The NVH (Noise Vibration Harshness) behavior of modern vehicles
becomes more and more important - especially in terms of new
powertrain concepts, like in hybrid electric or full electric
vehicles. There are many tools and methods to develop and optimize
the NVH behavior of modern vehicles. At the end of the development
process, subjective ratings from road tests are very important. For
objective analyses, different approaches based on artificial neural
networks exist. One example is the AVL-DRIVE™ system, a
driveability analysis and benchmarking system which allows, based
on a very small set of sensors, an adequate objective rating of the
vehicle's driveability. The system automatically detects
driving maneuvers and rates the driveability.
This article presents a method which is able not only to rate
different maneuvers and the behavior of the vehicle but also to
detect phenomena and causes in the domain of NVH. In terms of
effort, one main requirement was to use the same sensor set as the
driveability evaluation system and no additional equipment.
Basis for the method is a large database consisting of about 104
NVH phenomena. In this database the causes and phenomena are linked
to concrete driving maneuvers. That is permissible because most
phenomena can only occur within one specific maneuver. One example
is the so-called CLONK (powertrain phenomenon), which only appears
during a load change. The phenomena are identified by using
patterns of characteristic frequency ranges. So the user
automatically gets the rating on the one hand and information about
possible causes on the other.
This method will be illustrated by the examples of the humming
of the climatic compressor and the coolant pump as well as the
vibrations during the restarting of the combustion engine in a
hybrid test vehicle. Basis for the validation of this method is
data from experiments on the track and the acoustic roller test
bench at the IPEK - Institute of Product Engineering Karlsruhe.