Data-Based Parameterization of Thermal Models of a Gearbox

Conventional gear units are installed in almost all vehicles manufactured and developed these days in particular Battery Electric Vehicles (BEV). Despite the already very high sophisticated development of gearboxes, the focus on optimizing mass and efficiency is still important in development. Especially for BEV the efficiency becomes very important. A remarkable feature of a high efficient transmission of a BEV is the low mass and resulting reduced thermal masses. An additional significant challenge is posed by generally higher engine speed of BEVs, which leads to an increase of heat influx in the gearboxes. Overall, the thermal stability of the transmission units decreases, thus the importance of the temperature monitoring is increased. Oftentimes a model is needed for this, especially for topics such as thermal design with regard to temperature distribution or cooling effects on the surface. Under certain conditions having such a model on a test bench facilitates the avoidance of setting a temperature measuring point. In the scope of this work, a potential solution is shown to infer the inner temperature of the gearbox by using measurable temperature on the outer surface. The surface temperatures of the gearbox can be measured on a test bench by sticking sensors onto it. Furthermore using thermal imaging cameras or other methods based on infrared enable contact free measurements. In order to be able to deduce the inner temperature prevailing in the gearbox from these easily determined surface temperatures, a well-known and frequently used thermal model is created, which is based on the simple foundation of thermal networks. For this concrete application, a thermal network with a small number of thermal nodes and therefore low complexity is created. One node represents the point to be observed inside the gearbox, the others are either a heat sink or heat source or a point where the temperature is known or measured. Even these relatively simple models reveal a large number of parameters. These need to be determined, because each thermal source or sink as well as each transition needs to be parameterized. Determining these parameters demands detailed knowledge of the particular transmission unit. To avoid this, a method from the field of machine learning is applied. For this purpose, the gearbox to be modelled or an identical gearbox is analyzed on a test bench. Both, external and internal temperatures are recorded. In addition, data such as speed and power, which are usually available anyway, are required. Based on a sufficiently large data set, the optimization criterion for the approach is created. A method, which is based on evolutionary algorithms is verified using the example of a single-stage differential transmission. For this purpose, measurements on a driveline test bench are collected and evaluated. A model and its parameters are determined and subsequently compared with temperature measurements, derived from the inside of the transmission unit.