A Bayesian Approach to Transfer Path Analysis

In transfer path analysis (TPA), achieving precise results in matrix inversion often presents challenges. Mainstream approaches to address this issue include overdetermining the system and applying regularization. However, overdetermining requires additional indicators, complicating the process and making it impractical for applications with limited sensor mounting space. On the other hand, regularization techniques are frequently criticized for their lack of stability and accuracy. In this paper, a Bayesian framework is applied to classical TPA; the feasibility of the approach is analyzed both within an experimental setup that mimics the engine mounted on the suspension with resilient connections as well as corresponding numerical simulation. Also, to explore the possibility of applying TPA to objects in smaller shapes, an attempt is made to evaluate the Bayesian TPA when relaxing the constraints of an over-determined system.