Deep learning has shown success in several applications involving pattern recognition, expert systems, and scientific discovery. However, existing methods struggle with industrial applications, which are often challenged by non-ideal datasets. In many cases, the datasets are small, poorly labeled, noisy, or have unbalanced class distribution or any combination of such problems. In this Master's research, we propose a generative adversarial network (GAN) strategy that is able to circumvent limitations imposed by tiny datasets. As a case study, we use the extrapolation of corrosion in automobiles and feed our deep learning framework with only a few dozen images instead of the thousands to million images commonly found in many computer vision problems. In order to handle such a reduced dataset, we use one GAN for the rust level and one for the rust texture. The rust level GAN is conditional on random samples from the dataset and uses an additive random noise in the latent space to add variability to the generated rust level maps. The rust texture GAN adds shades of brown to the outputs of the rust level GAN. Loss functions are carefully designed to produce a robust training scheme for both GANs. In addition, given the significantly reduced size of our dataset, it is unfeasible to break down the data into training, validation, and test sets. We overcome this limitation by using the discrepancy between the generated and target distributions of the rust level and texture intensities as a way to monitor the convergence of training. The resulting models can ingest an image with a car having no corrosion and generate an image of this car with parts exhibiting varying degrees of corrosion (from mild to moderate to severe).

Composite materials are being more widely used today by aerospace, automotive, and a number of other commercial industries because of their advantages over conventional metals. Composites are finding applications ranging from bicycle frames to the proposed High-Speed Civil Transport (HSCT). Determining the response to a variety of damage mechanisms is necessary for a complete understanding of the total use environment of composite structures. The objective of the research presented here is to provide a method of quantifying the amount of damage in composite materials for a number of different damage scenarios. Components which have non-visible damage, but have degraded performance, are of interest. at this level of damage, the safety margin designed into the structure may be compromised.;Nondestructive Evaluation (NDE) is a field of measurement physics where energy is imparted to a material and information is obtained from observing how the energy interacts with the system. Many different forms of energy can be used to obtain useful information from these measurements: acoustic, thermal, x-ray, optical, and electromagnetic. Among the many various techniques available, ultrasonic Lamb waves offer a convenient method of evaluating these composite materials. as a material is damaged, the elastic parameters of the structure change. Since the Lamb wave velocity depends on these properties, an effective tool exists to monitor damage in composites by measuring the velocity of these waves. Additionally, Lamb wave measurements are beneficial because they can propagate over long distances and are sensitive to the desired in-plane elastic properties of the material.;Presented in this study are the results involving the investigation of a variety of damage mechanisms (fatigue, thermal, and thermal-mechanical) using the Lamb wave technique. Two fatigue studies were conducted which showed that the change in modulus and change in velocity of the Lamb wave squared follow the same general trend. The Lamb wave velocity was also observed to decrease with increasing crack density. For the thermal damage study, the results showed that the velocity of the lowest order symmetric Lamb mode dropped significantly for extended thermal damage. When the experimental results were compared to model calculations, good agreement was observed for both fatigue and thermal damage. Finally, for thermal-mechanical damage, it was found that the Lamb wave technique was also able to predict a local defect in a specimen, which was later found to have a large delamination zone.;The Lamb wave velocity is a quantitative measurement and it has been shown by this work to be an effective tool in monitoring different types of damage in composites. Since the Lamb wave velocity depends on a variety of material properties, an ideal technique exists to monitor composites as damage is incurred. With the continued development of damage assessment techniques such as the Lamb wave method, the safety of such structures can be assured.

Many different kinds of physics are used in sensors that produce time-domain signals, such as ultrasonics, acoustics, seismology, and electromagnetics. The waveforms generated by these sensors are used to measure events or detect flaws in applications ranging from industrial to medical and defense-related domains. Interpreting the signals is challenging because of the complicated physics of the interaction of the fields with the materials and structures under study. often the method of interpreting the signal varies by the application, but automatic detection of events in signals is always useful in order to attain results quickly with less human error. One method of automatic interpretation of data is pattern classification, which is a statistical method that assigns predicted labels to raw data associated with known categories. In this work, we use pattern classification techniques to aid automatic detection of events in signals using features extracted by a particular application of the wavelet transform, the Dynamic Wavelet Fingerprint (DWFP), as well as features selected through physical interpretation of the individual applications. The wavelet feature extraction method is general for any time-domain signal, and the classification results can be improved by features drawn for the particular domain. The success of this technique is demonstrated through four applications: the development of an ultrasonographic periodontal probe, the identification of flaw type in Lamb wave tomographic scans of an aluminum pipe, prediction of roof falls in a limestone mine, and automatic identification of individual Radio Frequency Identification (RFID) tags regardless of its programmed code. The method has been shown to achieve high accuracy, sometimes as high as 98%.

