Fault Tolerant Neural Aided Controller for Multi Degree of Freedom Structures Experiencing Online Sensor Failure

Abstract

This study validates an adaptive control algorithm capable of compensating for online sensor failure. Online failure is a relevant problem when considering actively damped, multi-story smart buildings experiencing a disturbance event. In recent years, Artificial Neural Networks (ANNs) have proven very efficient in pattern classification and control applications. In this study, the unique application of ANNs involving Radial Basis Functions (RBFs) combined with H∞ optimal control has demonstrated three significant characteristic advantages: (1) real time adaptability, (2) optimal convergence and computation time, and (3) most importantly, no offline training. The novelty of the proposed controller is elucidated by performing disturbance rejection tests involving a scaled two degree of freedom shear frame subjected to a combined H∞ and ANN control. A bench scale structural model is instrumented with piezoelectric sensors and actuators. After the onset of a first mode disturbance, the structural frame is subjected to a complete sensor failure. The proposed controller is shown to enhance the performance of a baseline H∞ controller in the presence of sensor failure.

Real time detection of stiffness change using a radial basis function augmented observer formulation

“…where the dof subscript indicates the degree of freedom or floor where measurements are being taken. Considering a particular degree of freedom's output for the faulty system, y = y dof produced by C = C dof , the baseline observer in equations (5) will produce corresponding outputŷ =ŷ dof . By storing p + 1 points of data, a vector y ∈ p+1 andŷ ∈ p+1 for the kth instant can be created.…”

“…These additions make it suitable for online convergence while maintaining a compact network. Applications of the EMRAN formulation for online adaptive sequential learning applications involving building structures have been presented by Narasimhan et al [24] and Contreras et al [5,6].…”

Damage detection under varying temperature using artificial neural networks

“…As a powerful tool for pattern recognition and classification, one nonparametric approach, the use of artificial neural networks (ANNs), has been widely employed to detect damage in structures regardless of temperature. Different types of neural networks, such as back‐propagation neural networks (BPNNs), self‐evolving networks, Bayesian neural networks, and ensemble of networks, have been successfully applied to various bridges.…”