Railroad defects is a major challenge that has received attention amongst railway specialists in recent years. All types of defects are observed to be deleterious to railroad safety if not attended to. In particular, squat defects are produced by wheel-rail dynamic impact, leaving a depression mark on the rail surface. In this research, deep neural networks are used to classify thousands of kilometers of railway track into two binary classes: squat(p) and no-squat(n). In order to mitigate large class imbalance, we consider several natural sampling and data augmentation methods. We demonstrate that data augmentation and segmentation techniques can yield a considerable improvement compared to traditional re-sampling approaches.

Rail geometry defects constitute a major cause of accidents in the United States. Geometry related accidents are often very severe and damaging. While rail geometry-caused derailments continue to rise according to Federal Railroad Administration (FRA) data, track quality analysis remain effectively unchanged. The use of TQI or track quality index takes a narrow view of track assessment by focusing on quality without considering safety. The bipartite analysis of track quality and safety results into two maintenance types: routine and corrective maintenance respectively. This study aims to create a hybrid index that combines both element of safety and geometry quality to predict only one maintenance regime based on track condition. It is an initial step towards the big picture of creating indices that will be iterated based on maintenance savings and defect probability thresholds. This study employs a linear and nonlinear dimension reduction technique that expresses the probability distribution of observations based on the similarity or dissimilarity in their embedded space whilst also maximizing the variance in data. This study found application in principal component analysis (PCA) and T-Stochastic neighbor embedding (TSNE) for separating geometry defects from higher dimensional space to lower dimensions. Results show that while both techniques effectively reduces track geometry data, PCA yields a potential defect probability threshold in spite of TSNE being a better geometry defect predictor.

Rail defects constitute a major cause of derailments in the United States. Derailments are often very fatal and disastrous. While rail defect-caused derailments are becoming an increasing concern, the analysis of rail defects remain relatively unchanged. Weibull analysis has historically been the "go-to" distribution when it comes to analyzing rail defects irrespective of the type, tonnage, size or age. In other words, the analysis is not only defect-based, the defect distribution varies from section to section. "No free-lunch" theorem in statistical learning provides the basis to assert that a single statistical technique is not guaranteed to perform excellently on all defect types because all techniques will perform almost equally if their performances are aggregated on all possible data types. Hence, the use of Weibull distribution is not only ideal for all defect types. It also does not provide a network infrastructure manager the information needed to allocate resources for each defect type. More so, Weibull distribution like most other distributions belong to the beta family which is trained, fitted and tested on the same data-set. In this study, we present an agglomerated machine learning of defects as an alternative to Weibull Analysis of rail infrastructure. This methodology considers different rail defect prediction techniques using Stacking Ensembles in a way that training and testing is done on different data splits. We considered a database of over 20 miles of rail defect and track geometry data spanning approximately 5 years. Our results provide an infrastructure tool developed to assist rail owners and infrastructure managers a decision-making rationale for resource allocation and maintenance rather than the micro-level decision making influenced by the Weibull for limited sections of track.

Track geometry has a considerable effect on rail travel comfort and safety and deteriorates with age and tonnage. In order to maintain the track geometry quality, maintenance activities such as tamping, stone blowing and ballast undercutting are usually employed. However, these activities are ineffective if the underlying cause of track deformation such as subgrade failure is not addressed. Geosyn-thetics such as geocells and geogrids can be placed in the subballast which strengthens the layer, lowers the stresses on the weak subgrade and invariably enhances track geometry quality. Machine learning techniques are becoming increasingly imperative in processing and analyzing of large volumes of track geometry data which exhibit the classical attributes of big data. Several unsupervised and supervised learning techniques were used to analyze the effect of geocell installation on track geometry quality. Cluster analysis was used to group the track geometry data with major clusters found to differ by surface and alignment features. Principal component analysis was employed as an effective dimension reduction tool to simplify the track geometry data based on the proportion of variance explained. Supervised learning techniques such as multiple linear regression, decision tree regression, random forest regression and support vector regression were subsequently used to estimate and predict the effect of geocell installation on the track geometry quality. Random forest regression was found to the best performing model for both the original and dimensionally-reduced data.

The National Bridge Inventory (NBI) has detailed information on over 600,000 bridges nationwide. The data, which spans a period of more than 20 years can be very useful for analyzing and modeling bridge performance. Previous analysis methods employ a 2-dimensional view of data which may result in the loss of subtle trends and changes in the data. Bridge data is inherently multidimensional and as such there is an added advantage in analyzing it in its natural multidimensional state. This paper focuses on the use of a multiway data analysis approach known as tensor decomposition to analyze the structural deficiency rate with respect to states, bridge structural types and time. The tensor decomposition approach is able to reveal clusters and patterns which are not easily perceptible when using conventional statistical tools.

As Big Data becomes part of railroad data analysis, there are many challenges which need to be addressed by the railway industry. This extended abstract highlights some of the challenges from specific examples in railway engineering. This work does not present the challenges of dealing with Big Data in general which is beyond the scope of this paper. The examples provided in this extended abstract cover both the engineering and the management of railroad applications.

Big data is becoming increasingly important in various fields; railway engineering is no exception. The use of advanced analysis tools will lead to improved reliability and safety in railway systems. This paper addresses how metaheuristics can be used as an optimization technique to accurately analyze large data in railway engineering. Contributions in both optimization and application in railway engineering are also mentioned. Also, future research towards data analysis in real-life problems is discussed.