Health Monitoring of Hydraulic System Using Feature-based Multivariate Time-series Classification Model

Aircrafts are equipped with hydraulic systems, which powers flight controls, flaps, breaks, track the landing gear, and change the blade angles. Hydraulic pump is widely known to be the most demanding component in a hydraulic system. The paper discusses methods to devise a feature-based machine learning (supervised) model to monitor the health of critical hydraulic pump component. The approach includes two key phases: [1] feature extraction, [3] feature classification. In feature extraction phase, statistical features are explored. The features are extracted using a single-window approach (traditional approach) as well as the segmented windowing approach from raw sensor data with a given state of health. For the classification part, we come up with a window-based classifier which takes the window feature of the current segment into account. Inputs to the proposed classifier are the feature vectors extracted from raw sensor data from the current window as well as the window feature. The impact of the induced noise on multivariate time-series data prediction is very significant to assess for precise prediction. In this paper, we inspect the effect of induced noise on feature extraction and classification of multivariate time-series data prediction. We assess the model performance on the original and noise-induced multivariate time-series data. Experimental results are analyzed and discussed when comparing the proposed method against traditional approach and conclusions are drawn on their basis.