Proceedings of the 12th IMA International Conference on Modelling in Industrial Maintenance and Reliability

Lucía Bautista, Christophe Bérenguer,  I.T. Castro, Laurent Doyen, Olivier Gaudoin

https://doi.org/10.19124/ima.2023.01.1

Abstract

Railway tracks degrade and may eventually break down due to several operational and environmental impacts that affect the rails’ reliability. The most  common type of railhead defect is called the rail squat, which cost Network Rail an estimate of approximately 3.9 million pounds annually. Squat defects are minor subsurface laminations that run diagonally down the running surface and spread laterally and longitudinally over and along the rail tracks (Li, Zili et al., 2008a; Li, Zili et al., 2008b; Li, Z., 2009). The occurrence of squat defects has a significant impact on the track performance, leading to speed restrictions, delays, and cancellation of in-service train operations and hence penalties for infrastructure owners. To ensure the performance and efficiency of service operations and more reliable railway infrastructure, the UK is investing in railway modernization projects to meet this demand (Rail director April 2023, 2023; MyBib Contributors, 2019).

This study uses a Dynamic Multiple linear regression to model the relationship between squat defects and influential parameters such as track length, maximum permissible speed, maximum axle load, estimated million gross tonnage, tamping frequency, rail grinding frequency, and corrugation frequency detection. A hazard model is proposed and used to predict transition probabilities between defect and failure.

Abstract

Degradation modeling is a paramount issue in failure time prediction or maintenance planning of complex systems. The stochastic-process-based models show great flexibility in describing the failure mechanisms caused by degradation [1]. The aim of degradation modeling is to select the best model from a set of models, that captures the features of the underlying degradation process. However, collected data and stochastic processes used for the modeling do not always satisfy the classical properties required to apply goodness-of-fit tests. Most of these tests require large-size data under the independent and identically distributed hypothesis which are usually not valid for reliability and degradation data.

Depth Function has been introduced to extend the notion of the median to multi-variate random variables as well as functional data, refer to [2] and [3]. More precisely, functional depth measures the position of a path in the envelope of all available paths. Lopez-Pintado and Qian [4] proposed a depth-based global envelope test which is a statistical test that rejects the null hypothesis if the observed path X is not completely inside the envelope.

Authors NG.D. Santos, B.M. Alkali, A.J. Kumar, O. Niculita

https://doi.org/10.19124/ima.2023.01.3

Abstract

Railway tracks degrade and may eventually break down due to several operational and environmental impacts that affect the rails’ reliability. The most common type of railhead defect is called the rail squat, which cost Network Rail an estimate of approximately 3.9 million pounds annually. Squat defects are minor subsurface laminations that run diagonally down the running surface and spread laterally and longitudinally over and along the rail tracks (Li, Zili et al., 2008a; Li, Zili et al., 2008b; Li, Z., 2009). The occurrence of squat defects has a significant impact on the track performance, leading to speed restrictions, delays, and cancellation of in-service train operations and hence penalties for infrastructure owners. To ensure the performance and efficiency of service operations and more reliable railway infrastructure, the UK is investing in railway modernization projects to meet this demand (Rail director April 2023, 2023; MyBib Contributors, 2019).

This study uses a Dynamic Multiple linear regression to model the relationship between squat defects and influential parameters such as track length, maximum permissible speed, maximum axle load, estimated million gross tonnage, tamping frequency, rail grinding frequency, and corrugation frequency detection. A hazard model is proposed and used to predict transition probabilities between defect and failure. This research evaluated six years of data acquired from network rail on squats defects,
grinding maintenance, and corrugation faults across the UK railway network. The dynamic Markov model is used within the scope of the hazard model to determine the transition probabilities of squat defects propagation. A complete squat data analysis is performed by comparing the efficiency of the rail gridding maintenance regime to the cumulative squat defect frequency against the number of repair operations. An example simulation is shown to anticipate the time to breakdown of railway track systems.

