Previous agent-based models

Finally, has ABM been used in modeling maintenance strategies? Not very often. Ac-tually, it is found that AB technology has been used more in designing electronic con-trol systems than in modeling. For example Shen et al. (2006, p. 427) found in their literature review that over hundreds of scientific papers have been published in be-tween 1998-2006 and on applying AB technology to intelligent manufacturing appli-cations. These are, for example, representing physical manufacturing resources, such as machines, as intelligent agents that optimize their behavior themselves in relation to their objectives and communicate automatically with other machines. In mainte-nance management, AB technology has been similarly used. For example, Cerrada et al. (2007, p. 313-328) have developed an electronic control system that can autono-mously coordinate maintenance activities. These applications actually go on to much deeper level than pure ABM: these are very complex and automated information sys-tems.

Pure AB simulation models are much fewer. In this study, only two papers were found to use ABM in optimizing maintenance strategies. In some other papers, for example in Wang (2010, p. 239-246), “agents” are used as model entities and they try to optimize maintenance tasks. But Wang’s model is not an actual AB model: ABM techniques are not used and the model entities do not share the common

characteris-tics of agency presented in the chapters 2.2.1 and 2.2.2. Actually, many papers refer to the entities in their models as “agents”, but this does not mean ABM has been used as the research methodology.

The two models, in which ABM was actually used, are briefly compared in the tables 9 and 10 in terms of what elements of ABM have been used and how they take stra-tegic aspects of maintenance into account. The overview and design concepts are compared, excluding model details.

Kaegi et al. (2009, p. 1417-1421) have developed a model for studying feasibility of ABM for analyzing maintenance strategies; and MacKenzie et al. (2012, p. 89-98) have applied ABM in aircraft maintenance manning and sortie generation. In general, it can be said that these two models are somewhat narrow in their perspective on maintenance management. The model made by Kaegi et al. is very simplified and on-ly tries to offer an example, how to possibon-ly make up a basic AB maintenance model.

The model made by MacKenzie et al., on the other hand, is rather narrow in its scope, as the application of ABM has been to simulate maintenance of a single military air-craft. In both of these models, several maintenance strategy decision elements are missing, such as maintenance modifications and performance measurement. Overall, the models deal with rather operational than strategic affairs. But the main finding from both models is the same: ABM is very suitable for modeling maintenance ac-tivities. Especially MacKenzie et al. have successfully applied ABM in a very com-plex decision making situation, in which many other modeling tools would not have been able to cope with this complexity.

Table 9. Comparison of two AB models on maintenance management from ABM’s point of view.

The paper

Analyzing maintenance strategies by agent-based simulations: A feasibility study (Kaegi et al. 2009)

Application of agent based modelling to aircraft maintenance manning and sortie generation (Mac-Kenzie et al. 2012)

Purpose

To show the potential and feasibility of ABM in modeling maintenance strategies and in risk and reliability analysis.

To apply ABM in simulation of sortie generation pro-cess that constitutes of various inspection, repair and preparation actions for military aircraft.

State varia-bles and scales

There are two types of agents in the model called “units” (machines to be maintained) and “operators” (maintenance workforce).

The possible states for the units are: “oper-ating” (works properly) and “non-operating” (requires maintenance). And the possible states for the operators are: “idle”

(waiting for maintenance tasks), “on the way” (moving to a unit) and “repairing”

(at the unit and repairing it).

The environment is spatially modeled as a circle of units at a certain distance away from the home office of operators. The operators move between the units and home office. Time steps, time horizon, size of the environment, and size of the cells in the environment are not explicitly stated.

There are four types of agents in the model called

“production supervisor”, “expediters”, maintenance agents”, and “aircraft agents”. The production super-visor has two states: “available / planning” and “Ex-ceptional Release signoff”. Expediters have three states: “checking for aircraft”, “checking for assigned jobs” and “assigning jobs”. The maintenance agent has two states: “busy” and “available”. The aircraft agent has two states: “Non-Mission Capable” and “Fully Mission Capable”.

The environment is not spatially modeled. That means there are supporting structures and routines simulating the environment and execution logic. The size of the time step is not stated, but the time horizon is finite.

