Continuous-time predictive-maintenance scheduling for a deteriorating system

Abstract

A predictive-maintenance structure for a gradually deteriorating single-unit system (continuous time/continuous state) is presented in this paper. The proposed decision model enables optimal inspection and replacement decision in order to balance the cost engaged by failure and unavailability on an infinite horizon. Two maintenance decision variables are considered: the preventive replacement threshold and the inspection schedule based on the system state. In order to assess the performance of the proposed maintenance structure, a mathematical model for the maintained system cost is developed using regenerative and semi-regenerative processes theory. Numerical experiments show that the-expected maintenance cost rate on an infinite horizon can be minimized by a joint optimization of the replacement threshold and the a periodic inspection times. The proposed maintenance structure performs better than classical preventive maintenance policies which can be treated as particular cases. Using the proposed maintenance structure, a well-adapted strategy can automatically be selected for the maintenance decision-maker depending on the characteristics of the wear process and on the different unit costs. Even limit cases can be reached: for example, in the case of expensive inspection and costly preventive replacement, the optimal policy becomes close to a systematic periodic replacement policy. Most of the classical maintenance strategies (periodic inspection/replacement policy, systematic periodic replacement, corrective policy) can be emulated by adopting some specific inspection scheduling rules and replacement thresholds. In a more general way, the proposed maintenance structure shows its adaptability to different possible characteristics of the maintained single-unit system.

Evaluating maintenance policies by quantitative modeling and analysis

“…CM policy does not require any information, algorithm and instrumentation; on the contrary, a database of the components' failure times is needed for the application of the PM policy, which also does not involve the adoption of any instrument; the CBM approach is founded on component monitoring and requires more or less refined software and hardware tools, depending on the particular case study; PrM relies on prognostics, which may be performed by means of more or less refined algorithms and technologies, whose development/employment necessarily requires a certain knowledge of the degradation process. In this respect, the task of deriving a model of the degradation process is usually more complicated than just statistically describing the binary transition from a good state to a failed state (Grall et al 2002). Thus, when setting a maintenance strategy the first step is the identification of those policies which are really applicable, given the information and resources available.…”

“…PrM can be regarded as an improvement of CBM: the knowledge of the current degradation state of the component (diagnostics) is complemented with a prediction of its future behavior and thus of its Remaining Useful Life (RUL) (e.g., Fan et al 2011, Grall et al 2002, Lu et al 2007, You et al 2010. The accurate estimation of the RUL provides time to opportunely plan and prepare the repair or the replacement of the component, e.g., by delaying the maintenance to the next planned plant outage, by provisioning with spare parts only at time of necessity, by optimizing staff utilization, while remaining acceptably confident that the system will not fail before maintenance.…”

Nonperiodic Inspections to Guarantee a Prescribed Level of Reliability

“…The proposed model considers a nonperiodic inspection policy, the reason for this being that it is a more general approach and often results in policies with lower costs, particularly in cases where high costs of lost production are taken into consideration. Rather than considering a dynamic programming problem as did Newby and Dagg (2004), the optimization problem is simplified by using an inspection scheduling function m as introduced in Grall et al (2002). The scheduling function is a decreasing function of d (R τ i ), the amount by which the threshold is decreased, and determines the amount of time until the next inspection time…”

Auto-steered Information-Decision Processes for Electric System Asset Management

“…A failuretime cdf F(t) is induced by a specified model for c(t,e;β), the environment e, and a definition of failure (usually a specified value c f , beyond which failure is said to have occurred). Stochastic behavior in c(t,e;β) can be captured either by using a stochastic process model (e.g., [5]) or by driving a deterministic model with a stochastic environmental model (e.g., [6]). As new condition information is received for a given unit, it is possible to update the failure probability for that unit.…”

“…Figure 1 illustrates the structure of the problem and facilitates description of how we intend to approach its solution. We overview intended implementation of the 5 Layer 1, The power system: The prototype will center on a continuously running model of the Iowa power system using network data provided by local utility companies using a commercial-grade (Areva) simulator. Layer 2, Condition sensors: As indicated by the taller "Condition History" cylinder at the far right of layer, 3 campus substations will be equipped with sensors, communication equipment, and servers to provide a benchmark prototype for hardware implementation.…”