As the worldwide aviation fleet continues to age, methods for accurately predicting the presence of structural flaws, such as hidden corrosion and disbonds, that compromise air worthiness become increasingly necessary. Ultrasonic guided waves, Lamb waves, allow large sections of aircraft structures to be rapidly inspected. However, extracting quantitative information from Lamb wave data has always involved highly trained personnel with a detailed knowledge of mechanical waveguide physics. In addition, human inspection process tends to be highly subjective, slow and prone to errors. The only practical alternative to traditional inspection routine is a software expert system capable of interpreting data with minimum error and maximum speed and reliability. Such a system would use the laws of guided wave propagation and material parameters to help signal processing algorithms automatically extract information from digitized waveforms. This work discusses several practical approaches to building such an expert system.;The next step in the inspection process is data interpretation, and imaging is the most natural way to represent two-dimensional structures. Unlike conventional ultrasonic C-scan imaging that requires access to the whole inspected area, tomographic algorithms work with data collected over the perimeter of the sample. Combined with the ability of Lamb waves to travel over large distances, tomography becomes the method of choice for solving NDE problems. This work explores different tomographic reconstruction techniques to graphically represent the Lamb wave data in quantitative maps that can be easily interpreted by technicians. Because the velocity of Lamb waves depends on the thickness, the traveltimes of the fundamental modes can be converted into a thickness map of the inspected region. Lamb waves cannot penetrate through holes and other strongly scattering defects and the assumption of straight wave paths, essential for many tomographic algorithms, fails. Diffraction tomography is a way to incorporate scattering effects into tomographic algorithms in order to improve image quality and resolution. This work describes the iterative reconstruction procedure developed for the Lamb Wave tomography and allowing for ray bending correction for imaging of moderately scattering objects.

This dissertation concentrates on the development of two new tomographic techniques that enable wide-area inspection of pipe-like structures. By envisioning a pipe as a plate wrapped around upon itself, the previous Lamb Wave Tomography (LWT) techniques are adapted to cylindrical structures. Helical Ultrasound Tomography (HUT) uses Lamb-like guided wave modes transmitted and received by two circumferential arrays in a single crosshole geometry. Meridional Ultrasound Tomography (MUT) creates the same crosshole geometry with a linear array of transducers along the axis of the cylinder. However, even though these new scanning geometries are similar to plates, additional complexities arise because they are cylindrical structures. First, because it is a single crosshole geometry, the wave vector coverage is poorer than in the full LWT system. Second, since waves can travel in both directions around the circumference of the pipe, modes can also constructively and destructively interfere with each other. These complexities necessitate improved signal processing algorithms to produce accurate and unambiguous tomographic reconstructions. Consequently, this work also describes a new algorithm for improving the extraction of multi-mode arrivals from guided wave signals. Previous work has relied solely on the first arriving mode for the time-of-flight measurements. In order to improve the LWT, HUT and MUT systems reconstructions, improved signal processing methods are needed to extract information about the arrival times of the later arriving modes. Because each mode has different through-thickness displacement values, they are sensitive to different types of flaws, and the information gained from the multi-mode analysis improves understanding of the structural integrity of the inspected material. Both tomographic frequency compounding and mode sorting algorithms are introduced. It is also shown that each of these methods improve the reconstructed images both qualitatively and quantitatively.

The aerospace industry is beginning to use advanced composite materials for primary load bearing structures and their failure mechanisms must be better understood to predict their behavior in service. The Combined Loads Tests (COLTS) facility is being constructed at the NASA Langley Research Center to characterize these failure mechanisms. Laser based ultrasonic NDE can monitor the samples during dynamic loading without interfering with the structural tests. However, the constraints of implementing laser ultrasound in a structures laboratory reduces the efficiency of the technique. The system has to be "eye-safe" because many people will be present during the structural tests. Consequently, laser light has to be delivered through fiber optics and a significant amount of light is lost. Also, the nature of the composite materials makes laser ultrasonic inspection difficult. The composites of interest are formed from woven layers that are stitched through the laminate thickness and bound in a resin matrix. These materials attenuate ultrasound strongly and exhibit a high degree of scattering.;Generation mechanisms in laser based ultrasound must be better understood to improve generation efficiency and consequently improve the signal-to-noise ratio. Although some experimental and theoretical studies have been conducted to characterize generation mechanisms, more extensive work is needed for composite materials. Specifically, we are concerned with generation mechanisms in thick, stitched composite structures. We describe a theoretical and experimental investigation of laser generated ultrasound in complex composite materials. We first develop a mathematical model describing the thermoelastic generation of ultrasound in a general anisotropic material. We then present a wide range of experimental data investigating the effects of laser and material parameters on the generated ultrasound. We specifically consider the relationship between laser pulse width, laser wavelength, and material composition. Finally, we compare this data to our mathematical model.

