This study investigates adaptive neural network (NN) state feedback control and robust observation for an active suspension system that considers parametric uncertainties, road disturbances and actuator saturation. An adaptive radial basis function neural network is adopted to approximate uncertain non‐linear functions in the dynamic system. An auxiliary system is designed and presented to deal with the effects of actuator saturation. In addition, since it is difficult to obtain accurate states in practice, an NN observer is developed to provide state estimation using the measured input and output data of the system. The state observer‐based feedback control parameters with saturated inputs are optimised by the particle swarm optimisation scheme. Furthermore, the uniformly ultimately boundedness of all the closed‐loop signals is guaranteed through rigorous Lyapunov analysis. The simulation results further demonstrate that the proposed controller can effectively suppress car body vibrations and offers superior control performance despite the existence of non‐linear dynamics and control input constraints.

Vehicle performance is largely affected by the properties of the suspension system, where semi-active suspension has been widely used in mass production of vehicles owing to its characteristics such as internal stability and low energy consumption. To solve the contradiction between ride comfort and road handling, road estimation based semi-active suspension has received considerable attention in recent years. In order to provide accurate estimation for advanced control strategies applications, this paper aims to develop a new method that can provide precise road class estimation based on measurable suspension system response (i.e. sprung mass acceleration, unsprung mass acceleration and rattle space). The response signal is first decomposed using wavelet packet analysis, and features in both time and frequency domains are subsequently extracted. Then, minimum redundancy maximum relevance (mRMR) is utilized to select superior features. Finally, a probabilistic neural network (PNN) classifier is applied to determine road classification output. The most representative semi-active control strategy, i.e. skyhook control, is used to validate this method, and simulation results with varying conditions including different control parameters and sprung mass are compared. The results show that unsprung mass acceleration is most suitable for road classification, and more robust to varying conditions in comparison to other responses.

This paper presents a comprehensive comparison and analysis for the effect of time delay on the five most representative semi-active suspension control strategies, and refers to four unsolved problems related to semi-active suspension performance and delay mechanism that existed. Dynamic characteristics of a commercially available continuous damping control (CDC) damper were first studied, and a material test system (MTS) load frame was used to depict the velocity-force map for a CDC damper. Both inverse and boundary models were developed to determine dynamic characteristics of the damper. In addition, in order for an improper damper delay of the form $t+τ$ to be corrected, a delay mechanism of controllable damper was discussed in detail. Numerical simulation for five control strategies, i.e., modified skyhook control SC, hybrid control (HC), COC, model reference sliding mode control (MRSMC), and integrated error neuro control (IENC), with three different time delays: 5 ms, 10 ms, and 15 ms was performed. Simulation results displayed that by changing control weights/variables, performance of all five control strategies varied from being ride comfort oriented to being road handling oriented. Furthermore, increase in delay time resulted in deterioration of both ride comfort and road handling. Specifically, ride comfort was affected more than road handling. The answers to all four questions were finally provided according to simulation results.

Inspired by unsupervised feature learning and deep learning, this paper provides a novel classification method for advanced suspension system based on Deep Neural Networks (DNNs). Sparse autoencoder and softmax regression are chosen to form deep structure and the parameters are trained by deep learning. Aiming at showing the superiority of DNNs based road classification method, a simulation of a B-class vehicle with skyhook control is performed in CarSim, and three measurable system responses, i.e., centre of gravity (C.G.) of sprung mass acceleration, rattle space and unsprung mass acceleration are chosen and three independent classifiers are established. Simulation results show that the classifier using unsprung mass acceleration has the highest accuracy and better performance than existing methods. Because of the adaptive learning ability and the deep structure, the proposed method can save work and provide higher classification accuracy.

Considering the importance of increasing driving safety, the study of safety is a popular and critical topic of research in the vehicle industry. Vehicle roll behavior with sudden steering input is a main source of untripped rollover. However, previous research has seldom considered road excitation and its coupled effect on vehicle lateral response when focusing on lateral and vertical dynamics. To address this issue, a novel method was used to evaluate effects of varying road level and steering wheel input on vehicle roll behavior. Then, a 9 degree of freedom (9-DOF) full-car roll nonlinear model including vertical and lateral dynamics was developed to study vehicle roll dynamics with or without of road excitation. Based on a 6-DOF half-car roll model and 9-DOF full-car nonlinear model, relationship between three-dimensional (3-D) road excitation and various steering wheel inputs on vehicle roll performance was studied. Finally, an E-Class (SUV) level car model in CARSIM^® was used, as a benchmark, with and without road input conditions. Both half-car and full-car models were analyzed under steering wheel inputs of 5°, 10° and 15°. Simulation results showed that the half-car model considering road input was found to have a maximum accuracy of 65%. Whereas, the full-car model had a minimum accuracy of 85%, which was significantly higher compared to the half-car model under the same scenario.

This paper presents a novel electrical variable stiffness device (EVSD) and its application in seat suspensions. The EVSD is inspired by the conventional mechanical variable stiffness device (VSD) and the force-current analogy, which applies an inductor and a variable resistor to simulate the functions of the spring and variable damper in the mechanical VSD. The electrical system has less physical limitation than the mechanical one in the practical application. Besides, the EVSD requires no power supply except for control signals at milliwatts level. The test result shows the stiffness controllability of the proposed EVSD, and matches with the system model in the simulation. Seat suspensions have been extensively utilised in vehicles, especially heavy-duty ones, to protect drivers' health and improve the ride comfort. An EVSD based seat suspension is investigated in this papers. The seat suspension applies the suspension deflection, circuit current and seat acceleration, which are all easy to measure with sensors, as feedback to estimate the mismatched disturbance in the system; and a $H_\infty$ controller with compensation of the mismatched disturbance is designed for the seat suspension. The experiments validate the effectiveness of the proposed seat suspension and controller. This EVSD based seat suspension consumes ignorable energy and has great potential in practical applications.