MiFaD: A Microphone Fault Dataset

This dataset contains MEMS microphone recordings for research on sensor fault detection in embedded and IoT systems. It includes recordings representing real fault conditions such as normal, clipping, spike, and stuck, created by inducing acoustic overload and undervoltage scenarios. It also includes synthetic fault types representing bias and drift, which simulate gradual or subtle sensor degradation. All recordings are organized according to their fault category and prepared for fixed-window analysis, including 2-second segments that have been shown to improve machine-learning performance. The dataset supports the development and evaluation of lightweight models for real-time fault classification, including CNN-based approaches and hybrid methods that combine CNN feature extraction with classical classifiers. It provides a reproducible resource for advancing sensor reliability studies and developing fault-detection solutions for resource-constrained environments. >> Code maintained on Gitlab << @article{Talayoglu2025Lightweight, title={Lightweight AI for Sensor Fault Monitoring}, author={Talayoglu, Bektas and Velde, Jerome Vande and Silva, Bruno da}, journal={Electronics}, pages={30}, year={2025}, publisher={MDPI}}