27th Annual INCOSE International Symposium (IS 2017) Adelaide, Australia, July 15-20, 2017
Towards a Modular System Dynamics Approach for Modelling Military Workforce Planning Problems Victoria Jnitova Research student, UNSW, Canberra
Sondoss ElSawah Capability Systems Centre UNSW Canberra
[email protected]
[email protected]
Michael Ryan Capability Systems Centre UNSW Canberra
[email protected] Copyright © 2017 by Victoria Jnitova, Sondoss ElSawah and Michael Ryan. Published and used by INCOSE with permission .
Abstract. Military workforce planning and management is increasingly drawing upon Modelling and Simulation (M&S) to enhance decision making. This project is to develop and validate an M&S approach suitable for modelling workforce dynamics ‘from cradle to grave’, offering end users a decision-support mechanism which is intuitive to use and tailorable to any specific problem within the military workforce context. The project is based on multiple case studies where generic System Dynamics ‘building blocks’ (BBs) are progressively developed, tested and validated. The Royal Australian Navy (RAN) Electronic Technician basic training pipeline is the first case study of the series. The earlier problem scoping and conceptualisation conducted by the authors is continued in collaboration with the RAN stakeholders to develop process models for the three existing pipeline variations. These conceptual models are then used to facilitate effective stakeholder engagement, and to inform the design of the preliminary working set of BBs.
Introduction Background Decision making in the Royal Australian Navy (RAN) workforce planning and management context is influenced by increased budgetary pressure and the need to implement and adjust to rapidly evolving new technologies, present and future assets operational focus and increased operational tempo. One promising way to address these challenges and to provide objective evidence to link decision options to possible costs and benefits under different scenarios is to use a System Dynamics (SD) based model-building approach. SD is a powerful modelling toolbox for analysing complex and dynamic systems. The SD toolbox includes conceptual and numerical models that can be used to map and simulate feedback interactions, delays, and non-linear relationships to support system design and problem solving, including the ability to: • integrate social and technical elements, • integrate physical and information views of the system, • model hierarchical systems, • model feedback interactions, non-linear relationships, and delays, and • integrate tools from other modelling paradigms (e.g. discrete event) in a hybrid framework. SD models use fundamental concepts of ‘stocks’ (also known as accumulators or levels) and ‘flows’ (also known as rates) to represent key processes that derive problem behaviour. Simulating a model gives insights into how these processes change in response to endogenous (i.e. generated inside the model) and exogenous (i.e. input drivers to the model) factors. Building and analysing SD models
help stakeholders understand what drives the dynamic behaviour of the system and the outcomes of decisions.
Issues Associated with the Use of SD Methodology Despite SD potential in promoting systems understanding, the uptake of SD as a learning and decision-support tool is still limited, especially in systems design and engineering areas. In this subsection, we present some of the issues that limit the use of SD, and how they motivate this research. To produce a reliable model to support insightful analysis, a modeler needs first to have an extensive experience in SD modelling and domain knowledge. Second, SD models are developed to test an a priori defined problem structure (known as dynamic hypothesis). A user can run experiments to change model parameters and to observe changes in a system’s performance. The SD rigid structure is limiting in situations where there is uncertainty about a problem structure and a decision maker is interested in exploring a system’s behaviour under a wide range of system designs and policy variants. To examine an alternative structure requires a model redevelopment, which comes with additional cost and time. Third, SD models can be difficult to understand especially by non-technically trained managers. Here, we use the terms ‘decision makers and managers’ to denote an individual or a group who are interested in making decisions to operate and improve a system. Fourth, models grow in complexity (both detail and dynamic). This may exert cognitive load on modelers and end users, and may compromise learning insights to be derived from a model. Despite this, large models are often support system understanding and decision making at an appropriate and acceptable level of detail (Jacobson et al., 2008). There is an ongoing search amongst M&S applications’ developers and practitioners to find an ultimate solution to the modelling issues described above. One of the M&S approaches that is gaining popularity in the last decade is using building blocks (BB) for SD modelling. Modelling issues and how they can be addressed using the BB approach is outlined at Table 1. Table 1. Modelling issues and how they can be addressed using the proposed BB approach Modelling issues
Capabilities offered by a modular (BB) SD approach
The modelling process is time consuming as Ability to use reusable and proven-to-work model models are built from scratch building blocks or components to speed up the process, reduce the burden of model testing, and improves confidence in the model output The modelling process starts with a priori Ability to assemble modelling building blocks to structure. This limits experimentation to generate and run specific model instances. changing model parameters. This is limiting when the purpose is to explore various design and policy variants. Models are difficult to understand especially by Ability to use domain-specific modelling components, non-technically trained managers. that separate the end user from the inner model details, to build and experiment with models Models grow in complexity (i.e. both details and Ability to use function-centered components may dynamic), which exert cognitive load on the reduce the risk of adding unnecessary elements since modeller and end user, and may compromise the each building block should add something to the overall learning insights to be derived from the model. model representation, otherwise it can be excluded from the model.
