A Non-linear approach to dynamic torque modelling for expedited engine characterisation, control and calibration

Abstract

Abstract Traditional approaches to engine torque mapping apply steady or quasi-steady-state approaches for data collection. Engine actuators are moved to some set-point and the engine is left to stabilise. Data is then recorded at fixed actuator settings over a period of time and averaged, the size of the averaging ‘window’ is often arbitrarily chosen. Such methods are time intensive and inefficient with the majority of the time available for data collection wasted. In this paper, a novel and fully transient characterisation methodology is introduced. Dynamic excitation signals are used in this work to significantly increase the rate of data collection and settling time (after an initial warm-up) is no longer required. Engine dynamic behaviour is described using first-order conditionally linear repeated measurement models, these reflect the inherent structure of the torque data that is a consequence of the experimental method employed. Model parameters are determined using Maximum Likelihood Estimation (MLE). To abstract steady-state data from the dynamic models (a requirement of legacy controllers) the models are extrapolated, in time, for fixed input settings. The efficacy of the extrapolation procedure is clearly demonstrated through direct comparison with actual steady-state sweeps utilised for validation of the method. The approach is shown to be seven times faster than conventional methods based on slow ramps or steady-state test.