TinyML Gearbox Fault Prediction on a $4 MCU

 

A new project that Sumit Kumar posted on Hackster.io. He has been immersed in various TinyML projects, which involve creating super tiny machine learning models and running them on small devices without turning to the Cloud.

This approach gives a number of advantages, such as:

• Internet independence
• No waste of energy and money on data transfer
• Advanced privacy and security 
• Ability to make cheap AI-driven devices 

In this project, Sumit implemented the ability to predict the damage in a gearbox based on data from 4 vibration sensors by using a tinyML model embedded into a simple $4 MCU (Raspberry Pi Pico). To train the model, he used a dataset from Kaggle. He trained the model automatically using a free-to-use platform, Neution.AI.

You can check the metrics of the model accuracy and size in the table below.

To see how Sumit got such outcomes, check out a full step-by-step guide on Hackster.io