Abstract
The gearbox is an accessory drive, used widely for power transmission in industries and vehicles. Since its invention in the twentieth century, it brought major changes in the field of mechanical engineering. At the same time as the gearbox evolution continues, people are focusing on customized operation with less maintenance cost. Instead of the traditional approach (scheduled and unscheduled maintenance), the industry is looking for condition-based preventive maintenance. Therefore, it is important to monitor the health condition of the gearbox. This paper presents three different architectures to diagnose the gearbox vibration signal between healthy and damaged condition. Fifteen wavelet features are extracted from the segmented signal and tested with infinite latent feature selection (ILFS) algorithm to find useful features based on ranking. Feature classification was done using a support vector machine (SVM) algorithm. The ideology of the round-robin technique was implemented in architecture-2. The result shows that, among the three developed architectures, the first architecture with discrete wavelet transform (DWT—1D) followed by the SVM model is providing better classification accuracy than the other two architectures. The results were presented with 100 Monte Carlo runs.