A GENERIC SPLIT PROCESS MODEL FOR ASSET MANAGEMENT DECISION-MAKING

Transcription

1 A GENERIC SPLIT PROCESS MODEL FOR ASSET MANAGEMENT DECISION-MAKING Yong Sun, a * Colin Fidge b and Lin Ma a a CRC for Integrated Engineering Asset Management, School of Engineering Systems, Queensland University of Technology, Brisbane, QLD 4001, Australia. b Faculty of Information Technology, Queensland University of Technology, Brisbane, QLD 4001, Australia. * Corresponding author: y3.sun@qut.edu.au, Tel: (61 7) , Fax: (61 7) Efficient asset management requires optimal decision making for asset repairs, renewal and replacement. Decisions concerning management of large-scale, long-lived engineering assets are made over a wide range of time scales. Strategic plans are made for periods of years while daily operational decisions may need to be made within a space of mere minutes. Furthermore, when making strategic decisions, we usually have the luxury of time to explore alternatives, whereas routine operational decisions must often be made with no time for reflection. Here we present an innovative split decision-making strategy which distinguishes meta-level decision making, i.e., deciding how to make decisions, from the information gathering and analysis steps required to make the decisions themselves. Key Words: Decision, Asset management, Time scale, Decision making 1 INTRODUCTION Optimal asset management plays an increasingly important role in today s industry. Asset Management (AM) involves making numerous decisions with significant technical and financial consequences, so making optimal decisions has become a priority within industrial organisations. To ensure that decisions are made efficiently and on a scientific basis, an effective AM decision-making is needed. Such a is defined as a set of linked (often interrelated) activities, and the sequence of these activities that are necessary for making optimal AM decisions, normally within the context of an organisational structure and resource constraints. Rhodes [1] indicates in general terms, but particularly in industrial settings, the decisionmaking consists of finding a series of actions which will lead to a desirable outcome and preferably to the most desirable outcome. With the pressing demand for effective integrated AM, not only specific decision-making models, but also a generic AM decision model is needed. This can be used to define executive-level workflow, the required analysis tools, and data input and output requirements for developing an integrated AM decision support system. It can also be used as a template for AM practitioners to enable them to customise their own decision-making es for specific AM activities. Despite increasing attention on Asset Management decision-making es in recent years, there are few practical publications in this area. Development of a generic AM decision-making has proven difficult due to the complex nature of AM decisions. Some decisions need to be made over a long term, such as annually, whereas others need to be made within a much shorter period, such as within hours or even minutes. In addition, AM decisions often involve multiple roles in an organisation and have various, sometimes conflicting, objectives. Existing publications often focus on a specific part of the decision-making. For instance, Wanyama and Homayoun [2] presented a for automated agent negotiation, and Khan and Haddara [3] presented an approach for planning asset maintenance based risk. Some models have been developed for specific enterprises. For example, Caron et al. [4] presented a for strategic decision making regarding financial assets in engineering and contracting companies. A notable exception is the general decision model for infrastructure project management defined by the New Zealand National Asset Management Steering (NAMS) Group [5]. This model has a much more complete consideration of AM decision-making activities and allowance for multiple criteria. However, it focuses specifically on infrastructure project

