Collaborative prognostics in Social Asset Networks
Collaborative prognostics in Social Asset Networks
复制标题
DOI：
10.1016/j.future.2018.02.011
复制
发表时间：
2019-03
期刊：
影响因子：
--
通讯作者：
A. S. Palau;Zhenglin Liang;Daniel Lütgehetmann;A. Parlikad
中科院分区：
文献类型：
--
作者：
A. S. Palau;Zhenglin Liang;Daniel Lütgehetmann;A. Parlikad
随着物联网（IoT）技术的普及，资产已经获得了通信、处理和传感能力。作为回应，资产管理领域已经从整个机队的故障模型转向个性化的资产管理。个性化的模型很少真正分布，并且往往无法利用资产车队的处理能力。这导致了难以扩展的机器学习集中模型，这些模型通常必须在准确性和计算能力之间找到折衷。为了克服这一点，我们提出了一种新的理论方法，在社会物联网的协同作战。我们介绍了社会资产网络的概念，定义为具有传感，通信和计算能力的合作资产网络。在所提出的方法中，通过传感器从介质中获得的信息被合成为健康指标，该健康指标确定资产的状态。每个资产的健康指标根据由三个参数确定的等式进行演变。资产被赋予方程的形式，但他们不知道自己的参数值。为了获得这些值，资产使用该公式对其健康指标数据执行非线性最小二乘拟合。使用这些估计的参数，它们通过相似性度量互连到协作资产的子集。我们展示了如何通过简单地交换他们的估计，网络资产能够精确地确定他们的健康指标动态和降低维护成本。这是在真实的时间内完成的，没有集中的库，也不需要大量的历史数据。我们将社会资产网络与典型的自学习和舰队范围的方法进行了比较，并表明社会资产网络具有更快的收敛速度和更低的成本。这项研究作为一个概念性的证明，为解决维修问题的协作动力学的潜力，并可以用来证明这样一个系统在真实的工业车队的实施。
With the spread of Internet of Things (IoT) technologies, assets have acquired communication, processing and sensing capabilities. In response, the field of Asset Management has moved from fleet-wide failure models to individualised asset prognostics. Individualised models are seldom truly distributed, and often fail to capitalise the processing power of the asset fleet. This leads to hardly scalable machine learning centralised models that often must find a compromise between accuracy and computational power. In order to overcome this, we present a novel theoretical approach to collaborative prognostics within the Social Internet of Things. We introduce the concept of Social Asset Networks, defined as networks of cooperating assets with sensing, communicating and computing capabilities. In the proposed approach, the information obtained from the medium by means of sensors is synthesised into a Health Indicator, which determines the state of the asset. The Health Indicator of each asset evolves according to an equation determined by a triplet of parameters. Assets are given the form of the equation but they are not aware of their parametric values. To obtain these values, assets use the equation in order to perform a non-linear least squares fit of their Health Indicator data. Using these estimated parameters, they are interconnected to a subset of collaborating assets by means of a similarity metric. We show how by simply interchanging their estimates, networked assets are able to precisely determine their Health Indicator dynamics and reduce maintenance costs. This is done in real time, with no centralised library, and without the need for extensive historical data. We compare Social Asset Networks with the typical self-learning and fleet-wide approaches, and show that Social Asset Networks have a faster convergence and lower cost. This study serves as a conceptual proof for the potential of collaborative prognostics for solving maintenance problems, and can be used to justify the implementation of such a system in a real industrial fleet.