Machine Learning to Determine Handle Force and Direction Using Strain Gauge Measurements

This paper introduces the theory and design for a push handle strain gauge system using machine learning to evaluate force and direction. The Able Innovations ALTA ™ patient transfer system is a new alternative to sling transfers, but the new system is heavy, and healthcare workers will require power assistance to move the system through the facility and to position the transfer system. Handles on the system provide the expected push bars that staff currently use on gurneys and the proposed sensor system measures the applied force magnitude and direction applied to these handles. Machine Learning is used to analyze sensors measurement to predict the angle and magnitude of the force. The results show that the strain gauge sensor provides a linear relationship for applied forces without requiring detailed calibration of each of the sensors. Machine Learning was used to address asymmetry in the mechanical design so that forward/backward and lateral forces could be combined into a combined force magnitude and direction. The results are presented for forces in 8 different directions with 2 gasket materials, 2 handle mount methods and an applied force range of 0 to 3 kg (typical for expected human effort). The best performance machine learning method is the decision tree which predicts the direction of force with an accuracy of99.1 % and force magnitude with an accuracy is 99.8%.