Thursday, March 18
4:30 - 5:30 PM
Room 1001 EECS
Friction is a highly nonlinear process that depends on many physical parameters. In this research, we focus on identifying the mechanism of the friction process of lubricated lip seal, sliding against steel shaft at low speed, where "low speed" means a sliding speed where the lubricant film is on the order of surface roughness dimensions or less.
Lubricated sliding lip seals are important components in many hydraulic devices. The requirements imposed by today's high precision machines motivate the precise simulation of friction between these seals and sliding components. The objective of this research is to develop models which successfully simulate the friction process with the velocity data of the sliding shaft, and the lip seal friction data as the input and output signals, respectively.
The complexity of friction process makes it very difficult to develop a physically based models. This motivates us to use an empirical system identification techniques. In this research, we take a macroscopic point of view, and instead of incorporating (microscopic) physical parameters into the model, we assume the system consists of nonlinear and linear components whose characteristics can be described by nonlinear functions or scalars. There are many advantages of this approach as follow. The modeling process can be much simpler than that based on physics, and often gives a model with fewer parameters. To some extent, the macroscopic parameters could give more intuitive interpretation of the friction process than the physically based models.
In this research, we present two different approaches of nonlinear system identification. As our first approach, we develop Hammerstein type models. Because of the non-stationary nature of friction process and time invariant assumption of the model, an adaptive algorithm cannot be used for the estimation of the model parameters. We use least squares gradient search algorithms to estimate the models parameters. As our second approach, we present state space model. In this model, we estimate some internal signals which are not accessible for measurement. For the estimation of these internal signals, Extended Kalman filter is used. Since both the internal signals and the model parameters are unknown, they must be estimated at the same time. The method used for the simultaneous estimation is called an "Extended Least Squares". A recursive Extended Least Squares algorithm for our state space model is presented.