Experimental Analysis for Sensor Reduction to Depict Real-Time Applications Through Regression Techniques

The expenses and maintenance requirements rise along with the number of sensors. To solve this problem, strategically placing sensors in real-time applications is essential. This work presents a novel approach that leverages regression analysis to identify representative and non-representative sensors, leading to the removal of redundant sensors without compromising accuracy. The proof-of-concept study employed a fan as the test object to collect sensor data. However, the proposed method can be extended to real-time scenarios such as vehicles, engines, turbines, and gym or sports equipment. Establishing an approximate correlation between representative sensors (close to the item) and non-representative sensors (placed at a distance) can eliminate the need for a larger number of sensor placements. It utilizes different regression models and analyzes the gathered data to determine the use of representative and non-representative sensors. The evaluation process involves techniques such as correlation matrix and eigenvalues to identify usual sensors. A machine learning regression method is used to predict non-representative sensor outputs, including linear regressions (LR) and Gaussian regressions (GR). A mean square error (MSE) is calculated to compare the accuracy of predicted values obtained from both regression techniques. The proposed method offers several benefits, including simplified maintenance, reduced costs, and improved performance with fewer sensors. Accurately identifying representative sensors provides a practical solution for optimizing sensor deployment in real-time applications.