Session: 02-06-01: Machine Learning in Structural Dynamics and Aeroelasticity

Paper Number: 152210

152210 - A Sensor Placement Approach to Monitor Vibrations Based on Data-Driven Principal Component Analysis Techniques 

Vibration testing in aircraft structures is a critical but expensive process, involving a large number of sensors to accurately model real-life vibrations. Traditionally, these sensors are distributed uniformly, resulting in high costs and inefficiencies. This research aims to address these challenges by applying Principal Component Analysis (PCA) to optimize sensor placement during vibration testing. By optimizing the number of sensors needed without compromising the accuracy of vibration data, this work seeks to significantly lower costs and simplify the testing process for aerospace structures.

Existing sensor placement methods rely on lengthy iterations to identify optimal sensor locations, while PCA, an established dimension-reduction Machine Learning technique, offers the potential to streamline this process. Although PCA has been widely researched for structural health monitoring, its application for optimizing sensor placement in vibration testing remains unexplored. This research fills that gap by introducing a novel approach that leverages PCA to optimize sensor quantity with high test accuracy, thus pioneering a new, cost-effective methodology for vibration testing in aerospace.

The proposed sensor optimization will be based on a data-driven methodology that, given a large dataset containing all possible locations for sensors identified by a finite element model, will identify the best locations to monitor the stresses and strains of the system. The inputs of the database will represent all possible sensor locations and contain the quantities measured at each possible location. The output of the database will be identified by applying a PCA approach to the vibration data that characterizes the system as a whole. The data reduction methodology will identify which inputs in the database are more important in representing the output. 

The final paper will contain a presentation of the proposed sensor placement methodology based on Principal Component Analysis, and a numerical validation of the approach to an Alouette Helicopter blade. Initially, a finite element model of the Alouette helicopter blade will be created using Ansys Mechanical. This model will generate a dataset containing possible sensor locations. PCA will be applied to this dataset to identify the optimal sensor locations based on their ability to monitor the dynamic stresses and strains of the system. This approach will reduce the dataset, and output the most critical sensor locations for accurate vibration analysis. Once the optimal sensor locations are determined, the numerical validation will compare the ability of the system to monitor the stresses and strains when sensors are placed at the PCA-optimized locations and when sensors are placed in a traditional uniform distribution.

This project contributes to the aerospace field by introducing a novel method for optimizing sensor placement in vibration testing, which is crucial for ensuring safety and performance in aircraft structures. The outcomes of this research are expected to provide a more efficient, cost-effective sensor placement strategy, making advanced vibration analysis more accessible to aerospace manufacturers and researchers.

Presenting Author: Shrihari Arunachalam San Jose State University

Presenting Author Biography: Shrihari Arunachalam, a junior Aerospace Engineering student at San Jose State University, specializes in sensor optimization and vibration analysis using Principal Component Analysis (PCA). He also designs aerodynamic components for his Formula SAE team and has presented research at the AIAA Student Conference.