gearbox_fault_detection

Gearbox Fault Detection

Overview

The Gearbox Fault Detection application is an Edge AI solution that classifies gearbox operating conditions into two states—healthy operation or broken tooth fault—using vibration sensor data in real-time. This enables predictive maintenance on embedded devices and prevents costly equipment failures in industrial settings.

Problem and Solution

• Gearbox failures are a leading cause of unplanned downtime in industrial machinery
• Traditional condition monitoring requires manual inspection or complex on-site diagnostics
• Early fault detection prevents catastrophic failures and extends equipment lifespan
• Edge AI enables real-time fault diagnosis directly on resource-constrained microcontrollers without cloud connectivity

Key Performance Targets

• Real-time classification of gearbox operating conditions
• High accuracy on vibration-based fault detection
• Low memory footprint suitable for MCU deployment

System Components

1. Hardware:

• MSPM0G5187 microcontroller with integrated NPU https://www.ti.com/product/MSPM0G5187

2. Software:

• Code Composer Studio 12.x or later
• MSPM0 SDK 2.10.00 or later
• TI Edge AI Studio

Dataset Information

The example uses the SpectraQuest Gearbox Fault Diagnostics Simulator dataset:

• Source: Gearbox Fault Diagnosis - Kaggle
• Classes: 2 (Healthy, Broken Tooth)
• Samples per class: 10 operating conditions (0-90% load)

Each CSV file contains:

• Columns: a1, a2, a3, a4 (4 vibration sensors placed in four different direction)
• Rows: ~88,000 samples per file
• Naming Convention: {b|h}30hz{load}.csv

Feature Extraction Pipeline

1. Sensor Input: 4-channel accelerometer data
2. Windowing: Extract fixed-length windows (256 samples) from continuous vibration data
3. Single frame input to model

Model Architecture Options

The example includes multiple optimized 1D CNN-based models targeting different deployment scenarios:

Model	Parameters	Flash (bytes)	Description
GearboxFault_model_1.2k	~1,174	~8,233	4-layer network with progressive channel reduction (12→12→8→8)
GearboxFault_model_1.5k	~1,914	~8,878	3-layer network with constant channels (16→16→16)

Both models use:

• INT8 quantization for reduced memory footprint
• 4-channel input, 256-sample sequences

Model Performance

• Accuracy: 97-100% depending on model selection
• Both models achieve >97% accuracy with varying resource trade-offs

Training and Deployment Process

NOTE: Running the config yaml handles everything including dataset loading, feature extraction, training, quantization, and compilation.

1. Training:

   • Use TI Edge AI Studio (GUI) or tinyml-tensorlab (CLI)
   • Batch size: 32-64
   • Training epochs: 20-30
   • Optimizer: Adam with learning rate 0.001

2. Quantization:

   • INT8 quantization for reduced model size
   • Maintains accuracy while enabling MCU deployment

3. Compilation:

   • TI Neural Network Compiler converts trained model
   • Generates model artifacts for device deployment

How to Run

After completing the repository setup, run the following command from the tinyml-modelzoo directory:

Windows:

.\run_tinyml_modelzoo.bat examples\gearbox_fault_detection\config_MSPM0.yaml

Linux:

./run_tinyml_modelzoo.sh examples/gearbox_fault_detection/config_MSPM0.yaml

References

• SpectraQuest Gearbox Dataset on Kaggle
• TI Neural Network Compiler User Guide
• TI Model Training Guide
• EdgeAI Software Guide