Validation of a Data-Based Thermal Model of a Gearbox

Abstract

"Oil sump temperature in conventional gearboxes is one of the most important technical framework conditions in terms of research, development and especially in testing of gearboxes. Nevertheless, it is not always possible to set a suitable sensor or to access sensor data with sufficient quality. The need for a simulated oil sump temperature is, among others, for such purposes. In a previous work, an approach for this is presented. A thermal model of a gearbox is generally created on the basis of lumped-element models. Therefore, the thermal masses, such as oil, gears and the gearbox housing, are modeled as lumped-capacitance and the thermal transitions between them as thermal resistances. In addition, there are thermal sources due to heat generation and thermal sinks due to cooling. In principle, a high level of detailed knowledge would be required to parameterize this lumped-element model. To circumvent this, a data-based approach for parameter identification is chosen. For this purpose, an optimization problem is formulated which aims at minimizing the error between simulated and measured values. The methods of genetic algorithms are used to solve the optimization problem. This computationally expensive step is done in advance, apart from the actual use of the model. Thus, a model with little detailed knowledge and without extensive model building can be generated. The preliminary conceptual paper focusses on gearboxes of battery electric vehicle (BEV) with low thermal masses. In contrast, transmissions in conventional and in hybrid electric vehicles (HEV) are steadily increasing in size, number of gears, and thermal mass. The generality of the approach described above is shown by modeling a very large gearbox in the present work. For the specific case, even a gearbox with a coolant/oil heat exchanger is considered. Beside this extension of the model, the former approach has additionally been improved regarding the accuracy of the results. By varying the measured and simulated model inputs, the parameters can be determined more accurately. Furthermore, the final simulation model is upgraded with these additional model inputs as well, which leads to a significant improvement of the simulation results. The model complexity even decreases. These results base on a large measurement data set, which is necessary for the identification of the parameters. In addition, measurement data is needed for the validation of the method. An extensive series of measurements have been taken on an electric powertrain test bench and is presented within this work. Both, generic measurement runs, which are ideal for system identification, and runs similar to a real driving data were recorded."