Machine learning approaches to estimate suspension parameters for performance degradation assessment using accurate dynamic simulations

Abstract

The suspension is one of the most vital systems in a vehicle. Its performance degrades over time due to road conditions. The suspension parameters of a moving vehicle are difficult or sometimes impossible to measure within the desired level of accuracy due to high costs and other associated impracticalities. In this work, we comprehensively investigate various machine learning (ML) methods to estimate the suspension parameters for assessing performance degradation. These methods include particle swarm optimization backward propagation, radial basis function neural network, generalized regression neural network, deep belief network, wavelet neural network, Elman neural network, extreme learning machine, and fuzzy neural network. During the training process, the vehicle states, calculated using a semi-recursive multibody model, are used as the inputs to predict the stiffness and damping coefficients of the suspensions. The semi-recursive multibody model considers the dynamic properties of all the components, which enables accurate vehicle states and characteristics. In addition, we compare the performance of the ML methods by using the reference data (multibody model data). The results show that the ML approaches can estimate accurate stiffness and damping coefficients in real-time. • An accurate multibody model is used for simulations to acquire vehicle states in real-time. • ML methods are created to assess the suspension performance degradation using vehicle states. • The capabilities of ML methods are explored for performance degradation assessment.