Original Article Statistical and frequency analysis of vibrations signals of roller bearings using empirical mode decomposition Parbant Singh and SP Harsha Abstract In the present work, defect detection in rolling bearing using empirical mode decomposition of vibration signal data has been done. Higher order statistical parameters viz root mean square, kurtosis, skewness, and crest factor are utilized to diagnose bearing fault. Simulated bearing defects as spall on outer race, on roller, and on inner race are used for the study. For experimental study, four different load and speed combination have been chosen to widen the acceptability of results. The effect of bearing speed on statistical parameters is also studied. Effectiveness of signal decomposition by the empirical mode decomposition method has been established by the results. Kurtosis and crest factor values of decom- posed and unprocessed signals have been selected and empirical mode decomposition-based values are shown as effective parameters for defect identification. The crest factor and Kurtosis of outer race defect show greater sensitivity to the load and speed variations, while the skewness of same defect shows its insensitivity to load and speed variations. Keywords Empirical mode composition, intrinsic mode function, statistical analysis, kurtosis, crest factor Date received: 17 January 2019; accepted: 5 April 2019 Introduction Nowadays, instead of following conventional con- cepts of preventive and corrective maintenance, indus- trial research is focused on diagnosis of machine as well as machine components health. Aim of all these maintenance types, either preventive or diagnostic, is to ensure consistent functioning of mechanical sys- tems. An effective fault diagnosis technique requires an accurate and precise feature extraction method- ology and a proper classifier as well. 1 In various industries like power plants or automo- bile industry, one of the major and important equip- ment category is rotary machines. In these rotary machines, rolling element bearings are crucial compo- nent whether it is a small rotary device or a heavy duty machine and defect in these bearings, is primarily major cause of equipment breakdown. These break- down results not only in system shutdown but may also cost human casualties. Hence, effective defect diagnosis in bearings is important for rotary machine system functioning. 2 Vibration signature analysis is primarily used tech- nique for defect detection in bearings. These bearing defects may be classified as point defects and distrib- uted defects. Geometrical imperfections in bearing components are the basic cause of vibrations. Source of these imperfections is manufacturing inac- curacies and wear during operations. Lynagh et al. 3 developed a computational model to predict the bearing vibrations and its sidebands due to manufac- turing inaccuracies and model was validated using experiential evidences. Each bearing component has its significant information in vibration signals and there are numerous techniques to analyze these signals to extract the information. 4 Patra et al. 5 presented a mathematical model to describe the effect of speed to study the nonlinear dynamics, for both balanced and unbalanced condition of a cylindrical roller bearing. Many researchers have paid attention for develop- ment of techniques, like wavelet transform, envelop analysis, empirical mode decomposition (EMD) method, time–frequency distribution, statistical fea- ture extraction, multi-scale morphological filters for rotary machines. Some other techniques used by researchers for identification of rotary machine Proc IMechE Part K: J Multi-body Dynamics 0(0) 1–15 ! IMechE 2019 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/1464419319847921 journals.sagepub.com/home/pik Vibration & Noise Control Lab, Mechanical and Industrial Engineering Department, IIT Roorkee, Roorkee, Uttarakhand, India Corresponding author: Parbant Singh, Vibration & Noise Control Lab, Mechanical and Industrial Engineering Department, IIT Roorkee, Uttarakhand 247667, India. Email: parbantsinghsandhu@gmail.com