Optimal sensor placement using data-driven sparse learning method with application to pattern classification of hypersonic inlet

Abstract

Placing the right amount of sensors in key locations is critical for system monitoring. In real applications, the determination of sensor placement is a compromise between monitoring performance and the costs of installation and maintenance. Given the well-interpretability of sparse learning, this paper proposes an efficient data-driven method to obtain the optimal sensor subset from the entire candidate sensor set. In order to make our model more robust to outliers and overcome the limitation of inconsistent coefficients for multiple class optimization problem, our proposed method introduces a special norm to realize the similar sparse structures of coefficients. Considering that the redundant data cannot effectively improve the real-time condition monitoring performance of engineering systems, our proposed method also includes a redundant information elimination model, which is rarely investigated in data-driven methods for optimal sensor placement problem, and this elimination model is designed by exploring the diversity of measurement data of different sensors. What’s more, we provide an alternating iteration algorithm to solve the non-smoothness convex problem of our proposed data-driven method, and the proof of its convergence has also been presented. The optimal sensor subset can be determined by the rank of the coefficients obtained by the alternating iteration algorithm. Finally, the effectiveness and feasibility of our proposed method are verified by a large number of experiments, including validation experiments on benchmark data sets and a real engineering example on the inlet model of hypersonic aircraft engine.