This paper was produced for the 2019 NAFEMS World Congress in Quebec Canada
The powertrain is supported by the vehicle chassis through mounts. Dynamic displacements at the active side of the mounts, induced by powertrain operation, are critical to NVH. The transmission of these displacements, through the mounts and into the vehicle, can cause vibrations at the steering wheel, the seat tracks and at the floor pan; it can also lead to structure-borne noises inside the passenger's compartment. The active side mount bracket can amplify the dynamic displacements, induced by powertrain operation, in particular when the bracket has a resonant frequency within the powertrain operating range; reducing the compliance of the active side mount bracket translates into lower mount displacements and consequently improved NVH. Factors influencing the bracket compliance include: material, mass and stiffness distribution and the interface to the powertrain.
It is very important to have a method that predicts the active side mount bracket early during the development cycle of this component. Typically, FEA-based methods are used by NVH CAE engineers to assess the bracket compliance and drive the design to meet the required NVH performance. However, reduction in design lead time, increase in engineering efficiency and First Time Right Design methodology require an alternative method to predict the mount bracket compliance and drive its design faster than ever before. Examination of newly developed technologies and disruptive philosophies reveals the potential for machine learning algorithms to be the virtual tool of choice to predict the bracket compliance while fulfilling the previously mentioned requirements.
This paper attempts to examine the application of machine learning in the prediction of mount bracket compliance. Unfortunately, machine learning requires a large sample size of compliance data associated with different bracket designs. Consequently, FEA models are generated, using adaptive meshing, to predict the mount bracket compliance for a large number of design configurations. Several input variables are used to define each of the design configuration; these variables include: (1) mount bracket mass distribution, (2) geometry of bracket bolting pad, (3) bolt mount pad attachment pattern, (4) mount pad centroid location, (5) distance of mount pad centroid location to mount elastic centre location, (6) amount of attachment points and (7) material properties. The FEA analysis, performed for each bracket configuration, predicts the bracket maximum compliance in the X, Y and Z directions and the associated resonant frequencies. This paper uses multi-label supervised classification algorithms in order to handle the complexity of input data and associated FEA results. Based on the generated data, this paper will evaluate the effectiveness of several machine learning algorithms, including Gaussian, Kriging and Neural Networks. Comparisons between FEA-based and machine learning based predictions are made to select the best machine learning algorithm for mount bracket compliance predictions.
The developed machine learning model is offered as a virtual tool to predict the mount bracket compliance based on critical bracket design parameters. This model can also be used to understand the sensitivity of the bracket compliance to the different design parameters and eventually establish potential design guidelines for mount bracket dynamics compliance.