2 management. On the other hand, Rhodes [1] presented a very generic five-step decision-making model: (1) gathering data and information, (2) finding an exhaustive set of possible, (3) allocating to each of these a degree of desirability, (4) selecting the best option, and (5) verifying the option. However, this model is highly abstract, making it difficult to directly apply in practice. In addition, neither the NAMS Group s model nor Rhodes model consider the different time scales relevant to different kinds of decisions. Also, these existing decision-making models, which mix basic decision-making activities with decision information generation and analysis activities, are not suitable for every kind of AM decision. An effective fully-generic AM decision-making model has yet to be developed. This paper addresses this issue by developing a generic split decision-making model: one that addresses both the basic AM decision-making and the specific needs of the AM decision context. Our model is based on an analysis of the characteristics of typical industrial AM decisions while taking into account the NAMS Group s decision model and Rhodes five-step model. There are a number of modelling techniques applicable to modelling AM decisionmaking es. We favour simple flowcharts in this paper because they are well-established, and familiar to most engineers and business managers. The rest of the paper is organised as follows. AM decision types and their characteristics are analysed in Section 2. A multiple-scale decision-making framework is described in Section 3. Following this, a generic split AM decision-making is developed in Section 4. Some of the issues associated with its design are discussed in Section 5. A case study is presented in Section 6. The conclusion is presented in Section 7. 2 ASSET MANAGEMENT DECISION TYPES To develop an effective generic Asset Management decision-making model, it is essential to first understand AM decision types and their corresponding characteristics. AM decisions can be classified using different criteria such as their relevant time scale and the organisational levels involved. With respect to the relevant time scale, we recognise the following four types of decisions: 1) AM strategic decisions. Such decisions include defining AM objectives, consistent with business objectives, and developing long term AM strategic plans for deciding on asset operational and maintenance and capital investment policies. Asset renewal decisions often belong to this category. AM strategic decisions are normally made annually or every five years or over an even longer period. 2) AM technical decisions. This type of decision includes developing AM plans, based on overall strategic plans, to determine major asset preventive maintenance and upgrading activities, as well as operational regimes. This type of decision is typically made annually, but it can be made quarterly, or monthly. 3) AM implementation decisions. This type of decision includes scheduling asset operational and maintenance activities, workforce allocation, expenditure and material delivery timetables based on AM plans for the short term, such as the next week or month. 4) Reactive decisions. This type of decision is needed when unplanned events occur, e.g., a component fails or there is an unexpected peak in demand. These decisions often have to be made in a short time half an hour to a day to decide, for instance, whether the failure-related assets should be shut down, and/or should be repaired immediately. Since reactive decisions need to be made in a short time, detailed technical and cost analyses usually cannot be conducted. Therefore, to ensure the accuracy of the decisions, the potential situations, the corresponding costs and the appropriate responses are often defined in the AM strategic or technical planning stages in advance. With respect to the organisational roles involved, we recognise the following three categories of decision: 1) Executive level decisions. This type of decision is normally made by the board of an enterprise or its CEO to decide on asset management policies, asset O&M strategies, capital projects and the asset maintenance budget. 2) Managerial level decisions. This type of decision is normally made by general managers or local office managers to determine the asset operation plan, maintenance job priorities, inventory levels, workforce allocations and/or maintenance budgets. 3) Operational level decisions. This type of decision is normally made by a site director or engineers to decide on maintenance/repair types, locations and procedures. Some reactive decisions need to be made by these people. Thus, we need to consider that AM decisions operate over different time scales and focus on and involve a wide range of personnel and maintenance activities. 2

3 3 MULTIPLE-SCALE DECISION-MAKING FRAMEWORK In our approach the relationship among the different Asset Management decision types, roles and time scales is described using a multiple-scale decision-making framework (Figure 1). An asset management decision framework is a conceptual model or guideline on how to make AM decisions effectively through proper integration of various asset management decision models and methodologies. Organizational hierarchy Decision hierarchy Time scale Executive level Define strategic plan for asset management 1 5 years or even longer Managerial level Plan operations and maintenance activities Monthly to annually Operational level Perform regular operations and maintenance activities Weekly to monthly Respond to unplanned events (immediately) Hours to days Short-term events influence long-term planning Long-term plans influence shortterm Figure 1. A typical multi-scale decision-making framework Figure 1 shows that different types of decisions are not isolated. They have interactions with each other, e.g., repeated needs to make short-term corrective repairs to a particular asset may lead to a change in its long-term maintenance strategy. On the other hand, short-term decisions have to be in compliance with the long-term strategies of an organisation. An AM decision-making has to enable decision makers to deal with these interactions effectively. 4 SPLIT AM DECISION-MAKING PROCESS MODEL A generic Asset Management decision-making model has to address the different time scales of different decision types and the interactions among these decision types. However, previous approaches do not solve this problem effectively. For example, although the decision model presented by the New Zealand National Asset Management Steering (NAMS) Group is comprehensive, it is more suitable for AM strategic decisions rather than implementation decisions or reactive decisions. In particular, when making a reactive decision, there is normally not enough time to conduct sophisticated analyses. This type of decision often needs to be made straight away based on observed information such as failure modes. Also, existing decision models cannot cope with AM decisions with different time scales because they normally mix basic decision-making activities, such as defining decision and selecting the best option, together with meta-level 3

