approach is grounded in a discrete-time state space formulation of the boiler thermal dynamics, in which the effective heater power is treated as a latent state evolving as a random walk. This augmented formulation allows the Kalman filter to reconstruct the effective power from temperature measurements. The noise covariances governing the stochastic model are unknown and are learned offline from a set of experimental heating curves via the Expectation-Maximisation (EM) algorithm, combining a Kalman forward filter with a Rauch-Tung-Striebel smoother in the E-step and deriving closed-form parameter updates in the M-step. The learned model is then deployed as a steady-state Kalman filter for real-time monitoring. Experimental results on a real machine demonstrate that the estimated effective power is sensitive to differences in the thermal energy actually delivered by the heating element, enabling fault detection at an earlier stage compared to threshold-based approaches on raw temperature measurements. The simulated temperature trajectory obtained from the filter closely matches the measured one across both warm-up and full use-cycle conditions, validating the identified model.
Le lavastoviglie ad uso industriale operano in condizioni termiche esigenti e devono rispettare rigorosi standard igienici che richiedono temperature dell'acqua elevate e costanti per l'intera durata di ogni ciclo di lavaggio. Nel tempo, fenomeni di degrado come il deposito di calcare sull'elemento riscaldante riducono la potenza termica effettivamente trasferita all'acqua, compromettendo progressivamente la capacità della macchina di soddisfare tali requisiti. La rilevazione precoce e affidabile di questi guasti riveste quindi una rilevanza sia operativa che igienico-sanitaria. Questa tesi sviluppa un sistema di rilevazione di guasti basato su modello per il boiler di una lavastoviglie industriale, progettato e validato in collaborazione con Electrolux Professional. L'approccio si fonda su una formulazione in spazio di stato a tempo discreto della dinamica termica del boiler, in cui la potenza efficace dell'elemento riscaldante è trattata come uno stato latente che evolve secondo un cammino aleatorio. Tale formulazione aumentata consente al filtro di Kalman di ricostruire la potenza efficace a partire dalle sole misure di temperatura. Le covarianze del rumore che governano il modello stocastico sono ignote e vengono stimate offline da un insieme di curve di riscaldamento sperimentali tramite l'algoritmo Expectation-Maximisation (EM), combinando un filtro di Kalman in avanti con uno smoother di Rauch-Tung-Striebel nell'E-step e derivando aggiornamenti in forma chiusa dei parametri nell'M-step. Il modello identificato viene poi impiegato come filtro di Kalman per il monitoraggio in tempo reale. I risultati sperimentali su una macchina reale dimostrano che la potenza efficace stimata è sensibile alle differenze nella potenza termica effettivamente erogata dall'elemento riscaldante, consentendo la rilevazione del guasto in anticipo rispetto ad approcci basati su soglie applicate direttamente alle misure di temperatura. La traiettoria di temperatura simulata ottenuta dal filtro riproduce fedelmente quella misurata sia nelle fasi di riscaldamento che in un ciclo d'uso completo, validando il modello identificato.
Rilevazione dei guasti per macchine industriali mediante un approccio basato su Expectation-Maximization
FUSARI, NICOLÒ
2025/2026
Abstract
Industrial undercounter dishwashers operate under demanding thermal conditions and must comply with strict hygiene standards that require sustained high water temperatures throughout every wash cycle. Over time, degradation phenomena such as limescale deposition on the heating element reduce the effective thermal power delivered to the water, progressively impairing the machine's ability to meet these requirements. Early and reliable detection of such faults is therefore of both practical and safety relevance. This thesis develops a model-based fault detection system for the boiler of an industrial undercounter dishwasher, designed and validated in collaboration with Electrolux Professional. The approach is grounded in a discrete-time state space formulation of the boiler thermal dynamics, in which the effective heater power is treated as a latent state evolving as a random walk. This augmented formulation allows the Kalman filter to reconstruct the effective power from temperature measurements. The noise covariances governing the stochastic model are unknown and are learned offline from a set of experimental heating curves via the Expectation-Maximisation (EM) algorithm, combining a Kalman forward filter with a Rauch-Tung-Striebel smoother in the E-step and deriving closed-form parameter updates in the M-step. The learned model is then deployed as a steady-state Kalman filter for real-time monitoring. Experimental results on a real machine demonstrate that the estimated effective power is sensitive to differences in the thermal energy actually delivered by the heating element, enabling fault detection at an earlier stage compared to threshold-based approaches on raw temperature measurements. The simulated temperature trajectory obtained from the filter closely matches the measured one across both warm-up and full use-cycle conditions, validating the identified model.| File | Dimensione | Formato | |
|---|---|---|---|
|
Fusari_Nicolò.pdf
embargo fino al 06/07/2027
Dimensione
7.1 MB
Formato
Adobe PDF
|
7.1 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/109272