Rotating Machinery, Optical Methods & Scanning LDV Methods, Volume 6

The need to validate simulation models of complex mechanical structures continues to grow in importance to increase efficiency in the design process. This is especially true for nonlinear structures (such as composite panels and jointed components) where it is critical to use an accurate full-field measurement method. Multipath Doppler vibrometry (MDV) embodies the most recent breakthrough in dynamic characterization. It delivers high-quality vibration data consistently even for the most adverse environments and provides the test engineer with a reliable tool that is fast and easy to set up. The test object is measured in its unaltered form as it is not necessary to apply any surface treatments. MDV can be applied to a single point measurement and for 1D as well as 3D scanning vibrometry tests. The benefits are illustrated in this paper by various application examples.

During operation, wind turbine blades are subjected to varying loads, making them vulnerable, reducing performance, and increasing mechanical failures. Furthermore, blade maintenance is both time-consuming and costly. Optical methods, such as DIC, are attractive to measure deformations on this kind of structures because they do not require electrical wiring, they avoid mass loading effects, and they can be easily configured to measure large test articles. In this paper, a novel optical technique based on DIC is described for performing full-field measurements by using contact-less techniques. In this paper, the possibility of using low speed cameras DIC to perform modal analysis and to validate and update finite element models is investigated, testing a scaled wind turbine blade in different conditions. The results obtained with DIC are compared with the ones obtained with accelerometers and correlated with the numerical solutions.

Nuclear power stations typically run electrical machines on a conservative hard-life basis, i.e., by specifying the number of operating hours and start-up/shut-down cycles before replacement. Not only is this approach costly, but it does not provide through-life performance information to extend operations or understand the failure modes. While there is a large amount of research on fault detection within induction motors, little has been conducted on extracting true load-related electrical and modal signatures that properly reflect the behavior of the machines. Many types of industrial plants use large rotating machines to convert electrical input into mechanical loads. The focus of this work is induction motors. These induction motors require three phase power and rotate at the slip frequency which is slightly slower than the synchronous power grid frequency. The light load and low slip cause operating conditions that present a significant challenge to structural health monitoring. The overwhelming influence of the power grid frequency on the stator current data makes it difficult to analyze sidebands which are used to detect rotor faults. Researchers have yet to devise an effective strategy for isolating the adjacent sidebands to enable the detection of rotor faults in lightly loaded induction motors, such as those in nuclear power plants. This work focuses on applying a modified extended Kalman filter to the stator current data and performing spectral subtraction to remove the closely coupled power grid frequency to observe the sidebands of the motors. The results for the analyses showed that the Kalman filter was able to create a model of the power grid frequency allowing for the signal to be removed from the data. This work also analyzes the subtraction of the rotor frequency from the acceleration data with the addition of a continuous wavelet transform prefilter.

Event-driven silicon retina imagers are useful for structural dynamics applications as they are low-power, high-information bandwidth, and high-dynamic range devices. Currently, event-driven imagers can detect light intensity changes associated with an LED blinking at 18 kHz, which indicates event-driven imagers are capable of capturing dynamic motion associated with the commonly accepted frequency range of human hearing, 20 Hz–20 kHz. As the development and utilization of event-driven imagers advances, it is reasonable to expect the upper bound of measurable frequency to increase. Therefore, event-driven imagers have the potential to be a viable replacement for high-speed imagers in the field of structural dynamics. The majority of statistical techniques currently developed for structural dynamics assume data is captured with a uniform sampling rate. This convention is problematic when using event-based imagers because event-based sensors generate send-on-delta data. The widespread use of these imagers will require the development of an acceptable technique for converting event data to time-series data, which can then be sampled uniformly. Another challenge with utilizing event-driven imagers is the inherent and significant variations exhibited between pixels on the same imager during the generation of an event. These variations need to be better understood and examined. Investigative research has revealed that certain aspects of the two problems, data conversion and pixel variation, have been addressed by the neuroscience community. The spike-based data analysis techniques that have been developed in the neuroscience community may have applicability to structural dynamics in the context of event-driven imagery. These practices will then be used as inspiration for developing event-based data processing techniques to tie event-based data to existing structural dynamics analysis frameworks.

