Towards Active Car Body Suspension in Railway Vehicles

Licentiate thesis, 2009

Today, most railway suspension systems are passive. The most wide-spread exception is active car body tilt systems, which are mounted in some high-speed trains. Replacing some of the passive suspension components with active could reduce the weight and cost of the vehicle. It may also improve passenger comfort without increasing the deflections within the suspension, or, similarly, allow the vehicle to be run at higher speeds or on less smooth tracks, with comfort and deflection kept at today’s levels. This thesis deals with background studies of a model of a railway vehicle, aiming towards actively controlling its vertical secondary suspension, i.e. the part of the suspension that is fitted vertically between the bogie frame and the car body. First, some requirements on the actuator, e.g. maximum forces, are studied, for some cases of replacing passive components with active. Those cases are: removing the antiroll bars, removing the pneumatic systems of the air-spring, and both combined, in all cases adding 2 actuators in the vertical direction for each bogie. The forces the actuators have to be able to deliver are high, but still within reason to implement. Also, the possibility to use a single-input single-output (SISO) control design is studied. It is found that neither input/output pairing, nor using stationary decoupling matrices, gives any promising results that a SISO control design could be based on. The coupling between the inputs and outputs is found to be both very high, and very frequency dependent. To make multiple-input multiple-output (MIMO) control design a feasible choice, the original nonlinear model with 330 states is linearized, and different methods of reducing this model are studied. A model reduction algorithm was developed, that was better suited to this problem than the two standard methods it was compared to. The new algorithm is both less computationally demanding, and for this model produces reduced models, that have gain curves that are closer to those of the full linear model, within the interesting frequency region. Finally, an attempt is made at designing a linear quadratic (LQ) control, and the difficulties with that control strategy on this particular model are discussed. Additional work is needed to fully understand the model, and to find a control law that offers an advantage over the fully passive system.