Authors Pedro Dias Longhitano, Christophe Bérenguer, Benjamin Echard

https://doi.org/10.19124/ima.2023.01.4

Abstract

The problem of assigning missions to a fleet of electric vehicles has been thoroughly studied in the literature under the name of Electric Vehicle Routing Problem (EVRP), however most versions of such a problem do not take battery degradation into account. One of the main difficulties of battery degradation aware routing is the necessity of including State of Health (SoH) models that are designed for different operation conditions and the necessity of including a realistic State of Charge (SoC) model. In this paper we present a version of the EVRP that includes battery degradation as well as the inclusion of decision variables such as speed and acceleration limitations. We illustrate our approach through numerical experiments.

Authors J. Corker, R. Remenyte-Prescott, M. Eskandari Torbaghan, J. Ninic

https://doi.org/10.19124/ima.2023.01.5

Abstract

The U.K. Road network is a complex and dynamic system, managed by a federated network of organizations, often with stringent constraints on resourcing. Remediating deterioration of the network is both costly and politically sensitive; resilience, sustainability, and utilisation impacts must be constantly balanced by asset managing organizations. An approach for localized prediction of deterioration is developed in this paper, combining remote sensing and automated survey data into a GIS-based model to support decision making. In addition, predictive models are proposed in this paper, based on random forest regression and classification approaches, which use the proposed data model to create localized deterioration profiles. The machine learning model has been trained with and validated against data from a 21km length of the UK Major Roads network, using ~10 years of condition data and authoritative traffic, cartographic, and environmental data from several UK government agencies. Both the regression model and the classifier can accurately predict condition metrics, described in industry standards. This approach allows for early detection and mitigation of pavement failure and support maintenance operations on the network whilst minimizing disruption and maximizing return on investment.

Author D. Silkworth

https://doi.org/10.19124/ima.2023.01.6

Abstract

The identification of latency is perhaps one of the most valuable contributions to safety and reliability engineering. Latent component events are characterized by hidden failures (or faults) which represent conditions under which additional failures will combine to propagate system failure, or undesired events. Similar to exposure time with the non-repairable model, latent events have exposure over an interval between inspections accompanied by any needed repair. The longer the interval, the higher the probability that a latent component will be in a failed state. Scheduling and even designing a system to permit inspection, proof or validation tests becomes key to risk reduction. In this presentation a small system exposing several opportunities for inspection of latent events will be modeled. By utilizing the R environment, a series of inspection intervals for each procedure is built in a matrix so that the fault tree can be constructed multiple times in a programed loop accessing alternate protocols for inspection in each pass. A resulting table identifies the likely improvement in failure intensity for the undesired event depending on each potential inspection regime.

Authors Wenxu Li, Tieling Zhang, Richard Dwight

https://doi.org/10.19124/ima.2023.01.7

Abstract

This work presents the implementation details of a simulation-based model proposed in an earlier work by the authors where no implementation was covered. This model aims to support the railway industry in their decision-making on selecting the most cost-effective track form for a particular application by comparing life cycle costs of the available options. It is based on degradation conditions or life time distributions of track components and pre-set thresholds for initiating various maintenance actions on these track components. Major maintenance actions on major track components are considered for both ballasted track and ballastless track. Data obtained under various application contexts are sourced from the literature and the industry, complemented by manufactured data when no information is found. Data from miscellaneous sources are considered not an issue for the purpose of this study which is to demonstrate the feasibility of the proposed approach. By calculating the number of required maintenance actions first and the maintenance costs for ballasted and ballastless tracks afterwards, the work proves that the proposed model has achieved its intended function. Trials of the model on actual data from local rail industry are still underway. With these trails, performance of the model to support actual track form selection decisionmaking for a specific application will be further tested.

Author D. Silkworth

https://doi.org/10.19124/ima.2023.01.8

Abstract

A large data set, admittedly synthesized, was provided for demonstration of analysis techniques to be used to satisfy management requests for warrantee analysis. Discussion will be given to the specific challenges that can occur with such big data and particular management demands. Modeling is performed in the open source R environment utilizing the WeibullR package.