Digital image correlation (DIC) has been widely accepted in the vibration community for extracting strain and displacement using noncontact optical techniques. Due to the nature of DIC, the preparation of a test structure with an applied pattern is important for obtaining accurate results. Investigation into patternless optical methods would be beneficial, and it would be ideal if a test structure no longer needed pretreatment prior to optical testing. Recently in the literature, phase-based motion magnification (PMM) has been utilized to exaggerate subtle motion for structural identification. In this work, template matching is used to correlate a template facet over a series of magnified images. Following the determination of a template facet, virtual red, green, and blue (RGB) targets are placed along the principal direction of displacement. Particles are then randomly generated and used to find the RGB-coded targets and clustered to obtain sub-pixel displacements that can be used for frequency extraction of magnified data. Application of the template match particle filter (TMPF) approach will further enhance noncontact sensing, in addition to providing a more efficient way of processing optical data. This method is implemented to experimentally characterize parameters of two structures (i.e., a cantilever beam and bridge) having both high and low frequencies.

Structural health monitoring (SHM) is imperative to the safety of structures, but can be costly and limited to a few locations for coverage. Many advancements have been made in the field of SHM over the past years such as the use of wireless sensors, implementation of compressed sensing, and event-based monitoring. These advancements are all pursued with the combined goal of collecting relevant data from the structure in a cost-effective manner, as well as taking into account the limitations placed on the system including size, energy consumption, safety, and bandwidth. Developments in wireless sensor systems enable greater coverage of structure monitoring since wired systems are costly to install. For wireless systems, it is ideal to use as little power as possible to process signals to reduce the sensor battery use and costly battery replacement. Previous research has focused on compressive sensing to reduce the size of data transferred while maintaining high-fidelity signal analysis. More recent work has focused on event-based monitoring, which collects data based on non-uniformly spaced triggers. Both compressive sensing techniques and event-based monitoring focus on improving sampling for multi-sensor systems, but are restricted to lower sampling rates or are prone to missed events. In order to further advance the application of compressive sensing techniques and event-based monitoring, this project will focus on triggering schemes for multi-sensor systems with signals above the Nyquist frequency of the individual sensors, assuming the signal is present across all sensor recordings. The problem with detecting frequencies above the Nyquist frequency is that the signals will alias and show up as a disguised frequency. Using existing sensors to determine the true frequencies of the signal requires a new method of detection. Multiple sensors were placed around a structure of interest and the optimal time delay of each sensor was determined through simulation and affirmed by experimentation.

While phase-based motion magnification has shown great success in enhancing sub-pixel motion, super-pixel motion has proven more difficult to work with. As “small motion” is usually a major assumption in the derivation of optical flow, large-scale super-pixel motion has been shown to cause artifacts and limit the usability of the method in both magnification and extraction applications. Viable methods do exist and have shown promise for scenarios where unimportant large motions are present in an otherwise ideal video. Likewise, current rules of thumb suggest utilizing a pyramid approach, where the images are downsampled until the motion becomes sub-pixel and thus avoids the issue entirely. However, the approaches for effectively removing these large-scale, super-pixel motions neglect their potential importance for objects which exhibit both super- and sub-pixel motions of interest. Further, for such objects, downsampling could degrade and even obliterate the relevant sub-pixel motion. Therefore, an approach is conceived and developed exploiting this degradation in conjunction with complexity pursuit (a blind source separation technique) to allow for more effective and purposeful processing using the complex steerable pyramid. With greater information in the beginning, rules of thumb and temporal bandpassing can be utilized more effectively to extract viable motion measurements in both regimes of pixel motion.

Digital image correlation is a well-established method for estimating the full-field displacement of two-dimensional surfaces by comparing pairs of images. The basic idea is to attach a texture or speckle pattern to the surface, and track the features of this pattern through the grayscale values of the image pixels. All this assumes a conventional camera that records image intensity at every pixel in every frame, which is inefficient for sharp transient events. In watching a balloon burst, for example, one needs a high frame rate to capture the details of the burst, but storing such a frame rate for the longer, slower expansion of the balloon before it bursts costs a lot of memory. It would be convenient, in applications such as monitoring structures, to store dense information only during dynamic events, not between them.

Silicon retinas are an alternative sensor type that record events rather than pixels. The raw data for such a device are a sequence of 4-tuples: the time at which each event occurred, the horizontal and vertical pixel coordinates containing the event, and whether the event involved increasing or decreasing intensity. Events can be reassembled into buckets corresponding to each pixel and a time interval corresponding to any desired camera shutter to produce frames analogous to those of a conventional camera. This project assesses whether it is feasible to combine digital image correlation or a similarly developed computer vision technique with the efficient storage of a silicon retina, via such converted frames.

The test article for this experiment was a latex band with a painted speckled pattern, mounted into the stationary and moving ends of a frame and subject to cyclical stretching. The image of this band was recorded simultaneously on conventional and silicon retina detectors through a beam splitter. Analysis of the resulting data showed that the silicon retina frames allow feature tracking with close to the quality of a conventional camera, but the computed displacements are consistently smaller. The surface has to accelerate before there are enough changes to register on the silicon retina, and this initial motion at either end of the oscillation is not recorded in the converted frames. This work demonstrates that silicon retina imagers have potential for persistent surveillance applications where there is a need to record sparsely occurring transient deformations over long time periods.

