Smarter Devices, Smaller Models: Unlocking Embedded AI Through Efficient Compression

This MathWorks project focuses on building an AI-powered fault diagnosis system for smart appliances using vibration sensor data and deep learning. The project explores how model compression techniques such as pruning, projection, and quantization can make neural networks smaller and faster for embedded applications while preserving diagnostic accuracy. The work includes sensor-data preprocessing, model development, performance evaluation, and analysis of accuracy, memory footprint, inference speed, and compression trade-offs, with potential extensions into embedded code generation and alternative model architectures.