Authors Yaxin Shen, Mitra Fouladirad, Antoine Grall

https://doi.org/10.19124/ima.2023.01.9

Abstract

The aim of this paper is to discuss the problem of modelling and optimizing condition-based maintenance policy for a deteriorated photovoltaic (PV) panels impacted by environmental conditions. In this research, the effect of dust and temperature on the performance of PV panels is investigated by stochastic process modelling. Dust is considered as the most important cause of PV degradation, while temperature as an influence fluctuation factor. Both preventive and corrective actions are considered according to information of system state inspections. Inspection actions can be periodic or non-periodic. Maintenance policies are optimized to determine the preventive threshold and cleaning frequency. The performance of the maintenance policies is studied through numerical implementations. The optimal maintenance policy leading to the minimal cost criterion is obtained

Authors A. Raza, V. Ulansky

https://doi.org/10.19124/ima.2023.01.10

Abstract

Condition-based maintenance (CBM) refers to proactive maintenance carried out based on the system’s condition. In this study, we developed mathematical equations to calculate maintenance effectiveness indicators, such as achieved availability, inherent availability, mission availability, average maintenance costs per unit of time, and operational probability of failure-free operation. We considered a system with a single-component structure, where the type of imperfect inspection corresponds to CBM, the perfect repair is employed, and only wear-out failures occur in the system. Mathematical models have been derived for cases involving multiple imperfect inspections and an arbitrary time-to-failure law. We determined the optimal number of inspections by the criterion of maximum achieved availability for a specific stochastic degradation process, considering both condition-based and corrective maintenance. Our research demonstrates that CBM significantly improves achieved availability and reduces the required inspections

Authors Wen Wu, Sergio Cantero-Chinchilla, Rasa Remenyte-Prescott, Darren Prescott, Ali Saleh, Manuel Chiachio Ruano, Dimitrios Chronopoulos

https://doi.org/10.19124/ima.2023.01.11

Abstract

Structural health monitoring (SHM) systems involve implementing damage identification strategies to determine health states of structures. However, it is important to pay close attention to the degradation of the SHM itself, especially the effect of sensor degradation on the reliability of the SHM. This paper aims to formulate a general framework for evaluating SHM reliability that takes sensor failures into account. The framework involves modelling the degradation process of sensor network using Petri nets (PNs) and calculating the expected information gain/loss of the sensor network based on Bayesian inverse approach. The PNs allow for taking account of the sensor location and the number of sensor failures. A Bayesian inverse procedure is used to calculate the expected information loss due to sensor failures, which integrates a damage localization and a damage identification scheme to update information about model parameters. The proposed framework is demonstrated for a plate with a part-thickness hole, monitored by an ultrasonic guided wave monitoring system. The given framework could be adapted to other monitoring systems. The proposed model is able to predict the health condition state of the SHM, which can be included in asset management models for various industries.

Authors S. Tolo, J. Andrews

https://doi.org/10.19124/ima.2023.01.12

Abstract

The Dynamic and Dependent Tree Theory (D2T2) was developed in order to tackle the limitations of traditional Fault Tree approaches, providing tools for modelling realistic system features, such as complex maintenance strategies and component dependencies. However, when these embrace large sets of individual components (or basic events in the fault tree structure), the design of elaborate simulation models may be required, putting strain on the analysist.

This study proposes a generalization of the D2T2 methodology, aimed at simplifying the dependency modelling of multiple components. The approach consists of computing the sub-trees involved in the dependency and using the analysis results as input of a Petri net model capturing the dynamic of their relationships. This is in turn estimated and the results fed back into the fault tree framework, resulting in a nested Petri net-fault tree framework.