Process operators when maintenance is required (the unit is in non-operating-state) by send-ing them a message at random time. An operator (chosen randomly) who is availa-ble (in idle-state) moves (on the way -state) to the unit that sent the message and repairs (repairing-state) the unit. When the unit is repaired, it goes back to operating-state, while the operator checks whether there are other non-operating units to be repaired. If there are, the operator picks one randomly and repairs it, if not, then the operator goes back to idle-state. After some time again, a non-operating unit informs that it requires maintenance and an operator goes and re-pairs the unit, and so on. From these pro-cesses the costs of equipment downtime and performing of maintenance tasks are derived.

The process of sortie generation constitutes of several stages: aircraft landing, parking and recovery, aircraft serving, unscheduled or scheduled maintenance, mis-sion preparation, prelaunch inspection(s), and aircraft launch. The process described and modeled in the arti-cle is rather complicated, so it is not described here in full detail.

Basically, the agents interact and try to accomplish the tasks making up various portions of the sortie genera-tion process. The producgenera-tion supervisor provides gen-eral oversight and direction to the other agents. It is the only agent having true global awareness, and it makes the majority of decisions in the model. These include job prioritization, determining flying schedule, mainte-nance prioritization and scheduling. The expediters allocate maintenance agents to their assigned tasks. The maintenance agents serve as assignable resources. The aircraft agents are purely reactive and serve to keep track of sortie characteristics and some maintenance activities.

Table 10. Comparison of two AB models on maintenance management, considering AB design concepts and the extent to which maintenance strategy elements have been taken into account.

The paper Analyzing maintenance strategies by agent-based simulations: A feasibility study (Kaegi et al. 2009)

Application of agent based modelling to air-craft maintenance manning and sortie gen-eration (MacKenzie et al. 2012)

Design concepts

 Emergence: systemic phenomena are related to the operators and how they pick units to be maintained.

 Adaptation: operators are able to determine, whether they have to conduct maintenance tasks or not.

 Objectives: operators always try to repair non-operating units.

 Prediction: no predictive behavior.

 Sensing: all the operators know the states of all the units.

 Interaction: operators interact with units but operators and units do not interact with themselves directly.

 Stochasticity: units require mainte-nance at random times and operators pick random units to be maintained.

 Collectives: in modeling respect, there is no hierarchical behavior, but conceptually the operators belong to the same organization.

 Observation: not specified.

 Emergence: although the sortie generation process is rather linear, it’s highly complex and there are great variations in results.

 Adaptation: based on the aircraft state, maintenance tasks are prioritized, allocated and conducted by the agents accordingly.

 Objectives: each agent has its specific goals that they try to meet.

 Prediction: no predictive behavior.

 Sensing: agents have specific functions in the system, and none of them have full ability to sense all information.

 Interaction: interactions happen in hierar-chical form (from the supervisor finally to the aircraft agents).

 Stochasticity: aircraft equipment requires maintenance at random times.

 Collectives: the agents are in clear hierar-chy, and the overall behavior of the system is dependent on this structure.

 Observation: programmed to the model to

The number of maintenance and aircraft agents determines the output of maintenance activities, and the supervisor and the expediters take care ac-count. Agents operate on the aircraft location.

Maintenance technology

Not taken into account. Not taken into account.

Vertical integra-tion

Not taken into account. Not taken into account.

Maintenance organization

No clear organizational form. Clear hierarchical organization.

Maintenance policy and con-cepts

Not clearly stated, but can be considered to be CBM, as the units inform the opera-tors when maintenance is needed.

Not clearly stated, but can be considered to be than agents’ having access to information on the units’ states.

Planning and control is modeled by hierarchical organization. Planning and control systems are not explicitly modeled.

Human resources Not taken into account in any other way but taking operators as capacity.

Training of the workforce is modeled. Man-agement style is top-down.

Maintenance modifications

Not taken into account. Not taken into account.

Maintenance performance measurement and reward systems

Not taken into account. Not taken into account.

It would be beneficial to leverage thinking to the more strategic level from these models, and this is tried to be done in this study. Especially the maintenance strategy elements, which have been ignored in these models are tried to be included in the model in this study. But the best parts of the above models are also tried to be taken into account, in order not having to “invent the wheel” again.

3 THE AGENT-BASED MODEL DESCRIPTION

In this chapter, the AB model is introduced. Making the AB model followed the theo-retical guidelines presented before. Thus, the model is described by (1) defining its purpose and boundaries, (2) constructing the dynamic hypothesis, (3) formulating the model, and (4) testing it.