Please enable JavaScript.

Coggle requires JavaScript to display documents.

generate CFD cases for NN training - Coggle Diagram

- - - - Control optimization
        
        I would need to see what I could do about control which I have no idea about and will probably require a lot of time to learn and contribute something new.
        
        It is a good chance to learn control, will help with project control, and aligns with my final ambitious goal of having real time control of the WWTP ( both the reactors and secondary clarifiers are replaced with CFD NN surrogates and are communicating the best control strantegies in real time). I think this better aligns with my PhD trajectory but I am not sure if I have enough time for it
        
        my phd so far 1. paper surrogate for stormwater basin design 2. surrogate for stormwater hydrodynamic separator transient simulation 3. coupled WASTE water CLARIFIER CFD-asm1 vs separate CFD simulations or Coupled 1D clarifier-asm1 (tested using BSM1) 4. this current work that I am deciding on what to do in 5. cfd modeling of the biological reactor including asm1 coupled to the cfd modeling of the secondary clarifier 6. real time optimizing control of NN surrogate of biological reactor coupled to NN surrogate of secondary clarifier. however I dont think I will have enoguh time to do this 6th step. That why I am debating whether I should go for control in paper so i make sure i touched control if i dont have time for 6 (probably) or i should focus on design to jsut get it done and because control is probaly more meaningful in 6 than 4, which only has cfd of secondary clarifer and dont have transient figured out yet. then maybe i have enoguh time in the end for a little bit of 6. but also i dont want to work on design if it is meangless. thats why i need help deciding which way to go
        
        i was hoping to finish this paper quick and I am not sure if gonig this direction, would require more time than I have
        
        probably is better with tranient but my currently trained NN is steady state which works best with capcity/design
      - We generate 1000 designs that have different outer and and inner geoemtry or We generate 10 outer designs that have 100 inner geoemtry.
        
        Find geometry that maximizes capacity margin across one or more representative loading states.
        
        I could use Cost to decide on the cheapest upgrade, whether in volume or design upgrade but this is a lot of work to find cost, i think, i am not sure if it really is
        
        following Dr. Li's "Implementing machine learning to optimize the cost-benefit of urban water clarifier geometrics " paper
        
        Find geometry with two objective functions: highest capacity Qcrit and lowest volume with effluent below limits across one or more representative loading states. This means 10% larger tank, 50% more capacity. this way i make sure, minimal footprint without using cost, i think
        
        Find geometry with lowest volume with effluent below limits across one or more representative loading states. this way i make sure, minimal footprint is used while meeting effluent instead of using cost, i think
      - We generate 1000 internal designs for the same outer geoemtry
        
        Find geometry that minimizes effluent.
        
        I could do this by using lhs to generate 1000 secondary clarifiers internal geoemtries then i rank them according to lowest effluent then analyze the best cases.
        
        This means I can test this at different loading scenarios
        
        Assuming each scenario is the worest at a certain wwtp plant, whats best geoemtry for each scenario/plant?
        
        weaker because it doesnt include capcity which is a big thing?
        
        Find geometry that maximizes capacity margin across representative loading states.
        
        I could do this by using lhs to generate 100 secondary clarifiers internal geoemtries then i use a testing method to increase the SOR or SLR capcity while keeping Qras pump constant, assuming this is highest, and then rank them to the designs that have the highest capcity while still meeting the effluent limit
        
        loadings
        
        does this mean I use one loading scenario becuase it will be tested to increase both the loading scenario Qin and MLSS? How do I choose this one loading scenario?
        
        I could select mroe than one and increase each geoetry at each loading but this is too much work
        
        Assuming the same plant expereience all these scenarios, whats best geoemtry for the plant across all scenarios?
        
        strong because it includes both capcity and effluent
      - WWTP Loading scenarios
        
        Maybe sample some loading points from BSM1 or from measurements