The detection of cracks in sand moulds has long been a challenge for both safety and maintenance purposes. Traditional image processing techniques have been employed to identify and quantify these defects but have often proven to be inefficient, labour-intensive, and time-consuming. To address this issue, we sought to develop a more effective approach using deep learning techniques, specifically semantic segmentation. We initially examined three different architectures—U-Net, SegNet, and DeepCrack—to evaluate their performance in crack detection. Through testing and comparison, U-Net emerged as the most suitable choice for our project. To further enhance the model's accuracy, we combined U-Net with VGG-19, VGG-16, and ResNet architectures. However, these combinations did not yield the expected improvements in performance. Consequently, we introduced a new layer to the U-Net architecture, which significantly increased its accuracy and F1 score, making it more efficient for crack detection. Throughout the project, we conducted extensive comparisons between models to better understand the effects of various techniques such as batch normalization and dropout. To evaluate and compare the performance of the different models, we employed the loss function, accuracy, Adam optimizer, and F1 score as evaluation metrics. Some tables and figures explain the differences between models by using image comparison and evaluation metrics comparison; to show which model is better than the other. The conducted evaluations revealed that the U-Net architecture, when enhanced with an extra layer, proved superior to other models, demonstrating the highest scores and accuracy. This architecture has shown itself to be the most effective model for crack detection, thereby laying the foundation for a more cost-efficient and trustworthy approach to detecting and monitoring structural deficiencies.

Image-based techniques are a promising non-destructive approach for road pavement condition evaluation. The main objective of this study is to extract, quantify and evaluate important surface defects, such as cracks, using an automated computer vision-based system to provide a better understanding of the pavement deterioration process. To achieve this objective, an automated crack-recognition software was developed, employing a series of image processing algorithms of crack extraction, crack grouping, and crack detection. Bottom-hat morphological technique was used to remove the random background of pavement images and extract cracks, selectively based on their shapes, sizes, and intensities using a relatively small number of user-defined parameters. A technical challenge with crack extraction algorithms, including the Bottom-hat transform, is that extracted crack pixels are usually fragmented along crack paths. For de-fragmenting those crack pixels, a novel crack-grouping algorithm is proposed as an image segmentation method, so called MorphLink-C. Statistical validation of this method using flexible pavement images indicated that MorphLink-C not only improves crack-detection accuracy but also reduces crack detection time. Crack characterization was performed by analysing imagerial features of the extracted crack image components. A comprehensive statistical analysis was conducted using filter feature subset selection (FSS) methods, including Fischer score, Gini index, information gain, ReliefF, mRmR, and FCBF to understand the statistical characteristics of cracks in different deterioration stages. Statistical significance of crack features was ranked based on their relevancy and redundancy. The statistical method used in this study can be employed to avoid subjective crack rating based on human visual inspection. Moreover, the statistical information can be used as fundamental data to justify rehabilitation policies in pavement maintenance. Finally, the application of four classification algorithms, including Artificial Neural Network (ANN), Decision Tree (DT), k-Nearest Neighbours (kNN) and Adaptive Neuro-Fuzzy Inference System (ANFIS) is investigated for the crack detection framework. The classifiers were evaluated in the following five criteria: 1) prediction performance, 2) computation time, 3) stability of results for highly imbalanced datasets in which, the number of crack objects are significantly smaller than the number of non-crack objects, 4) stability of the classifiers performance for pavements in different deterioration stages, and 5) interpretability of results and clarity of the procedure. Comparison results indicate the advantages of white-box classification methods for computer vision based pavement evaluation. Although black-box methods, such as ANN provide superior classification performance, white-box methods, such as ANFIS, provide useful information about the logic of classification and the effect of feature values on detection results. Such information can provide further insight for the image-based pavement crack detection application.

No description available.

No description available.

Sample collection is a common task for mobile robots and there are a variety of manipulators available to perform this operation. This thesis presents a novel scoop sample collection system design which is able to both collect and contain a sample using the same hardware. To ease the operator burden during sampling the scoop system is paired with new semi-autonomous and fully autonomous collection techniques. These are derived from data provided by colored 3D point clouds produced via image-based 3D reconstructions. A custom robotic mobility platform, the Scoopbot, is introduced to perform completely automated imaging of the sampling area and also to pick up the desired sample. The Scoopbot is wirelessly controlled by a base station computer which runs software to create and analyze the 3D point cloud models. Relevant sample parameters, such as dimensions and volume, are calculated from the reconstruction and reported to the operator. During tests of the system in full (48 images) and fast (6-8 images) modes the Scoopbot was able to identify and retrieve a sample without any human intervention. Finally, a new building crack detection algorithm (CDA) is created to use the 3D point cloud outputs from image sets gathered by a mobile robot. The CDA was shown to successfully identify and color-code several cracks in a full-scale concrete building element. / Master of Science

