Improved Monitoring of Wind Speed Using 3D Printing and Data-Driven Deep Learning Model for Wind Power Systems

Abstract

This study presents a novel method for airflow rate (i.e., wind speed) sensing using a three-dimensional (3D) printing-assisted flow sensor and a deep neural network (DNN). The 3D printing of thermoplastic polyurethane can realize multisensing devices for different flow rate values. Herein, the 3D-printed flow sensor with an actuating membrane is used to simultaneously measure two electrical parameters (i.e., capacitance and resistance) depending on the airflow rate. Subsequently, a data-driven DNN model is introduced and trained using 6,965 experimental data points, including input (resistance and capacitance) and output (airflow rate) data with and without external interferences during capacitance measurements. The mean absolute error (MAE), mean squared error (MSE), and root mean squared logarithmic error (RMSLE) measured using predicted flow rate values by the DNN model with multiple inputs are 0.59, 0.7, and 0.18 for continuous test dataset without interference and 1.16, 3.95, and 0.73 for test dataset with interference, respectively. Compared to the prediction results using single-input cases, the average MAE, MSE, and RMSLE significantly decrease by 70.37%, 88.74%, and 72.26% for test datasets without interference and 51.91%, 53.01%, and 12.20% with interference, respectively. The results suggest a cost-effective and accurate sensing technology for wind speed monitoring in wind power systems.