An experimental study was performed in which the response of a fully populated printed circuit board was monitored during an environmental vibration test using both an accelerometer and a scanning laser vibrometer. The test article was subjected to a random vibration profile typical of electrical components during a vibration test. A response accelerometer was shown to significantly alter the dynamics of the primary flexible mode of several components on the printed circuit board when the accelerometer was attached to the component of interest. The results of this experimental study demonstrate that a laser vibrometer can provide accurate single-point response measurements and can characterize the operational shapes of the test article during a vibration test.

To perform modal analysis, a digital 3D reconstruction of the structure under investigation is required. The acquisition and creation of such reconstructions is very time-consuming and cost-intensive, especially for performing simulations, and is mainly created manually using CAD software. Measurement-based approaches often use much simplified reconstruction models, such as line models, because of this problem. Due to the rapid development of inexpensive RGBD scanners, such as the Intel RealSense or ZED depth cameras, it is obvious to use them also for the field of modal analysis.

In this publication, the performance of these depth cameras and of algorithms for semiautomated post-processing of the 3D scans is investigated, especially for use as a reconstruction basis for finite element analysis. In addition, the 3D reconstructions are also used and evaluated in experimental modal analysis applications. Especially for the algorithms to interpolate unmeasured nodes of the reconstruction, the quality of the reconstruction plays a special role.

The whole acquisition and analysis process consists of three steps and is integrated in the WaveImage software. It starts with the acquisition of the Intel RealSense and ZED depth information (RGBD) and the following online 3D reconstruction. After the acquisition of the reconstruction, the post-processing is performed, which is applicable for beginners as well as experts. Finally, the processed reconstructions are used and validated in the software WaveImage both in the finite element and in the experimental modal analysis.

We perform Operational Modal Analysis (OMA) through Frequency Domain Decomposition (FDD) on a test structure using a high speed camera. Using optical flow algorithms, brightness changes due to displacements of a test structure under load are translated into displacement data. By using a mirror, a split image is generated that enables a stereoscopic view of the structure and thus 3D information can be extracted from the high speed video footage.

This setup is used to determine the modal parameters of the structure, especially the mode shapes with high spatial resolution. The mirror that is used to gather a second point of view onto the image sensor enables a cost-efficient stereoscopic measurement compared to a two-camera setup. Moreover, it eliminates the need of synchronizing two separate signals. This is also why OMA is used instead of EMA: as no force signal needs to be recorded, we can perform a full-field modal analysis with inputs from only one single device.

The results are evaluated taking into account the accuracy as well as the complexity of the setup of the 3D measurements compared to a setup using triaxial accelerometers.

A three-dimensional (3D) continuously scanning laser Doppler vibrometer (CSLDV) system that contains three CSLDVs and an external controller is developed to conduct full-field scanning to measure 3D vibrations of an aluminum plate with free boundary conditions under sinusoidal excitation. A reference object parallel to the plane of the plate is used as the measurement coordinate system to obtain in-plane vibration components. Calibration among three CSLDVs in the 3D CSLDV system based on the geometrical model of its scan mirrors is conducted to adjust their rotational angles to ensure that three laser spots can continuously and synchronously move along the same 2D scan trajectory on the plate. The demodulation method is used to process the measured response to obtain operating deflection shapes (ODSs) of the plate. By using frequencies that are close to damped natural frequencies of the plate as sinusoidal excitation frequencies, four in-plane ODSs, including shear and longitudinal ones, are obtained in the frequency range between 0 and 5000 Hz. These ODSs are in good agreement with those obtained by traditional stepwise scanning and theoretical undamped mode shapes of the plate calculated from its finite element (FE) model. Modal assurance criterion (MAC) values between the first four in-plane ODSs from 3D CSLDV measurements and those from stepwise scanning measurements are larger than 95%. MAC values between ODSs from 3D CSLDV measurements and corresponding mode shapes from the FE model are larger than 90%. However, the 3D CSLDV system can scan more measurement points in much less time than 3D stepwise scanning.

