Comparison of independent component analysis and principal component analysis for remaining useful life prediction of rolling element bearings

Accurate prediction of the Remaining Useful Life (RUL) of rolling element bearings is essential for predictive maintenance and reliability engineering. This article presents a comparative analysis of Principal Component Analysis (PCA) and Independent Component Analysis (ICA) as dimensionality reduction techniques for constructing health indicators from vibration signals. Experiments were conducted using the publicly available IMS Bearing Dataset, which provides run-to-failure vibration measurements under realistic operating conditions. A set of time- and frequency-domain features was extracted from the raw signals and transformed using PCA and ICA to generate univariate health indicators. These indicators were subsequently modeled using machine learning approaches, including Random Forest, Support Vector Regression, Multi-Layer Perceptron, and a baseline Artificial Neural Network, for RUL estimation. The results demonstrate that ICA-based health indicators exhibit more monotonic degradation behavior and greater robustness to noise than PCA-based indicators, leading to improved prediction accuracy and stability. Notably, the Random Forest model combined with ICA achieved an R2 value of 0.91 and the lowest prediction errors, consistently outperforming PCA across all models. The proposed framework is computationally efficient and interpretable, making it suitable for industrial prognostics applications.