Please enable JavaScript.
Coggle requires JavaScript to display documents.
Opening - Coggle Diagram
Opening
Historical and Current knowledge
Gap in knowledge
Statement of need and consequence of not meeting that need
3 Objectives
2) examine the predictive capability of the secondary clarifier ML separately and the whole framework compared to CFD coupled simulations across a range of hydraulic and PM loading conditions, different geoemtries, and others
or 2) compare the expressivity and computational efficiency of learning formulation and deep learning architecture such as composite neural network (CPNN) and foureir neural operator
3) leverage the deep learning automatic differentiation capability for evaluating the steady secondary clarifier sensitivity to loading conditions and geoemtry changes
or visualize the field-resolved sensitivity of basin hydrodynamics and PM transport to variations in system geometries and loading conditions.
1) develop a plant-coupled NN (trained on secondary clarifier CFD) framework
1) examine the predictive capability of the secondary clarifier ML separately and the whole framework compared to CFD coupled simulations across a range of hydraulic and PM loading conditions, different geoemtries, and others
or compare deep learning architecture such as composite neural network (CPNN) and foureir neural operator for learning the secondary clarifier separately
2) Couple the trained NN to the biological reactor to develop a plant-wide simulation framework and validate with CFD
3) optimize internal geoemtry for one or more clarifier outer design (probably the one in BSM1) and one or more loading conditions (I will choose a representitve point from BSM1 or measurements papers) with the optimization objective being
solids or hydraulic capcity,while meeting effluent.
or effluent
4) Optional: optimize inner and outer geometry for one or more loading conditions (I will choose a representitve point from BSM1 or measurements papers) with two optimization objectives which are 1) outer volume, so biggest tank doesnt always win,
2) solids or hydraulic capcity,while meeting effluent.
or 2) effluent
3) Control is better with transient but my currently trained NN is steady state which works best with capcity/design so i am not going to do any control.
E is needed for D. by doing E to solve D, we get XX benefits
A NN surrogate is good for real time prediction, which enables utilizing this high-fidelity predictino for control
it also enables optimizing the clarifier geometry in reasonable time (before, each design would take weeks of runs for CFD and plantwide modeling to reach stead state, limiting the number of explored designs)
identify the lacks and where your research/this study comes in
A suffers from B
or we have A problem in B
or we don't understand B
Current 1D clarifier models cant capture non linear dynamics existing in the secondary clarifier
or Secondary clarifier has non linear dynamics that cant be resolved with simple ODE models
Current CFD simulations and optimizations papers doesnt preserve solids mass consistency between the plant components.
Clarifiers suffer from hydraulic problems
Control uses very simple representation of the clarifier, which 1) doesnt resolve the physics inside 2) results in the calrifier and reactor operating at conditions different from the higher-fidelity cfd ones
C is a great method. I explain the advantages it offers to face problem B
CFD is great for capturing these physics. CFD resolves these dynamics and can be connected to the WWTP as my latest research shows.
Coupling CFD to the biological reactor makes sure solids are consistent between plant components and that the clarifier is operating at more consistent solids with the predicted RAS
Design optimization is important to get more capacity and lower effluent, and other important metrics. I can cite papers.
C has a problem D. If only there was a method that could solve this problem
CFD is slow. 1) It takes days to have a prediction of separate CFD simulations and 2) weeks to get steady state prediciton of whole plant including the cfd (my last paper)
The number of explored designs is limited in each research paper
CFD cant be used for real-time control, unlike simpler 1D/ODE models used for plant-wide modeling and control
last paper shows that CFD simulation needs to be coupled to preserve solids mass consistency between the plant components. The reactor takes 50 days+ to reach a statistical steady state, making this simulation more unfeasable.
Plant-wide CFD coupling is slow. it takes weeks to get steady state prediciton of whole plant
exploring clarifier CFD geoemtry optimization becomes compuationally expensive
Design optimization through CFD is slow and expensive
E is our proposed method or research to solve problem D
A NN surrogate can provide CFD higher-fidelity while giving predictions in real-time
cite the previous research and show that you have a good understanding of the existing method
BSM1 and others
1D models and CFD for clarifier
F1,F2,F3
All design optimization papers
All plantwide control papers and methods
importance
wwtp is important/used for xx
WWTP plant-wide simulation/digital twin is important/used for xx (design/control)
what is a digital twin? can cfd be considered a digital twin? if yes then instead of talking on plant-wide simulations, we talk about digital twins
Optimizing clarifier design is important for xx
plant-wide control is important for xx.
I need to decide whether the paper is going to say we created this surrogate for quicker optimization or control or both. i originally did this for control but I had no idea on what to explore