Study the Vibration of Turning Aluminum 6500 Surface Roughness Based on Sound Chatter Using Machine Learning

This paper experimented with aluminum grade 6500 to predict surface roughness based on the projected workpiece on a CNC lathe machine. The operation used a 50 mm diameter solid bar, 240 mm long, divided into six equal segments. Each segment was machined at depths of 0.25 mm, 0.5 mm, 0.75 mm, 1.0 mm, 1.25 mm, and 1.5 mm. The actual cutting depths were analyzed based on SolidWorks G-code. Under constant feed rate and spindle speed, the sound chatter was recorded for machine learning. The librosa library enhanced the sound chatter utilizing 50 neurons and a 71051 input shape. In parallel, a batch size of 32 and the 10-epoch training over a 9-layer model achieved 50.4% accuracy. Besides, the spectrogram’s purple color indicated significant signal energy between 0 Hz to 64 Hz and 4096 Hz to 8192 Hz, while lighter color intensity showed weaker energy. The peak intensity of energy represents high vibration, and the weak intensity of energy is linked to the low vibration. Additionally, the Abacus simulation showed cutting depths of 0.25 mm and 0.75 mm, resulting in deformations between 8.713e -08 mm and -1.176e -00 mm. The higher deformation values corresponded to less chatter, while lower values indicated low vibration. Overall, deeper cuts on the aluminum grade resulted in peak frequencies associated with a smoother surface finish, whereas shallower cuts produced rougher machined surfaces.