Dual-Mode Adaptive Neural Observer Design for Pure-Feedback Switched Nonlinear Systems with Hybrid Measurement Defects

Abstract

This paper addresses the control challenge for  switched nonlinear pure-feedback systems (SNPS) under
 measurement imperfections and uncertain dynamics. A  neural observer framework simultaneously handles state
 estimation and unknown function approximation during  measurement defects, including data loss and sensor saturation.  Utilizing the mean value theorem, the non-affine structure  is transformed to enable systematic backstepping synthesis  without linearization. Dual adaptive strategies for normal  operation and data-loss scenarios are unified via probabilistic  measurement modeling. Stability guarantees are established  through average dwell-time constraints and switched Lyapunov  analysis, proving uniform ultimate boundedness (UUB) of  all closed-loop signals. Benchmark simulations demonstrate  the controller’s efficacy in maintaining tracking performance  amid intermittent measurements and subsystem switching.  This approach extends existing methods by integrating neural  approximation, switching logic, and defect compensation into a  unified architecture.