Optimization signal writing with machine learning assisted control

Abstract

The high-precision signal writing machine, experiencing a 0.1% failure rate due to discrete fourier transform (DFT) of position error signal (PES) exceeding control limits, can be improved with an appropriate controller gain. This paper combines machine learning (ML) classification and controller optimization to determine the suitable gain for the hard disk drive (HDD) signal writing process. The result from machine classification has a high potential for position error improvement, distinguishing them from those with obvious degradation. The identified machine classes with high potential for signal write quality improvement undergo controller optimization using a genetic algorithm (GA). The objective function considers gain crossover frequency, phase margin, and PES DFT at low frequencies. Experimental results demonstrate that the new controller gain enhances signal write quality of class 0 and class 3 by 14.68% and 17.18%, respectively, leading to a reduced failure rate down to 0.05%.