This repository contains the source code and configuration for an industrial-grade, edge-deployed autonomous monitoring and fault detection system. Engineered specifically for the Raspberry Pi 5 operating within strict determinism requirements, the architecture leverages true real-time multi-threading, hybrid dual AI models, physics-based signal processing, and direct hardware actuation to mitigate critical machinery failures with sub-40 millisecond latency.

The orchestrator bypasses the Python Global Interpreter Lock (GIL) by utilizing a hybrid multiprocessing and threading architecture. I/O operations (I2C polling, SQLite transactions) operate on the main priority thread, while computationally expensive tasks (FFT operations, TFLite inferences) are executed asynchronously in isolated processes. This guarantees a maximum end-to-end latency of less than 40ms.

Raw sensor telemetry is conditioned prior to inference via physics-based processing pipelines:

• Kalman Filtering: Implements a 1D state estimator to eliminate stochastic noise from sensitive analog readings (e.g., current draw).
• Z-Score Normalization: Standardizes inputs based on baseline statistics to ensure numerical stability during neural network propagation.
• Fast Fourier Transform (FFT): Converts high-frequency (1kHz) physical vibration buffers from the time domain into spectral frequency bins to detect mechanical anomalies such as bearing wear and rotor unbalance.

The anomaly detection engine relies on a concurrent dual-model approach, combined via a weighted fusion layer:

• LSTM Autoencoder (TFLite): A deep learning sequence model tasked with identifying gradual pattern drift over time by calculating reconstruction loss.
• Isolation Forest: A scikit-learn ensemble algorithm optimized for immediate point-anomaly and sudden spike detection.

• Hardware Watchdog Integration: Hooks directly into the Linux kernel via /dev/watchdog to guarantee an autonomous hard-reboot of the edge hardware in the event of a catastrophic OS or orchestrator hang.
• SQLite Event Logging: Maintains a persistent local database (storage/fault_history.db) logging every anomaly score, recovery action, and sensor state for post-incident forensic analysis.

This software stack requires a Raspberry Pi 5.

• DHT22 (Environmental Humidity): GPIO 4
• Relay Channel 1 (Primary Machine Power Cutoff): GPIO 17
• Relay Channel 2 (Secondary/Backup System Actuation): GPIO 27
• Audio Annunciator (Buzzer): GPIO 22
• Visual Annunciator (LED Status Panel): GPIO 23

Enable the I2C interface and configure the hardware watchdog module on the Raspberry Pi.

sudo raspi-config # Interface Options -> I2C -> Enable

# Load Watchdog Kernel Module
sudo modprobe bcm2835_wdt
echo "bcm2835_wdt" | sudo tee -a /etc/modules

Initialize the virtual environment or install dependencies directly to the system Python environment.

pip3 install -r requirements.txt

Generate synthetic baselines, train the LSTM and Isolation Forest topologies, convert the LSTM to .tflite, and serialize the models to the ai_engine/models/ directory. This step may be performed on an x86 workstation prior to edge deployment.

python3 ai_engine/model_trainer.py

Execution requires elevated privileges to access raw GPIO memory registers and the /dev/watchdog file descriptor.

sudo -E python3 main.py

To guarantee high availability, the orchestrator should be configured to initialize on boot and automatically recover from unexpected terminations via a systemd service.

1. Create the unit file: sudo nano /etc/systemd/system/fault-detection.service
2. Populate the configuration:

[Unit]
Description=High-Performance Fault Detection Edge Daemon
After=network.target

[Service]
ExecStart=/usr/bin/python3 /home/pi/Real-Time-Fault-Detection/main.py
WorkingDirectory=/home/pi/Real-Time-Fault-Detection
StandardOutput=inherit
StandardError=inherit
Restart=always
User=root
Environment=PYTHONUNBUFFERED=1

[Install]
WantedBy=multi-user.target

3. Enable and initialize the daemon:

sudo systemctl enable fault-detection.service
sudo systemctl start fault-detection.service
sudo systemctl status fault-detection.service