Gigala (Engineering design by reinforcement learning, genetic algorithms and finite element methods)
Are you interested in new ways of engineering design? This repository is an attempt to apply artificial intelligence algorithms for the purpose of engineering design of physical products. I combine numerical simulation like finite element analysis with artificial intelligence like reinforcement learning to produce optimal designs. Starting from 2018, my work has been focused on intelligent topology optimization in structural mechanics. I am constantly exploring different ways that AI can be applied to science and engineering.
Reinforcement learning is a global, gradient-free, non-convex, learning-based, generalizable topology optimization method suitable for practical needs. Sequential nature of reinforcement learning makes it also applicable to technological processes where it can provide manufacturing steps (agent-technologist), and to the design of complex machinery with moving parts (agent-builder).
With my diverse interests, I am using this repository as a testbed for my ideas to create software for artificial intelligence aided design. I hope that my work can inspire you to explore new ways that AI can be applied to your field.
At present, Gigala software consists of topology optimization module, and offshore pipelay dynamics module (now separated into Ocean Intella software). It uses artificial intelligence to assist an engineer in her design. You can use it as research or engineering analysis tool to design different physical components and elements.
For now, Gigala is a pre-industrial grade software capable of conceptual design for simple components like pins, hinges or small elements within the power of a personal computer. But I am persistently striving to improve on the benchmarks (resolution of a design space). At that, if supercomputers are leveraged, Gigala's algorithms should already be those of a full industrial grade.
Philosophy of the software:
- free (accessibility)
- open source (full customization)
- practical performance on your PC (low carbon footprint)
- developed in Python (widespread)
- use AI (modern)
Please check my Blog and ResearchGate for the specifics of the models and algorithms I use.
For citation please use:
- Reinforcement Learning Guided Engineering Design: from Topology Optimization to Advanced Modelling
- Practical Topology Optimization with Deep Reinforcement Learning and Genetic Algorithms
RL agent designing a cantilever:
To keep up to date with the project please check Gigala page.