云存储硬盘亚纳米级头盘界面的失效机理和可靠性预测技术

With the continuous increase of the storage areal density, current spacing between head and disk has decreased to sub-one-nanometer. Due to the effect of contact force, friction, electrostatic force, intermolecular force, a wear of head/disk can not avoid, which brings huge challenges to head-disk interface and hard disk. This project aims to deal with such challenges of head-disk interface in cloud hard disk drives. First, a dynamics model of sub-one-nanometer head-disk interface taking the coupling effect between head stack assembly and air bearing, and the interfacial forces into account was developed. To reveal the failure mechanisms of head-disk interface, the dynamic evolution of head-disk interface and the wear progression of head and disk were also investigated. Upon the mechanisms, a physics-of-failure model was built. The physics-of-failure model, however, has shortcomings in dealing with the reliability prediction due to the diversity of the material, and geometry in hard disk drives. To amend it, a fusion model that integrates both physics-of-failure model and data-driven model was built. Furthermore, a new prediction model based on the unscented Kalman filter, which can predict the dynamic behaviors of the performance degradation data and update the model parameters adaptively was developed. With the help of this model, the potential risks and residual life of hard disk drive can be predicted in time. This is a cutting edge research and is critical for engineering practice. It is also helpful to improve the quality and reliability of cloud hard disk drives.

随着存储面密度的不断提高，当今硬盘的头盘间距已降到1nm以下，由于微观接触力、摩擦力、静电力、分子间作用力的影响，头盘间的近接触损伤已在所难免，这对头盘界面、乃至整个硬盘的可靠性都提出了巨大的挑战。本项目以云存储硬盘亚纳米头盘界面为研究对象，建立亚纳米级头盘界面动力学模型，考虑磁头折臂和空气轴承的耦合作用以及各微观界面力的影响；研究头盘动力学特性演化规律和头盘表面微观组织的损伤扩展规律，揭示出亚纳米头盘界面的退化/失效机理，建立头盘界面的失效物理模型；提出失效物理模型和数据驱动模型融合方法来克服单纯使用失效物理模型会受到材料、几何特性等方面差异性影响的缺点；提出基于非线性滤波技术—unscented Kalman filter具有自适应性的实时故障预测模型，以预报硬盘潜在威胁和硬盘的残余寿命。本项目研究工作源于学科前沿和工程需求，对云存储硬盘质量改善和可靠性的提高具有现实意义。

近年来，随着大、智、移、云（大数据、智能化、移动互联网和云计算）的蓬勃发展，全球数据信息量已经呈指数式爆炸增长之势。硬盘作为这些海量数据的最主要存贮载体，一旦坏损，不但会造成重要数据的丢失还有可能会造成整个存储计算系统当机，从而给个人或企业造成不可估量的财产损失。本项目以云存储硬盘为研究对象，运用三维实体建模技术和模态参数辨识方法，构建了能综合考虑折臂组件、微观接触力、摩擦力影响的磁头高精度动力学仿真模型；在此基础上，利用高速摄像机等手段对硬盘跌落过程进行实验研究，通过高精度时频分析-改进的集成噪声的经验模式分解算法，揭示了快速冲撞引起的头盘瞬态接触的动力学特性；深入研究了亚纳米级头盘的退化/失效机理，利用扫描电镜对磁头不同磨损状态进行分析，揭示了头盘表面微观磨损萌生与扩展情况；利用近接触动力学模型与磨损实验，研究了高温、高频振动及微观作用力（如摩擦导致的静电力）诱导头盘界面失稳原因，进而确定了头盘界面微观诱导磨损机制，并提出了改善磁头磨损的有效方法；基于头盘界面振动响应信号，提出了改进的TR-LDA算法，实现了磁头磨损程度的有效识别；据此，提出了改进的Archard磨损模型，在建立磨损模型的基础上，考虑头盘磨损存在的动态、非线性等特征，提出了基于Wiener过程的磁头加速磨损模型，减少了磁头可靠性试验的时间和成本；通过马氏距离融合多维退化数据，结合3西格玛准则和Kalman滤波平滑方法，提出了基于两阶段策略的剩余寿命预测方法；在此基础上，提出了自适应粒子滤波方法和基于Rao-Blackwellized原理的切换状态空间退化模型，利用实时磨损数据对退化模型进行自适应更新，实现了硬盘剩余寿命的高精度预测。研究成果为云存储硬盘的质量提升和可靠性改善提供坚实的理论基础和实用技术。