Towards building digital twins for prognostics and health management (PHM) of industrial assets

随着数字和传感器技术的快速发展，为了提高可靠性、减少成本和浪费并最大限度地提高资产可用性，机器或过程健康状况的实时监测、诊断和预测正变得越来越普遍。十多年来，我们一直致力于机器健康状况监测、制造质量保证和医疗应用的诊断决策。我们开发了使用振动、声学、电流以及视觉和热成像信号进行分析并确定故障位置和严重程度的技术。近年来，我们已经开始开发一个建模框架来构建基于物理和数据驱动的综合混合模型。基于物理学的方法是必要的，以弥补在感官数据中缺乏迫在眉睫的证据。数据驱动的方法可以解决未建模的动态和不确定性。最终，我们的目标是建立一个集成的系统模型，代表一个物理系统的数字副本。这样的模型将能够准确地定位故障、确定根本原因、预测迫在眉睫的组件和系统故障，并做出最佳的纠正决策。在这项研究中，我们将开发一个概念验证集成系统的模型，它表示物理系统的数字副本。该系统模型集成了材料疲劳模型、传动系动力学模型、齿轮传动误差模型、轴承动力学模型和电机/发电机机电模型。我们将开发数据驱动模型来描述松散的连接、不良的润滑和故障导致的轴承共振。实验研究将在动力传动装置上进行。研究的最终结果是概念证明，可以被许多应用领域采用，包括先进制造、运输和关键资产健康管理的能源生产。

With the rapid advancement in digital and sensor technologies, real-time monitoring, diagnosis, and prognosis of machine or process health conditions are becoming increasingly common practices in an effort to improve reliability, reduce cost and waste, and maximize asset availability. For more than a decade, we have been working on diagnostic decision making for machine health condition monitoring, manufacturing quality assurance, and medical applications. We have developed techniques that use vibration, acoustic, electric current, and visual and thermal imaging signals to conduct analyses and determine the location and severity of faults. In recent years, we have started to develop a modeling framework to construct comprehensive hybrid physics-based and data-driven models. The physics-based approach is necessary to compensate for the lack of impending evidence in sensory data. The data-driven approach can address unmodeled dynamics and uncertainty. Ultimately, the goal is to build an integrated system's model that represents a digital replica of a physical system. Such a model will enable accurate localization of faults, root cause determination, impending component and system failure prediction, and optimal corrective decisions. In this research, we will develop a proof-of-concept integrated system's model that represents a digital replica of a physical system. The system model integrated component models including materials fatigue models, drivetrain dynamics models, gear transmission error models, bearing dynamics, and motor/generator electromechanical models. We will develop data-driven models to describe loose connections, poor lubrication, and fault-induced bearing resonance. Experimental studies will be carried out on a drivetrain apparatus. The end result of the research is a proof of concept that can be adopted by many application areas including advanced manufacturing, transportation, and energy production for critical asset health management.