Decentralized Output-only Modal Identification Techniques for Wireless Monitoring Systems

Sensor network technology entails the deployment of small, low-power sensor nodes that can communicate with one another. Since sensor networks can be used to monitor the condition of many types of architectural spaces, the technology has recently received significant academic and industrial attention. Amongst the potential applications, structural monitoring of large-scale civil structures is one of the most promising. With the recent advancement of low-cost and ultra-compact wireless sensing and data acquisition technologies, it has become feasible to apply dense arrays of wireless sensors in a single structure so as to identify its structural properties. In many cases, forced excitation of a civil structure in a controlled manner is difficult, especially when the size of the structure increases. Therefore, output-only modal identification methods are advantageous for use in characterizing system properties. This paper describes three output-only modal identification methods (peak picking, random decrement, and frequency domain decomposition) embedded in dense wireless sensor networks. The embedded software is designed using a decentralized computing architecture; such an approach is quite different from conventional centralized structural monitoring systems where algorithms are implemented in a centralized, but resource-rich data server. The decentralized computing architectures proposed in this work are scalable to a large numbers of sensors organized as a community that collectively carries out complex engineering analyses. Using the cantilevered balcony of a historic theatre as a test-bed structure, an array of twenty-one wireless sensors are deployed to collect acceleration response data during a series of vibration tests. The proposed algorithms for wireless sensors are effective in autonomously determining the balcony’s modal properties (including natural frequencies, modal damping ratios, and mode shapes). INTRODUCTION Structural health monitoring (SHM) has attracted wide-spread attentions in recent years leading to high expectations for this technology. In Japan, there are three motivations driving the development 1 Research Engineer 2 Assistant Professor of this technology: (1) public demand for technologies that can automatically diagnose the health of structures after a large earthquake; (2) the necessity of society to continue its use of old buildings and infrastructures which were built during period of high economic growth in Japan; (3) commercial demand for more quantitative evaluation of structural performance during due diligence in real estate transactions. Within the field of SHM, wireless sensors have emerged as a viable data acquisition technology offering cost lower than traditional wired monitoring systems. Wireless sensor nodes consist of three major functionalities: sensing, networking, and computing. In this study, an emphasis is placed on the computing functionality of the wireless sensors. Specifically, the present study is intended to modify widely employed modal identification methods so they can be implemented in a decentralized fashion upon wireless sensor networks. Three modal identification methods are explored: peak-picking, random decrement, and frequency domain decomposition methods. The decentralized modal identification methods as executed by a wireless sensor network are demonstrated during experimental testing of a historical theatre balcony. DECENTRALIZED SHM USING SENSOR NETWORKS Structural health monitoring strategies can be broadly classified into two categories: global and local monitoring. The global monitoring method focuses on the existence, or degree of damage present in the whole structure. In contrast, local monitoring methods are applied to pre-identified probable damaged locations (“hot spots”) for accurate long-term inspection. In the global monitoring methodology, structural condition is evaluated mainly by identifying and detecting the variance of mathematical model parameters (e.g., modal features) which express the global behavior of the structure. Most of the current global monitoring systems are designed based on conventional seismic observation systems, which are centralized systems gathering analog data from multiple acceleration meters via high-cost coaxial cables (Celebi 2002). In such a system (see Fig. 1a), data is also analyzed at the central data server. Large-scale systems suffer from complex installations and high costs. In addition, an excessive amount of data is concentrated at the data server; in many instances, the data is simply stored and left for future manual investigation by engineers. Over the last few years, many researchers have published detailed studies on sensor network technology which can be applied for system identifications and damage detections in civil structures (Lynch et al. 2006). Some of the weak points mentioned above for conventional SHM systems are resolved by making use of the best features of sensor networks. For example, decentralization of processing power allows data to by autonomously processed on the sensor nodes. Also, wireless communication eradicates any need for wiring in the structure which lowers system costs (see Fig. 1b). As a result of these features, sensor networks will be able to achieve more scalable monitoring systems defined by higher sensor densities. High sensor densities are desirable when applying the technology to large structures such as buildings and bridges. These high densities are driving innovation in how future monitoring system are deployed. For example, if the number of sensors increases by two or three times, the advantage gained is simply a greater amount of data available. However, if the number of sensors increases 10 or 20 times, then new system architectures would be necessary to extract detailed information from such vast sources of sensed data. While high sensor densities would permit the measurement of local structural responses, centralization of the monitoring system is not technologically possible. Rather, decentralized data processing architectures are needed to Fig. 1. Global monitoring system: (a) conventional centralized system; and (b) decentralized system with sensor networks Data/Analysis Server (Centralized control) Sensor (Accelerometer)