19th Asia-Pacific Vibration Conference (APVC2021)

Dynamic input forces of the power battery pack on an EV are usually required for structural integrity and reliability. Force reconstruction based on accessible acceleration signals is of great significance, since forces are often difficult, if is not impractical, to be directly measured in engineering. Problems occurs to force reconstruction, such as ill-posedness, uncertainties in structure parameters or measured responses, especially with a complex structure. In this paper, Bayesian algorithm is adopted to reconstruct in time history the multiply input forces of the power battery pack on an EV, whose vibrational modes are complex ones. The ill-posed problem during matrix inverse computation is addressed by an inherent regularization. Model uncertainties and introduced hyper-parameters are determined by Monte Carlo Markov Chain process. Error propagation induced by uncertain structural parameters and measurement noise is analyzed according to Bayesian inference. The error bounds and distribution of unknown input forces are finally obtained by delivering corresponding posterior probability density functions. The reconstructed forces have good consistency with the directly measured axial forces based on customized bolt force sensors.

Force reconstruction,Bayesian algorithm,Uncertain structure,Error propagation