SPECIAL FEATURE Fault Detection in Hard Disk Drives Based on a Semi Parametric Model and Statistical Estimators Lucas P. Queiroz 1 • Joao Paulo P. Gomes 1 • Francisco Caio M. Rodrigues 1 • Felipe T. Brito 1 • Iago C. Chaves 1 • Lucas G. M. Leite 1 • Javam C. Machado 1 Received: 3 February 2017 / Accepted: 9 April 2017 Ó Ohmsha, Ltd. and Springer Japan 2017 Abstract Detecting faults in Hard Disk Drives (HDD) can lead to significant benefits to HDD manufacturers, users and storage system providers. As a conse- quence, several works have focused on the development of fault detection algo- rithms for HDDs. Recently, promising results were achieved by methods using SMART (Self-Monitoring Analysis and Reporting Technology) features and anomaly detection algorithms. In this work, we propose a method for fault detection on HDDs that uses a Gaussian Mixture to model the behavior of healthy HDDs. After obtaining the similarity between a given HDD and this statistical model, an anomaly is detected when a statistical estimator computed over these dissimilarities exceeds a threshold. In addition to the proposed method, we also conducted an extensive evaluation of different statistical estimators. The proposed method, named Fault Detection of HDDs based on GMM and statistical estimators (FDGE) was & Joao Paulo P. Gomes joao.pordeus@lsbd.ufc.br Lucas P. Queiroz lucas.queiroz@lsbd.ufc.br Francisco Caio M. Rodrigues caio.rodrigues@lsbd.ufc.br Felipe T. Brito felipe.timbo@lsbd.ufc.br Iago C. Chaves iago.chaves@lsbd.ufc.br Lucas G. M. Leite lucas.goncalves@lsbd.ufc.br Javam C. Machado javam.machado@lsbd.ufc.br 1 LSBD-Department of Computer Science, Federal University of Ceara ´, Fortaleza, Brazil 123 New Gener. Comput. DOI 10.1007/s00354-017-0016-0