Proceedings of the Institution of Mechanical Engineers,Part O: Journal of Risk and Reliability

Instead of performing maintenance at fixed intervals, the operational efficiency of assets can be improved significantly by taking into account the variations in usage and operating environment of the asset. In that way, the traditional static maintenance policy is replaced by a dynamic maintenance policy. In the present work, this concept is demonstrated by modeling the failure behavior of a rather complex multi-component system, i.e. a navy frigate. For this system, several non-identical subsystems are included. Some of the systems are unique on board, while others have been made redundant, which means that the criticality of the subsystems to the mission capability of the frigate varies. Moreover, the variation in deployment of the frigate in terms of mission types and operating environments is translated into various usage profiles for the subsystems. Simulations are then performed to obtain the optimal maintenance policy in terms of interval length and preventive maintenance threshold, given certain requirements for the deployability of the frigate in a certain period. Moreover, the sensitivity of the results for different subsystem initial service life times and variations in usage profile are investigated. The results show a clear dependence of the optimal interval on the mission program of the frigate and the subsystem failure behavior.

This article addresses the relation between maintenance optimisation modelling and safety risk management. Most models for determining optimal maintenance strategies are based on a classical cost-benefit rationale, whereby all effects studied are transformed into monetary value and different concerns are balanced based on expected value calculations. If the failure of a system could have negative safety effects, there is a potential conflict between such an optimisation approach and the risk management principle saying that risk should be reduced to a level that is as low as reasonably practicable (ALARP); at least this is so if an interpretation of ALARP applies where verifying ALARP includes cost-benefit assessments but also extend beyond these. The ALARP principle is also typically implemented in combination with risk acceptance criteria, which could imply a restriction on the set of permissible maintenance policies and hence lead to suboptimal maintenance scheduling from an optimisation modelling point of view. This article considers the implementation of the ALARP principle, in combination with risk acceptance criteria, in the context of maintenance optimisation, addressing two different interpretations of the ALARP principle. A numerical example is included to show how one of these would affect the determination of an optimal policy. Two key points discussed are that risk reduction can be achieved in other ways than by adjusting the decision parameter of a given maintenance optimisation model, and that according to one of the interpretations the philosophy and practice of the ALARP principle extends beyond mechanical cost-benefit optimisation.

The geometry of railway track governs the quality of the ride for passengers and, should it deteriorate to an extreme state, can become a safety concern with potential derailment. The geometry is dependent upon a number of different factors, among which is the condition of the ballast. Several options exist to control the condition of the ballast, including manual intervention, tamping and stoneblowing. Ballast condition is monitored implicitly by measuring the geometry of the railway lines using a specially equipped measurement train. By analysing the data collected by this train, the deterioration process of the railway geometry can be understood. Using this understanding, mathematical models can then be constructed, which take account of the possible maintenance and renewal options to predict the track state. This model enables decisions to be made on the best or optimal strategy for maintenance and renewal of the ballast. This article describes a model of the track maintenance process for a railway network. Owing to their flexibility, the model is formulated using a Petri net and combines the deterioration, maintenance and inspection processes for a railway network containing a number of regions. There are a limited number of maintenance machines in the network and the maintenance in each region is organised independently. The model also takes account of the fact that the maintenance of track sections with severe levels of ballast deterioration must take priority over all other maintenance in the network and allows for opportunistic maintenance to be analysed.

We consider the problem of the evaluation of the maintenance policy of a component by means of degradation modeling. We assume that the stochastic laws governing the degradation process are uncertain, and so are the related parameters. We assume that the information available is in the form of qualitative judgment by an expert. We develop a representation framework based on possibility theory and the concept of fuzzy random variables. An example of application is given with reference to a medium-voltage circuit-breaker test facility.

Design to capacity is an engineering principle that is increasingly applied in chemical industry, among others owing to increasing plant sizes and associated investments. This principle aims to reduce over-capacity, over-sized buffers and excessive redundancy. Concurrently, a high level of availability is targeted over the entire production chain. The consequences of unavailability of highly interconnected chemical process plants can be significant because a technical disruption in one plant is able to spread over the entire production network. In chemical process plants not only technical equipment determines the availability, but also storage units, which are able to bridge times of planned or unplanned interruptions of production. To find a balance between the principle of design to capacity and high production availability, the influence of different design parameters, such as capacity of production units, redundancy concept and the size of storage units, have to be evaluated and integrated in the design process. In this article, we present an analytical method for availability evaluation based on extending Semi-Markov processes integrating storage units and multiple production lines. Semi-regenerative states are used to capture the characteristics of storage units, and an approach is proposed in this work to assign distributions for the remaining holding times in these states. The proposed modelling and analysis are demonstrated on two case studies.

