Evaluating LLMs for Predictive Maintenance: Gemini Ultra vs. GPT-4

Introduction

With Google's recent unveiling of Gemini Ultra, the AI world is buzzing with excitement. As a mechanical engineer passionate about predictive maintenance, I couldn't resist putting both AI giants to the test in a real-world engineering challenge: bearing fault diagnosis.

The Challenge

I presented both models with identical bearing parameters and vibration data:

Key observation: 12 distinct peaks in 0.1 seconds

Findings: GPT-4 demonstrated remarkable accuracy in diagnosing the bearing fault solely from the vibration signal description, significantly outperforming Gemini Ultra.