Aim and structure This paper is a part of an ongoing project which aims to develop an SD based modular approach for model building in the Military context to: (1) design and implement a set of BBs, and (2) validate the developed BBs in terms of their fitness-for-purpose for each specific problem. The BB approach has potential to: • expedite the modelling process by providing reusable and tested ‘plug-in’ components; • bridge the chasm between modelers and system engineers by providing high level domain objects; and • improve learning by providing users with a flexibility to build and experiment with models, without being overwhelmed with the model’s technical details. The ultimate aim of this study is to develop a generic SD approach suitable for use across the capability lifecycle to support the design, development, and evaluation of the methodology (i.e. process and tools), using a ‘multiple case studies’ approach. Each individual case study will be supported by a tailor-made simulation application using generic BBs to solve a specific workforce problem. Ongoing collection of the BBs, models and lessons learnt from their practical implementation would provide designers, system engineers, workforce planners and training specialists with a comprehensive database of BBs and BB-based M&S applications, that is intuitive to use, flexible to tailor to a specific need, and versatile to suit a wide variety of problem domains. The paper consists of the following sections: Introduction; BBs and their associated concepts and definitions; Literature review; Research methodology; Problem framing and scoping; Model’s conceptualisation and formulation; Reflections and future research directions.
BBs and their associated concepts and definitions There is a set of concepts that guide the discourse and practice surrounding the idea of identifying useful elementary and reusable BBs. These concepts spread across the fields of simulation, systems engineering, software development, and SD models. These concepts include: structure, design patterns, model building blocks, and software components. We make a start by defining these concepts and illuminating differences.
Structure Structure is a fundamental SD concept. In his classic work, Forrester (1990) defines systemic structure as the ‘smallest number of components, within which the dynamic behaviour under study is generated’, where ‘structures exist in many layers and hierarchies’. Within any structure there can be sub-structures. Forrester also identifies structure elements as follows: closed system boundary, levels (or accumulators that describe the system conditions at any particular time), and rates (or flows that describe processes influencing changes in the system conditions). Rates are substructures which are composed of: a goal, observed condition of a system, gap between goal and observed condition, and action to be taken based on the perceived gap.
Design patterns Design patterns concept goes back to the architect Christopher Alexander (1977) who recognised the pattern language that allows people to develop infinite combinations of architectural designs (e.g. buildings, towns, cities). A pattern is the basic rule which defines the language system. It is connected to lower or higher levels of hierarchy in the system, composing a network of patterns which all together builds up the holistic architecture. Design patterns gained popularity in computer science after publishing the book ‘Design Patterns: Elements of reusable Object-Oriented Software (Gamma et al., 2005). The book provides the following definition of a ‘pattern’: ‘a recurring solution to a common problem in a given context and system of forces’. It also proposes a ‘patterns catalogue’,
broadly dividing design patterns into ‘structure’ and ‘behaviour’ classes. Since the release of this revolutionary work, patterns continue to help us document design decisions and their rationale, reuse known constructs, and form a shared vocabulary for problem-solving discussions.
Building blocks Building blocks concept is used in simulation studies to describe the basic structures of models. BBs have the following characteristics (Valentin (2011); Verbraek (2002); Orfali, 1995) • self-contained (all the information needed to execute the block’s function is contained in the building block itself), • reusable, • replaceable (it can be substituted by another building block which have the same functionality and interfaces), • provide useful service or functionality to its environment, • exchange information with other blocks through precisely defined interface, • can be customised to any specific requirements where they are used, and • a BB design is independent from its implementation, which means it can be described as a conceptual model, function, and/ or an interface with other building blocks.
Software components Software components concept comes from software/systems engineering fields to describe fundamental entities, often in a distributed environment (Szyperski et al., 1999). Components are organised into software libraries. On the relationship between BBs and components, a component is seen as an implementation of a BB in a software environment, which means that the interfaces and functionality of components and BBs are the same (Verbraek and Valentin, 2008).
Literature review In this section, we discuss the effort made in SD modelling to improve the efficiency and effectiveness of the modelling process, with special focus on the idea of using generic and reusable BBs. To guide the BB development and implementation, Liehr (2001) proposes to categorise BBs into four structural levels (Figure 1).
Figure 1. Four levels of the BB structure We use these four levels to guide the literature review of previous work done in this area, as follows.
Level 1: Fundamental BBs Several SD software environments (e.g. Powersim Studio, Anylogic) have emerged over time with different capabilities and purposes. Due to these packages, the efficiency of modelling process has improved as they provide generic elements (stocks, rates, auxiliary variables) that can be easily accessed (usually through drag-and-drop functionality) to build models. This gives modelers freedom to build models by linking elements without having to define and code these elements. This is definitely a powerful feature for experienced modelers, but it does not fully address the modelling issues described in the introduction to this paper in regard to improving modelling efficiency and effectiveness for end users and invoice modelers. Still, modelers have to go through a lengthy process of building a model.
Level 2: Universal BBs Richmond (1985) was the first to use the term ‘atoms of structure’ to describe common structures in SD models. Eberlein and Hines (1996) used the term ‘molecule’, in the same sense as atoms, to describe and organise elementary BBs. Molecules are domain-independent structures that are used to develop models in different application areas. The first attempt to catalogue molecules resulted in defining and characterising about 50 molecules. Another related concept is an ‘archetype’ (Senge, 1990), which is a conceptual problem structure (usually represented using Causal Loop diagrams) that explains counter-intuitive and policy resistant behaviour of dynamic systems (e.g. limits to growth). Archetypes are developed after accumulations of experiences learning about dynamic behaviour across different problem domains (Paich, 1985). They are flexible and easy-to-understand owing to their high-level conceptual nature. The flip side is they have limited utility as operational blocks for developing large scale numerical model. Dowling (1995) details the limitations of transforming archetypes to simulation models.
