MS17.11 - Structural Health Monitoring

In the framework of the "Smart Tower" project, a joint effort between the VUB and Belgian grid operator Elia to investigate the feasibility of SHM for transmission towers. A transmission tower is instrumented with several types of sensors, including accelerometers, strain gauges, acoustic emission sensors and PZTs. In order to evaluate the approach, the tower was damaged in a controlled manner during the monitoring campaign to simulate common failure modes of this type of structure. Scenario’s included the removal of a single bolt, bending a cross brace, grinding a cross brace and ultimately the removal of a cross brace., and after a specified period, the structure was fixed. In this contribution, the focus is on the four tri-axial acceleration sensors installed on each face of the high voltage steel lattice tower. The main objective is to perform vibration-based anomaly detection. The proposed approach is based on operational modal analysis (OMA), where the modal parameters are first estimated using the poly-reference Least Squares Complex Frequency-domain (p-LSCF) algorithm. Each sensor is processed individually, which allows differentiating local and global modes between the different faces of the structure. Then, the identified modal parameters are tracked over time in a time window of 10-minute intervals. Each tracked modal parameter is first filtered by discarding data out of the mainstream of the mode. Then, the retained modes are normalized for varying environmental conditions. The resulting compensated modal frequencies from the four sensors are used to evaluate the dynamic behavior of the structure. Results show that a large number of modes can be detected. We also notice the existence of local and global modes, as local modes can have slightly different frequency value from 1 sensor to another. This allows the estimation of the symmetry of the structure, which can be used as a non continuous structure health evaluation. Moreover, the proposed methodology was able to detect nearly all the damage events. To some level the nature of a damaging event was even identifiable as the bolt removal failure modes only affected local modes at the high frequency part of the spectrum, and it is only detectable in the sensors installed in the same face where the damage was introduced. Meanwhile, the removal of brace affected global modes and thusly was detectable in the acceleration reading of all different four faces of the transmission tower.

Due to the lack of information regarding structural details and construction technology, finite element analysis based results may be dubious for aging structures. The results will not confirm the effect of soil structure interaction either since back analysis based assumptions may not exactly represent the behavior of soil and structure accurately. To this end, realistic material properties as well as dynamic characteristics are required for a reliable finite element model. We conducted ambient vibration measurements in a 50-year-old elevated water tank located in Kathmandu, Nepal. The tank was constructed using smooth bars before the codal provisions were introduced. Ambient vibration records were taken for empty, half, and full reservoir conditions. Parametric identification using the numerical algorithm for subspace state space system identification (N4SID) was performed to estimate the modal properties of a 14.43 m tall water tank. The fundamental vibration frequencies of the tank for empty, half, and full reservoir conditions are estimated as 1.1 Hz, 0.93 Hz, and 0.77 Hz, respectively. We also developed finite element models of the tank with and without soil structure interaction. The fixed base analysis resulted in the fundamental vibration frequencies of 1.21 Hz, 1.02 Hz, and 0.86 Hz, respectively for empty, half, and full reservoir conditions. On the other hand, while considering soil flexibility, the fundamental vibration frequencies were estimated as 1.08 Hz, 0.91 Hz, and 0.78 Hz, respectively for empty, half, and full reservoir conditions. The results highlight that system identification-based fundamental vibration frequencies are better captured by the model created incorporating soil flexibility. We conclude that soil structure interaction is vital to be considered for special structures such as water tanks in active seismic regions with loose soil deposits.

Fatigue in steel components is of major importance for the safety of infrastructures. Early detection of fatigue cracks in such structures means that appropiate repairs can be performed before the damage becomes too severe. Despite the need for a reliable early detection method for this type of damage, detection is notoriously difficult. Therefore, there is a need for innovative detection methods to improve or complement the current monitoring systems. In the search for a new method to monitor the early growth of fatigue cracks, the magnetic stray field generated by the magnetisation of the steel component becomes interesting because the strength and direction of the magnetic stray field are strain-dependent. The behaviour of the magnetic field during fatigue loading has received little research so far. Our main research question is if the magnetic stray can provide a sufficient information in the onset and development of this damage. To investigate the evolution of the magnetic stray field under fatigue loads, laboratory tests with a cyclic tensile load have been performed on several CT-specimens made of steel with different stress ratios. During each test, data were collected using four different measurement techniques: a conventional crack gauge (CG), digital image correlation (DIC), acoustic emission (AE) sensors and a magnetic field (MF) sensor. By comparing the data from these different measurement techniques, the effectiveness of magnetic field sensors in the early detection of fatigue crack growth is assessed.

The design of offshore wind turbines is driven by the fatigue limit state; critical fatigue details are often situated around or below the mudline, and particular interest lies with the fatigue accumulation at these locations. Performing dynamic response measurements below the mudline is a challenging task, largely related to the high costs of equipping a turbine at those locations as well as the risk of sensor failure during the installation process. However, measuring this data is extremely valuable for both fatigue assessments and validating model and data driven virtual sensing approaches that do not rely on sensors installed so close to the mudline. During the past decades fiber Bragg grating (FBG) based optical strain sensors have become more commonplace for fatigue monitoring applications of large structures. Despite many advantages over conventional strain gauges, in some cases FBG measurements can become corrupted by sudden, repeated, jumps in the signal, which manifest as spikes or step-like offsets in the data. These jumps are often referred to as peak-splitting. This particular type of failure mode is especially problematic when the data is intended for fatigue assessments, as the artificial jumps will introduce artificial stress ranges to the Rainflow-counting results, and thereby artificially inflate the computed damage. The data can easily be corrupted to a degree where the time domain data as well as damage related measures become un-usable. The latter has motivated the development of a reconstruction tool which removes peak splitting artefacts from the measurement signals. However the method also introduces a quasi-static drift, which can complicate the use in a fatigue damage oriented setting. For this contribution, a long-term measurement campaign, suffering from severe peak splitting on some of the optical strain sensors, will be used to demonstrate the potential of recovering severely damaged data. The data originates from the subsoil region of an operational offshore wind turbine foundation and has been recorded using 4 multiplexed FBGs. Furthermore, conventional strain gauges near interface level are available for reference purposes. The goal of this recovery is to make the data useable again for damage-based assessments. One month of corrupted FBG data has been processed using the peak splitting removal tool. Subsequently, damage equivalent moment (DEM) values have been computed for the corrupted FBG strain data, the conventional strain gauge data, as well as the reconstructed FBG strain data. Comparing the correlations in the computed DEM time series shows that, despite the heavy peak splitting, the corrupted data still contains physical information, and that a scale factor approach could be leveraged to make the data useable again. The scale factor is grating specific and has been derived using the time domain reconstructions. The results of this contribution show that FBGs with severe peak splitting, can still contain valuable information, and might be usable for damage related applications, as well as many time domain applications which are not concerned with the quasi static response, or accept a certain tolerance towards errors in the latter response regime.