This thesis introduces a multi-Bayesian framework for detection and classification of features in environments abundant with error-inducing noise. The approach takes advantage of Bayesian correction and classification in three distinct stages. The corrective scheme described here extracts useful but highly stochastic features from a data source, whether vision-based or otherwise, to aid in higher-level classification. Unlike many conventional methods, these features’ uncertainties are characterized so that test data can be correctively cast into the feature space with probability distribution functions that can be integrated over class decision boundaries created by a quadratic Bayesian classifier. The proposed approach is specifically formulated for road crack detection and characterization, which is one of the potential applications. For test images assessed with this technique, ground truth was estimated accurately and consistently with effective Bayesian correction, showing a 33% improvement in recall rate over standard classification. Application to road cracks demonstrated successful detection and classification in a practical domain. The proposed approach is extremely effective in characterizing highly probabilistic features in noisy environments when several correlated observations are available either from multiple sensors or from data sequentially obtained by a single sensor. / Master of Science / Humans have an outstanding ability to understand things about the world around them. We learn from our youngest years how to make sense of things and perceive our environment even when it is not easy. To do this, we inherently think in terms of probabilities, updating our belief as we gain new information. The methods introduced here allow an autonomous system to think similarly, by applying a fairly common probabilistic technique to the task of perception and classification. In particular, road cracks are observed and classified using these methods, in order to develop an autonomous road condition monitoring system. The results of this research are promising; cracks are identified and correctly categorized with 92% accuracy, and the additional “intelligence” of the system leads to a 33% improvement in road crack assessment. These methods could be applied in a variety of contexts as the leading edge of robotics research seeks to develop more robust and human-like ways of perceiving the world.

A load-enhanced methodology has been proposed to enable the in situ detection, localization, and characterization of damage in metallic plate-like structures using Lamb waves. A baseline-free load-differential method using the delay-and-sum imaging algorithm is proposed for defect detection and localization. The term “load-differential” refers to the comparison of recorded ultrasonic signals at various levels of stress. Defect characterization is achieved by incorporating expected scattering information of guided waves interacting with defects into the minimum variance imaging algorithm, and a method for estimating such scattering patterns from the measurements of a sparse transducer array is developed. The estimation method includes signal preprocessing, extracting initial scattering values from baseline subtraction results, and obtaining the complete scattering matrix by applying radial basis function interpolation. The factors that cause estimation errors, such as the shape parameter used to form the basis function and the filling distance used in the interpolation, are discussed. The estimated scattering patterns from sparse array measurements agree reasonably well with laser wavefield data and are further used in the load-enhanced method. The results from fatigue tests show that the load-enhanced method is capable of detecting cracks, providing reasonable estimates of their localizations and orientations, and discriminating them from drilled holes, disbonds, and fastener tightness variations.

<p></p><p>Visual inspection is the primary means of ensuring the safety and functionality of in-service bridges in the United States and owners spend considerable resources on such inspections. While the Federal Highway Administration (FHWA) and many state departments of transportation have guidelines related to inspector qualification, training, and certification, an inspector’s actual capability to identify defects in the field under these guidelines is unknown. This research aimed to address the knowledge gap surrounding visual inspection performance for steel bridges in order to support future advances in inspection and design procedures. Focusing primarily on fatigue crack detection, this research also considered the ability of inspectors to accurately and consistently estimate section loss in steel bridge members. </p> <p> </p> <p>Inspection performance was evaluated through a series of simulated bridge inspections performed in representative in-situ conditions. First, this research describes the results from 30 hands-on, visual inspections performed on full size bridge specimens with known fatigue cracks. Probability of Detection (POD) curves were fit to the inspection results and the 50% and 90% detection rate crack lengths were determined. The variability in performance was large, and only a small amount of the variance could be explained by individual characteristics or environmental conditions. Based on the results, recommendations for improved training methods, inspection procedures, and equipment were developed. Above all, establishment of a performance based qualification system for bridge inspectors is recommended to confirm that a satisfactory level of performance is consistently achieved in the field. </p> <p> </p> <p>Long term, managing agencies may eschew traditional hands-on bridge inspection methods in favor of emerging technologies imagined to provide improved results and fewer logistical challenges. This research investigated the potential for unmanned aircraft system (UAS) assistance during visual inspection of steel bridges. Using the same specimens as in the hands-on inspections, four UAS-assisted field inspections and 19 UAS-assisted desk inspections were performed. A direct comparison was made between performance in the hands-on and UAS-assisted inspections, as well as between performance in the two types of UAS-assisted inspections. Again, significant variability was present in the results suggesting that human factors continue to have a substantial influence on inspection performance, regardless of inspection method. </p> <p> </p> <p>Finally, to expand the findings from the crack detection inspections, the lower chord from a deck truss was used to investigate variability in the inspection of severely corroded steel tension members. Five inspectors performed a hands-on inspection of the specimen and four engineers calculated the load rating for the same specimen. Significant variability was observed in how inspectors recorded thickness measurements during the inspections and engineers interpreted the inspection reports and applied the code requirements. </p><br><p></p>