Level 3: Domain-specific BBs There have been calls for developing domain-specific BBs in SD, but very few attempts have been made. For example, LaRoche (1994) recognised the need to define key structure for mapping business processes, and used the term “Template-loop” to conceptualise a simplified manufacturing chain. Ahmed (1997) proposed architecture for a component catalogue to allow users and modellers to access readily developed components relevant to the problem area of their interest. Similar calls have been repeated by Myrtveit (2000) and Powers (2011).
Level 4: Project-specific model instances An increasing number of project-specific M&S applications involving patterns and modularity is developed. They are usually tailored to a specific problem, but could be re-used outside a project’s boundary if contextualised. The patterns approach is based on an underlying assumption that designers, software engineers in particular, rarely create whole new structures, but rather focuses on repetitive things they do over and over, using ‘re-usable’ patterns, whether they choose to articulate them or not (Kent, 1995). The patterns and their associated BBs are only useful, however, if the problem is understood. If it is not the case, the patterns are useless (Kent, 1995).
Research methodology Similar to other simulation-based modelling methodologies, SD model development may take a lot of time and resources. A typical modelling process cycle progresses through four iterative phases (Luna-Reyes and Andersen, 2003): (1) problem framing and definition (scoping), (2) model conceptualisation, (3) model formulation (design) and testing, and (4) model use and communication of results. Through this process, a modeler makes use of various data collection and system mapping methods to represent a problem of interest.
To motivate BB development and to support design, development, and evaluation of the methodology (i.e. process and tools), we use a ‘multiple case studies’ approach. Each individual case study of the project is supported by its own model development cycle. The Royal Australian Navy Electronic Technician Basic Training Pipeline (RAN ET BTP) is the first case study of the series, forming the first cycle of the research enquiry. The RAN ET BTP case study will result in a preliminary working set of BBs, that are rigorously tested and validated to ensure the BBs and their combinations work and produce valid and reliable results. The application will continue to grow by adding new cycles of case studies with added complexity and new BB types as required. The problem framing and definition, as well as work towards the model’s conceptualisation for our first case study (RAN ET BTP), was reported in our previous work (Jnitova, 2016). In this paper, we continue our work on the model’s scoping and conceptualisation, followed by the model’s initial formulation. The model’s testing, use and communication of results will be a subject of our future research.
Problem framing and scoping We determined the RAN ET BTP research context and model purpose in our previous work (Jnitova, 2016). To gain an insight into the patterns’ identification, types of problems and questions of analysis relevant to the domain, we conducted a literature review into the patterns embedded into applications of SD in workforce planning (Table 2). This activity informs transition between the steps of the model development cycle. The model’s physical and information constructs are designed to enable experiments’ conduct for a particular model purpose. Table 2. Patterns in SD [pure and hybrids] workforce planning models Paper
Problem/need
Model purpose
Experiments
Model constructs: physical and information
Andersen, 1982
Examine the interactions among the retirement, separation, and recruiting decisions on the workforce mix and costs
Analyse the overtime effects of different retirement policies on the workforce size and personnel costs.
Gradual changes in the % of separation rates
McGinnis, 1994
Evaluate the feasibility of Officer Professional Development alternatives, and to provide insights into the complex dynamics of Army personnel management
Examine the combined effects of changes in the promotion, command selection and work conditions’ policies on the workforce size, retention and professional development
Changes in promotion rates, command selection rates, military school selection rates, duration of assignments and system delays for promotions, assumption of command, and military schooling.
Thomas, 1997
Examine the interactions among the retirement, separation, and recruiting decisions on the workforce
Examine the combined effects of training, retirement, and recruitment policies on the workforce size
Change the number of training spots available, and the desirable size of workforce
Physical: A 4-stage aging chain • with outflow rates to model separation rates at each stage • push-type (once people spend a certain time in the stock, they are moved to the next) Information: Close the gap (function of the total levels of all stocks and desired workforce). Physical: A modular design includes aging chain modules to model the Officer Professional Development critical path, as well as policy module enabling changed parameters input; graphical analysis module providing graphical representation of the results, and reporting module, producing reports. Information: Close the gap (function of the total levels of all stocks and desired workforce). Physical: A 3-stage aging chain with outflow rates to model the separation rates at each stage Information: Close the gap (function of the workforce stock, separation rates from stocks, and the desirable workforce level)
Linard, 1999
Examine long term effects of ‘Boombust’ recruiting patterns on the training system after the initial event
Evaluate different employment policies
Changes in the percentage of retirements, promotion percentage on the workforce size.
Cavana, 2007
Understanding of the causal factors of poor retention and recruitment, their interconnections and complex relationships and to identify leverage points to develop a turn-around strategy Analyse complex organisational behaviour associated with a hierarchical structure in which the lowest level agents are characterised by continuous and discrete eventvariable dynamics and the highest level agents by heuristically based decision-making mechanisms. Forecasting for the right numbers and right kinds of people at the right places and the right times
Analyse the combined effects of outsourcing of base level repair and changes in promotion policy on the troops morale and attrition rates
balancing the number of people in each stock and the flows to and from each stock.
