Real-time Virtual Sensor for NOx emisions and
stoichiometric air-fuel ratio λ of a Marine Diesel Engine
Using Neural Networks

Δημητρακόπουλος, Παντελής; Dimitrakopoulos, Pantelis

dc.contributor.author	Δημητρακόπουλος, Παντελής	el
dc.contributor.author	Dimitrakopoulos, Pantelis	en
dc.date.accessioned	2020-06-03T09:43:36Z
dc.date.available	2020-06-03T09:43:36Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/50751
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.18449
dc.description	Εθνικό Μετσόβιο Πολυτεχνείο--Μεταπτυχιακή Εργασία. Διεπιστημονικό-Διατμηματικό Πρόγραμμα Μεταπτυχιακών Σπουδών (Δ.Π.Μ.Σ.) “Ναυτική και Θαλάσσια Τεχνολογία και Επιστήμη”	el
dc.rights	Αναφορά Δημιουργού - Παρόμοια Διανομή 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-sa/3.0/gr/	*
dc.subject	Recurrent Neural Networks	en
dc.subject	Timeseries prediction	en
dc.subject	Emisions	en
dc.subject	Internal Combustion Engines	en
dc.subject	Virtual Sensor	en
dc.title	Real-time Virtual Sensor for NOx emisions and stoichiometric air-fuel ratio λ of a Marine Diesel Engine Using Neural Networks	en
dc.title	Εικονικά αισθητήρια πραγματικού χρόνου για τον υπολογισμό των εκπομπών ρύπων ΝΟx και του στοιχιομετρικού λόγου αέρα-καύσης λ με νευρωνικά δίκτυα	el
heal.type	masterThesis
heal.classification	Internal Combustion Engines	en
heal.classification	Neural Networks	en
heal.classification	Machine Learning	en
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2019-11-01
heal.abstract	λ value and NOx mass species have been proven important variables to emissions control and reduction for marine diesel engines. Models for these quantities can substitute real sensors that are often cost ineffecient and faulty. On top of that, model based control for emissions must operate on the basis of fast and accurate models. Arti cial neural networks (ANN) are data based models that calculate predictions on relatively simple operations using non-linear functions. They appear in many avors and con gurations. In the present work, the time-delay neural network (TDNN) and recurrent neural networks (RNN) with inputs delay are investigated as models for λ value and NOx variables. Speci cally, virtual sensors were developed for marine diesel engine's turbine-out NOx emisions and λ value based on raw measurements from laboratory data acquisition. A method was developed to search for the optimal neural network con guration between time-delay neural network models and recurrent neural network models with inputs delay. Model inputs are decided based on traditional thermodynamic models, such as the Zeldovich mechanism for NOx formation, and the available quantities sensors in any inmarket marine engine. The resulting models capture the non-linear phenomenon of NOx formation and changes in λ value of the marine diesel engine. Calculations performance is fast and portability of the models is easy. The resulting neural network models are deployed on an ECU prototype machine and they are validated in real-time side by side with the real system's sensors. The validation tests include engine operating points within the training range but of different pattern. The real-time validation for the recurrent neural network models shows that their predictions stay consistent in most operating areas and the dynamic behavior of the emissions variables is captured and reproduced accurately. The recurrent neural network model for λ value was compared against a rst principle physics based virtual sensor with more accurate results. Therefore, the validation proves that the RNN models generalize adequately within the training range with the minimum possible complexity. On the other hand, the time-delay neural network models are more complex and they do not exemplify the same accuracy in the validation tests, especially in unknown to them operating areas. As a result, the recurrent neural network models with input delay are suggested to be used as benchmark models for emissions control applications.	en
heal.advisorName	Papalambrou, George	en
heal.advisorName	Παπαλάμπρου, Γεώργιος	el
heal.committeeMemberName	Κυρτάτος, Νικόλαος	el
heal.committeeMemberName	Kyrtatos, Nikolaos	en
heal.committeeMemberName	Παπαδόπουλος, Χρήστος	el
heal.committeeMemberName	Papadopoulos, Christos	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ναυπηγών Μηχανολόγων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	98 σ.	el
heal.fullTextAvailability	false