An improved and repeatable method for meshing golf club heads using finite elements in TrueGrid® was developed. Using solid brick elements through the thickness of the club head instead of shell elements better represents the many thickness variations throughout each section of a club head. This method also results in a high quality mesh at the center of the club head sections while still maintaining high quality at the edges. A simulation procedure was also developed to predict the acoustic pressure at a designated point in an acoustic medium of a golf club and ball impact using the BEM and Rayleigh methods in LS-DYNA®. The simulation time and computing power required for the impact are modest, while the acoustic simulation time and computing power are much greater. The Rayleigh method provides an alternative which can greatly reduce these requirements. The simulation of sound produced from the ball and a USGA COR plate, generic driver, and hybrid impact was accomplished with reasonable results. Experimental testing was performed using a USGA plate to validate the plate result. A simple tap test and an air cannon test were performed to record the acoustic response with a microphone. A Fast Fourier Transform was performed to obtain the frequency response. These two tests correlated with each other, indicating that air cannon procedures could be negated in favor of a much simpler tap test during prototype testing for acoustics. The simulation frequency responses showed similar results to the experimental tests, demonstrating that the procedure developed in this project can be a viable and effective method for determining the acoustic response of the golf club and ball impact.

<p>Condition monitoring and fault diagnosis of mechanical systems are two important issues that have received considerable attention from both academia and industry. Several techniques have been developed to date to address these issues. One category of these techniques which has been successfully applied in many industrial plants is based on the multiresolution multivariate analysis algorithms and more specifically the multi-scale principal component analysis (MSPCA). The present research aims to develop a multi-resolution multivariate analysis technique which can be effectively used for fault diagnosis of an internal combustion engine. Crank Angle Domain (CAD) Analysis is the most intuitive strategy for monitoring internal combustion engines. \comment{ as a cyclic system in which events at each cycle is correlated to a particular position of the crankshaft, this leads to analyzing the engine performance in angle domain (i.e. Crank Angle domain for engine) as very logical and intuitive strategy.} Therefore, MSPCA and CAD analysis were combined and a new technique, named CAD-MSPCA, was developed. In addition to this contribution, two indices were defined based on estimation of covariance matrices of scores and fault matrices. These indices were then employed for both fault localization and isolation purposes. In addition to this development, an interesting discovery made through this research was to use the statistical indices , calculated by MSPCA, for fault identification. It is mathematically shown that in case these indices detect a fault in the system, one can determine the spectral characteristics of the fault by performing the spectrum analysis of these indices. This analysis demonstrated the MSPCA as an attractive and reliable alternative for bearing fault diagnosis. These new contributions were validated through simulation examples as well as real measurement data.</p> / Master of Applied Science (MASc)

<p>This thesis tackles the behavior of two planar shaped opposing air jets. The word opposing means that the two air jets are positioned against each other to have a common center-line so that they impinge on each other. Opposing planar jets have several applications in industry which include: galvanization, chemical mixing processes, and combustion. Opposing planar jets are found to produce high level low frequency acoustic tones. In this study, the acoustic response of the opposing planar jets is investigated for different operating conditions. Acoustic tone analysis is performed for a wide range of jet exit velocities and separation distances between the two jets. Results show that the jets produce strong acoustic tones over the whole investigated range of test conditions. The acoustic response depends on the operating conditions of the jets and the acoustic tone is found to be generated by a self-sustained flow oscillation of the two jets.</p> <p>In order to further understand the reported oscillation phenomenon, particle image velocity (PIV) is utilized. Images of the oscillating flow field are acquired and image analysis is performed to obtain various flow dynamic properties. A computational fluid dynamics simulation is performed to help in highlighting the oscillation behavior of the jets. Results reveal that the jets exhibit anti-symmetric ‘’flapping’’ oscillation behavior. The jets are found to initially deflect away from their common centerline due to high stagnation pressure in the impingement region, while they deflect back to their common centerline due to cross stream oscillating velocity in the entrainment regions around the jets.</p> / Master of Applied Science (MASc)

Three types of acoustic nonlinearity parameters for solids are discussed. The results of measurements of these parameters for three polymers--polymethyl methacrylate, Polystyrene, and polysulfone--are presented.;The author has developed a new technique, using piezoelectric transducers directly bonded to the specimens, which allows the measurements of fundamental and second harmonics generated in the solids, and thereby the determination of nonlinearity parameter {dollar}\beta\sb3{dollar}, which is the ratio of a linear combination of second- and third-order elastic coefficients to the second-order elastic coefficient.;The second nonlinearity parameter, B/A can be determined from the temperature and pressure derivatives of the sound velocity. We derive its exact relationship for the case of solids. The results from the two techniques are shown to be consistent.;The pressure derivative of the sound velocity is also related to the Gruneisen parameter, which can be used to describe the anharmonicity of interactions in polymer molecules, especially of interchain vibrations. The interchain specific heat for these polymers is then calculated from the Gruneisen parameters; and the characterization of polymers by using these thermoacoustic parameters is discussed.