Traditional feature extraction and selection is a labor-intensive process requiring expert knowledge of the relevant features pertinent to the system. This knowledge is sometimes a luxury and could introduce added uncertainty and bias to the results. To address this problem a deep learning enabled featureless methodology is proposed to automatically learn the features of the data. Timefrequency representations of the raw data are used to generate image representations of the raw signal, which are then fed into a deep convolutional neural network (CNN) architecture for classification and fault diagnosis. This methodology was applied to two public data sets of rolling element bearing vibration signals. Three time-frequency analysis methods (short-time Fourier transform, wavelet transform, and Hilbert-Huang transform) were explored for their representation effectiveness. The proposed CNN architecture achieves better results with less learnable parameters than similar architectures used for fault detection, including cases with experimental noise.

Recurring expenses associated with preventative maintenance and inspection produce operational inefficiencies and unnecessary spending. Human inspectors may submit inaccurate damage assessments and physically inaccessible locations, like underground mining structures, and pose additional logistical challenges. Automated systems and computer vision can significantly reduce these challenges and streamline preventative maintenance and inspection.The authors propose a convolutional neural network (CNN)-based approach to identify the presence and type of structural damage. CNN is a deep feed-forward artificial neural network that utilizes learnable convolutional filters to identify distinguishing patterns present in images. CNN is invariant to image scale, location, and noise, which makes it robust to classify damage of different sizes or shapes. The proposed approach is validated with synthetic data of a composite sandwich panel with debonding damage, and crack damage recognition is demonstrated on real concrete bridge crack images. CNN outperforms several other machine learning algorithms in completing the same task. The authors conclude that CNN is an effective tool for the detection and type identification of damage.

One of the main challenges that the industry faces when dealing with massive data for failure diagnosis is high dimensionality of such data. This can be tackled by dimensionality reduction method such as principal components analysis, which usually results in an improved fault diagnosis. Other available techniques include auto-encoders and its variants denoising auto-encoders and sparse auto-encoders. Most recently, variational auto-encoders are one of the most promising techniques for unsupervised learning with successful applications in image processing and speech recognition. Differently from other auto-encoder methods, variational auto-encoders use variational inference to generate a latent representation of the data and impose a distribution over the latent variables and the data itself. In this article, we propose a fully unsupervised deep variational auto-encoder-based approach for dimensionality reduction in fault diagnosis and explore the variational auto-encoder capabilities for such a task. This is achieved by comparing the latent representations provided by variational auto-encoders to the ones from principal components analysis as well as when no reduction is performed in ball bearings' fault classification using vibration signals. To tackle massive sensor data, we propose different architectures for the variational auto-encoder's encoder and decoder that are based on deep neural networks and deep convolutional neural networks. Experiments are also carried out by varying the data preprocessing methods for generating spectrograms and hand-engineering features as well as the use of raw vibration data, the architecture of the neural networks fault classifiers operating on the latent representations from variational auto-encoder and principal components analysis, the degree of data dimension reduction, and the size of the available labeled data used for training the fault classifiers. The results show that variational auto-encoders are a competent and promising tool for dimensionality reduction for use in fault diagnosis and worth further exploring their capabilities beyond vibration signals of ball bearing elements.

Modal parameters such as natural frequencies and mode shapes are sensitive indicators of structural damage. However, they are not only sensitive to damage, but also to the environmental conditions such as, humidity, wind and most important, temperature. For civil engineering structures, modal changes produced by environmental conditions can be equivalent or greater than the ones produced by damage. This article proposes a damage detection method which is able to deal with temperature variations. The objective function correlates mode shapes and natural frequencies, and a parallel genetic algorithm handles the inverse problem. The numerical model of the structure assumes that the elasticity modulus of the materials is temperature-dependent. The algorithm updates the temperature and damage parameters together. Therefore, it is possible to distinguish between temperature effects and real damage events. Simulated data of a three-span bridge and experimental one of the I-40 Bridge validate the proposed methodology. Results show that the proposed algorithm is able to assess the experimental damage despite of temperature variations.

Artículo de publicación ISIAn hybrid real-coded Genetic Algorithm with damage penalization is implemented to locate and quantify structural damage. Genetic Algorithms provide a powerful tool to solved optimization problems. With an appropriate selection of their operators and parameters they can potentially explore the entire solution space and reach the global optimum. Here, the set-up of the Genetic Algorithm operators and parameters is addressed, providing guidelines to their selection in similar damage detection problems. The performance of five fundamental functions based on modal data is studied. In addition, this paper proposes the use of a damage penalization that satisfactorily avoids false damage detection due to experimental noise or numerical errors. A tridimensional space frame structure with single and multiple damages scenarios provides an experimental framework which verifies the approach. The method is tested with different levels of incompleteness in the measured degrees of freedom. The results show that this approach reaches a much more precise solution than conventional optimization methods. A scenario of three simultaneous damage locations was correctly located and quantified by measuring only a 6.3% of the total degrees of freedom

The effect of molecular weights and hydrolysis degrees (HD) of polyvinyl alcohol (PVA) on thermal and mechanical properties and crystallinity of polylactic acid (PLA)/PVA blends was investigated. Blends were prepared by the melt blending method using PLA/PVA ratios: 80/20, 90/10 and 97/3 wt. %. A single glass transition temperatures was observed for all PLA/PVA blends, suggesting the formation of binary compatible blends at concentration range studied. Thermogravimetric analysis results showed a better thermal stability for PLA/PVA blends containing PVA of higher Mw and HD. According to mechanical properties, low quantities of PVA (3 wt. %) do not affect the tensile strength of blends (irrespective of Mw and HD). However, as the PVA content increases, tensile strength tends to lower values, especially for blends with 20 wt.% of PVA, with 98% of HD.

No abstract

This article presents a novel impact identification algorithm that uses a linear approximation handled by a statistical inference model based on the maximum-entropy principle, termed linear approximation with maximum entropy (LME). Unlike other regression algorithms as artificial neural networks (ANNs) and support vector machines, the proposed algorithm requires only parameter to be selected and the impact is identified after solving a convex optimization problem that has a unique solution. In addition, with LME data is processed in a period of time that is comparable to the one of other algorithms. The performance of the proposed methodology is validated by considering an experimental aluminum plate. Time varying strain data is measured using four piezoceramic sensors bonded to the plate. To demonstrate the potential of the proposed approach over existing ones, results obtained via LME are compared with those of ANN and least square support vector machines. The results demonstrate that with a low number of sensors it is possible to accurately locate and quantify impacts on a structure and that LME outperforms other impact identification algorithms.