Internal combustion engine durability monitor : Identifying and analysing engine parameters affecting knock and lambda

Abstract: This study has been performed at Powertrain Engineering Sweden AB (PES), a fully owned subsidiary of Volvo Cars Group, which is constantly working to develop and improve internal combustion engines. As part of this work, durability tests are performed to analyse the impact of wear on the engines. At present, there is a strong focus on visual inspections after the engines have undergone durability tests. PES wants to develop a method where collected data from these tests can be used to explain how the phenomenon of knocking and the control of lambda changes over time. The study analyses one specific durability test and investigates the methodology of data analysis by using the open-source software platform Sympathy for Data, with an add-on developed by Volvo Cars Group, for data management, visualisation and analysis. To execute the analysis, engine parameters that affect these systems as well as parameters suitable to use as response variables are identified through literature studies of internal combustion engine fundamentalsas well as internal material, and knowledge acquired at the company. The result is presented in the form of an analysis generated by the node for partial least squares regression (PLSR) which is pre-programmed in Sympathy for Data as well as the images and graphs obtained as output. For knock, the signal for the final ignition angle was found to be suitable to use as the response variable in the PLSR. A suitable response variable for lambda was more difficult to identify, this is why both signals for the measured lambda and lambda adaptation are analysed. Studies of the internal material and knowledge highlighted the fact that several engine subsystems are highly dependent on each other and that even deeper research would be necessary to fully understand the process and identify the primary cause for the variations observed in the generated models. However, partial least squares regression was performed using parameters derived from literature reviews as input (predictors) in order produce regression models to explain the variance in sought response. Well-fitting models could be created with a varying number of latent variables needed for the different responses. The output obtained from the PLSR enables further studies of the specific cases as well as the methodology itself, hence, increase the use of data analysis with the help of the software used in the department for durability testing at PES.