Connected Freight Rail Rolling Stock: a Modular Approach Integrating Sensors, Actors and Cyber Physical Systems for Operational Advantages and Condition based Maintenance Prof. Raphael PFAFF (FH Aachen University of Applied Sciences)
Due to the long braking distance of railway systems and the high velocities achieved,
railway operation needs to rely on train control systems. At the foundation of these
systems are models to predict the motion of the trains, including their anticipated
braking curve. Depending on the infrastructure manager, these braking curves need
to be achieved with a given safety, which is typically in the rare event region of
In current settings, it is typical to develop these so called braking curves either by
physical modelling of the train followed by a Monte Carlo simulation or following a
heuristic approach, mostly based on the high level of safety over the past centuries.
However, higher developed train protection and control systems, such as the
European Rail Traffic Management System (ERTMS) or the Russian KLUB-U
System together with current efforts towards quantitative risk analysis, e.g. the
European Common Safety Methods, require a more formal approach to
communicate the braking curve of a train between rolling stock and infrastructure.
An a priori determined set of braking curves is feasible for trains running in fixed or a
limited number of formations, such as multiple unit trains, however in the freight
railway system due to its vast amount of different vehicles and possible train setups,
the determination of the braking curves is prohibitive.
In this work, a procedure is proposed to obtain the variation in braking force from
accelerometer data onboard the vehicles. In order to allow for the large amount of
data to be expected, the procedure is formalised in MapReduce form. The variation
of the braking force from the expected value determined in this way can then be
forwarded to a Monte Carlo Analysis using importance sampling methodologies to
allow for an online calculation of braking curves based on the safety requirements
communicated by the infrastructure manager. It is expected that such a procedure
yields shorter braking curves than the safety factors currently proposed, leading to
higher commercial speeds and thus higher infrastructure usage.
A Multivariate CUSUM Chart Handling Auto- and Cross-correlated Observations Ms. KyuYoung LEE, Chuljin PARK and Mi Lim LEE (Hanyang University, Hongik University)
Multivariate CUSUM charts have been widely used as statistical-process-control tools to detect out-of-control states of monitoring variables. Most of earlier studies regarding multivariate CUSUM charts assume that observations of monitoring variables are independent or correlated with a limited structure. In this paper, we suggest a multivariate CUSUM chart that can handle the observations auto- and cross-correlated. The chart estimates its control limit analytically and captures small mean shifts faster when compared to existing charts.
Condition Monitoring Using Compressive Measurement with Variance Considered Machine Algorithm Mr. Jun Young JEON, Myoung Jun LEE, Gyuhae PARK, To KANG and Soon Woo HAN (Chonnam National University, Korea Atomic Energy Research Institute)
In this study, compressive sensing approaches for condition monitoring are proposed to demonstrate their efficiency in handling a large amount of data and the improved damage detection capability. Compressive sensing is a novel sensing/sampling paradigm that takes much fewer samples compared to traditional sampling methods. For experiments, a built-in rotating system was used. Data were compressively sampled to obtain compressed data. For damage detection, we used the Variance considered machine (VCM) algorithm to classify failure modes of rotating systems. Compared to the performance of traditional approaches, including Fisher Discriminant Analysis and Support Vector Machine, the VCM showed the superior capability in classifying failure modes. The experimental results showed that the proposed method could effectively improve the data processing speed and the accuracy of condition monitoring of rotating systems.
Comparing the parameter estimation methods of Weibull distribution with censored lifetime Mr. JIHYUN PARK, Juhyun LEE and Suneung AHN (Hanyang University)
Weibull distribution is widely used in reliability engineering and lifetime analysis because of its flexibility in modeling both increasing and decreasing failure rate. Weibull distribution has shape parameter and scale parameter, and it is difficult to estimate the parameters due to the no-closed form of likelihood function. In recent years, there has been studied on the approximating parameter estimation methods based on the simulation. In this study, we use the approximating parameter estimation of Weibull distribution with censored lifetime. The methods which are applied in numerical example are Bayesian estimation method, maximum likelihood estimation, and Markov chain Monte Carlo. Accuracy of estimation methods is performed by the mean square errors of parameter estimator in simulation reducing the experiment time. In addition, it can be helpful to set the design of experiment considering the characteristics of Weibull distribution with censored lifetime.