Abstract
Abstract
This study presents an approach that combines variational mode decomposition (VMD) and relevance vector machine (RVM) as a prediction technique for surface roughness during turning. The method helps machinists detect irregularities during machining, such as looseness in machine parts or tool wear. The study uses work hardened EN8 steel and extracts vibration signal features into five modes. PSO optimized RVM prediction models are developed using these modes with and without combining cutting parameters. The most sensitive mode is selected for accurate surface roughness prediction. The results show that the first mode of decomposition when combined with cutting parameters provides the least mean square error of 0.1095. Surface roughness was found to be primarily influenced inversly by cutting speed and directly by feed rate respectively. The signal with high surface roughness value has high amplitude noise scattered over the large frequency range. The first decomposed mode of vibration signal increases noise with a large amplitude as surface roughness increases. The approach can be integrated into a microcontroller to regulate machine settings based on vibration data.