Recent automotive technological advancements mainly focus on improving fuel economy with satisfactory emission levels, leading to a significant increment of engine system complexity, especially diesel engines. This increases the number of engine control parameters, making the engine calibration process challenging and time-consuming using the conventional map-based approach. Note that engine calibration is a crucial step in achieving optimal engine performance with satisfactory emissions, and it is an expensive process in general. With the advancement and widespread adoption of machine learning methods for control applications, it is now possible to use a black-box model with intelligence to efficiently calibrate nonlinear systems without detailed knowledge of system dynamics. The surrogate-assisted optimization approach is an attractive way to reduce the total computational budget for obtaining optimal solutions. This makes it special for its application to practical optimization problems requiring a large number of expensive evaluations.The current research work focuses on the problem of performing engine calibration using the surrogate-assisted optimization approach. The objective is to find the trade-off curve between engine efficiency in terms of brake specific fuel consumption (BSFC) and its NOx emissions by efficiently optimizing various control parameters. The complete study is divided into three parts. The first part deals with modifying the original algorithm for efficiently handling the practical system with measurement noise. A new constrained handling algorithm is proposed for lower confidence bound (LCB) criteria that showed good performance for both deterministic and stochastic systems. Furthermore, two extensions based on the expected improvement (EI) criterion are proposed for handling stochastic multi-objective problems.After the methodology development for handling stochastic systems, the second part validates their efficacy for performing the engine calibration in a simulation setting. All three algorithms are compared to identify the best approach for its implementation on the actual engine experimental setup. Three control parameters, namely variable geometry turbocharger (VGT) vane position, exhaust-gas-recirculating (EGR) valve position, and the start of injection (SOI), are calibrated to obtain the trade-off between engine fuel efficiency performance (BSFC) and NOx emissions within the constrained design space. The simulation study identifies the lower confidence bound (LCB) criteria with the proposed constraint handling approach to work well in the stochastic setting, compared with the other two extensions. Therefore, this approach is used for the experimental evaluation of the proposed surrogate-assisted optimization for engine calibration.Finally, the third part is the experimental validation. It is the first step towards automating the entire engine calibration process. Experimental evaluations are performed on a 6.7L Ford diesel engine to validate the algorithm's efficacy. Problems with different complexity are formulated and evaluated using the proposed approach. Initially, a simpler problem with two control variables is formulated to get the confidence to perform the experiments using the proposed algorithm. Two variables: EGR valve position and VGT vane positions, are calibrated to obtain a trade-off between engine efficiency (BSFC) and NOx emissions. After observing promising results, the study is concluded with a more complicated three control variable problem. An external electrically assisted boosting device (eBoost) is added to the engine system to perform calibration. Results showed improved engine performance using the eBoost with a significant reduction in calibration effort in terms of the number of experimental evaluations. The study successfully demonstrated the application of the surrogate-assisted optimization approach to a practical engine system and opened the door to automate the engine calibration process with reduced calibration efforts.
Read
