Artificial neural network-based investigation of the effects of suspension dynamic properties on chassis vibrations

Abstract

Nowadays the comfort level, stability and safety of vehicle are the important parameters considered in comparative analysis while selecting the vehicle. This paper explores a development of mathematical model for investigating the effect of suspension dynamic properties on the chassis vibration. Equations of motion derived for the model with seven degrees of freedom system and solved by Newmark Beta method programmed in MATLAB software to obtain the value of peak acceleration of automobile chassis. Sprung mass, equivalent spring stiffness, equivalent damping coefficient of shock absorber, amplitude of excitation, velocity of excitation and wavelength of excitation were considered as a suspension dynamic properties. The data obtained from the mathematical model gives strong foundation for building an accurate predictive model using an artificial neural network (ANN). Use of GridSearchCv with KerasRegressor optimizes the layer size, adding cross validation improves ANN model by avoiding over fitting of limited data. The ANN model cross validation results obtained with mean acceleration of 4.54792 m/s² and standard deviation of 1.47530 m/s². For validation the experimental investigation of the chassis vibration performed on a car with monocoque chassis, Macpherson front suspension and twist beam rear suspension with coil spring considered and FFT analyzer used for the data collection. This model enables non linear mapping between suspension parameters and chassis response and having ability to handle different road inputs. With the validated theoretical model it is easy to generate large datasets for extreme conditions and it reduces time and cost of testing. This method forms a base to obtain more accurate prediction of the effect of suspension dynamic properties on the automobile chassis vibration to enhance the vehicle performance.