A Comparison of On-Engine Surge Detection Algorithms using Knock Accelerometers

Paper #:
  • 2017-01-2420

Published:
  • 2017-10-08
Citation:
Kerres, B., Cronhjort, A., Mihaescu, M., and Stenlaas, O., "A Comparison of On-Engine Surge Detection Algorithms using Knock Accelerometers," SAE Technical Paper 2017-01-2420, 2017.
Pages:
11
Abstract:
On-engine surge detection could help in reducing the safety margin towards surge, thus allowing higher boost pressures and ultimately low-end torque. In this paper, experimental data from a truck turbocharger compressor mounted on the engine is investigated. A short period of compressor surge is provoked through a sudden, large drop in engine load. The compressor housing is equipped with knock accelerometers. Different signal treatments are evaluated for their suitability with respect to on-engine surge detection: the signal root mean square, the power spectral density in the surge frequency band, the recently proposed Hurst exponent, and a closely related concept optimized to detect changes in the underlying scaling behavior of the signal. For validation purposes, a judgement by the test cell operator by visual observation of the air filter vibrations and audible noises, as well as inlet temperature increase, are also used to diagnose surge. The four signal treatments are compared with respect to their reliability as surge indicator and the time delay between surge onset and indication. Results show that the signal power in the surge frequency band has reasonably good properties as surge indicator. The normal Hurst exponent is problematic, since periodic vibrations from engine firing dominate the scaling behavior. Root mean square and the above mentioned scaling exponent do not measure vibrations caused by surge directly, but rather the reduction in housing vibrations due to the engine load drop. Nevertheless, it was found to be possible to design an indicator that gives good results based on the change in scaling behavior.
Access
Now
SAE MOBILUS Subscriber? You may already have access.
Buy
Select
Price
List
Download
$27.00
Mail
$27.00
Members save up to 40% off list price.
Share
HTML for Linking to Page
Page URL

Related Items

Technical Paper / Journal Article
2011-04-12
Training / Education
2018-09-17