Structural dynamic optimization of vehicle brake pad design for squeal noise reduction

A dynamic finite element (FE) model has been proposed for use in optimizing the vehicle brake pad design for squeal noise reduction. The proposed model is intended primarily to investigate the role of pad structure design on the vibration instability of the disc brake system, which leads to squeal problem. The stationary system model is initially assembled from FE components by applying appropriate connection stiffness and boundary conditions that are eventually validated by cross-point FRF experiments. Subsequently, friction coupling at the rotor-pad interface is investigated from an analytic viewpoint. The friction effect produces an asymmetric stiffness matrix that gives rise to complex conjugate eigenvalues. The positive real part of the eigenvalue indicates the existence of theoretical negative damping that tends to cause the modes to become unstable and generates squeal noise response. In the current parameter and eigensensitivity studies, beneficial pad design changes can be identified and implemented in the detailed FE model to predict as to whether a more stable dynamic response can be obtained. The improved pad design can be obtained by setting up an optimization problem for the complex-valued eigensolution. Using a neural network-based squeal noise evaluation technique, the complex-valued eigensolutions are further simplified to give a set of simulated scalar subjective ratings. Thus, the numerical problem reduces to the optimization of these scalar subjective ratings relative to the physical design variables. The process is accomplished in two steps. First, the optimization of prototype design features is performed with respect to simulated subjective responses using the FE model. Then, the physical design variables are optimized with respect to the approximated design response surface using a set of second-order polynomial functions. In this way, the optimal pad structure having less propensity to squeal can be determined. Finally, both the FE model and neural network (NN) simulation are verified by actual vehicle test data, which includes driver's subjective assessments and measured noise spectra from an interior microphone placed in the passenger compartment.