Input Output Selection Based on Robust Performance: An Active Suspension Application

In [45], a method for selection of actuators and sensors (Input Output (IO) selection) in linear control systems was studied. Its main shortcoming was the inability to effectively deal with structured uncertainties and robust performance. This is partially resolved by two IO selection methods studied here. The key idea is to eliminate actuator/sensor combinations (IO sets) for which a controller meeting a guaranteed level of robust stability or robust performance cannot be designed. The structured singular value plays a key role, so the report starts with a discussion on this concept and how it can be employed for control system design. The first method is closely related to the one in [45], i.e., for each IO set six conditions are checked for existence of a stabilizing KFt, controller. However, these " viability conditions " now apply for a generalized plant extended with scaling filters, which are introduced t o account for structure in the uncertainty block. The resulting condition for IO set viability is a sufficient one, mainly because the scalings are only determined for the IO set including all actuators and sensors and so may not be optimal for other IO sets. The second method is based on a necessary condition, which is mainly due to the fact that the stabilizing property of the controller is dropped. The viability conditions (in the form of linear matrix inequalities) are checked for various frequencies, using transfer matrix data of the generalized plant. A major shortcoming is the inability to effectively deal with uncertainties consisting of more than two full blocks. The two IO selection methods have been implemented in MATLAB and an active suspension control problem for a tractor-semitrailer combination served to evaluate them. A total number of 45 candidate IO sets are assessed for their ability t o meet a specified level of robust performance under one scalar uncertainty. Due t o its sufficiency, the first IO selection method rejects two viable IO sets, while, due to its necessity, the second method accepts two non-viable IO sets. Based on this single example, it is impossible to conclude on the methods' effectiveness in general. In the current implementation, the second IO selection method takes considerably more computation time than the first one; improvement of the efficiency of both met hods merits further investigation. A major topic for future research is IO selection for nonlinear systems. First, it must be revealed if …