Orientador: Gilberto Pechoto de Melo / Resumo: A identificação de trincas ainda é um desafio na área de monitoramento da integridade estrutural em eixos rotativos. Embora muitas técnicas e modelos tenham sido desenvolvidos, encontrar uma técnica eficiente que possa localizar uma única ou múltiplas trincas ao longo do eixo, ainda é um grande desafio. Nesse trabalho, um novo índice para localização de trincas em eixos rotativos é apresentado. A equação do movimento do rotor com trinca utiliza a notação em coordenadas complexas a fim de separar as contribuições dos modos de precessão direta e retrógrada. O índice SDI (shape and directivity index) é calculado para o rotor, cujo modelo matemático considera a variação instantânea da rigidez do elemento finito devido à abertura e o fechamento gradual da trinca, conhecido como efeito breathing. Através da manipulação do SDI no modelo de cores HSV (hue, saturation and value), desenvolve-se uma escala métrica, visualizada em um mapa de cores, que possibilita localizar a anisotropia causada pela trinca ao longo do eixo. Profundidade e posição da trinca, presença de múltiplas trincas, entre outros fatores que afetam a assinatura da trinca em outros métodos de identificação são analisados. Bons resultados demonstram a eficiência e robustez do novo índice para diversos casos de operação do rotor. Essa métrica de dano acrescenta uma contribuição para os métodos de localização de trincas em sistemas rotativos. / Abstract: Crack identification is still a challenge in the area of structural health monitoring dedicated to rotating shafts. Although many techniques and models have been developed, finding an efficient technique capable of locating a single or multiple cracks along the shaft is still a challenge. In this work, a new index for locating cracks in rotating shafts is proposed. The equation of motion of the cracked rotor uses notation in complex coordinates in order to separate the contributions of forward and backward precession modes. The SDI (shape and directivity index) is calculated for the cracked rotor, which mathematical model considers the instantaneous variation of the finite element stiffness due to the gradual opening and closing of the crack, known as the breathing effect. By manipulating the SDI in the HSV (hue, saturation and value) color model, a metric scale is developed to locate the anisotropy caused by cracks along the shaft, visualized on a color map. Depth and position of the crack, presence of multiple cracks, among other factors that affect the signature of the crack in other identification methods are analyzed. Good results demonstrate the efficiency and robustness of the new index for several rotor operation conditions. This damage metric contributes to crack localization methods in rotating systems. / Doutor

To increase efficiency, shafts are made lighter and more flexible, and are designed to rotate faster to increase the system's power-to-weight ratio. The demand for higher efficiency in rotordynamic systems has led to increased susceptibility to transverse fatigue cracking of the shaft. Shaft cracks are often detected and repaired during scheduled periods of off-line maintenance. Off-line maintenance can be expensive and time consuming; on-line condition monitoring allows maintenance to be performed as-needed. However, inadequate (or a lack of) monitoring can allow rapidly propagating cracks to result in catastrophic shaft failure. It is therefore imperative to develop on-line condition monitoring techniques to detect a crack and diagnose its severity. A particularly useful method for transverse shaft crack detection/diagnosis is vibration monitoring. Detection, and especially diagnosis, of transverse fatigue cracks in rotordynamic systems has proven difficult. Whereas detection assesses only the presence of a crack, diagnosis estimates important crack parameters, such as crack depth and location. Diagnosis can provide the operator with quantitative information to assess further machinery operation. Furthermore, diagnosis provides initial conditions and predictive parameters on which to base prognostic calculations. There is a two-fold challenge for on-line diagnosis of transverse fatigue crack parameters. First, crack characterization involves specifying two important parameters: the crack's depth and location. Second, the nature of rotating machinery permits response measurement at only specific locations. Cracks are typically categorized as breathing or gaping; breathing cracks open and close with shaft rotation, while gaping cracks remain open. This work concerns the diagnosis of gaping crack parameters; the goal is to provide metrics to diagnose a crack's depth and location. To this end, a comprehensive approach is presented for modeling an overhung cracked shaft. Two linear gaping crack models are developed: a notch and a gaping fatigue crack. The notch model best approximates experimentally manufactured cracks, whereas the gaping fatigue crack model is likely more suited for real fatigue cracks. Crack diagnosis routines are established using free and forced response characteristics. Equations of motion are derived for both crack models, including excitation due to gravity and imbalance. Transfer matrix techniques are established to expediently obtain the steady-state system response. A novel transfer matrix technique, the Complex Transfer Matrix, is developed to distinguish forward and backward whirl components. The rotor's angular response is primarily employed in this work for crack detection and diagnosis. The overhung shaft induces an increased sensitivity to variations in crack depth and location. In addition, an available overhung rotordynamic experimental test rig allows for comparison of the current analytic results to previously obtained experimental results. Under the influence of gravity, the steady-state response of the cracked system includes a prominent 2X harmonic component, appearing at a frequency equal to twice the shaft speed. The magnitude of the 2X harmonic is strongly influenced by the shaft speed. A resonant response occurs when the shaft speed reaches half of a system natural frequency. This work demonstrates that the profile of the 2X harmonic versus shaft speed is a capable diagnostic tool. Identification of the 2X resonance frequency restricts the crack parameters to certain pairs of location and depth. Following this limiting process, the magnitude of the 2X harmonic is used to identify the crack's depth and location. Orbital shapes at the rotor are discussed, as are orbital modes of the shaft deflection. Quantitative results and qualitative observations are provided concerning the difficulty of crack detection and diagnosis.