Riassunto analitico
This study investigates the feasibility of reproducing the non-linear and time-variant transmission behaviour of vibration dampers through recurrent neural networks (RNNs). Exploiting the capabilities of the TensorFlow library in Python, the research focuses on the two-valve, electronically controlled dampers used in the semi-active suspensions of Porsche SUVs Macan and Cayenne. Using only synthetic data generated, starting from real track data, on a testing machine, a Neural Network model was trained to establish the relationship between input variables (displacement, velocity, acceleration, and current) and the damper force output. The goal is to create a black-box model capable of predicting the transient force profile and hysteresis behaviour of the damper based on these input variables. A comprehensive dataset of 3600 seconds for each vibration damper was utilized for training the neural network. The quality of the modelling procedure was evaluated within the context of a full vehicle simulation. Additionally, a comparative analysis was conducted between the results obtained from the Neural Network and the currently adopted damper modelling techniques, which rely on look-up tables. To enhance model accuracy, parameters influencing the network structure and training process were individually considered and optimized. This research contributes to advancing the understanding and application of machine learning in reproducing complex automotive dynamics. The results of this investigation showcased a significant advancement in accuracy compared to the currently employed Look-up Tables. Graphical representations of the time evolution of the force and metrics such as Mean Average Error, Mean Root Square Error, and Standard Deviation demonstrate a notable improvement when utilizing the neural network-based modelling approach. These findings underscore the substantial potential that neural networks bring to the simulation of vibration dampers, paving the way for enhanced accuracy and reliability in automotive dynamics modelling.
|