4 decision-support information generation and analysis activities, such as identifying project objectives and identifying potential failures. When making AM decisions based on these models, one has to implement both the basic decision-making activities and the decision-support information generation and analysis activities. In practice, however, whereas basic decision-making activities are necessary for all AM decisions, not every AM decision needs to implement the information generation and/or analysis activities. For example, once failure modes have been identified, decision makers only need to use the results of this for subsequent decision making. They do not need to repeat the decision support work in each decision making task. In addition, implementation of the information generation and/or analysis activities often takes a long time which is not suitable for those AM decisions which need to be made within a short period such as reactive decisions. To address these differences between the basic AM decision-making activities and their supporting information generation and/or analysis activities, we divide the overall AM decision-making into a basic decision-making, which focuses on decision-making activities only, and a number of decision-supporting information acquisition and generation es, which provide inputs for decision making. This division leads to the concept of a split AM decision-making model (Figure 2). Based on the above concept and taking into account the NAMS Group s decision model and Rhodes five-step model, our generic split AM decision-making model is structured as shown in Figure 3. The first step in this model is to identify an AM decision which needs to be made. Asset Management involves numerous decisions, from routine maintenance planning to how to respond to an unexpected failure. Different decisions need different information and analyses. Basic AM decision making Decision input inquiry Decision required information Information acquisition/ generation es to support AM decisions The second step is to identify the objectives and the constraints for making the decision. Figure 2. Overview of our split AM decision-making Accurately recognising the decision objectives and constraints is imperative because these objectives and constraints define the criteria for optimising the decision. The objectives of a specific AM decision have to be in compliance with the overall business objectives. Therefore, to identify the objectives and constraints, one often needs to conduct a number of analyses which may take a long time to complete. These analyses are not suitable for those decisions that need to be made within a short time period. Fortunately, although every decision must be based on a clear understanding of the objectives and constraints, this does not necessarily mean that the objective and constraint analyses have to be conducted during each decision making event. Instead the analysis of objectives and constraints can be completed in advance, based on the experience and knowledge of domain experts, in order to produce a set of prepackaged decision which can be applied quickly based on the current system or asset state only. To allow this, the AM decision objective and constraint identification has been separated from the basic decision-making in Figure 3. The third step is to gather the health status and operational information of assets which are associated with the identified decision. This step is essential for all AM decision making. It includes identifying asset failure modes and causes, assessing assets current conditions and predicting the next failure times of the assets or their future reliability changes. The impact of AM decisions on improving asset health conditions also needs to be analysed. Asset health assessment and prediction is often time-consuming since it typically involves gathering and analysing historical data for a large number of assets. For the same reason as mentioned above, this long-term asset health assessment and prediction is separated from the short-term basic decision-making in our model. The fourth step is to identify all potential decision. Decisions imply choices among alternatives [6]. Therefore, identification of decision (alternatives) is a crucial step in a decision-making. In engineering asset management, some are discrete (e.g., replace a component) while others are continuous (e.g., increase the frequency of inspections). After the decision have been identified, one then needs to shortlist them against deal-breaker rules, i.e., eliminate those that cannot meet the objectives and/or constraints. When there a large number of or there are continuous, obtaining a shortlist of decision becomes difficult. te in Figure 3 that there is a feedback loop from decision verification (at the end of the ) to decision option identification. If all shortlisted decision are unsatisfied, the decision makers may come back to look at those previously discarded. In some cases, discarded become viable again because of changes to the decision objectives and constraints. 4

5 1. Identify an AM decision Have the decision objectives and constraints been identified? AM decision objectives /constraints identification 2. Gather AM decision objectives/constraints information Have asset conditions been identified? 3. Gather asset condition and operation information Asset health assessment and prediction Modify AM decision objectives BASIC DECISION MAKING PROCESS Have the decision been identified? 4. Gather decision Have the decision been ranked? 5. Gather option ranking information 6. Select the best option and check the decision parameters Do the decision parameters need to be optimised? Has the relationship between decision parameters and known objectives been quantitatively analysed? Option identification Option ranking Relationship analysis DECISION INFORMATION ACQUISITION/ GENERATION PROCESSES 7. Gather the relationship information 8. Identify the optimal decision parameters Are all outcomes of the decision known? Simulation /sensitivity analysis 9. Verify the decision Is the decision satisfied? Are any other available? 10. Output the decision Figure 3. Our generic split asset management decision-making 5

