AI-Based Optimization of Fluid Flow Through Pipes

A student-laptop project that combines pipe-flow physics, AI surrogates and optimisation. Physics solvers (Darcy-Weisbach with Colebrook-White friction, minor losses, pump power, the Global Gradient Algorithm for looped networks and a small finite-difference CFD solver for developing laminar flow) generate the training data and check every AI answer. A neural network replaces the implicit Colebrook equation (mean error 0.0571 %), and a physics-guided random-forest surrogate predicts the head loss of a complete pipeline with 0.075 % test error. The surrogates drive a particle-swarm search for the economic pipe diameter and a surrogate-assisted genetic algorithm for least-cost network design, which reaches the best known cost of the two-loop benchmark (419,000 $). A web application with five database options stores analyses and optimisation runs; the same algorithms are also written as MATLAB scripts. All data are generated by the physics solvers - no experimental data set is used.

Category: Mechanical Engineering / Fluid Mechanics / Machine Learning Surrogates / Optimisation / MATLAB and Python

Mechanical Engineering Project M09 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.