Abstract
When developing complex mechatronic systems, efficient and automated methods and processes play an important role in compensating the increasing time and cost pressure. High complexity leads, among other things, to a high level of possible error sources in development and operation. Consequently the demand to monitor such systems while operating on the one hand and in the validation activity on the test bench on the other hand has increased in order to detect and avoid possible system errors in advance. This allows a considerable reduction of test bench time and most notably the test bench costs. Damage to the test bench as well as unintended destruction of pre-series test items is not acceptable these days. Generating incorrect data due to unknown effects, especially in long-term test runs, bears an additional risk. This poses new challenges to test bench operators that cannot be solved with conventional tools of test bench monitoring. New tools, that recognize critical states in time and thereby reduce or even avoid the downtime of test benches and the number of damaged prototypes, have to be created. For this purpose advanced monitoring mechanisms must be developed and tested.
A method for model based test bench monitoring is introduced in this paper. This method combines conventional monitoring models with artificial neural networks in an innovative way. This approach solves the trade-off between flexibility in the day-to-day test bench operation and the traceability of the source of defect. In addition to preventing damage from test items and test stand, the presented method also guarantees the validation of measured data. The following examples affect the quality of measured data:
incorrect sensor signals
temperature influence
parasitic friction in the testing device o bearing friction e.g. in the engine when measuring the loading via motor current o bearing friction between torque test ports
slippage in connecting parts
The control process has to recognize these causes and display them to the test bench operator. Moreover an offline analysis of the cause of error, which allows access to collective expert knowledge, is being integrated in order to detect the sources of occurring errors.
In collaboration with the Kröhnert Infotecs GmbH this concept is being translated into a monitoring system. The prototypical implementation takes place at powertrain test bench of IPEK – Institute of Product Engineering. With this example the whole workflow from system analysis and creation of the monitoring model all the way to the operation on a test bench is validated under real circumstances.
This method also qualifies for being transferred into the simulation, into hardware-in-the-loop environments as well as into the vehicle itself. Because of this continuity and modularity it is for example now possible to monitor the interaction of new functions in the controller with already existing hardware components in the validation process.
Keywords: Test bench; simulation; condition monitoring; artificial neural networks; model based condition monitoring;