Examine the effect of different detailing policies and different crew assignment strategies, and other organisational strategies and their positive/negative consequences on an individual sailor’s performance, decision making, motivation etc.
Changes in: • type of duty • training schedule • recruitment rate • advancement; • reenlistment retirement
Physical: Push type aging chain Information: Close the gap (function of the workforce numbers at each stock, components’ interactions and the desirable workforce level
Analyse the overtime effects of varying the attrition and recruitment rates policies on the workforce size, performance and cost Identify and analyse the linkages (cause and effect) between the enrollment rates, higher attrition rates, lower availability of the reservists to take the training courses on time, and lower availability of the
Change in • recruitment rate, • attrition rate
Physical: 3 levels of aging chains from 2 to 4 stages; Mixed pushpull type Information: Close the gap (function of the total levels of all stocks and desired workforce)
Change in • recruitment rate, • promotion rate • instructor availability
Physical: A number of 3-4-stage aging chains; Mixed push-pull type Information: Close the gap (function of the total levels of all stocks and desired workforce).
Garagic, 2007
Garza , 2007
Mehmood, 2007
Examine the gap between demand and available number of reservists for most of the trades.
Physical: A 4-stage aging chain Mixed push-type (once people spend a certain time in the stock, they are moved to the next) and pull type (people cannot pass to the next level unless a position is available) Information: Close the gap (depending on the policy formulation, function of the total levels of all stocks and desired workforce). Physical: Casual loop diagrams for job morale, job satisfaction, civilian vs. military pay etc. Information: Close the gap (function of the total levels of all stocks and desired workforce).
Skraba, 2007
Change training lead time, ratio, separation rates, and desirable workforce size at each rank
Physical: A 3-stage aging chain Information: Close the gap ((function of the workforce stock at each ranks and the desirable workforce level)
Test the effects of different training and personnel management policy on the workforce size.
Change training lead time, Instructor to Trainee ratio, separation rates, and desirable workforce size at each rank
Examine interactions amongst the planning flexibility and training capacity and the uncertainty and volatility of separation rates Determine shortages in personnel numbers and skill levels in light of future needs
Examine effects of forecasted workforce targets on the workforce size and costs
Changes in the forecast workforce numbers whilst the separation rate remains high
Physical: 4 identical layers of a 2-stock aging (bidirectional) chain. The bidirectional structure captures the failure of employees to progress from one level to another. Information: Close the gap (function of the workforce stock at each ranks and the desirable workforce level) Physical: A two-stage aging chain of workforce Accumulator stock for the availability of beds Information: Close the gap (function of the workforce stock and the desirable workforce level)
Examine future effect from the changes to the FFG WEE scheme of complement on training throughput
Changes of schemes of complement options by increasing the number of concurrent untrained billets without affecting the rest of the billet structure to estimate the resulting training outcomes
Armenia, 2015
Determine shortages and redundancies in personnel numbers and skill levels in light of future needs
Examine the combined effects of changes in the retirement and promotion policies on the workforce size and composition
changes in: • the % of retirements, • promotion rules • desirable recruitment.
Johnstone, 2015 (a)
Examine planning and scheduling problems within an aviation training continuum
The SD model provides the school’s targeted demand to the DES model
Changes of policy and planning effect on the training continuum, both in the short and long term
Wang, 2007
Markham, 2008
Evans, 2009
Understand effects from human resources transitions in large organisation prediction of a particular dynamic of particular rank members Gain insight into unknown effects from increased training demand at high rank
instructors to offer the training courses on time, in order to understand the possible reasons for the gap Test the effects of different training and personnel management policy on the workforce size.
Physical: A five-stage aging chain. Mixed push-pull type (once people spend a certain time in the stock, they are moved to the next) and pull type (people wait an instructor to become available to get them through to the next level) Information: Close the gap (function of the workforce stock, instructor availability, and the desirable workforce level) Physical: A 4-stage aging chain Mixed push-pull type (once people spend a certain time in the stock, they are moved to the next) and pull type (people cannot pass to the next level unless a position is available) Information: Close the gap (function of the total levels of all stocks and desired workforce). Physical: aging 2 stages chain; Mixed push-pull type Information: Close the gap ((function of the workforce numbers at school and squadron
Johnstone, 2015 (b)
Tustanovski, 2015
Strategic decision making for an aviation training continuum that is going through major change.
Examine the sustainability of the military forces with respect to their size and composition. The focus is also on the demographic trends and their implications on the force
Examine the effects of changes in the training continuum and school restructure to estimate recovery time after the policy change or disturbance Examine the effects of changes in recruitment pool on the workforce size
Changes in the training student intakes and pass rates
Changes in the number of recruits in response to different population growth scenarios and levels of acceptable military services.
stocks and the desirable workforce level) Physical: 2-stage aging chains. SD-SD hybrid “push-pull” design - flow control to quantify transience and estimate recovery time after a policy change or disturbance. Information: Close the gap (function of the total levels of all stocks and desired workforce) Physical: A 3-stage aging chain with outflow rates to model the separation rates at each stage Information: Close the gap (function of the total levels of all stocks and desired workforce).
