Motor Fault Detection and Classification

This notebook implements a two-stage fault detection and classification system for motor vibration data using LightGBM models.

Data Preprocessing

Feature Extraction

• Uses wavelet decomposition (bior3.1 wavelet, level 4) to extract time-frequency domain features
• Features extracted from approximation and detail coefficients:
  • Statistical features: mean, std dev, variance, RMS
  • Shape features: kurtosis, skewness
  • Zero crossing rate and mean crossing rate
  • Signal entropy
  • Hilbert transform features
• Total 273 features extracted from multiple vibration sensor signals

Data Preparation

• Uses MAFAULDA dataset containing vibration measurements
  • Dataset publically available at: https://www02.smt.ufrj.br/~offshore/mfs/page_01.html
  • Contains machinery fault simulator data with multiple fault types
  • Includes accelerometer measurements from different sensor positions
• 10 fault classes labeled A-J:
  • Normal (A)
  • Imbalance (B)
  • Horizontal misalignment (C)
  • Vertical misalignment (D)
  • Overhang/underhang bearing faults (E-J)
• Train-test split: 80-20
• SMOTE applied to handle class imbalance

Model Architecture

Stage 1: Binary Classification

• LightGBM binary classifier to detect fault vs normal operation
• Parameters:
  • learning_rate: 0.05
  • num_leaves: 31
  • n_estimators: 100

Stage 2: Multi-class Classification

• LightGBM multi-class classifier to identify specific fault type
• 9 fault classes (excludes normal operation)
• Same hyperparameters as binary model
• Uses one-vs-rest approach

Results

Binary Classification Performance

• Accuracy: 100%
• Perfect separation between normal and faulty operation
• Confusion matrix shows zero misclassifications

Multi-class Classification Performance

• Accuracy: 99.48%
• Excellent discrimination between different fault types
• Confusion matrix shows minimal misclassification between fault classes

Usage

The trained models can be used to:

1. Detect presence of fault (binary classification)
2. If fault detected, identify specific fault type (multi-class)

Example usage:

# Load saved models
binary_model = joblib.load("lgbm_binary_model.joblib")
multi_model = joblib.load("lgbm_multi_model.joblib") 

# Make predictions
binary_pred = binary_model.predict(X_test)
if binary_pred == 1:
    fault_type = multi_model.predict(X_test)

The high accuracy of both models demonstrates the effectiveness of the wavelet-based feature extraction and two-stage classification approach for motor fault diagnosis.

Link to Kaggle Notebook: https://www.kaggle.com/code/ayushraj911/vibrational-analysis-of-motor