Surface finish and morphology play a vital role in the reliability and life of cam-roller or slider mechanisms, with improvement due to lesser contact pressure-based fatigue and failure as surface finish improves. An experimental setup using magnet-based abrasive finishing techniques is designed and developed to conduct studies on a double camshaft follower arrangement that resulted in less than 300 nm finish at an enhanced finishing rate. Finishing rate improves greatly as speed of rotation of the workpiece increases with relatively bigger abrasive particles while the best finish is obtained at lower speeds and particle sizes with progression of the finishing cycles. Subsequently, using the results as the training dataset, a machine learning system is developed to overcome the limitations of surface measurement techniques and human interventions, to predict outcomes like number of cycles, speed of rotation of workpiece, time to reach the desired finish and particle size to be used for the given material of workpiece, initial surface roughness and final desired surface finish. The predictions from the completely trained system are compared with the physics-based model and the real data and the error is found to be within 20 nm which has led to development of a highly reliable system.
Read full abstract