6 The fifth step is to rank the decision based on decision criteria which are determined according to the decision objectives and constraints. In modern asset management, AM decisions often involve multiple actors, different objectives and constraints, i.e., AM decision making usually belongs to multiple criteria decision problems. As a result, ranking decision is often difficult. To address this issue, various decision option ranking models and methodologies have been developed, such as Decision Trees, the Analytic Hierarchy Process, and so on. These techniques can effectively assist in AM decision ranking. However, using them correctly typically requires a sound knowledge of how they work and, in particular, an understanding of their limitations. In addition, in most cases, it also takes a significant amount of time to conduct decision option ranking analyses, so the ranking is also separated from the basic decision-making in our model. The sixth, seventh and eighth steps are to optimise the decision parameters. After the fifth step, the decision have been ranked. Then one can determine a best option based on the rankings. However, the determination of the best option does not necessarily mean that the AM decision can be finalised because AM decisions are so complex. In practice, further analysis may be needed to optimise the parameters which are associated with the selected decision option. For example, when the reliability of an asset is lower than an acceptable level, a number of maintenance activities can be applied to improve its reliability, including conducting preventive maintenance or renewing the whole asset. If a decision to renew an asset is made, then one needs to further decide on the optimal renewal time. To address these issues, our split decision-making model has additional steps in which we need to identify data availability and then conduct optimisation analysis using an appropriate optimisation model or method based on the decision objectives and constraints which have been identified in the second step. The ninth step is to verify the decision. This step is the last action in our split AM decision-making model. It usually involves a number of what-if analyses to ensure that the decision chosen is robust. Once the decision has been validated, it becomes the final one. If the selected decision option is not satisfactory, the decision makers need to modify the objectives or reconsider the decision. 5 DISCUSSION Our split decision-making can effectively solve the problems of different time-scales in AM decision making. Decision makers need to go through the basic decision-making for every decision made, but they do not necessarily need to go through all the information generation and/or analysis es at the same time. They can do information collection and analyses, such as cost analysis and failure prediction, whenever they like and use the accumulated knowledge to inform later decisions. The basic decision-making enables decision makers to consider the decision inputs systematically so that the required inputs can be prepared in advance. This capability is very useful for making lower level decisions, especially reactive decisions. The split is also beneficial for decision support software design. It allows us to develop a core decision-making module implementing the basic decision-making, and a number of separate analysis modules to acquire or generate inputs for decision making. The core module and the analysis modules need to be loosely coupled only, making the overall software development easier. Users of the resulting system need the core module and some selected analysis modules so that they can do some simple and common analyses themselves. When the users need more sophisticated and/or unusual analyses, they can access other analysis modules through stand-alone application software or web-based services. Since the analysis modules are loosely linked to the core module, they can be modified and extended without affecting the core module. For comparison, we present a simplified view of our in Figure 4, and the NAMS infrastructure management decision-making model in Figure 5. Comparing Figure 4 with Figure 5, we can see that the NAMS Group s decision starts with identifying project objectives because it was designed specifically for infrastructure project management. However, our split decision-making starts with identifying an AM decision and then identifying the objectives of the decision. This arrangement enables our to accommodate more generic AM decisions. In addition, our requires not only identifying the decision objectives, but also identifying the constraints for making the decision because the identification of the constraints is crucial for decision optimisation. The second major difference is that the NAMS Group s decision has one feedback loop from reviewing to defining project objectives, whereas our split decision-making has two feedback loops. One is from the verification step back to identifying the decision objectives and constraints. This feedback loop is similar to that in the NAMS Group s decision. In addition to this loop, our split adds another feedback loop from verifying the decision to defining decision. As mentioned above, some AM decisions have numerous. The difficulty exists in initially and exhaustively identifying all potential. Therefore, reviewing is often necessary. 6

7 Identify an AM decision Identify asset conditions Define decision Identify decision objectives and constraints Modify AM decision objectives Select a decision option Optimise decision parameters Do the decision parameters need to be optimised? Verify the decision Output the decision Is the decision satisfied? Are any other available? Figure 4. A simplified version of our generic split decision-making Define project objectives Does the problem relate to an existing asset? Identify potential failures Modify project objectives Identify the nature of the opportunity Define decision Define the criteria for failure Review Analyse against multiple criteria Can a preferred option be selected from the remaining? Complete financial analysis Figure 5. A simplified version of the NAMS Group s decision-making for infrastructure projects 7