Model conceptualisation We developed process models of the RAN ET BTP variations, as identified by Jnitova (2016), to assist in conceptualisation and to improve communication with the key stakeholders, who often have no or limited modelling experience. These models proved to be extremely useful for the stated purpose due to their use of common logic (sequential progression of the activities from start to finish) and everyday workplace language. There are multiple representations of process models, with the one selected for this research is Business Process Model Notation (BPMN). It uses a graphical representation of a business process using standard objects and rules defining available connections between the objects (Kenneth, 2012). BPMN components used to construct the RAN ET BTP variations process models are activities (rectangles), gateways (diamonds) and connectors (arrows indicating sequence flow). The three RAN ET BTP variations as identified by Jnitova (2016) are as follows: Pre-2015, Current (2015) and Post-2016. These variations represent three different strategies for the RAN ET BTP conduct, where the policy changes result in the changes of the pipeline key components and their sequences. Each of these strategies is expected to lead to different effects in terms of training cost, duration, location of delivery and approach to consolidation training. The three process models have been developed in consultation with the ET category managers for each pipeline variation. The process model for the Pre-2015 variation is at Figure 2(a).
Acronyms used at Figures 2a and 2b: ET ITT
– Electronic Technician Initial Technical Training course
CTJ Journal
– Competency Task
S1, S2, S3 – Specialisations (and also Specialisation courses at Figure 2b) Class 1… – Platform’s class (e.g. ANZAC Frigate) ASTC – Advanced Skills Training Course EACs
– Equipment Application Courses
AB ET
– Able Seaman Electronic Technician
Figure 2(a). Pre-2015 RAN ET BTP variation The Current (2015) variation is similar to the Pre-2015, with an exception of the CTJ component, which has been removed from the RAN ET BTP in 2015. The ET Certificate III component previously achieved at the CTJ has been added to the ET ITT, increasing the course’s duration as a result. Similar to the Current (2015) variation, the Post-2016 variation retains achievement of the ET Certificate III ashore via the ET ITT. It also introduces a new component: specialisation courses, to be conducted on completion of the ET ITT. The ASTC, based on the ET Certificate IV qualification, is moved from the RAN ET BTP to the intermediate training pipeline, conducted post-initial minimum period of service, when a new service period agreement has been reached between a sailor and the RAN. ASTC will become a prerequisite for promotion to a Leading Seaman. The process model for the Post-2016 variation is at Figure 2(b).
Figure 2(b). Post-2015 RAN ET BTP variation
Conceptualisation continues with further detailing of the system elements identified during process modelling to infrastructure and control, and commencement of the conceptual design of BBs, including consideration of their structural components, functionalities and interfaces. Together with the BB conceptual models, the process models are instrumental for the model design, where M&S execution is facilitated by translation of common logic of BPMNs into the SD logic of ‘stocks and flows’.
Model formulation The purpose of this phase is to translate the qualitative model into a working formal quantitative model. This includes clear definition of stocks, flows, auxiliary variables, parameters, and equations. To build the SD conceptual model, the authors apply the steps proposed by McLucas (2005). These steps are as follows: Step 1: Formulating dynamic hypotheses for each sector of the pipeline variations; Step 2: Capturing and prioritising business rules; and Step 3: Draw an SD conceptual model.
Step 1 The dynamic hypothesis captures identification of state variables and their associated rate variables, physical structure and operation of the feedback mechanisms and influences from the combined auxiliary variables on the rate variables (Forrester, 1994). The variables and rates as identified by Jnitova (2016) are at Table 3. Table 3. RAN ET BTP variables and rates Variables a. Numbers of Recruits b. Instructor numbers/ hours c. Training components: types, number per year, duration, number of students per component; d. Position types (trainee/ junior ET Sailor positions); e. Policy inputs into RAN ET BTP step changes; f. Ranks, and g. Levels of confidence and satisfaction of the RAN ET BTP graduates
Rates a. b. c. d. e.
Failure rate Recruitment rate Separation rate Retention rate; and Waiting periods between the courses – pipeline delays rate.
Each year a pre-determined number of ET recruits of Seaman* rank commence their ET training as members of the RAN Training Force, starting with the ET ITT. They complete their basic training with gaining their RAN Trained Force status, promotion to Able Seaman, ET qualification, specialisation and class allocation according to which they proceed to their first work posting. The courses selection and consequent posting are subject to specialisation and platform class allocation, with some courses are common for all RAN ET BTP trainees, and some associated with specialisations and platform classes, splitting the RAN ET BTP into a number of respective streams. All three variations of the RAN ET BTP end up for each intake on completion of the EACs.
