Fatih Camci

System Maintenance Scheduling With Prognostics Information Using Genetic Algorithm

Condition based maintenance (CBM) aims to balance two extreme sides (i.e., corrective maintenance (CM), and preventive maintenance (PM)) by observing and forecasting the real time health of machines. Recent developments in CBM revealed promising technologies for advanced fault detection, and forecasting. Traditional maintenance scheduling in CBM is based on the threshold setting on forecasted failure probability, or remaining useful life (RUL) for individual components. However, this approach may not give the best result for the system, because individual components are inter-related, and mutually dependent. It is not uncommon in systems that turning off a machine due to failure or maintenance causes other machinery or components to be turned off. Designing a comprehensive tool that optimizes availability & cost of the whole system incorporating prognostics information is crucial to fully benefit from CBM. The goal of this paper is to emphasize this need by demonstrating scenarios in CM, PM, and CBM; and to present a solution that optimizes system availability, and cost with system-maintenance constraints using genetic algorithms. The proposed tool acquires the forecasted failure probability of individual components from the prognostics module, and their reliability expectations after maintenance. The tradeoff between maintenance & failure is quantified in risk as the objective function to be minimized. The risk is minimized utilizing genetic algorithms for the whole system rather than individual components. The results of the proposed tool are compared with PM, CM, and CBM in which prognostics information of components are analyzed individually.

A Simple State-Based Prognostic Model for Railway Turnout Systems

The importance of railway transportation has been increasing in the world. Considering the current and future estimates of high cargo and passenger transportation volume in railways, prevention or reduction of delays due to any failure is becoming ever more crucial. Railway turnout systems are one of the most critical pieces of equipment in railway infrastructure. When incipient failures occur, they mostly progress slowly from the fault-free to the failure state. Although studies focusing on the identification of possible failures in railway turnout systems exist in literature, neither the detection nor forecasting of failure progression has been reported. This paper presents a simple state-based prognostic (SSBP) method that aims to detect and forecast failure progression in electromechanical systems. The method is compared with Hidden-Markov-Model-based methods on real data collected from a railway turnout system. Obtaining statistically sufficient failure progression samples is difficult, considering that the natural progression of failures in electromechanical systems may take years. In addition, validating the classification model is difficult when the degradation is not observable. Data collection and model validation strategies for failure progression are also presented.

Feature Evaluation for Effective Bearing Prognostics

Rolling element bearing failure is one of the foremost causes of breakdown in rotating machinery. It is not uncommon to replace a defected/used bearing with a new one that has shorter remaining useful life than the defected one. Thus, the prognostics of bearing plays critical role for increased availability and reduced cost. Effective prognostics highly depends on the quality of the extracted features. Diagnostics is basically a classification problem, whereas prognostics is the process of forecasting the future health states. The quality of the features for classification has been studied thoroughly. However, the evaluation of the quality of features for prognostics is a relatively new problem. This article presents an evaluation method for the goodness of the features for prognostics and presents results on bearings run until failure in a laboratory environment. Copyright

Robust kernel distance multivariate control chart using support vector principles

It is important to monitor manufacturing processes in order to improve product quality and reduce production cost. Statistical Process Control (SPC) is the most commonly used method for process monitoring, in particular making distinctions between variations attributed to normal process variability to those caused by ‘special causes’. Most SPC and multivariate SPC (MSPC) methods are parametric in that they make assumptions about the distributional properties and autocorrelation structure of in-control process parameters, and, if satisfied, are effective in managing false alarms/-positives and false-negatives. However, when processes do not satisfy these assumptions, the effectiveness of SPC methods is compromised. Several non-parametric control charts based on sequential ranks of data depth measures have been proposed in the literature, but their development and implementation have been rather slow in industrial process control. Several non-parametric control charts based on machine learning principles have also been proposed in the literature to overcome some of these limitations. However, unlike conventional SPC methods, these non-parametric methods require event data from each out-of-control process state for effective model building. The paper presents a new non-parametric multivariate control chart based on kernel distance that overcomes these limitations by employing the notion of one-class classification based on support vector principles. The chart is non-parametric in that it makes no assumptions regarding the data probability density and only requires ‘normal’ or in-control data for effective representation of an in-control process. It does, however, make an explicit provision to incorporate any available data from out-of-control process states. Experimental evaluation on a variety of benchmarking datasets suggests that the proposed chart is effective for process monitoring.

