Drives and Controls Magazine
AI learns how to control equipment faster and cheaper
Published:  14 February, 2018

Mitsubishi Electric has developed a technology that uses model-based artificial intelligence (AI) to control equipment autonomously. Using AI, the technology constructs models of a system’s dynamics through repeated trial-and-error, and then uses these models to learn the control rules needed to reach pre-defined goals automatically.

Mitsubishi predicts that the technology, which is still under development, will in future avoid the need for humans to develop programs to teach control actions. It will therefore cut the time and cost of developing such programs.

The company has demonstrated the technology using a circular maze on a tilting platform, where the objective is to guide a ball to the centre of the maze by tipping and tilting the platform. The technology successfully learned how to achieve this without needing any human programming.

Mitsubishi Electric is now planning to develop the technology further by enhancing its AI capabilities to improve the speed and accuracy of the autonomous learning. It will also use 3D simulations of the controlled equipment to reduce the time it takes to learn control methods.

How conventional control technology (top) compares with Mitsubishi's AI-based control programming technology (bottom)

Mitsubishi has two patents pending for the technology in the US, and six outside the US.

The technology is part of Mitsubishi’s Maisart (Mitsubishi Electric AI creates the State-of-the- ART in Technology) programme.