2009 ARS, North America

Showcase of Current University Research

On Friday, June 12th, distinguished university faculty from across the country will present significant aspects of their research programs. This portion of the Symposium program has been designed to allow university researchers and Symposium participants to explore mutually beneficial avenues of research that are of interest and relevant to today’s practitioner. Presentations will be delivered from 8:00 a.m. to 3:15 p.m. and the program concludes with a round table discussion at 3:30 p.m.

The showcase of current university research is organized by the ReliaSoft Risk, Reliability, and Maintainability Research Alliance, which has been established through the Department of Industrial Engineering at the University of Arkansas and receives support and cooperation from ReliaSoft Corporation. The goal of this research alliance is to foster university and industry partnerships for meeting research challenges and needs identified by practicing engineers.

Session U-1
8:00 to 9:00 a.m. Friday June 12, 2009

Adaptive Prognostics: Overview and Recent Developments
Nagi Gebraeel
H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology

This presentation is divided into three parts. The first part provides an overview of the adaptive prognostic methodology presented in the 2008 ARS. This adaptive degradation-based prognostic technique utilizes sensor-based condition/health monitoring information to compute and continuously update, in real-time, residual lifetime probability distributions of partially degraded systems and their components. In contrast to conventional reliability techniques where the mean time to failure (MTTF) is a fixed (or time-based) value, this methodology leads to the evaluation of dynamically evolving MTTFs that are based on the latest degradation state of the system.

The second part of this talk will present recent developments targeted towards improving the adaptive prognostic methodology. Some of these developments focus on: 1) increasing the accuracy of predicting the remaining useful life for signals with low signal-to-noise ratio (i.e. noisy sensor signals), 2) addressing challenges related to the scarcity of degradation data, and 3) solving scalability aspects for implementing these novel prognostic models in large scale applications such as fleet maintenance management. Finally, the presentation will be concluded by a demonstration of the facilities available in the Prognostics Systems Laboratory at Georgia Tech. These facilities provide various platforms for testing and validating our methodological developments using real-world data.

Session U-2
9:15 to 10:15 a.m. Friday June 12, 2009

Reliability Growth Planning for Fast Time-to-Market Products
Tongdan Jin
Department of Engineering, Mathematics, and Physical Sciences, Texas A&M International University

When a new system design is driven by the need for faster time-to-market, reliability growth planning can be utilized to achieve the reliability goal after the shipment. For those systems, extended in-house reliability testing becomes difficult due to the compressed design cycle. This presentation will introduce an empirical method to predict and plan reliability growth considering both surfaced and latent failures that would emerge after the shipment. A surfaced failure mode is referred to as the failure with known failure mechanisms, while a latent failure mode is the one that did not occur until the system had operated in the field for a certain period of time. Analytical models will be used to forecast the occurrence rate of latent failures.

Based on the proposed method, the product manufacturer can proactively implement corrective actions against critical failure modes before they escalate to a high level. A discussion between the fix effectiveness of a failure mode and corrective action cost is also provided. The method can be applied to many electronic system designs where a tight design schedule is required in order to gain the market window.

Session U-3
10:30 to 11:30 a.m. Friday June 12, 2009

Assessing the Potential Impact of Prognostics for Improving System Performance
C. Richard Cassady
Department of Industrial Engineering, University of Arkansas

The use of prognostics for the purpose of condition-based maintenance has recently received an increased amount of interest from many industries. These methods use data derived from sensors applied to a system to make some assessment of system health that leads to an estimate of remaining life. Based on this remaining life estimate, a preventive maintenance action may be taken. Ideally, such an action will take place instantaneously before failure (just-in-time) so that failure is avoided and no system uptime is lost unnecessarily. The primary challenge associated with prognostics is developing a system health assessment methodology that is both economically feasible and statistically valid as a means of predicting the remaining system life.

The focus of this presentation is on the second aspect of this challenge – statistical errors. The objective of this research is to demonstrate a method for assessing the impact of prognostics on system performance when the prognostic methods are subject to statistical errors. To achieve this objective, a discrete-event simulation model is used to assess the performance of a system under three maintenance policies: 1) run-to-failure maintenance, 2) scheduled preventive maintenance, and 3) condition-based maintenance (prognostics). Various levels of prognostic error are modeled, including the ideal case in which prognostics are perfect. The results of this experimentation are used to address three questions: 1) How much can perfect prognostics improve system performance beyond scheduled preventive maintenance? 2) How bad do prognostics have to be to make things worse than scheduled preventive maintenance? 3) How bad do prognostics have to be to make things worse than run-to-failure maintenance?

Session U-4
1:00 to 2:00 p.m. Friday June 12, 2009

Optimizing Spare Parts Inventory Using Genetic Algorithms
Lance Luttrell
Graduate Student in the Department of Industrial Engineering, University of Arkansas

Spare parts play a crucial role in maintenance and reliability of systems ranging from machine failures to supply chain systems. The demand for these parts is often sporadic and difficult to predict. Analytic methods can be used in systems that reflect many basic assumptions including Poisson demand and restrictions such as no lateral supply between same-level echelons. These assumptions are relaxed in simulation approaches which allows for more accurate modeling of real world complexity. The presentation will discuss the uses of genetic algorithms in optimizing the inventory levels for spare parts in such systems.

Session U-5
2:15 to 3:15 p.m. Friday June 12, 2009

Condition-Based Maintenance Models for Civil Infrastructure Subject to Cracking
Thomas Yeung
L’Ecole de Mine, Nantes, France

Cracking in civil infrastructure presents a serious threat to its reliability and integrity. To date, relatively few mathematical models exist for optimizing inspection and maintenance decisions of structures subject to cracking. We develop degradation and condition-based maintenance optimization models for structures subject to two different kinds of cracking processes: those with cracks that may be repaired individually and those that cannot. In the first case, we utilize the combination of a Poisson and gamma process to account for the tremendous amount of uncertainty and difficulty in predicting the proliferation of cracks. Based on this degradation model we present two condition-based maintenance optimization models. The first is a continuous-time approach based on Markov semi-renewal theory and the second utilizes discrete stochastic dynamic programming, i.e. Markov decision process (MDP). In the second class of cracks, we define the state of the system based on a percentage of total degradation (cracking) and the rate of change of the degradation. In addition to doing nothing and a complete renewal action, we consider several imperfect maintenance actions corresponding to different thicknesses in resurfacing of the structure. Prior imperfect maintenance models do not consider the scenario where imperfect maintenance not only restores the system to a state less than "as good as new" but also under a new deterioration law. In this case, we also consider that the rate of degradation after imperfect maintenance will be altered based on the action performed and the state of the system prior to maintenance. Here, we also utilize an MDP framework to determine the optimal action for each state of the structure. We utilize cracking in pavement of roads and highways as a motivating example in this presentation; however, this research has application to all kinds of civil infrastructure (e.g. buildings, bridges, etc.).

Round Table Discussion
3:30 to 5:00 p.m. Friday June 12, 2009

The program will conclude with an interactive round table discussion designed to foster open communication among the participating university researchers and Symposium attendees. The focus of the discussion will be to share information about the ways in which current university research can be applied to meet real-world challenges and to target specific areas in which additional solutions are needed. The discussion will also address the specific role that the ReliaSoft Risk, Reliability, and Maintainability Research Alliance can play in focusing the expertise available in academic reliability programs on specific corporate initiatives.