Railway Wheelset Active Control and Stability via Higher Order Neural Units

Abstract

This article investigates an unconventional approach to solving the control of lateral displacement for railway bogie wheelsets using recurrent higher order neural units (HONUs). Although studies addressing control of independently rotating wheelsets have shown promising results, they are rarely applied by railway manufacturers. Research and developments in modern bogie design are trending toward active yaw control design as an extension to conventional wheelsets mechanics, particularly for higher speeds. We investigate a model-reference architecture for active control via setpoint tracking of lateral displacement. Then, a new HONU sliding mode architecture is derived to solve convergence for zero lateral displacements in higher running speeds which is a more profoundly complex issue in maintaining minimal hunting motion. Starting from the property of nonlinear polynomial architecture of HONUs with in-parameter linearity, we derive a time-variant state-space representation via nonlinear identical decomposition. Then, an input-to-state stability (ISS) approach is applied to prove the local asymptotic convergence of the applied algorithm in each state point and the bounded-input-bounded-state stability of the entire nonlinear adaptive control loop. Using ISS theory, we also prove the global asymptotic stability of the HONU sliding mode controller for the actively controlled wheelset system. The techniques are validated by simulations and a real roller rig system.