Spark Plug Condition Classification – Edge Impulse Project

This project uses Edge Impulse’s Image Classification (Transfer Learning) block to identify and classify spark plugs based on visible defect types. It is designed for predictive maintenance applications where detailed condition labeling is important for diagnostics or alerting.

Note
This dataset also works for FOMO-AD i.e. each fault class as an anomaly and healthy as no anomaly
HF- https://huggingface.co/datasets/eoinedge/spark-plug-classification 👉 Check out the companion project: Spark Plug Anomaly Detection project

Project Overview

• Task: Image Classification
• Input: Still images or frames extracted from video
• Classes: Multiple spark plug conditions (e.g. healthy, mechanical damaged, oil / carbon fouled)
• Image resolution: 96x96

Use Case

The model is trained to recognize and classify individual spark plug conditions based on visual appearance. This allows for automated identification of specific failure modes such as oil fouling, mechanical damage, or carbon build-up — supporting fault diagnosis or maintenance scheduling.

Dataset Structure

Each condition has its own labeled folder. These folders are used for both training and testing.

frames/
├── healthy/
├── mechanical_damage/
├── oil_fouled/
├── carbon_fouled/

Each folder contains still images or extracted video frames of spark plugs representing that condition.

Impulse Design

Input Block

• Type: Image
• Dimensions: 96x96 pixels
• Resize mode: Fit short axis

Learning Block

• Type: Transfer Learning (MobileNetV2 0.35)
• Output: Classification (multi-class)

Model Training Steps

1. Upload samples from all class folders into your Edge Impulse project.
2. Use the training tab to balance the dataset across labels.
3. Train the image classification model using Edge Impulse Studio.
4. Evaluate the model performance in the Model Testing tab using a labeled test set.