Abstract
In this study a Neural Network based fault tolerant control is proposed to accommodate oil leakages in a magnetorheological suspension system based in a half car dynamic model. This model consists of vehicle body (spring mass) connected by the MR suspension system to two lateral wheels (unsprung mass). The semi-active suspension system is a four states nonlinear model; it can be written as a state space representation.
The main objectives of a suspension are: Isolate the chassis from road disturbances (passenger comfort) and maintain contact between tire and road to provide better maneuverability, safety and performance. On the other hand, component faults/failures are inevitable in all practical systems, the shock absorbers of semi-active suspensions are prone to fail due to fluid leakage but quickly detect and diagnose this fault in the system, avoid major damage to the system and ensure the safety of the driver.
To successfully achieve desirable control performance, it is necessary to have a damping force model which can accurately represent the highly nonlinear and hysteretic dynamic of the MR damper. To simulate parameters of the damper, a quasi-static model was applied, quasi-static approaches are based on non-newtonian yield stress fluids flow by using the Bingham MR Damper Model, relating the relative displacement of the piston, the frictional force, a damping constant, the stiffness of the elastic element of the damper and an offset force.
The Fault detection and isolation module is based on residual generation algorithms. The residua r is computed as the difference between the displacement signal of functional and faulty model, when the residual is close to zero, the process is free of faults, while any change in r represents a faulty scheme then a wavelet transform, (Morlet wave function) is used to determine the natural frequencies and amplitudes of displacement and acceleration signal during the failure, this module provides parameters to the neural network controller in order to accommodate the failure using compensation forces from the remaining healthy damper. The neural network uses the error between the plant output and the neural network plant for computing the required electric current to correct the malfunction using the inverse dynamics function of the MR damper model.
Consequently, a bump condition, and a random profile road (ISO 8608) described by the power spectral density (PSD) of its vertical displacement, is used as disturbance of control system. The performance of the proposed FTC structure is demonstrated trough simulation. Results shows that the control system could reduce the effect of the partial fault of the MR Damper on system performance.
Adaptive Neural Network Control of Zero-Speed Vessel Fin Stabilizer Based on Command Filter