In the past few years, computer vision-based applications are widely used to determine damages in structural components. However, due to the involvement of certain levels of manual measurements, there still exists potential for improvements in the quantification of damage caused in a structure. Therefore, in this study, key-point point detection using fiducial markers is used to reduce the error in quantification of overall displacement and deformed profile of structural components (e.g., beams). This study demonstrates the use of a fiducial marker-based approach to quantify the deformation and bending profile in beam elements without physically disturbing the specimen. The visual fiducial marker (e.g., AprilTag) system has been extensively used in robotics for purposes ranging from localization of robots to increasing precision while tagging different objects in the robot’s surroundings. These AprilTag are attached to the surface of the specimen before loading. In this research, while the load is applied on the beam, the behavior of the specimen is then recorded by using a high-resolution digital camera during the experiment. Camera calibration using a checkerboard is performed to determine the extrinsic parameters (e.g., camera pose). With the help of attached AptilTags on the surface of the specimen, pixel values at the four corners and the center of each of the tags are detected during the experiment. These pixel locations are then used to get the corresponding world coordinate system by minimizing the loss function. Unlike normal bundle adjustment algorithms, camera parameters are treated as known parameters during the optimization process. Primary results show potential of proposed method in obtaining robust measurements.

This paper addresses the analysis of torsional dynamic behaviour in a magnetic transmission, where the torque is transferred in a contactless way between two coaxial rotors with permanent magnets through the interaction with a modulator element holding ferromagnetic poles. This transmission device is called planetary magnetic gear (PMG), due to its topological and functional similarity with a planetary mechanical gearing device, from which the same working principles are derived.

A test bench for testing the magneto-mechanical efficiency of the PMG prototype has been designed and realised. The PMG planetary arrangement allows the possibility to test different configurations, as regards the input/output power, and the prototype dynamic behaviour has been tested as a torque multiplier or as a speed multiplier. No-load tests have been performed, evaluating the torque losses due to bearing dissipations inside the transmission, proving that the efficiency is practically independent from the power direction, in contrast with the traditional mechanical transmissions.

The results here presented can be considered as an overview of a wider activity since, a design of experiment (DOE) with different loads, different speeds at the two sides of transmission and different transmission gear ratios will be further investigated, to assess the independency of efficiency from external conditions.

Waveforms of propagating flexural waves can reveal plentiful information about anomalies caused by damage-wave interactions, and such anomalies can be used for damage identification. However, they can be masked by the interference of measurement noise and may be able to indicate only a fraction of the extent of the damage. In this paper, an effective noise-robust damage identification method is proposed. It extracts local anomalies based on two-dimensional curvature propagating flexural waves (2D-CPFW). To alleviate adverse effects of measurement noise on calculating 2D-CPFW, the continuous wavelet transform with a second-order Gaussian function is used as a differentiation operator. Three two-dimensional quantities, including the curvature of the 2D-CPFW, Teager energy of 2D-CPFW, and Teager energy of the curvature of the 2D-CPFW, are defined and they can intensify local anomalies caused by the existence of damage. To obtain a complete identification of damage extent based on the anomalies, a two-dimensional accumulative damage index is defined. A convergence index is introduced to determine the number of waveforms to be included when calculating the damage index. Effectiveness and noise-robustness of the proposed method are investigated in a numerical example of a damaged plate. Results verify that the proposed method is effective and noise-robust in identifying the location and extent of damage.

Infrastructures such as tunnels and bridges in the United States are facing severe degradation. According to the National Tunnel Inventory, the tunnels’ condition needs to be evaluated by a condition state method periodically, which includes detecting and measuring the delamination in the concrete liner due to safety issues. For concrete tunnels in poor conditions, delamination is usually detected using sounding tests (using hammer) and quantified manually (with tape measures and sketch). The quantification process of identified delaminated areas is time-consuming and uneconomical. An alternative approach for quantifying the detected delaminated areas is proposed and validated in this case study. A series of images were extracted from a video recorded from a vehicle traveling through a tunnel. Images were scaled, then converted to binary images, and processed with a pixel-based quantification algorithm. The quantification algorithm can take into consideration the curvature of the surfaces to obtain accurate quantification of surface areas. The delamination dimensions evaluated by the algorithm were verified by the manual quantification method. It is believed that this method could be combined with other structure testing methods for the interdisciplinary perspective of the structural condition.