8 6 A CASE STUDY The proposed model has been applied for pipeline renewal decision making in an Australian water company. Step 1: Identify the AM decision. In this case study, the AM decision is to decide the optimal renewal time for each pipeline in a pipeline network of the company. Step 2: Identify the objectives and constraints associated with pipeline renewal. Since the objectives and constraints have not been identified before, we worked them out through an AM decision objectives/constraints identification, which is not presented in this paper due to the limited space. The objective in maintaining the whole pipeline network was to minimise the expected total maintenance cost including repair costs and renewal costs. Production losses could be ignored in this case study. The identified critical constraint was customer satisfaction which was measured by the number and duration of service interruptions. Different pipelines had different values for these two measurements. Another critical constraint was the maintenance budget limit. Step 3: Predict pipeline health and operational conditions. The existing pipelines were built about 40 years ago and were dynamic aging systems. Historical failure and operational data were collected and used to assess current health conditions and predict future changes of the pipelines through an asset health assessment and prediction, again not presented in this paper. The health conditions of the pipelines were represented by failure probability functions which were used to estimate the expected total maintenance cost and the expected number of service interruptions. The duration of service interruptions needs to be calculated based on the repair performance of the company, however, for this situation, past experience showed that such durations were tolerable. Step 4: Gather maintenance. There was only one option (renewing the pipelines) which was suggested by the maintenance manager in the company. Step 5: Check the decision parameters. The renewal time was a decision parameter which needs to be optimised. 1. Identify an AM decision 2. Identify the objectives and constraints associated with pipeline renewal 3. Predict pipeline health and operational conditions 4. Gather maintenance 5. Check the decision parameters 6. Optimise the renewal times 7. Verify the decision using sensitivity analysis and uncertainty analysis Is the decision satisfied? 8. Output the decision Modify AM decision constraints Are any other available? Figure 6. Pipeline renewal decision-making Step 6: Optimise the renewal times. The aim of optimising the renewal times was to minimise the expected total maintenance cost of the pipeline network while meeting the requirements for customer satisfaction. Step 7: Verify the decision using sensitivity analysis and uncertainty analysis. Since the renewal costs were only estimated, albeit by domain experts, conducting a sensitivity and uncertainty analysis was essential to ensure the decision to be robust. When the determined renewal time for a pipeline was satisfied, we accepted it and the decision making loop for this pipeline was closed. Otherwise, we went back to review and modify the customer satisfaction target and the budget limit and repeat the above decision procedure until we had worked out the renewal times for every pipeline in the network. From this case study, it can be seen that the split AM decision-making can be used for pipeline renewal decision making effectively and efficiently. In fact, a software tool based on the above pipeline renewal decision-making has been 8

9 developed for the company. On the other hand, the NAMS Group s decision-making is not effective in this case since it does not include a function to optimise decision parameters (i.e., renewal times). In addition, when deciding to renew the pipeline network, the expected number of failures rather than potential failure modes was the major concern of the company. 7 CONCLUSION Asset management involve various AM decisions which have different characteristics. These decisions have different time scales and focuses. They also involve different personnel and AM activities. With respect to the time scale, AM decisions can be classified into four categories: AM strategic decisions, AM technical decisions, AM implementation decisions and reactive responses. These four types of decisions have very different time scales ranging from years to half an hour. The existing decision-making models which require implementing basic decision activities as well as decision information generation and analysis activities sequentially are not suitable for all of these different types of AM decisions, and hence cannot be used as a sufficiently generic AM decision-making model. Here we have presented a generic split asset management decision-making model. In this new model, a basic decision-making which focuses solely on decision making activities has been separated from the decision-supporting information acquisition and generation es which provide inputs for making decisions. When making an AM decision, one has to go through the basic decision-making, but is not always necessary to go through all the decision information acquisition and generation es. The split model has been developed to address the issue that AM decisions have different time scales and involve different roles. Our split AM decision-making model is more applicable to the wide range of decision making activities required in large-scale engineering asset management. It can be used as a reference model so that industrial personnel can quickly develop their own customised decision-making for different specific AM activities. The model is also useful as a framework for developing an integrated AM decision support software system; we plan to develop a prototype of such a system in future work. 8 REFERENCES 1. Rhodes, P.C., Decision Support Systems: Theory and Practice. 1993, Henley-on-Thames: Alfred Waller Limited. 2. Wanyama, T. and Homayoun Far, B., A protocol for multi-agent negotiation in a group-choice decision making. Journal of Network and Computer Applications, (3): p Khan, F.I. and Haddara, M.M., Risk-based maintenance (RBM): a quantitative approach for maintenance/inspection scheduling and planning. Journal of Loss Prevention in the Process Industries, (6): p Caron, F., Fumagalli, M., and Rigamonti, A., Engineering and contracting projects: A value at risk based approach to portfolio balancing. International Journal of Project Management. In Press, Corrected Proof. 5. Limited, M., Optimised Decision Making Guidelines: A sustainable approach to managing infrastructure. 2004, Thams: NZ National Asset Management Steering Group. 6. Holloway, C.A., Decision Making under Uncertainty: Models and Choices. 1979, Englewood Cliffs, NJ: Prentice- Hall, Inc. 7. Sun, Y., Ma, L., and Fidge, C. Optimization of economizer tubing system renewal decisions. in Proceedings of International Conference on Infrastructure Systems Rotterdam, Netherlands: IEEE, Submitted. Acknowledgments This research was conducted within the CRC for Integrated Engineering Asset Management, established and supported under the Australian Government s Cooperative Research Centres Program. 9