PSI - Issue 14

Vamsi Inturi et al. / Procedia Structural Integrity 14 (2019) 937–944 Vamsi, Sabareesh, Vaibhav/ Structural Integrity Procedia 00 (2018) 000–000

938

2

1. Introduction Gearbox is a critical component in the transmission system of the wind turbines. Each and every tooth of the gear contributes towards the transmission load and failure of a single gear tooth could yield to the disruption of power. Wind turbines are often subjected to variable wind loads and harsh environmental conditions and therefore, the transmission system of the wind turbine experiences severe fluctuating loads which results in failure of the components. Generally, gear faults can be classified into two categories: local fault and distributed fault. Tooth crack & single tooth wear fall into local fault category whereas, misalignment & eccentricity comes under the distributed fault category [1]. Detecting the gear faults at incipient stages is essential as they immediately lead to fatal secondary faults. Hence, developing a diagnostic system which detects the gear faults at initial levels is a challenging task. In this study, an attempt has been made to develop a fault diagnosis system which can detect the gear faults at the initial stages. Every diagnostic scheme comprises a data collection, signal processing and feature extraction systems. Now a-days, Condition Monitoring (CM) has become the reliable maintenance strategy in diagnosing the wind turbine gearbox faults. Researchers have used the vibration, acoustic signal, motor current, oil debris analysis and thermography to detect possible faults in the system. Among all, vibration, oil debris and acoustic signal analysis have been used to diagnose the gear faults. The acquired raw data signal is processed using signal processing approach to elicit the maximum amount of diagnostic information in the form of features/descriptors. Wavelet analysis has been extensively used for the feature extraction as it provides good resolution and varying window size features [2]. It can also be used for analyzing the non-stationary signals. Wavelet transform decompose the original signal in such a way that, each scale corresponds to a particular frequency band. Empirical Mode Decomposition (EMD) is another widely used approach for feature extraction [3]. The main advantage of EMD method over wavelet method is its ability to decompose a complex signal without knowing an initial basic signal, simultaneously determining the level of decomposition on its own from the nature of the decomposed signal [3]. The extracted features are then supplied as an input to feature classification algorithm to distinguish the faults. As vibration is an explicit measure of machine structural dynamics, vibration analysis provides an effective description about the fault in the system. Accelerometers are used to record vibration signatures. Ahmed et al. [4] performed vibration analysis on a single stage spur gearbox to detect the tooth root crack fault and examined the variation in the trend of statistical features (RMS, kurtosis and crest factor) to distinguish between healthy and faulty gear. Li et al. [5] executed vibration analysis on a single stage gearbox to detect the local gear crack. The authors have used continuous wavelet analysis to extract the transient information present in the raw data signal. Sailendu et al. [6] analyzed the vibration signatures to estimate the fault size of gear tooth root crack fault by decomposing the raw vibration signal using discrete wavelet transform and extracting the statistical features to distinguish between the various levels of root crack. Jaouher et al. [7] decomposed the vibration signals using EMD and extracted the intrinsic mode functions (IMFs) to distinguish between different types and severity of faults in a roller bearing. In recent years, acoustic signal analysis has gained popularity in diagnosing the faults due to its higher and sensitivity. Generally, microphones have wider band width than the accelerometers. Therefore, the acoustic signals contain more useful information than vibration signals. Bayder & Ball [8] performed acoustic signal analysis on a single stage helical gearbox and reported that acoustic signals are effective for the early detection of local gear faults compared to vibration signals. Amarnath & Praveen [9] conducted acoustic signal analysis to detect the local gear and bearing faults. The authors decomposed the acquired acoustic signals into IMFs using EMD and extracted the kurtosis values to quantify the different faults. Amarnath and Praveen [10] detected the local faults in helical gears of a two stage gearbox using acoustic signals. The authors implemented EMD on to the acquired acoustic signals to extract the IMFs. Further, higher order statistical features such as skewness, RMS and kurtosis were extracted from the IMFs and the trend in kurtosis values was examined to categorize them into healthy and faulty cases. Implementing a single CM technique alone to diagnose the faults may not give proper health status about the machine components. Now-a-days, integrated CM programs are becoming popular because of their ability to diagnose various types & levels of faults and robustness. It was reported that, two CM techniques are executed individually, can diagnose about 30-40% of the faults [11]. However, integrating two or more CM techniques makes the diagnostic system more reliable and thereby resulting in an effective and comprehensive maintenance scheme. Adrain et al. [12] supplemented the temperature information with the vibration data to diagnose the rotating machinery faults and the extracted statistical features were classified using Principal Component Analysis (PCA). It