Automated Denoising Technique for Random Input Signals using Empirical Mode Decomposition (EMD)-Stabilization Diagram

¹ Institute of Noise and Vibration, Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur, Malaysia
² School of Mechanical Engineering, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia

^* Corresponding author: muhd_danial200@yahoo.com

Abstract

The present paper deals with the novel approach of filtering technique using hybrid of empirical mode decomposition technique with stabilization diagram, that autonomously implemented within Matlab. Noise or unwanted signal is always present in the data and a bad signal-to-noise can cause a severe error in modal parameter extraction. With the recent developments of automated procedures without user interaction for the operational modal analysis (OMA), the corrupted input signals turn out to be a big issue in obtaining reliable results of automated modal parameter identification. The appearance of noise or unwanted modes due to environmental effects could affect the actual structural modes selection. There is a significant issue regarding “noise” (or spurious) modes and eliminating them from the raw signal remains to be solved and requires a lot of interaction with an expert user. In the parametric modal analysis, oversizing of a modal model is usually performed to minimize the bias on modal estimates by getting all physical modes in the frequency range of interest and help to obtain a good model fit to the data. However, this will introduce noise modes. Thus, authors take advantage of tools, such as the stabilization diagram, to illustrate, and decide, if a mode is physical or not. This selection is not a trivial task, but it may be difficult and time consuming depending on the quality of data, the performance of the estimator and the experience of the user. Since the extensive interaction between tools and user is inappropriate for monitoring purposes, image clustering tool is introduced to separate the physical poles from the others with short response time and low computational efforts compared to the available clustering algorithm. Meanwhile, Empirical mode decomposition (EMD) is then introduced to break down a signal into various components without leaving the time domain and purposely used for filtering. These are a great combination as well as an effective procedure in producing a good input signal that free from unwanted modes which can cause disruptive decision making for the actual modes selection.