A data-driven method to enhance vibration signal decomposition for rolling bearing fault analysis. (15th December 2016)

• Record Type: Journal Article
• Authors: Grasso, M.; Chatterton, S.; Pennacchi, P.; Colosimo, B.M.
• Is Part Of: Mechanical systems and signal processing. Volume 81(2016)
• Page Start: 126
• Page End: 147
• Publication Date: 2016-12-15
• Subjects: Empirical Mode Decomposition -- Combined Mode Functions -- Vibration -- Bearing -- Fault detection; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2016.02.067
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: Health condition analysis and diagnostics of rotating machinery requires the capability of properly characterizing the information content of sensor signals in order to detect and identify possible fault features. Time–frequency analysis plays a fundamental role, as it allows determining both the existence and the causes of a fault. The separation of components belonging to different time–frequency scales, either associated to healthy or faulty conditions, represents a challenge that motivates the development of effective methodologies for multi-scale signal decomposition. In this framework, the Empirical Mode Decomposition (EMD) is a flexible tool, thanks to its data-driven and adaptive nature. However, the EMD usually yields an over-decomposition of the original signals into a large number of intrinsic mode functions (IMFs). The selection of most relevant IMFs is a challenging task, and the reference literature lacks automated methods to achieve a synthetic decomposition into few physically meaningful modes by avoiding the generation of spurious or meaningless modes. The paper proposes a novel automated approach aimed at generating a decomposition into a minimal number of relevant modes, called Combined Mode Functions (CMFs), each consisting in a sum of adjacent IMFs that share similar properties. The final number of CMFs is selected in a fully data driven way, leading to an enhanced characterization of the signal content without any information loss. A novel criterion to assess the dissimilarity between adjacent CMFs is proposed, based on probability density functions of frequency spectra. The method is suitable to analyze vibration signals that may be periodically acquired within the operating life of rotating machineries. A rolling element bearing fault analysis based on experimental data is presented to demonstrate the performances of the method and the provided benefits. Highlights: The study proposes a data-driven approach to enhance multi-scale signal decomposition. It works by automatically convert the Intrinsic Mode Functions (IMFs) into a minimal number of final modes. A probability density-based criterion is proposed to cluster IMFs that share similar properties. Benefits in terms of information synthesis and signal characterization can be achieved. The methodology is demonstrated by means of real data in a cylindrical roller bearing diagnosis application.