Title Numeričko modeliranje podmorskih ispusta uz primjenu strojnog učenja : doktorski rad Title (english) Numerical modelling of submarine outfalls with the application of machine learning : doctoral thesis Author Marta Alvir Mentor Lado Kranjčević (mentor)Committee member Siniša Družeta (predsjednik povjerenstva)Committee member Stefan Ivić (član povjerenstva)Committee member Severino Krizmanić (član povjerenstva)Granter University of Rijeka Defense date and country 2024-06-12, Croatia Scientific / art field, TECHNICAL SCIENCES Universal decimal classificationUDC ) 62 - Engineering. Technology in general Abstract Prilikom rada podmorskih ispusta dolazi do ispuštanja otpadne vode u morsku vodu. S
obzirom da takva voda u sebi sadrži različite vrste nečistoća poput bakterija, toksina, metala,
velike količine soli te može imati značajne varijacije u temperaturi, potrebno je postići što bolje
miješanje kako bi se minimizirali negativni efekti na morski ekosustav.
U doktorskom radu je analizirano ponašanje podmorskih ispusta korištenjem numeričkih
modela i strojnog učenja. Pri tome su predložene četiri
... More tehnike koje doprinose razumijevanju
ponašanja i smanjenju negativnog utjecaja na okoliš, te pružaju skup inovativnih metoda koje se
mogu primjenjivati kod planiranja izgradnje, praćenja rada podmorskih ispusta te određivanja
njihovog utjecaja na okoliš.
Otpadna voda najčešće ima drugačiju gustoću od morske vode, što prouzrokuje različite uzgonske efekte. Ovisno o gustoći otpadne vode, mogu se analizirati pozitivni ili negativni uzgonski mlazovi pri čemu su u ovom radu promatrana oba slučaja. Rezultati numeričkih simulacija
uzgonskih mlazova izrađenih u OpenFOAM-u uspoređeni su s prethodnim eksperimentalnim
podacima. S obzirom da navedeni problem zahtijeva nestacionarne simulacije miješanja fluida s različitim svojstvima, provođenje takvih simulacija zahtijeva značajne računalne resurse.
Iz tog razloga, predložen je pristup reduciranja broja varijabli koje se računaju u numeričkim
simulacijama pri čemu se na temelju poznatog saliniteta i temperature računa gustoća otpadne
vode, te se dalje promatra koncentracija kao udio s obzirom na gustoću. Na taj način se smanjuje
broj diferencijalnih jednadžbi koje se rješavaju. Numeričke simulacije s navedenim pristupom
pokazale su veoma dobro poklapanje s eksperimentalnim podacima, te su navedeni modeli korišteni u daljnjoj analizi.
Kod projektiranja rada ispusta, nužno je odabrati adekvatne parametre strujanja i odgovarajuću geometriju ispusta kako bi se minimizirali ekološki učinci. Zbog toga je razvijena metodologija
za izradu i analizu negativnih uzgonskih mlazova na temelju strojnog učenja i SHAP metode
(engl. Shapley Additive Explanations), pri čemu se baza podataka razvila na velikom broju numeričkih simulacija. Isprobano je više različitih uvjeta strujanja, te nekoliko modela strojnog
učenja. Najbolji rezultati postignuti su korištenjem umjetne neuronske mreže, dok se najutjecajniji parametar na ponašanje mlaza pokazala brzina mlaza.
Kako bi se dobili realni uvjeti strujanja i modelirao rad podmorskih ispusta uključujući morske
struje i ostale parametre, razvijen je postupak jednosmjernog ugnježđivanja twoLiquidMixingFoam OpenFOAM modela u globalni ROMS model (engl. Regional Ocean Modeling System)
strujanja mora. Postupak je primijenjen na stvarnom podmorskom ispustu u Rijeci za ljetni
i zimski period te je analizirano širenje onečišćenja. Rezultati su pokazali da ovakav pristup
može biti vrlo koristan kod većih brzina u nižim slojevima mora, pri čemu uobičajeni postupci
koji zanemaruju brzine strujanja mora, nisu adekvatni.