Step 2 The business rules outlining the relationships between the variables and basic mechanisms of the pipeline dynamics are listed in the order of priority and are as follows: • although there are provisions for step changes for aspects such as policy changes, the RAN ET BTP model is linear and contains only a few critical parameters requiring special consideration; • the pipeline production is regulated by the recruitment policy, with the recruitment numbers adjusted every six months based on the following: o capability needs (platforms commissioning and de-commissioning including transition provisions; new skills requirement to use new technologies etc.);
•
•
o pipeline production capacity (maximum number of trainees per course, number of courses per year, resources (including instructors) availability etc.); and o compensate the separation/ transition numbers; training at sea is phasing out gradually replaced by the training ashore, leading to a significant decrease of the training costs, associated with the use of operational resources for training purposes. The qualitative factors such as levels of confidence of newly qualified personnel when using operational equipment for the first time could also be affected; failure / separation (or transition) from the RAN ET BTP occurs at the ET ITC only; and
Step 3 To develop our SD conceptual model, we used a stocks and flow diagram (SFD) to represent the structure of RAN ET TBP pipeline (Jnitova, 2016). In a SFD, stocks represent the entities that accumulate in the system, for example, the number of trainees undertaking a course or waiting to repeat the training. Flows represent the rates of change that influence stocks, for example, trainees starting or completing a course. Developing a SFD is an essential step for understanding the problem structure. It provides basis for building a numerical SD model. Development of the preliminary BB working set is conducted using the process models of the RAN ET BTP variations presented at the previous section, as well as BBs conceptualisations proposed by Elsawah (2016). The following preliminary set of three BBs for physical asset stock domain of the RAN ET BTP is proposed for further development and validation: • Resource BB, • Training component BB, and • Streaming BB. Resource BB. This BB is a contextualised and simplified version of the generic physical asset module proposed by ElSawah (2016), which models the most likely physical pathways a particular training resource type can take. In the RAN ET BTP context it depicts resource types such as instructors, facilities and equipment, with the focus on these resources availability. The SD construct for this BB is a basic four stocks representation comprised of Potential Resource, Active, Inactive and Former Resource stocks. The processes for maintaining the Potential Resource stocks are different for instructors, facilities and equipment. ‘Potential Instructors’ stocks are an integral part of the training pipeline, with current ET trainees acquired during the recruitment process and specialists employed within the RAN are potential future instructors subject to achieving a certain phase in their career progression. ‘Facilities and Equipment’ potential stocks are either refilled from new acquisition and sustainment projects, or result from resource sharing and competing with other uses. ‘Active Resource’ is available for use for the RAN ET BTP operations, whilst ‘Inactive Resource’ is temporarily unavailable for the operations and mission conduct, which could impact the pipeline performance if not mitigated (e.g. replaced by other resource or repaired). At the end of the resource lifecycle, a resource becomes permanently unavailable for active use (separation from the Navy for instructors, and disposal/ retirement for facilities and equipment). The ‘Becoming Inactive’ and ‘Returning to Active’ rates control the flow between the facilities/ equipment resource stocks. They need to be synchronised to ensure the ‘Active Resource’ stock remains in a steady state. The rates are influenced by the activities such as repair and/ or replacement that require additional cost, resources and time. The flow for the ‘Instructional resource’ is different. ‘Active Instructor’ resource does not flow to ‘Inactive’, but to a ‘Potential’ resource stock, with the rates between the stocks are dictated by the posting cycle, assigned priorities and required availability. This complex resource dynamics calls for trade-off decisions. The external and internal drives such as competition for the additional resources would affect the availability of the additional resources if required, potentially leading to a backlog of inactive and potential resources waiting to be activated, and ultimately the pipeline becoming non-
operational in the worst-case scenario. The stock and flow representation of the BB for the ‘Equipment and Facilities’ resource is at Figure 3(a). Figure 3(a). ‘Equipment and Facilities’ Resource BB– SFD representation Training component BB. The initial conceptualisation of this BB has been conducted by Jnitova (2016). The process model development and stakeholder contribution led to the BB update. The BB depicts dynamics associated with the training components such as courses and consolidation training. A generic training BB for any course number ‘N’ includes three stocks (training waiting to commence, completed training and training waiting to repeat) and the following four rates: training failure, training repeat, training completion and separation. In reality, in the case of the basic training pipeline
under consideration, only the first course of the training continuum might result in training failure and separation, whilst all other courses present a linear unobstructed progression to the end of the pipeline. As such, the failure and separation rates for all but the ET ITT to be set at ‘0’. The model’s utility in this instance is to support decision making when selecting the pipeline variations, to project the possible consequences and efficiencies gained from selected options’ implementation. The BB could also be used to support recruitment decisions, dictated by the capability’s current and future needs and by the retention levels across the ET ranks. As it is envisioned to adopt this BB for use in the other case studies, these functions become more beneficial to a decision maker compared to the current case study. The SFD for the Training BB of the RAN ET BTP is at Figure 3(b).
Figure 3(b). Training BB – SFD representation Streaming BB. The conceptual model for the Streaming BB has been first developed by Jnitova (2016) to model streaming the ET trainees into the three specialisations that exist within the ET category, also called ‘career streams’ by the category stakeholders. The career streams are associated with their specific sets of training components and employment sub-categories. The BB allows splitting a single stream pipeline into a required number of parallel streams, to which Training Component BBs corresponding to specific courses and training activities, could be ‘plugged in’. A platform class allocation can also use the Streaming BB as it requires streaming into the class specific categories. After class streaming has been completed, trainees commence their stream specific EACs, associated with a particular class of the Navy platforms. This often occurs in parallel with the first posting, so the training is tailored to the job to be performed after completion of the class specific EACs. The streaming is often specific to the basic training pipelines, and is rarely used for the intermediate and advanced training levels. The Streaming BB SFD representation is at Figure 3(c).
