"Source Code, Models, and Results for Knowledge Distillation and INT8 Quantization of Turbofan RUL Prediction on ESP32-S3"

"This dataset provides the complete reproducibility package for the paper \"From Deep Learning to Microcontroller: Knowledge Distillation and INT8 Quantization for Real-Time Remaining Useful Life Prediction in the Internet of Things.\" It includes: (1) 20 Python experiment scripts covering knowledge distillation, nine-architecture baseline comparison, quantization-aware training, ablation studies, cross-domain validation, and statistical analysis; (2) trained PyTorch checkpoints and TFLite models for the teacher TTSNet, student StudentTTSNet (FP32 and INT8), and GRU baseline; (3) all experimental results as JSON files with per-seed metrics across five random seeds; (4) preprocessed C-MAPSS FD001 turbofan degradation data; and (5) ESP32-S3 firmware source code for real-time on-device inference. The package enables full reproduction of all tables and figures reported in the paper, including the finding that temperature-scaled knowledge distillation is mathematically inert for regression tasks, INT8 QAT achieves 3.98x compression with only 1.6% accuracy degradation, and hardware profiling reveals that deployment feasibility\u2014not desktop accuracy\u2014is the binding constraint for edge AI."