Abstract
Microscrews measuring M2.0 or smaller are widely used in the assembly of smartphones, tablets, and wearable devices. To assemble delicate components, screw-tightening torques of less than 100 mNm and accuracies of ±5% are typically required. In recent years, high-performance electric screwdrivers have been developed to satisfy these requirements. However, even under strict torque control, the actual screw-fastening force remains unclear. Hence, screw tightening must be performed under various conditions using the torque method to understand the axial-force distribution of the screw and to determine the tightening conditions that satisfy the required fastening force. Directly measuring the axial force in microscrews is challenging because of difficulties in embedding strain gauges or other devices, which consequently complicates the optimization of screw-fastening conditions. Hence, this study proposes a system that accurately estimates the axial force during microscrew tightening by utilizing IoT sensing technology and a machine-learning-based regression prediction model. We design a system that senses and captures multivariate time-series data and develop a machine-learning regression prediction model based on acquired data to accurately estimate the axial force of the screws. Experimental verification using M2 screws shows that the proposed system achieves an axial-force-estimation accuracy of approximately 3%.
