Estimation, Identification and Data-Driven Control Design for Hard Disk Drives

The demand for online storage has been increasing significantly during the last few years. Hard disk drives are the primary storage devices used in data centers for storing these online contents. The servo assembly of the dual-stage Hard Disk Drive (HDD) is composed of the Voice Coil Motor (VCM) and the Mili-Actuator (MA), where the VCM is responsible for coarse positioning at low frequency regions and the MA is responsible for fine positioning at high frequency regions. Controlling these two actuators is very critical in precision positioning of the read/write head, which is mounted at the edge of the servo assembly. In this dissertation, the precision positioning of the head during the self-servo writing process as well as feed-forward and feedback controls in the track following mode are considered. This dissertation discusses three control design methodologies for hard disk drives servo systems, in order to improve their performance as well as their reliability. The first is a state estimator for non-uniform sampled systems with irregularities in the measurement sampling time, which estimates the states at a uniform sampling time. The second is an online uncertainty identification algorithm, which parameterizes and identifies the uncertain part of transfer functions in a dual-stage HDD. The third is a frequency based data-driven control design methodology, which considers mixed H_2/H_infinity control objectives and is able to synthesize track following servo systems for dual stage actuators utilizing only the frequency response measurement data, without the need of identifying the models of the actuators.The state estimator design for non-uniform sampled systems with irregularity in the measurement sampling time is considered, where it is proposed to design an observer to estimate the states at a uniform sampling time. This observer is designed using a time-varying Kalman filter as well as a gain-scheduling observer. The Kalman filter has the optimal performance, while the gain-scheduling observer requires relatively lower computational power. Simulations are conducted involving the self-servo writing process in hard disk drives, where performance as well as computational complexity of these two observers are compared under different noise scenarios.Uncertainties in system dynamics can change the closed loop transfer functions and affect the performance or even stability of the control algorithm. These uncertainties are parameterized as stable terms using coprime factorizations, and are identified in an online fashion. The uncertainty identification, in comparison to the complete transfer function identification, requires less computational power as well as a smaller order for the identified transfer function.The proposed online uncertainty identification algorithm is utilized to factorize and identify the uncertain part of transfer functions in a dual-stage Hard Disk Drive (HDD). The dual-stage actuators' gains and resonance modes are affected by temperature variations, which in turn affect all closed loop transfer functions. Therefore, these transfer functions must be periodically updated in order to guarantee the convergence and stability criteria for the adaptive Repeatable Run-Out (RRO) following algorithm proposed in [61, 62]. Experimental results conducted on a hard disk drive equipped with dual-stage actuation, confirm the effectiveness of the proposed identification algorithm.A frequency based data-driven control design considering mixed H_2/H_infinity control objectives is developed for multiple input-single output systems. The main advantage of the data-driven control over the model-based control is its ability to use the frequency response measurements of the controlled plant directly without the need to identify a model for the plant. In the proposed methodology, multiple sets of measurements can be considered in the design process to accommodate variations in the system dynamics. The controller is obtained by translating the mixed H_2/H_infinity control objectives into a convex optimization problem. The H_infinity norm is used to shape closed loop transfer functions and guarantee closed loop stability, while the H_2 norm is used to constrain and/or minimize the variance of signals in the time domain.The proposed data-driven design methodology is used to design a track following controller for a dual-stage HDD. The sensitivity decoupling structure[34] is considered as the controller structure. Thecompensators inside this controller structure are designed and compared by decoupling the system into two single input-single-output systems as well as solving for a single input-double output controller