Well … not really, but it made you look?
A research team at South Ural State University in Chelyabinsk Russia is working a neural network design for making the shock absorbers in your car work better than ever before. There are three ways to do this.
- Passive (old school) springs or gas pistons that really don’t work that well and merely make the ride less jarring than having no shock absorbers at all.
- Adaptive (or semi-active) shock absorbers that respond to the up and downs based on an algorithm that simply tries to compensate for the latest bump. They are slow and inaccurate, but still way better than simple spring or gas shock absorbers.
- Active shock absorbers that are so smart and fast that their AI can take over the world. OK … maybe that’s an exaggeration, but they have a thing called Perceptrons in their neural network.
A Perceptron is essentially a souped up black box. In this case, it’s actually a multilayer perceptron, but who cares unless you’re an engineer.
You have some number of inputs [input layer] like position, speed, and force that get crunched by algorithms in the black box [hidden layer] such that predictions obtain in the output layer. The image below is from the paper submitted to the journal Energies.
What you’re looking at is called a graph, which will make sense late in an undergraduate academic experience for STEM Nerds. Trust me for now.
Anyway, the concept of a Perceptron was first proposed by the American Psychologist, Frank Rosenblatt in 1958. Yes … Psychology. A decade or so later, Minsky and Papert (1969) refined the model a bit and here we are today. I might have skipped a little history.
You can read all of the gory matrix algebra and calculus details in the linked paper below, or just take it for granted that these simulations will someday lead successfully to active shock absorber controls that make your car ride on a cloud. Well, not actually on a cloud, just feels like one. You know what I mean.
Alyukov, A.; Rozhdestvenskiy, Y.; Aliukov, S. Active Shock Absorber Control Based on Time-Delay Neural Network. Energies 2020, 13, 1091.