Next generation telecommunication core networks are typically based on the Third Generation Partnership Project Internet protocol (IP) multimedia subsystem (IMS). Their planning and deployment must take into account the occurrence of random failures causing performance degradations, in order to assess and maintain a high level of quality of service. In particular, IMS signalling servers can be modelled as repairable multi-state elements where states correspond to different performance levels. This article provides an evaluation of IMS signalling network performance in long runs in terms of two metrics adopted in the practice, such as the number of call set-up sessions that the network can manage at the same time and the call set-up delay. A semi-Markov model has been adopted for the IMS servers, which allows as well for non-exponential probability distributions of sojourn times, as often observed in real networks. Furthermore, a redundancy optimization problem is solved in an IMS-based realistic scenario, to the aim of minimizing the deployment cost of a telecommunication network with a given availability requirement.

The dynamic flowgraph methodology is a promising way to find the prime implicants of a top event for a dynamic system possibly containing digital subsystems. This article demonstrates how to express dynamic flowgraph methodology models as logic programs, and top events as queries to those programs, in a natural and comprehensible way. Computation of the logic program lists the prime implicants of a top event in the system. We also present and implement an algorithm for computing the probability of the top event from its prime implicants. Together, computation of prime implicants and calculation of top event probability from these constitute a complete way of finding a system’s failure probability. Logic programs, implemented in this article in the leading logic programming language Prolog, enable rapid prototyping of dynamic flowgraph methodology models. The logic programming framework introduced here could also be utilized in teaching dynamic flowgraph methodology in risk analysis courses.

Two well-known modelling approaches are in use in probabilistic risk assessment: fault tree linking and event tree linking. The question of which modelling approach is most appropriate for specific applications has been extensively, if not emotionally, debated among experts in the past two decades, addressing both modelling and quantification issues. In this article, we determine their degree of equivalence and build ‘methodological bridges’ between the two approaches from a mathematical and algorithmic perspective. We show that, under certain conditions, both modelling approaches are equivalent. Since both fault tree linking and event tree linking approaches are subject to limitations and approximations, established bridges make it possible to formulate important recommendations for probabilistic risk assessment practitioners and quantification engine developers.

Sensors are being increasingly used to monitor the functional state of complex systems. Sensors are used to make observation of physical quantities. The measured quantities are expected to provide information about the state of the system, its subsystems, components, and internal and external physical parameters. A complex system normally requires many sensors to extract required information from the sensed environment. In most cases, the problem of optimal sensor placement is difficult, because it requires optimization under uncertainty. This research developed new algorithms for sensor placement optimization under uncertainty and utilized them in a new system reliability monitoring approach. The overall methodology is designed to answer important questions such as how to infer the reliability of a system based on a limited number of sensor information points at certain subsystems (upward propagation); how to infer the reliability of a subsystem based on knowledge of the reliability of a main system (downward propagation); how to infer the reliability of a subsystem based on knowledge of the reliability of other subsystems (distributed propagation); and what are the optimum locations of sensors to provide the best estimate of system reliability.

This article presents a study of key foundational issues in relation to two methodologies developed and used for assessing national risk events in Norway and the Netherlands. Both these methodologies are, to a large extent, probability based, but there are considerable differences in the approaches adopted. The main aim of the article is to point to these differences and relate the methodologies to existing theories for risk conceptualisation and description. Neither of the two methodologies are able to create a strong conceptual platform for the risk assessment, largely because they fail to clearly define and relate the core concepts of risk, probability and uncertainty. The treatment of uncertainty is another problematic issue, and we argue that both methodologies, but especially the Norwegian one, suffer from a severe weakness: key aspects of uncertainty are not adequately reflected. Finally, we question the way the Dutch methodology has integrated value aspects into the risk assessment.