Figure 3(c). Streaming BB – SFD representation The purpose, input and output for the BBs for the RAN ET BTPs are at Table 4. Table 4. BBs purpose, input and output BB Resource Availability
Training
Streaming
Purpose to simulate resource (instructors, facilities, equipment) availability in response to changes in the acquisition, retirement, maintenance and failure processes to simulate training duration and successful completion rate in response to course actual duration, delays, sequencing and failure process to simulate the pipeline split into the parallel streams
Input acquisition rates, retirement rates, failure rates and maintenance rates
Output resource availability/ unavailability
course duration, trainee numbers, failure rates, separation rates, sequencing and delays number of students ready to be streamed, streaming rates
Course/training pipeline duration per student, and a number of successful students pipeline’s parallel streams containing a desired/ not desired number of trainees
The BB set transition into the executable SD model capable of processing the input data and of producing a required output will require interfaces definition and also include development of the ‘decision making’, ‘execution’ and ‘performance measures’ domains (ElSawah, 2016). When developed, the working BB set will be presented to the Navy stakeholders for validation during our future laboratory experiments and series of the validation workshops. Test and validation of the BB working set will conclude the first case study and will prove the concept if successful. It is envisioned to follow up the first case study with the case studies for the RAN ET Intermediate, Marine Technician (MT) and Avionic Technician (AT) training pipelines.
Reflection and future research direction The paper introduces a modelling project, targeting development of a BB based SD approach to support decision making in the military context. The paper describes the project’s methodology supported by a literature review and introduces the initial phases of the project’s first iteration of the Research and Development (R&D) cycle using the RAN ET BTP case study, conducted for the purpose of proving the concept. Initial difficulties experienced when communicating with the stakeholders are overcome by introduction of the process models, which proved to be extremely beneficial to successful communication with the non-M&S specialists, and a valuable contributor to the conceptualisation process. It is envisioned that this strategy will continue to be used effectively in the future, with the process models assisting in development, validation and testing of the BB sets. The RAN ET BTP working set of BBs requires only a small number of the BBs compared to the intermediate and advanced ET training levels, as well as the other category training such as MT and AT. The other consideration is to incorporate BBs into the modelling that capture human decision making process and the seemingly irrational behavior that happens when people get into the mix, by accounting for irrationality, and for soft variables such as motivation, stress, working conditions etc. Soft variables present a challenge for modelling because numerical data is often unavailable or nonexistent, making them hard to validate and substantiate. By solving how to incorporate soft variables
into the BBs and modelling process, we may discover hidden value propositions and be able to better accommodate models for their purpose. The BB based applications are constructed in a way that is open to expansion to remove/ replace or add new BBs to the model when required. As the RAN ET BTP case study progresses through the phases of the R&D cycle, the preliminary BB will be tested and validated in collaboration with the RAN stakeholders. This project will continue with the ultimate purpose of development an M&S approach, capable to facilitate tailor-made solutions to specific problems, by selecting and combining generic BBs readily available to all potential users via comprehensive and validated BB library.
References Alexander, C., Ishikawa, S. and Silverstein, M., 1977, A pattern language, New York: Oxford University Press Ahmed, U., 1997, A process for designing and modelling with components, In Proceedings of the 1997 International System Dynamics Conference, Istanbul, Turkey, System Dynamics Society Andersen, D.F. and Emmerichs, R., M., 1982, Analysing US military retirement policies, available online at http://sim.sagepub.com/, accessed at UNSW library 19 Nov 16 Armenia, S., Centra, et.al., 2012, Military workforce dynamics and planning in the Italian airforce, In: 30th international Conference of Systems Dynamics Society, St. Gallen, Switzerland, 1: 123-160 Cavana, A.Y., Boyd, D.M. and Taylor, A.J., 2007, A systems thinking study of retention and recruitment issues for the New Zealand army electronic technician trade group, Systems Research and Behavioral Science 24(2):201-216 Creately website, at ‘http://creately.com/blog/diagrams/business-process-modeling-techniques/’, [accessed online 1 Nov 16] Dowling, A. M., MacDonald, R. H., and Richardson, G.P., 1995, Simulation of systems archetypes, In Proceedings of the 1995 International System Dynamics Conference, vol. 2, pp. 454-463, Tokyo Eberlein, E. and Hines, J., 1996, Molecules for modelers, In Proceedings of the International System Dynamics Society. Cambridge: System Dynamics Society Elsawah, S., Ryan, M., 2016, ‘A Modular Approach to Dynamic Modelling for Capability Planning’, Proceeds from the Australian Simulation Congress, SimTecT 2016 Evans J., R., 2009, Optimising training of ET in FFGs, Internal RAN research for NPCMA Workforce Planning Task Forrester, J.W., 1990, Principles of systems, Portland: Productivity Press Forrester, J.W.,1994, System Dynamics, System Thinking and Soft OR, System Dynamics Review, vol. 10, No.2-3, pp. 245-256 