1 Diagnosis of gear faults by cyclostationarity Thameur Kidar 1,2 , Marc Thomas 1 , Mohamed El Badaoui 2 and Raynald Guilbault 1 1 Department of Mechanical Engineering, École de Technologie Supérieure. 1100, Notre-Dame street West, Montreal, H3C 1K3, Quebec, CANADA. 2 University of Lyon, University of Saint Etienne, LASPI EA-3059. 20 Ave de Paris, 42334 Roanne Cedex, FRANCE. {thameur.kidar, mohamed.elbadaoui}@univ-st-etienne.fr {marc.thomas, raynald.guilbault}@etsmtl.ca Abstract Gearbox maintenance is not an easy task and conventional diagnostic techniques do not always provide suitable indication about failures. In this paper, tooth crack failures have been experimentally investigated. The mechanical system is composed of a two-stage gearbox with spur gears. Two depths of gear cracks are treated. Cyclostationnarity analysis of order 2 has been conducted to some experimental signals delivered by a test bench under two loads by using an electromagnetic brake. The information has been extracted by using the squared envelope of the signal. The results show that the cyclostationarity approach is advisable for extracting the amplitude variation at the characteristic fault frequencies related to the defect. Experimental applications for diagnosis of tooth crack demonstrate that this technique is powerful and effective in feature extracting and fault detection for gearboxes. The crack signature was shown very sensitive to the crack severity whatever the applied load. 1 Introduction The mechanisms with gears play an important role in various industrial applications. Gearbox transmission systems are one of the fundamental and most important parts of machinery and are employed in industrial sectors worldwide. Gearbox transmissions may undergo excessive stresses that led to wear and malfunctions, the appearance of defects and consequently may stop production. Frequently working in severe conditions, gears are subjected to several types of defects, especially on their teeth [1], and their maintenance is conditioned to an adequate monitoring. The early detection and diagnosis of gear faults is therefore crucial to prevent the system from malfunctions that could cause damage or entire system halt [2]. Up to now, fault diagnosis of industrial gearboxes has received intensive study for several decades, and vibration analysis of these machines can detect certain types of defects (unbalance, misalignment, blade pass, friction, noise, fluctuation, turbulence, etc.) [3]. Determination of each of these types of faults constitutes in itself a powerful monitoring technique. Some monitoring methods applied to gears are simple to use like the scalar descriptors (RMS, Peak, Kurtosis, etc.) [4] and spectral analysis (Fast Fourier Transform, envelope analysis, cepstrale analysis, etc.) [5]. However these methods are not able to distinguish between the nonlinear and the nonstationary behaviours due to the fluctuation of the velocity and load or the random phenomena (variation of wear on the surface of teeth or the opening and closing of cracks). The diagnosis consequently requires more sophisticated signal processing techniques. It is important to determine the dynamic loads on all their components (axis, teeth, bearings...), the level and variation of local stresses in these components or identify transient phenomena. In this paper, the theory of cyclostationary is applied to our model as a tool for the diagnosis of gear defects since there are a combination of periodic and random processes due to the machine’s rotation cycle and interaction with random phenomenas [6]. This approach is chosen because it is very well suited for distinguishing the faults of rotary machines for several reasons [7]: first, the occurrence of a fault in a rotating component will typically produce a repetitive variation of vibration energy, secondly because the cyclostationary framework makes it possible to localise precisely a fault within the machine kinematics and