Week 03

Jan 16

In DDA, there is negative pressure in leak junction
In PDA, the pressure at leak junction does not go negative (correctly in 15 min)
In PDA, the results are unstable when time step is small (5 min, 10 min)
Does demand pattern got apply to emitter coefficient?
How is the household demand got model when there is leak in PDA?
Base demand inp file
take pressure and back the calculate the demand Pressure depended demand vs Pressure Driven Analysis
Pressure dependent demand (Q=CP^exp) flow at joint is dependent on pressure. Will still can use DDA run the simulation. Demand will be satisfy first. then the pressure at each node is calculated. Some pressure will have to go below zero in order to satisfy the demand. (not realistic)
PDA (pressure driven analysis) - pressure will need to be satisfy first.
How do we still get negative pressure when base demand (= 0) and only have emitter = 0. (PDA) (J-217, J-235) (issue)
Need to zip up the results
https://epanet22.readthedocs.io/en/latest/input_keywords.html?highlight=emitter#emitters
- Actual demand reported in the program's results includes both the normal demand at the junction plus flow through the emitter
Back calculate the emitter coefficient based on the demand at 00hr and 48hr. Run the simulation at time step = 15min when base demand =0, the DDA and PDA give the same results.
Diurnal pattern in pressure driven analysis
include my PACP certification when emailing someone about CCTV.

submit the job using the csv file created from dnv streamlit and see if it works
send the email to Deven about the feature needed
test the sample code for wandb on compute canada

Making sure frame grabbing script create standardized image size
- ffmpeg
- Resize in opencv vs Resize when training
  - opencv instead of cropping to resize the image to 470, 700 standard size
  - Save the image to the best resolution and resize in in the training (do this first)
Transfer learning: using dnv models and further training it with SD1 images
- load_learner() to load previous models
- fastai
- does it has to be the same image size to further training
Try resnet50 vs resnet101 compare the computing times. Look at some other pre-trained models as well.
XResnet
Find latest research paper that compare pre-models in the similar domains. Start with model used from SewerML guy
UBC library search and Elsveria search on sewer cctv resnet?

An associated tool – the influent wastewater generator model (Matlab) – is also freely available (contact Dr Krist V. Gernaey). A preliminary software version of an extended BSM2 including a catchment model and sewer system model (published in Environmental Modelling and Software, volume 78, pp. 16-30) is also available.
catchment & sewer network simulation model to benchmark control strategies within urban wastewater systems.

use nbdev2 for notebook
create ubc account for wandb for academy. Asking how we can share with other students and a supervisor
making the results public when running with sengv1
data analysis. Image processing
in what way my research thesis contribute compare to the thesis paper by Moradi. How mine is different. How would we phrase it so that my work is different
- look at it what he did for image argumentation
- what is the overall performance
- how did he weight different type of defects
how different architecture impact the performance of the model
run the job with the gray scale to the existing image. There are two way PILImageBW vs gray scale.
PowerPoint slide for Sean (construct around what we got)
Have a slide issues with data
Use model with image augmentation to predict on SD1.

When you used PILImageBW it change the image channel from 3 channels to 1 channel
Will need to try the prediction on SD1 to see if they will be problem if the precition images have 3 channels
F1 and F2 and how I applied weigth for each labels
distribution of data