FDE-OKECA: Fault Data Enhanced Optimized Kernel Entropy Component Analysis

In modern industrial systems, fault diagnosis is crucial for ensuring safety, stability, and efficiency. Traditional physics - based methods struggle with increasing process complexity, prompting growing attention to flexible and adaptable data - driven approaches. The Kernel Entropy Component Analysis (KECA) algorithm is a standout feature extraction and dimensionality reduction technique for nonlinear data. Yet, conventional KECA algorithms have fault - diagnosis limitations like underutilized fault data and noise sensitivity. This paper presents the Fault Data Enhanced Optimized Kernel Entropy Component Analysis (FDE-OKECA) algorithm. It constructs two sub - models via the OKECA framework, extracting kernel entropy components from normal and fault data. Bayesian inference fuses statistical parameters from these sub - models for fault detection and diagnosis. The normal data sub - model extracts components through data standardization, kernel matrix construction, eigenvalue decomposition, and rotational optimization. The fault data sub - model uses the OKECA-LDA algorithm to boost fault feature separability. TE process simulations show FDE-OKECA improves fault detection and type identification versus traditional OKECA, with fewer false alarms.