Virtual Torque Sensing: A Model-based Approach for Indirect Measurement of Dynamic Operational Loads on Mechatronic Powertrains

Modern mechatronic powertrains are characterized by an ever increasing power density and high-dynamic operation. In addition, high demands are put on the energy efficiency and the behaviour in terms of noise and vibrations. This is true for powertrains in industrial machines, wind turbines and vehicles.

The increasingly strict requirements, together with a constant pressure on shortening development times, have led to a growing use of models in the design phase of mechatronic systems. State-of-the-art modelling techniques allow to simulate the complex vibration behaviour of mechanical components and systems in detail. However, a purely model-based approach often fails to answer decisively how the machine will behave in actual operating conditions. The remaining uncertainty is usually due to poorly known (time-varying) model parameters or an unknown excitation to which the operational system is subjected. The subsequent discrepancy between model and reality may lead to unwanted vibrations and noise emission, a lack of performance because of sub-optimal control or - in extreme cases - the failure of a component or the entire system.

In principle, such uncertainty can be greatly reduced by instrumenting the powertrain and analysing the measured data. The most relevant dynamic quantities such as forces and moments can only be measured in highly intrusive manners, though. As a result, such measurements are almost never feasible to obtain on operational machines.

This work has aimed to progress towards more insight in the operational behaviour of mechatronic powertrains by combining model and sensor information. It makes use of existing techniques concerning model-based estimation of states and inputs. The a−priori unknown input here corresponds to the torque with which the powertrain is excited by an unknown load. The choice for focusing on the torque estimation follows from the fact that this quantity is of prime importance to the powertrain operation but can be measured directly only in rare cases. The here composed model combines knowledge driveline that  provides the connection to its load. The multiphysical nature of the model allows to combine also measurements from the electrical and mechanical domains.

The resulting virtual torque sensor has been validated experimentally on a here developed and purpose-built test setup. The design of this setup is such that the mechatronic powertrain dynamics are decoupled from those of its surrounding frame, as well as from the environment. Thus, the powertrain model can be identified and employed without precise knowledge of the frame dynamics. The focus of this work is on the high-dynamic character of the unknown torsional excitation. It has been shown that an electrical actuator model is in most cases insufficient for identifying the load torque accurately in the structural excitation bandwidth of 0 Hz to 200 Hz. On the other hand, an electro-mechanical powertrain model does allow to reach this goal.

Next to the necessity of the mechanical model, this work has clarified the influence of available instrumentation on the achievable bandwidth of the virtual torque sensor. Classical criteria for making a suitable selection of sensors do not provide a useful conclusion in this respect. By considering the linearized closed-loop bandwidth of the input estimation, the added value of rotational acceleration sensors becomes visible. Through the use of these sensors, the useful bandwidth of the virtual torque sensor is increased significantly.