Empirical Mode Decomposition for Fault Diagnosis of Multi-Component Systems

This paper presents the research work investigating the applicability of and possible improvements to the Empirical Mode Decomposition (EMD) signal processing technique for simultaneously monitoring multi-component systems. A modified EMD algorithm was developed in MATLAB R2013b software with the input being time signal data from each data collection point for all components. To test and validate the approach a fault simulator test rig was used to gather data. Fault classification accuracy is evaluated using an Artificial Neural Network (ANN) classifier model. A final classification accuracy of 87.5% was obtained without fault localization for each component and 62.5% obtained with four classes of fault localization.