Figure 1: The locations of the eight tilt sensors, labelled TL1-8, on the bridge. An Echo State Network Approach to Structural Health Monitoring A. J. Wootton, C. R. Day, P. W. Haycock Keele University Staffordshire United Kingdom {a.j.wootton, c.r.day, p.w.haycock}@keele.ac.uk Abstract—Echo State Networks (ESNs) have been applied to time-series data arising from a structural health monitoring multi-sensor array placed onto a test footbridge which has been subjected to a number of potentially damaging interventions over a three year period. The time-series data, sampled approximately every five minutes from ten temperature sensors, have been used as inputs and the ESNs were tasked with predicting the expected output signal from eight tilt sensors that were also placed on the footbridge. The networks were trained using temperature and tilt sensor data up to the first intervention and subsequent discrepancies in the ESNs’ prediction accuracy allowed inferences to be made about when further interventions occurred and also the level of damage caused. Comparing the error in signals with the location of each of the tilt sensors allowed damaged regions to be determined. Keywords—Structural Health Monitoring; Echo State Networks; Reservoir Computing Applications; Wireless Sensor Networks I. INTRODUCTION Intelligent analysis of data collected from reinforced concrete structures using non-destructive techniques is becoming increasingly important given the increased availability and use of non-invasive structural health monitoring (SHM) sensor networks that can monitor our built civil engineering infrastructure in real time. Such analyses have the potential to provide structural engineers with a more accurate picture of the dynamics of structures and an indication of the presence of defects within a structure: a task which hitherto has been difficult and costly. Early detection of defects within a structure not only helps to increase the serviceable life of a structure and prevent structural failure, but also reduces repair costs due to a reduction in the severity of any defects. In this study, a type of Recurrent Neural Network (RNN), the Echo State Network (ESN), was applied to a very large, multi-dimensional, longitudinal, time-series dataset composed of sensor readings from a real-world civil engineering structure that was subjected to a number of deliberate interventions, many of which were likely to undermine the structure’s integrity. This study built on previous work, where the present authors compared the ESN approach to the commonly used NARMAX model and found that the ESNs are more versatile and better suited to damage detection [1, 2]. A. The NPL Footbridge The UK’s National Physical Laboratory (NPL) footbridge project was set up as a means of developing new SHM sensor technologies and the methods for processing large time-series datasets that arise from wireless sensor networks. The project was centred on a concrete footbridge that was built in the 1960s and underwent normal use for nearly 50 years prior to the beginning of the project. In 2009 it was taken out of use and was embedded with a plurality of sensors, which took data readings at regular five minute intervals over a three year period. The work we present here is concerned with the data produced by ten temperature sensors and eight tilt sensors, delivering 365 376 sensor readings collected between January 2009 and May 2012. Fig. 1 shows the spatial arrangement of the tilt sensors on the bridge. Note that sensors 7 and 8 are attached to the two columns of the bridge and that it is a standalone structure, allowing weights to be suspended from either of the two cantilevers. This is due to the fact that for the duration of the study the bridge’s sole purpose was for field testing. During the course of the study, the bridge was subjected to damage and repair cycles, detailed in full by Livina [3]. There have been other studies looking into the data provided by the project, but as of yet none has been able to accurately detect events and characterise any consequential long term damage to the bridge [3--6].