Prediction of surface roughness and cutting force induced during rotary ultrasonic bone drilling via statistical and machine learning algorithms

The ultrasonic-assisted drilling of bone is gaining significant attention in orthopedic research and clinical applications due to the advantage of reducing thermal necrosis, cutting force, and microcracks. Mechanical and surface morphological damage due to cutting force and surface roughness rise during bone drilling may result in implant failure during osteosynthesis. It is difficult to predict and measure cutting force and surface irregularities during real-time medical orthopedic surgery. Therefore, the present work uses the application of different machine learning models to predict cutting force and surface roughness induced during bone drilling. The predictive results of machine learning models were compared with statistical analysis. During the study, rotary ultrasonic drilling is performed on the pig femur bone. The surface roughness and cutting forces are monitored using a surface roughness tester and a dynamometer respectively. Predictive models were developed by drilling parameters such as spindle rotational speed, abrasive girt size, and feed rate. Various machine learning algorithms such as ridge regression, lasso regression, support vector regression, multi-linear regression, and artificial neural networks were leveraged and compared at different error metrics to provide a robust predictive model. It was observed that ridge regression has the least error metrics compared to other machine learning algorithms during surface roughness prediction. Moreover, the most accurate model for predicting cutting force was support vector regression. The error metrics for statistical analysis were comparatively higher than the machine learning algorithms. Therefore, machine-learning algorithms are preferable for adequate prediction of surface roughness and cutting force induced during bone drilling compared to statistical analysis.