This is the first post on my new website. I'm not sure what I'll use the site for yet, but I'll figure it out as I go along.
For now, the purpose will be to record things I think are useful and to document things I am learning.
I'm currently interested in exploring how the simplification of computer programming and rise of artifical intelligence might impact my field - process engineering.
I'm interested in this because I think AI will profoundly change much of what process engineers do in the (not too distant) future.
I also have a feeling the future might already be here!
I have just asked ChatGPT to size a pump using the following information:
Description:
Water at 20C is pumped through a piping system at a rate of 400 litres per minute.
The piping system consists of:
- 150 metres of 3" Schedule 40 pipe
- 1 off 2.5" globe lift check valve with wing guided disc installed with reducers in 3" pipe having a total resistance coefficient of 27 velocity heads
- 1 off 3" gate valve
- 4 off 90 standard elbows
- 1 pipe exit
- Inlet elevation at pump centre line
- Outlet elevation 120 metres above pump centre line elevation
Fluid density = 998.2 kg/m3 Fluid viscosity = 0.98 cP
Requirement:
Find the total discharge head at flowing conditions and the brake power required for a pump having an efficiency of 70%.
Here is what ChatGPT did (in real time):
Total discharge head = 127 metres
Total brake power required = 12 kW
The results are the same as presented in Crane TP410M "Flow of Fluids" Example 4-15 where the problem was originally taken from.
Well in less than 2 minutes an AI model has been able to size a pump and document it's reasoning and calculations. All I did was input a brief description of the system and what I wanted to know. The model did it's own research to determine the calculations required and to figure out what additional information it needed e.g. friction factors for fittings.
The speed and accuracy is impressive to me. Especially since I used a general purpose AI model and give absolutely no context to my request.
OK, this was a simple example. And the outputs are still the responsbility of the human engineer. But for how long?
These AI models are only going to get better and cheaper.
What happens when different models are used to cross-check each other?
How many human engineers are required if everyone has a team of AI engineers?
