How ciaranm1999/thesis-edge-ai works
Thesis-Edge-AI is a multi-node cold-chain mould-monitoring research system. Sensor nodes collect environmental readings; ESP-NOW delivers them to a master, which forwards them over UART to Raspberry Pi collection and camera-based labelling. Offline preparation and training produce datasets and models deployed to ESP32 benchmarks for inference or on-device learning, with energy measured using a PPK2. The README documents the end-to-end workflow and capabilities; source excerpts and the index clarify selected dataset/model dependencies, but do not establish every runtime connection.
Main entry points: RaspberryPi Batch data · energy_analysis.ipynb Energy profiling · mould_prediction_dataset.h Prediction dataset · tinyol_benchmark.cpp TinyOL adaptation
Components 23
Sensing and transport
- Sensor nodes
ESP32/src/Node_Main.cpp - Master node
ESP32/src/Master_Node_Main.cpp
Edge inference
- AIfES inference
ESP32/src/aifes_inference.cpp - TFLM inference
ESP32/src/tflm_inference.cpp - TinyOL adaptation
ESP32/src/tinyol_benchmark.cpp - AIfES on-device training
ESP32/src/aifes_training_benchmark.cpp
Collection and ML pipeline
- UART collector
RaspberryPi/scripts/uart_data_collector.py - Camera labelling
RaspberryPi - Batch data
RaspberryPi - Live dashboard
RaspberryPi/scripts/web_dashboard.py - Dataset preparation
ML_Training/data_preparation/prepare_dataset.py - Model training
ML_Training/model_training/train_model.py - TinyOL backbone training
ML_Training/model_training/train_tinyol_backbone.py - Prediction dataset
ML_Training/esp32_datasets/mould_prediction_dataset.h - Training corpus
ML_Training/esp32_datasets/combined_training_dataset.h - Held-out dataset
ML_Training/esp32_datasets/held_out_dataset.h - AIfES weights
ML_Training/esp32_datasets/aifes_weights.h - TinyOL weights
ML_Training/esp32_datasets/tinyol_weights.h - TFLM model
ML_Training/esp32_datasets/tflm_model.h
Analysis and reporting
- Energy profiling
ML_Training/energy_analysis.ipynb - Batch analysis
RaspberryPi/analysis/batch_analysis.ipynb - Thesis report
ThesisReport/Report/main.tex
Other components
- Cold-chain environment
Connections 29
- Cold-chain environment → Sensor nodes: provides readings
README.md - Sensor nodes → Master node: sends ESP-NOW
README.md - Master node → UART collector: forwards UART
README.md - UART collector → Batch data: logs readings
README.md - Camera labelling → Batch data: pairs images
README.md - Batch data → Dataset preparation: supplies batches
README.md - Batch data → Live dashboard: serves live data
- Dataset preparation → Prediction dataset: exports dataset
README.md - Dataset preparation → Model training: prepares splits
README.md - Model training → AIfES weights: exports weights
ML_Training/model_training/train_model.py - Model training → TFLM model: exports model
ML_Training/model_training/train_model.py - Model training → TinyOL backbone training: provides backbone data
README.md - TinyOL backbone training → TinyOL weights: exports weights
- AIfES inference → AIfES weights: loads weights
ESP32/src/aifes_inference.cpp - AIfES inference → Prediction dataset: evaluates samples
ESP32/src/aifes_inference.cpp - TFLM inference → TFLM model: loads model
ESP32/src/tflm_inference.cpp - TFLM inference → Prediction dataset: evaluates samples
ESP32/src/tflm_inference.cpp - TinyOL adaptation → TinyOL weights: loads backbone
ESP32/src/tinyol_benchmark.cpp - TinyOL adaptation → Held-out dataset: adapts on samples
ESP32/src/tinyol_benchmark.cpp - TinyOL adaptation → Prediction dataset: evaluates samples
ESP32/src/tinyol_benchmark.cpp - AIfES on-device training → Training corpus: trains on corpus
ESP32/src/aifes_training_benchmark.cpp - AIfES on-device training → Prediction dataset: evaluates samples
ESP32/src/aifes_training_benchmark.cpp - AIfES on-device training → AIfES weights: uses dimensions
ESP32/src/aifes_training_benchmark.cpp - AIfES inference → Energy profiling: measured by
README.md - TFLM inference → Energy profiling: measured by
README.md - TinyOL adaptation → Energy profiling: measured by
README.md - AIfES on-device training → Energy profiling: measured by
README.md - Batch data → Batch analysis: analyses batches
README.md - Energy profiling → Thesis report: informs research