Condition Monitoring of Gearbox of an Ic Engine Using Vibration Analysis Through Signal Processing and Machine Learning Techniques

Abstract

Fault diagnosis of the internal combustion engine gearbox is extremely important for enhancing the efficiency of the engine and preventing the failure of connected components. Bearings and gear elements are the primary components of a gearbox, which operate in a variety of dynamic conditions with varying load and speed. Because of these severe operating circumstances, gear tooth and bearing problems occur in gearbox parts. If these flaws are not addressed, the result is a catastrophic breakdown of the gearbox, which is extremely costly and also causes additional risks in the industry. Monitoring the state of the gearbox while the engine is operating is critical to preventing damage to the other components of the engine, which is extremely useful in order to minimize component loss. As a result, it is important to select an effective and efficient technique for monitoring gearbox health without interfering the engine running. This research focuses on the condition monitoring of an engine gearbox utilizing vibration signals with signal processing and artificial intelligence approaches. The gearbox is investigated in both healthy and simulated defective conditions, such as gear tooth damage and bearing defects, which occur mostly during operation. The vibration signals from the gearbox are collected in both healthy and defective conditions and these signals are then analyzed to determine the state of the gear and bearing. The current research work is divided into two stages. The initial part of the work involves identifying/detection of gearbox conditions by analyzing vibration signals using basic signal processing techniques. To identify gearbox conditions, signal processing methods such as time-domain analysis, frequency domain analysis, time-frequency domain analysis, cepstrum analysis and wavelet analysis are used. Employing vibration signals, frequency domain analysis gave significant information on the gearbox condition. Even while signal processing methods give diagnostic information. Assessing the signals needs expertise in the area and these approaches are not suitable for studying nonstationary signals. Machine learning/deep learning is one of the best alternatives for building an effective condition iv monitoring system for developing an autonomous fault detection system for gearboxes based on artificial intelligence technologies. In the second phase, artificial intelligence models are used to investigate gearbox conditions based on vibration signals. Machine learning approaches are divided into three stages: feature extraction, feature selection and feature classification. Statistical features, empirical mode decomposition (EMD) features and discrete wavelet transform (DWT) features are extracted from the vibration signals. These extracted features are given as input to the decision tree-J48 algorithm for selecting significant features. The classifiers such as support vector machine (SVM), K-star random forest are used to classify the conditions of gearbox elements using selected features. Fault diagnosis using vibration signals are carried out by making use of different set of features and classifiers with selected features from the decision tree technique. The drawback of manual feature extraction method is time consuming, laborious, requires expertise to understand the features for different set of signals. To address these issues, deep learning techniques such as convolution neural network (CNN), residual learning, softmax classifier and long short-term method (LSTM) are used to develop an automatic feature extraction method for fault diagnosis of gearbox. Outcome of the machine learning techniques showed that, vibration signal-based fault diagnosis provided better classification accuracy in classifying the gearbox conditions. Present research work has demonstrated that discrete wavelet features served as best features among all other features such as statistical and EMD features. It was also observed that K-star algorithm provided better classification accuracy in comparison to other classifiers such as SVM and random forest algorithm. Also, results obtained from deep learning techniques provided promising classification accuracy by adopting automatic feature extraction techniques such as CNN, residual learning and stacked LSTM algorithm. Based on the research work, it is proposed that the combination of wavelet feature with K-star algorithm as a classifier is the best feature- classifier pair for diagnosis of gearbox conditions using vibration signals.