MODELING AND EXPERIMENTAL ANALYSIS OF MULTIFAULTS IN A TAPERED ROLLER BEARING SYSTEM

Abstract

Rolling element bearings are widely employed in the industrial and domestic machines and appliances due to their ease in mounting and efficient operational features. About 80% bearings found in the industrial machines happens to be the rolling bearings. Numerous factors contribute to the deterioration of these bearings, primarily including wear, ageing, environmental influences, improper installation, inadequate lubrication, and material fatigue. A defective bearing frequently leads to reduced efficiency or, in some instances, significant injury to the machine. Consequently, health monitoring and fault diagnosis have gained significant attention in recent years, and they can be performed utilising various information, including acoustic emission, stress waveforms, oil analysis, temperature, and vibration. The prevalent method employed for defect detection is vibration monitoring and analysis, which provides critical insights into anomalies occurring within the internal structure of bearings. This study investigates the vibration-damping behaviour of four different types of antifriction bearings through experimental and computational analyses. Experiments are conducted on a customized rotor-bearing test rig with data acquisition through the OROS NV-Gate software at both the drive end (DE) and non-drive end (NDE). The damping characteristics are evaluated through static and dynamic analyses. Additionally, the study employs Taguchi and ANOVA methods to assess the effects of load and rotational speed on the vibrations and noise of healthy tapered roller bearings (SKF30206). The Taguchi method, which integrates statistical and mathematical techniques, is used to establish relationships between input parameters and system response. Twenty-seven experiments have been performed using the L27 design of experiments (DOE) approach, considering two factors (speed and load) at three levels. Further experiments are conducted on tapered roller bearings with inner race, outer race, roller, and compound defects. A DOE framework comprising 64 experiments is designed by incorporating two continuous and one categorical factor at four levels. Beyond traditional signal processing techniques, soft computing methods are explored vii to automate defect detection partially. The efficiency and relevance of this study are significantly enhanced by integrating machine learning and classifier algorithms for automated fault diagnosis. Time and frequency domain features were extracted from 800 signal datasets to ensure optimal model performance and ranked using one-way ANOVA and Kruskal-Wallis selection techniques. Initially, machine learning models are trained and tested for automated defect classification, followed by a comparative analysis of different classifiers available in MATLAB. The results indicate that tapered roller bearings exhibit superior damping capacity compared to cylindrical, spherical, and self-aligned bearings. The contact between rolling elements and raceways played a crucial role in the damping behaviour of antifriction bearings. For both healthy and defective TRBs, vibration response data— measured in terms of root mean square (RMS), kurtosis, and noise levels (Leq) are analyzed to evaluate performance. The study highlights the effectiveness of combined parametric effect analysis with DOE and the Taguchi method in predicting the behaviour of tapered roller bearings within rotor-bearing systems. Furthermore, computational analysis demonstrates that feature selection through ranking mechanisms significantly enhances machine learning model efficiency. Among various classifiers, the highest defect classification accuracy is achieved using the top 10 features ranked by the Kruskal–Wallis test with classification accuracies of 79.0%, 86.6%, 92.9%, 97.6%, 81.9%, and 64.4% for linear, quadratic, cubic, fine, medium, and coarse models, respectively. The Kruskal–Wallis test outperforms the One-way ANOVA in feature selection and further improves the classification accuracy. This integrated approach offers a robust predictive framework for assessing the performance of TRB (tapered roller bearings) and automating defect diagnosis in rotor-bearing systems.