The solution proposed is described and demonstrated through its application to a simple case study involving a safety critical subsystem

Authors S Lunt, J Andrews

https://doi.org/10.19124/ima.2023.01.13

Abstract

Traditional Fault Tree Analysis (FTA), known as Kinetic Tree Theory (KTT), was derived by Vesely [1] in the 1970s to model and analyse engineering systems. The tree structure provides a clear visual representation of the causes of system failure in terms of component and software failures and human errors. FTA has 2 stages, qualitative analysis which involves identifying the necessary and sufficient causes of system failure, i.e., the minimal cut sets and quantitative analysis which involves calculating the system unavailability, system failure frequency and measures of importance.

Author S Reed

https://doi.org/10.19124/ima.2023.01.14

Abstract

Reliability importance measures, such as risk achievement worth and criticality, are used to quantify and rank components with respect to the influence they have on system reliability. However, there are applications where measuring and ranking the influence on system reliability that parts of a system within regions of the physical space it occupies, instead of named components, may be more useful. Examples include measuring the importance of fire zones in an offshore oil platform and “behind armour debris” zones from projectile impacts on an armoured fighting vehicle. Spatial reliability importance measures are introduced as ways of measuring the influence on system reliability from this perspective. They can be used to support decisions such as optimising the physical layout of system components, focusing improvements on regions with greatest reliability effect, and prioritising regions for fault diagnostics in failed systems. The spatial importance measures are demonstrated by analysing the importance of regions around different points in a 2-terminal grid network.

Authors Soufian Echabarria, Phuc Doa, Hai-Canh Vun, Bastien Bornand

https://doi.org/10.19124/ima.2023.01.15

Abstract

The proton exchange membrane fuel cell (PEMFC) is a critical and essential component of a zeroemission electro-hydrogen generator. An accurate prediction of its performance is important for optimal operation management and preventive maintenance of the system. However, the prediction is not easy because the PEMFCs have complex electrochemical reactions with multiple nonlinear relations between operating variables as inputs and voltage as output. In this paper, we propose an efficient prediction approach based on XGBRegressor and Tree-structured Parzen Estimator. In addition, to better select relevant features, Kernel Principal Component Analysis and Mutual Information are jointly used. The proposed approach allows consideration of the dynamic operating conditions of the PEMFC. To test and validate the robustness of the proposed approach, a data-set of ten PEMFCs was used. Furthermore, a comparison study with traditional machine learning models, such as artificial neural networks and support vector machine regression, is investigated. It was shown that the proposed approach provides better results.

Authors Sumit Sood, Phuc Do, Nicolae Brînzei, Benoît Iung, Jean-François Pétin

https://doi.org/10.19124/ima.2023.01.16

Abstract

The objective of this paper is to study the effectiveness of an opportunistic condition-based maintenance strategy for the sea water pumps in a nuclear power plant. For the presented study a primary seawater system consisting of four sea water pumps, supplied by three water intake chambers, have been considered. There exists functional dependency and state interaction between these pumps due to the operation policy. This results in the degradation of the pumps at different rates. In order to ensure the availability of the pumps for the effective operation of the plant, an optimal maintenance strategy is required. To that end, both functional dependency between the pumps and state interactions between the components of the pumps are firstly modelled. An opportunistic condition-based maintenance strategy considering these dependencies is then proposed. The proposed opportunistic maintenance policy allows consideration of the quality of the maintenance actions. A cost model is then developed to evaluate the performance of the proposed opportunistic maintenance policy. Long term simulation results show the effectiveness of the proposed model and opportunistic maintenance policy.

