APPLICATIONS OF SURROGATE FINITE ELEMENT MACHINE LEARNING APPROACH FOR STRUCTURAL MONITORING

Current maintenance intervals of mechanical systems are scheduled a priori based on the life of the system, resulting in expensive maintenance scheduling, and often undermining the safety of passengers. This problem is particularly relevant in the development of autonomous vehicles, especially in the concept of urban air mobility. The actual usage of the vehicle will be used to predict stresses in the structure and therefore to define maintenance scheduling. Supervised regression machine learning algorithms are used to map a reduced set of data coming from real-time measurements of a structure into a detailed/high-fidelity finite element analysis (FEA) model of the same system, therefore creating a surrogate of the finite element model. The paper will present applications of the approach to a one-dimensional beam structure, modeled with finite element methods. Based on the response of the beam measured at a few reference locations, the surrogate finite element approach determines the entire response of the beam at all spatial locations (displacements, velocities, accelerations, stresses, strains) using neural networks. The FEA-based machine learning approach estimates the stress distribution over the entire system during operations, thus improving the ability to define ad-hoc, safe and efficient maintenance procedures. The effect of type of input features and output and their relationship on the performance of the neural network is discussed, as well as the effect of the beam boundary conditions on network performance.