Failure diagnostics for railway point machines using expert systems

Maintenance is an inevitable reality in industry. Maintenance of a system usually involves maintenance of multiple components with multiple failure modes, each of which may require different maintenance policy (i.e., corrective (CM), preventive (PM), or condition based maintenance (CBM)). A maintenance policy may be best for one component and the worst for the other (CM may be best for a very cheap and non-critical component and the worst for a critical one). This paper presents an economical analysis method that identifies the best maintenance policy for a failure mode and/or component of a system.

Comparison of genetic and binary particle swarm optimization algorithms on system maintenance scheduling using prognostics information

Recent technical advances in condition-based maintenance technology have made it possible to not only diagnose existing failures, but also forecast future failures, which is called prognostics. A common method of maintenance scheduling in condition-based maintenance is to apply thresholds to prognostics information, which is not appropriate for systems consisting of multiple serially connected machinery. Maintenance scheduling is defined as a binary optimization problem and has been solved with a genetic algorithm. In this article, various binary particle swarm optimization methods are analysed and compared with each other and a genetic algorithm on a maintenance-scheduling problem for condition-based maintenance systems using prognostics information. The trade-off between maintenance and failure is quantified as the risk to be minimized. The forecasted failure probability of serially connected machinery is utilized in the analysis of the whole system. In addition to the comparison of a genetic algorithm and binary particle swarm optimization methods, a new binary particle swarm optimization that combines the good sides of two binary particle swarm optimizations is presented.

Failure prediction on railway turnouts using time delay neural networks

Turnout systems on railways play critical role on reliability of railway infrastructure. Identification and prediction of failures on mechanical systems have been attracting researchers and industry in recent years. Condition based maintenance focuses on failure identification and prediction using sensory information collected real-time from sensors embedded on electro-mechanical systems. This paper presents neural network based failure prediction algorithm on railway turnouts.

State-Based Prognostics with State Duration Information

Failure prediction (i.e. prognostics) is critical for effective maintenance because it greatly impacts the competitiveness of organizations through its direct connection with operating and support costs, system availability, and operational safety. In recent years, research has focused on state-based prognostics that forecast future progression by first identifying the current state. The duration spent in a state is a factor that influences the expected time to be spent in that state in the future; however, previous works on state-based prognostics have ignored the effect of duration. Hidden Markov Models are the most famous state-based prognostics methods in the literature with practicality problems such as computational complexity, requirement of excessive data, and dependency on initialization. This paper presents a new, simple and easy to implement state-based prognostic method using state duration information. The presented method is applied to two real systems (railway turnout systems and drill bits), and the results are compared with the existing methods presented in the literature. The results show that the presented method outperforms the existing methods. Copyright

Rethinking Future of Utilities: Supplying All Services through One Sustainable Utility Infrastructure

Sustainable Utility Infrastructure Fatih Camci,†,* Bogumil Ulanicki,‡ Joby Boxall, Ruzanna Chitchyan, Liz Varga, and Ferhat Karaca† †IVHM Centre, Cranfield University, Bedford, U.K. ‡Department of Engineering, De Montfort University, U.K. Department of Civil and Structural Engineering, University of Sheffield, U.K. Department of Computer Science, University of Leicester, U.K. Complex Systems Research Centre, Cranfield School of Management, U.K.

CHANGE POINT DETECTION IN TIME SERIES DATA USING SUPPORT VECTORS

Change Point Detection in time series data is of interest in various research areas including data mining, pattern recognition, statistics, etc. Even though there are several effective methods in the literature for detecting changes in mean, and an increase in variance, there are none for decrease in variance. Effective detection of decreased variance has been reported as future work in earlier papers. In addition, most, if not all, methods require some model like AR to fit into the time series data in order to extract noise information, which is assumed to be independent and identically distributed (i.i.d.) and follow standard normal distribution (white noise). Thus, effectiveness of the methods is tied to the fitness degree of the AR model to the time series data. This paper presents a change point detection method based on support vectors that targets changes in mean and variance (including variance decrease) without any assumption of model fitting or data distribution. The data is represented by a hyper-sphere in a higher dimensional space using kernel trick. The change is identified by the change in the radius of the hyper-sphere. A comparison of this method with other methods is presented in the paper.