Distributed adaptive diagnosis of sensor faults using structural response data
Distributed adaptive diagnosis of sensor faults using structural response data
复制标题
DOI：
10.1088/0964-1726/25/10/105019
复制
发表时间：
2016-09
影响因子：
4.1
通讯作者：
K. Dragos;K. Smarsly
中科院分区：
文献类型：
--
作者：
K. Dragos;K. Smarsly
无线结构健康监测（SHM）系统的可靠性和一致性可能会受到传感器故障的影响，从而导致校准错误、数据损坏甚至数据丢失。已经提出了几种用于故障诊断的研究方法，称为“分析冗余”，用于分析不同传感器输出之间的相关性。在无线SHM中，大多数分析冗余方法需要将数据集中存储在服务器上进行数据分析，而其他方法则利用无线传感器节点的机载计算能力，直接在机载分析原始传感器数据。然而，由于无线传感器节点的电力资源有限，使用原始传感器数据会造成操作限制。本文提出了一种基于处理后结构响应数据的分布式自主传感器故障诊断方法。加速度响应数据的傅里叶振幅之间的内在相关性，在峰值对应于结构的特征频率，用于诊断在给定结构条件下的异常传感器输出。人工神经网络（ANN）是一种完全由数据驱动的分析冗余方法，不需要任何关于被监测结构或SHM系统的先验知识，它被嵌入到传感器节点中，以完全分散的方式实现协同故障诊断。将分布式分析冗余方法应用于无线SHM系统，并在实验室实验中进行了验证，证明了无线传感器节点能够以最小的数据流量准确有效地自诊断传感器故障。除了能够进行分布式自主故障诊断外，嵌入的人工神经网络还能够适应结构的实际情况，即使在结构发生变化的情况下也能保证准确高效的故障诊断。
The reliability and consistency of wireless structural health monitoring (SHM) systems can be compromised by sensor faults, leading to miscalibrations, corrupted data, or even data loss. Several research approaches towards fault diagnosis, referred to as ‘analytical redundancy’, have been proposed that analyze the correlations between different sensor outputs. In wireless SHM, most analytical redundancy approaches require centralized data storage on a server for data analysis, while other approaches exploit the on-board computing capabilities of wireless sensor nodes, analyzing the raw sensor data directly on board. However, using raw sensor data poses an operational constraint due to the limited power resources of wireless sensor nodes. In this paper, a new distributed autonomous approach towards sensor fault diagnosis based on processed structural response data is presented. The inherent correlations among Fourier amplitudes of acceleration response data, at peaks corresponding to the eigenfrequencies of the structure, are used for diagnosis of abnormal sensor outputs at a given structural condition. Representing an entirely data-driven analytical redundancy approach that does not require any a priori knowledge of the monitored structure or of the SHM system, artificial neural networks (ANN) are embedded into the sensor nodes enabling cooperative fault diagnosis in a fully decentralized manner. The distributed analytical redundancy approach is implemented into a wireless SHM system and validated in laboratory experiments, demonstrating the ability of wireless sensor nodes to self-diagnose sensor faults accurately and efficiently with minimal data traffic. Besides enabling distributed autonomous fault diagnosis, the embedded ANNs are able to adapt to the actual condition of the structure, thus ensuring accurate and efficient fault diagnosis even in case of structural changes.