Optimization-enhanced machine failure classification using critical sensor features and hybrid learning models with...

Area of Science:

• Industrial engineering and manufacturing.
• Machine learning and artificial intelligence.
• Predictive maintenance and reliability engineering.

Background:

• Industrial machine failures disrupt production, increase costs, and reduce equipment reliability.
• Accurate failure prediction is crucial for proactive maintenance and operational efficiency.
• A milling-machine dataset with 9,976 observations and 14 attributes was utilized.

Purpose of the Study:

• To develop and optimize a hybrid machine learning framework for predicting machine failures.
• To enhance predictive accuracy using hyperparameter optimization.
• To identify key machine condition variables influencing failure.

Main Methods:

• Developed Light Gradient Boosting (LGBC), Random Forest Classification (RFC), and XGBoost Classification (XGBC) models.
• Employed the Tasmanian Devil Optimization (TDO) algorithm for hyperparameter tuning.
• Evaluated models using accuracy, sensitivity, and specificity metrics.

Main Results:

• Optimized LGBC (LGTD) achieved a top predictive accuracy of 0.985.
• The optimized hybrid ensemble (LG-RF-TD) reached 0.979 accuracy.
• Torque and process temperature were identified as critical failure predictors.

Conclusions:

• Optimization significantly enhances machine failure prediction accuracy.
• The TDO algorithm effectively optimizes machine learning models for industrial applications.
• The proposed framework offers a robust solution for proactive industrial maintenance.