Hybrid HC-PAA-G3K for Novelty Detection on Industrial Systems

Piecewise aggregate approximation (PAA) provides a powerful yet computationally efficient tool for dimensionality reduction and feature extraction. A new distance-based hierarchical clustering (HC) is now proposed to adjust the PAA segment frame sizes. The proposed hybrid HC-PAA is validated by a generic clustering method ‘G3Kmeans’ (G3K). The efficacy of the hybrid HC-PAA-G3K methodology is demonstrated using an application case study based on novelty detection on industrial gas turbines. Results show the hybrid HC-PAA provides improved performance with regard to cluster separation, compared to traditional PAA. The proposed method therefore provides a robust algorithm for feature extraction and novelty detection. There are two main contributions of the paper: 1) application of HC to modify conventional PAA segment frame size; 2) introduction of ‘G3Kmeans’ to improve the performance of the traditional K-means clustering methods. Keywords— Piecewise aggregate approximation, Hierarchical clustering, G3Kmeans, Novelty detection, Industrial gas turbine.