Za praćenje rada podmorskih ispusta i validaciju numeričkih modela, predložena je primjena satelitskih snimki. U te svrhe, analizirane su dostupne metode i vrste snimki koje se mogu
koristiti. Uspostavljena je metodologija za obradu slika iz Sentinel-2 satelita koja je primijenjena na ispust u Rijeci. Navedeno omogućava analizu rada podmorskog ispusta tijekom dužeg
perioda kako bi se moglo planirati buduće zahvate, uočiti eventualna oštećenja i pratiti širenje
onečišćenja. Napravljena je usporedba snimki i okolišnih parametara. Rezultati obrade satelitskih snimki uspoređeni su s numeričkim modelom, te je dobiveno podudaranje lokacije pojave
onečišćenja.
Navedeni set tehnika služi za poboljšanje planiranja i analizu rada podmorskih ispusta, čime
se utječe na smanjenja onečišćenje u obalnom morskom području. Less Abstract (english) During the operation of submarine outfalls, wastewater is released into seawater. Given that
such water contains various impurities such as bacteria, toxins, metals, and large amounts of salt
and can have significant temperature variations, achieving the best possible mixing is necessary
to minimize the negative effects on the marine ecosystem.
In the doctoral dissertation, the behavior of submarine outfalls was analyzed using numerical models and machine learning. Therefore, four
... More techniques were proposed that contribute to
understanding behavior and reducing negative environmental impact. They provided a set of
innovative methods that can be applied in construction planning, monitoring the operation of
submarine outfalls, and determining their impact on the environment.
Wastewater usually has a different density than seawater, which causes different buoyancy
effects. Positive or negative buoyancy jets can be analyzed depending on the density of wastewater, and both cases were observed in this paper. The results of numerical simulations of buoyant
jets modeled in OpenFOAM were compared with previous experimental data. Since the stated
problem requires non-stationary simulations of fluid mixing with different properties, performing such simulations requires significant computer resources. Hence, an approach of reducing
the number of variables was proposed, which are calculated in numerical simulations, whereby
the density of wastewater is calculated based on the known salinity and temperature, and the concentration is further observed as a proportion with respect to the density. In this way, the number
of differential equations to be calculated is reduced. Numerical simulations with the mentioned
approach showed a very good agreement with the experimental data, and the mentioned models
were used in further analysis.
When designing the outfall operation, it is necessary to choose adequate flow parameters and
the appropriate geometry of the outfall to minimize the environmental effects. For this reason, a
methodology was developed for creating and analyzing negative buoyant jets based on machine
learning and the SHAP method (Shapley Additive Explanations), whereby the database was developed on many numerical simulations. Several different flow conditions and a few machine
learning models were tested. The best results were achieved using an artificial neural network,
while the most influential parameter on the behavior of the jet proved to be the velocity of the
jet.
In order to obtain actual flow conditions and model the operation of submarine outfalls, including sea currents and other parameters, a one-way nesting procedure of the twoLiquidMixingFoamOpenFOAM model was developed in the global ROMS model (Regional Ocean Modeling
System) of sea flow. The procedure was applied to a real submarine outfall in Rijeka for the
summer and winter conditions, and the spread of pollution was analyzed. The results showed
that this approach can be very useful for higher velocities in the lower layers of the sea, where
the usual procedures that ignore the velocities of sea currents are not adequate.
The application of satellite images is proposed to monitor the operation of submarine outfalls
and validate numerical models. For these purposes, available methods and types of images
that can be used were analyzed. A methodology was established for processing images from
the Sentinel-2 satellite, which was applied to the actual outfall in Rijeka. The aforementioned
enables the analysis of the work of the submarine outfall over a longer period in order to be able
to plan future operations, spot any damage and monitor the spread of pollution. A comparison
of recordings and environmental parameters was made. The results of the processing of satellite
images were compared with the numerical model, and the location of the pollution occurrence
was matched.
The mentioned set of techniques is used to improve the planning and analysis of the operation
of submarine outfalls, thereby influencing the reduction of pollution in the coastal marine area. Less Keywords
podmorski ispusti
uzgonski mlazovi
OpenFOAM
satelitski nadzor
strojno učenje
SHAP
Keywords (english)
submarine outfalls
buoyant jets
OpenFOAM
satellite monitoring
machine learning
SHAP
Language croatian URN:NBN urn:nbn:hr:190:027585 Promotion 2024 Study programme Title: Postgraduate university (doctoral) study Type of resource Text Extent VI, 128 str. : 30 cm File origin Born digital Access conditions Open access Terms of use Created on 2024-06-13 07:15:33
