IMU-based Autonomous Solar Tracking System with Soft Actuators: A Machine Learning Approach

Increasing energy-efficiency and cost-effectiveness is an important research topic towards more economic solar panels. In this paper, we focus on developing a new data-based control design approach that adaptively adjusts the orientation of solar panels, built upon low-cost soft actuators, to increase their energy harvesting efficiency. In particular, the proposed approach is composed of four steps. First, we propose a calibration algorithm (CaA) that automatically creates a database of pneumatic input combinations with a correlating orientation for the yaw and pitch of a vertically mounted solar panel, as measured by an onboard Inertial Measurement Unit (IMU). Second, we design a machine learning algorithm that interpolates a finer granularity of the CaA output. Third, we propose a search algorithm that finds the corresponding set of data for the current Sun azimuth and elevation from the output of the machine learning algorithm. Forth, we propose a control algorithm that utilizes the output, obtained from the search algorithm, to apply the correct pneumatic input such that the solar panel can track the location of the Sun. To show the performance of the proposed approach, we conduct field tests to show that the solar panel with adaptively adjusted orientation under the proposed method can yield much higher power output. Another advantage of the proposed approach is the applicability of the approach to other solar harvesting systems, i.e., roof-mounted solar panels, because the data-based approach does not require physical models to be known or approximated.