A Neural-Networks-Based Adaptive Disturbance Rejection Method and Its Application to the Control of Hard Disk Drives

Abstract

This paper presents a neural-networks-based disturbance rejection adaptive scheme for dealing with repeatable and nonrepeatable runout simultaneously. The effectiveness of this method is demonstrated empirically on a commercial hard disk drive where the adaptive disturbance rejector is added to a baseline linear time-invariant (LTI) controller. The adaptive scheme can be broken into two subsystems: one subsystem is designed to suppress the repeatable runout (RRO) and the other to attenuate the residual disturbance and nonrepeatable runout (NRRO) by the use of radial basis functions. Two different methods for RRO suppression are employed in conjunction with the neural-networks-based NRRO rejector. The first one is an adaptive feedforward disturbance rejection scheme. The second is a repetitive controller. In both cases the neural modeled disturbance rejector is adapted online further increasing the track-following performance by as much as 6.4%. Experimental results of the schemes at various locations of the hard drive are included to demonstrate the general applicability of the approach on commercial drives. The total reduction of the error during track-following is measured to be as much as 25.4% respect to the baseline LTI controller.

Lift force control of a flapping-wing microrobot

“…In this article, we introduce the notion that the reference signal r(k) in Fig. 5 can be seen as an output disturbance, and therefore, that the reference-following problem considered here is very similar to the disturbance rejection case in [12]. Note that since u(k) is filtered trough P(z),α(k) andβ (k) are not estimates of a(k) and b(k), respectively.…”

“…Note that since u(k) is filtered trough P(z),α(k) andβ (k) are not estimates of a(k) and b(k), respectively. Nonetheless, as explained in [12], the ideas on stability and convergence, for the input disturbance case, discussed in [11] and references therein, apply to this case.…”

“…Disturbance profiles of this kind are reminiscent of the repeatable runout described in the hard disk drive (HDD) literature (see [11], [12], [10] and references therein). Thus, it is possible that the reasons for the appearance of harmonic disturbances in this case are similar to ones in the HDD case.…”

“…Considering this design choice, a natural control strategy is the implementation of algorithms specialized in dealing with the tracking and rejection of periodic signals. In this category, there are the internal model principle (IMP) [5] based algorithms such as those in [6], [7], [8], [9], [10] and other related articles, and also the adaptive feedforward cancelation (AFC) algorithms such as those in [11] and [12] and references therein.…”

“…Since the AFC algorithm is a disturbance rejection scheme, here, the reference signals to be followed are treated as disturbances to be rejected. As in [11] and [12], the frequencies of the relevant signals are known while the amplitudes and phases are assumed unknown. The idea of treating the amplitudes and phases of sinusoidal references as unknowns seems counterintuitive.…”