Extended neural network-based scheme for real-time force tracking with magnetorheological dampers

Abstract

This paper validates numerically and experimentally a new neural network-based real-time force tracking scheme for magnetorheological (MR) dampers on a five-storey shear frame with MR damper. The inverse model is trained with absolute values of measured velocity and force because the targeted current is a positive quantity. The validation shows accurate results except of small current spikes when the desired force is in the vicinity of the residual MR damper force. In the closed-loop, higher frequency components in the current are triggered by the transition of the actual MR damper force from the pre-yield to the post-yield region. A control-oriented approach is presented to compensate for these drawbacks. The resulting control force tracking scheme is validated for the emulation of viscous damping, clipped viscous damping with negative stiffness, and friction damping with negative stiffness.The tests indicate that the proposed tracking scheme works better when the frequency content of the estimated current is close to that of the training data. The LuGre friction model was adopted as observer in [18] to solve the force tracking task. This short review on existing real-time force tracking schemes shows that the nonmodel-based Heaviside step function approach only works with force feedback; the Dahl model leads to precise current estimation only in special cases, and the LuGre friction model can only be used as observer because this approach cannot be inverted. In this study, the real-time force tracking task without force feedback is therefore investigated by the adoption of soft-computing techniques that allow modeling the inverse MR damper dynamics directly and thereby circumvent the aforementioned drawbacks.Besides the fuzzy logic approach that is used in [19] to directly estimate the inverse MR damper behavior, the neural network method is seen to be able to capture the nonlinear inverse MR damper dynamics. Recurrent neural network models were constructed in [20] to emulate both the forward and inverse MR damper dynamics. Both models take the previous values of the estimated force and voltage, respectively, as inputs among other states. The validation of the inverse model shows some high frequency spikes in the estimated voltage. In the work of Xia [21], the inverse MR damper behavior was modeled by using a multilayer perceptron optimal neural network and system identification. The inverse model is trained and validated with simulated data taking into account the previous values of the voltage. The authors of [22] studied the modeling of the inverse MR damper dynamics using feedforward and recurrent neural networks. For the targeted constant voltage of 1.5 V, the predicted voltage showed spikes of up to AE0.17 V at a frequency that seems to be equal to the damper motion frequency. If the targeted voltage is a sinusoidal function with constant frequency, the time history of the estimated voltage does not describe a pure sine, and thereby, the resulting error is larger than AE0.17 V. In contrast, the forward mo...

Amplitude and frequency independent cable damping of Sutong Bridge and Russky Bridge by magnetorheological dampers

“…The residual force constraint is also visible in the current time history shown in Figure (c) for the time window 10 to 22 s where the displacement amplitude is increased from 20 to 30 mm. When the desired force is constrained by the residual force of the MR damper, the desired current is zero .…”

Adaptive block backstepping control for civil structures with unknown parameters subjected to seismic excitation

“…In this context an interesting force tracking control scheme for MR dampers is described in [9] which is derived by a control-oriented mapping approach to reduce the modeling effort of the inverse MR damper behavior. Reference [10] validates numerically and experimentally a new neural network-based real-time force tracking scheme for MR dampers on a shear frame.…”

“…But the proposed controller (7) only depends on the estimation vector, b θ, which is defined by the designer as (8). 3 It is possible to rewrite (10)…”

Technical Note: Active and Semi-Active Strategies to Control Building Structures Under Large Earthquake Motion

“…Mitchell et al [2012] proposed a wavelet-based fuzzy neurocontroller algorithm for the hazard mitigation of seismically excited buildings employing passive viscous liquid dampers and an ATMD. Weber et al [2013] validated a neural networkbased real-time force tracking scheme for the use on a 5-story building equipped with MR dampers; test results indicated that the proposed tracking scheme works better when the frequency content of the estimated current is close to that of the training data. Pnevmatikos and Gantes [2010] proposed a 2-stage activated-deactivated diagonal braces-based structure equipped with active variable stiffness devices; the control algorithm determines the bracing stiffness based on the frequency content of a range of anticipated earthquake signals, achieving a significant reduction in the response of the controlled structure.…”