Structural dynamics provide critical information for structural health monitoring (SHM), such as changes in the modal behavior which indicate damages. However, for complex systems with noisy operational environments, many factors may influence the estimation of natural frequencies and other modal domain SHM features, such as varying mass distribution of bridges and the variation due to fluctuating temperature and unideal boundary conditions. For this reason, allying the mode shapes with the natural frequencies to forecast damages would pose a more robust solution. Among the techniques existent to perform damage detection, data-driven models, such as machine learning algorithms, are becoming widely used currently. For mode shape extraction, convolutional neural networks (CNN) have been applied to imagery data, allowing to extract full-field mode shapes with a denser spatial resolution (quasi-full field) of the structure if compared to traditional hardware. Combining CNN with long short-term memory (LSTM) network will associate the temporal dependency of the frames with its features which will be more specific for SHM decision-makings. In addition, for the circumstances with low vibration amplitude and subpixel image resolution, applying phase-based motion estimation (PME) and phase-based motion magnification (PMM) allows to extract the natural frequencies with subtle motion magnified at the resonances aiding to emphasize the dynamic features desired. As the training of the deep learning model, a lab-scale truss structure was adopted with different load conditions in order to obtain the required data, and the performance is cross-validated.

Current computer video-based vibration modal analysis approaches typically decompose video frames into representations and then adjust them that allow to magnify motions to extract motion representations for vibration modal analysis. Their decomposition usually relies upon handcrafted designed kernels, such as the complex steerable kernels, which typically may not be optimally designed for the extraction of subtle motions specially in higher frequency domains. In this paper, optimal decomposition kernel is learned and designed directly from baseline dataset images using deep convolutional neural network (CNN) models. Each subpixel of an image obtained from a digital camera is included when computing the spatiotemporal information, which serves similar to an individual motion sensor to acquire the modal frequencies of a vibrating structure. A hybrid-attention-LSTM-based deep convolutional neural network architecture is developed to take advantage of attention and LSTM blocks to discover subtle motions from a specific source to visualize high resolution of dynamic properties of the structures in the existence of high amounts of noise. The idea of transfer learning is utilized to transfer the knowledge previously learned to new limited dataset. Transfer learning is used to take advantage of limited existing dataset to avoid underfitting in the training of the network, considering the current publicly available modal frequency datasets are insufficient to train a generalized network. The proposed deep learning architecture is designed in such a way that has capability of transferring the trained model from baseline dataset on a simple structure to a complicated structure using transfer learning perspective. After training, the model takes the video of a vibrating structure as input and outputs the fundamental modal frequencies. By showing reliable empirical results, the proposed model is autonomous, efficient, and accurate.

Most of the critical defects in structural components are invisible at the surface, mainly throughout early stages of deterioration, causing their timely detection to be a challenge. Assessing the actual and accurate 3D form and extent of interior defects is a complicated and also cumbersome task, unexpectedly with the developments in NDE techniques. Unlike the majority of traditional methods based on specialized forms of surface-penetrating waves or radiation imaging, this research uses optical cameras for full-field sensing of surface strains and deformations using the 3D-DIC technique as the basis for damage identification. This data-rich representation of behavior of the structural component is then leveraged in an inverse mechanical problem to reconstruct the underlying subsurface abnormalities. The inverse problem is solved through a topology optimization formulation that iteratively adjusts a fine-tuned FEM of the structure to infer abnormalities within the structure.

Recently illustrated the feasibility of detecting and reconstructing the existence of 3D defects within small-scale structural components such as coupons using the proposed idea, this work focuses on expanding on the work by the authors to reveal that the proposed idea can be employed on large-scale structural components using the rich data from full-field image-based measurements to enable the identification of a more detailed picture of the internal defects. Thus, the goal of this research is to demonstrate the practicability and investigate the performance of the previously proposed method through an experimental program in which a set of large-scale structural components such as steel beams with and without buried defects are tested with full-field DIC sensing. A corresponding set of research steps with an increasing level of sophistication are designed to assess the capability of the approach to estimate steel material properties then to extent to infer the 3D shape of embedded defects. Upon completion, this research is expected to demonstrate the feasibility and practicality of the proposed subsurface structural component condition assessment technique and pave the way for its future implementation in existing structures.

The phase-based motion magnification, which decomposes video frames into a set of kernels, is a recent video processing technique that has been developed to perceive the undetectable subtle motions in videos within a specific frequency range. However, the parameter determination in designing the optimal kernel characteristics for video motion magnification is challenging, which needs to be addressed, in terms of the center frequency and bandwidth, especially when the structure is geometrically complex and the structural vibration modes are not well separated in the frequency domain. Their decomposition usually is determined by handcrafted designed kernels, such as the complex steerable pyramids, and Gabor wavelets, typically may not be optimally designed kernels for the extraction of subtle motions in a specific scenario. In this paper, optimal decomposition kernel is learned and designed directly from baseline dataset images acquired from existing videos using deep convolutional neural networks (CNNs) approach. Many of these responses resemble Gabor wavelet filters and Laplacian filters, which suggests that the proposed deep network learns to extract similar information as done by the complex steerable filters. By contrast, the texture kernel responses show many blurring kernels.