Authors E. T. Bacalhau, L. Casacio, F. Barbosa, Felipe Yamada, L. Guimarães

https://doi.org/10.19124/ima.2023.01.17

Abstract

The development of new technologies and the improvement of systems made solar photovoltaic (PV) energy generation grows exponentially in the last years. Research into failure and degradation mechanisms has become necessary for obtaining efficient and reliable systems. However, unsolved challenges remain concerning safety, unforeseen outages, and high operation and maintenance (O&M) costs. Early detection of problems is essential to provide reliability and avoid production losses over time. Performing preventive maintenance can anticipate faults and limit unplanned downtime as it is based on history and probability of failure. This work aims to increase PV plants’ operational performance by improving the methodologies for O&M in PV systems. A Markov Decision Problem model is adapted to a Reinforcement Learning approach to recommend preventive maintenance actions in PV systems considering equipment degradation, such as cables and switches in the inverter. The methodology allows to explore the economic and energy production benefits of the detection, prevention, and mitigation techniques applied to PV power production. Case studies consider large-scale scenarios and show that the approach can be applied to create a long-term horizon planning maintenance policy.

Authors Trung Thanh N. Thai, Viet Cuong Pham, Phuc Do

https://doi.org/10.19124/ima.2023.01.18

Abstract

The effect of control action is crucial to system reliability. However, it has been not significantly investigated in the literature, particularly for balanced systems which require strict control to maintain balance restriction. The purpose of this paper is to present a reliability assessment and adaptive control for a balanced system considering actuators degradation. The degradation state of an actuator process affects vice versa to the control actions by forcing them to increase generating stress on the actuator in order to maintain the desired setpoint. In that way, first we propose a novel degradation model for actuators considering the impact of control actions. An adaptive control allowing considering the degradation state of actuators is then developed. Illustrated with the use case of quadcopter, the reliability of the system has been analyzed and effects of proposed control policy have been demonstrated with simulation.

Authors Taofeeq Alabi Badmus, Darren Prescott, Rasa Remenyte-Prescott

https://doi.org/10.19124/ima.2023.01.19

Abstract

The existing Generalised Stochastic Petri Net and modified Bayesian Stochastic Petri Net (GSPN-mBSPN) methodology has demonstrated improved modelling capabilities for fault diagnosis in dynamic systems with feedback control loops. However, the GSPN-mBSPN approach uses predefined input conditional probability tables (iCPTs) for fault diagnosis, limiting its usage in dynamic fault diagnosis due to the time required to define and populate the iCPT entries accurately. This paper presents an algorithm to automatically generate iCPT tables, enhancing the modelling capability of the GSPN-mBSPN approach for fault diagnosis of dynamic systems under time-varying conditions. The GSPN module in a GSPN-mBSPN model of a dynamic system is analysed and structured into sub-net modules, representing system components, monitoring parameters, and interactions. These sub-net modules provide data structures for the iCPT tables, describing the working and failure states/modes of system components, states of the monitoring parameter, and causal relationships between system component states and observable process parameters. The algorithm populates the entries of the iCPT tables based on the analysis of the sub-net modules. Application of the algorithm to a water tank level control system demonstrates improved speed and accuracy in generating iCPTs for dynamic fault detection and diagnosis applications using GSPN-mBSPN approach.

Authors I.Kıvanç, C.Fecarotti, N.Raassens, G.J.vanHoutum

https://doi.org/10.19124/ima.2023.01.20

Abstract

Many original equipment manufacturers offer customized after-sale service contracts to provide personalized services to their customers. To support them in this practice, we develop a quantitative approach to craft maintenance policies which minimize system maintenance costs over a finite horizon. We consider a system of multiple heterogeneous components subject to degradation. We introduce the concept of semi-urgent maintenance, which involves planning maintenance actions within a short time frame to prevent the higher costs associated with corrective maintenance. This approach complements traditional preventive maintenance methods and contributes to a more cost-effective maintenance strategy. The maintenance decision problem consists of determining the interval between scheduled visits at system level, and the thresholds triggering preventive and semi-urgent replacement at component level. We employ a decomposition approach based on which we optimize the maintenance policy per component by solving a Markov decision process via dynamic programming, and use an iterative procedure to optimize the interval of scheduled visits for the system. We present both technical and computational results, the latter showing that the two-threshold policy leads to lower total costs, less corrective maintenance, and longer maintenance intervals than a single-preventive threshold.