MULTI-VIEW FAILURE ANALYSIS ON FLEET LEVEL MULTIVARIATE TIME SERIES DATA
Publication Date: 2019-Jan-31
The IP.com Prior Art Database
The invention relates to a method for fleet level failure analysis. The method may include an algorithm for selecting key features relevant to the failure from multivariate temporal data. The selected key features may aid in root cause analytics. The features may include information regarding fault signatures timestamps (labels) and sensors relevant to the final failure. The algorithm minimizes a loss function using a multi-view failure analysis on fleet level multivariate time series data (MAMT) algorithm. The method filters wrong labels using dynamic and directional label diffusion method. The proposed method consumes less time and has lower space computational complexity. The method can be used to select the key features and the most relevant instances in time by correcting the initial positive labels.