Gamma, E., Helm, R., Johnson, R., Vlissides, J., 1994. Design Patterns: Elements of Reusable Object-Oriented Software, Addison Wesley Publishing Company, Reading, Massachusetts Garagic D, Trifonof I, et.al., 2007, An Agent Based Modelling approach for studying manpower and personnel management behaviors, In: Proceedings of the 2007 Winter Simulation Conference Garza A, Kumara S, et.al., 2014, System Dynamics based manpower modeling, In: Proceedings of the 2014 Industrial and Systems Engineering Research Conference, pp. 3683-3693 Jacobson, J.J., Malczynski, L. and Tidwell, V., 2008, Very Large System Dynamics Models–Lessons Learned, In Proc. 2008 Intl. Conf. System Dynamics Soc., July, pp. 20-24 Jnitova, V., Elsawah, S., Ryan M., 2016, ‘Modelling Complexity of RAN Training Pipelines: A System Dynamics Approach’, Proceeds from the Australian Simulation Congress, 2016 Johnstone, M, Novak A, et.al., (a), 2015, A Multi-level approach to planning and scheduling resources for aviation training, Centre for Intelligent Systems Research, Deakin University, VIC, Australia and DSTG
Johnstone, M, Le V, et.al, (b), 2015, Modelling a helicopter training continuum to support system transformation, In: Interservice /Industry Training, Simulation, and Education Conference (I/ITSEC) Kenneth, J., S., 2012, Business process modelling with BPMN: Modelling and designing business processes course book using the business process model and notation specification, version 2.0, Admaks Training and Publications Kent, B., 1995, Design Patterns: Elements of Reusable Object-Oriented Software, IBM Systems Journal; 1995; 34, 3 Liehr M., 2001, Towards a Platform-Strategy for System Dynamics Modeling: Using Generic Structures Hierarchically, In Proceedings of the 2001 System Dynamics Conference. Atlanta, System Dynamics Society La Roche, U., 1994, A basic business loop as a starting template for customized business-processengineering, International Systems Dynamics Society, Sterling Linard K.T., Blake M. and Paterson D., 1999, Optimising workforce structure - the System Dynamics of employment planning, In: Proceedings of the 1999 Systems Dynamics Conference Luna-Reyes, L.F., Andersen, D.L., 2003, Collecting and analysing qualitative data for system dynamics: methods and models, System Dynamics Review, 19(4), pp.271-296 Markham, J.Y., 2008, Framing user confidence in a Systems Dynamics model: the case of workforce planning in a New Zealand army, PhD Thesis, Victoria University of Wellington McGinnis, M.L., Kays, J.L. and Slaten, P., 1994, Computer simulation of US Army Officer professional development, In: Proceedings of the 1994 Winter Simulation Conference McLucas, A.C., 2005, System Dynamics Applications: A Modular Approach to Modelling Complex World Behaviour’, Argos Press, Canberra, Australia Mehmood, A., 2007, Application of System Dynamics to human resource management of Canadian naval reserves, College of Engineering, U.A.E., 2007, Available at http://www.systemdynamics.org/conferences/2007/proceed/papers/MEHMO253.pdf (accessed 25 June 16) Myrtveit, M. and Bean, M., 2000, Business modelling and simulation, Peer reviewed journal Wirtschaftsinformatik, 2000, Vol.42(2), pp.156-160 Powers, R., 2011, An Object-Oriented approach to managing model complexity, Master thesis, University of Bergen, Norway Orfali, R., Harkey, D., and Edwards, J., 1995, The essential distributed objects survival guide, John Wiley & Sons, Inc. Paich, M., 1985, Generic structures, in: System Dynamics Review, Vol. 1, No. 1: 126-132 Richmond B., 1985, STELLA: software for bringing system dynamics to the other 98%, In Proceedings of the 1985 International Conference of the System Dynamics Society, Keystone, CO; 706–718 Senge, P., 1990, The fifth discipline: the art and practice of the learning organization, Random House Business Books, London Skraba, A., Kljajic, M., Knaflic, A. et.al., 2007, Development of a human resources transition simulation model in Slovenian Armed Forces, Available at: http://www.systemdynamics.org/conferences/2007/proceed/ (accessed online on 7 Mar 16) Szyperski, C., Bosch, J., & Weck, W., 1999, Component-oriented programming, In European Conference on Object-Oriented Programming, pp. 184-192, Springer Berlin Heidelberg Thomas, D.A., Kwinn, B.T., McGinnis, M. et.al., 1997, The U.S. Army enlisted personnel system: a System Dynamics approach, In: Proceedings of the 1997 IEEE International Conference on Computational Cybernetics and Simulation, pp. 1263-1267 Tustanovski, E., Baj, M.P. and Vrankic, I., 2015, Testing the sustainability of the Croatian military forces: A system dynamics approach, Croatian Operational Research Review, CRORR 6: 55–70 Wang, J., 2007, ‘A System Dynamics Simulation model for a four rank military workforce’, Land Operations Division report, DSTO-TR-2037
Valentine, E., 2011, ‘Effective simulation studies using domain specific simulation building blocks’, thesis in logistics an economics, Technical University of Delft, Gildeprint Drukkerijen Verbraeck, A. and Valentin, E.C., 2008, Design guidelines for simulation building blocks, In Proceedings of the 40th conference on winter simulation, pp. 923-932, Winter Simulation Conference Verbraeck A.; Y. Saanen; Z. Stojanovic et.al., 2002, ‘Chapter 2: What are building blocks?’, in: A. Verbraeck; A. Dahanayake (eds.) Building blocks for Effective Telematics Application Development and Evaluation, pp.8-21
Biographies LCDR Victoria Jnitova is qualified education specialist with an extensive experience in vocational training as the Training Systems Officer in the Royal Australian Navy. Last year she has completed with excellence her Masters of Systems Engineering degree at UNSW@ADFA. Currently she is a part of the Navy Modelling and Simulation team with her main focus being on the Defence M&S strategic policy and M&S acquisition projects. Victoria Jnitova also co-authored a conference paper focused on M&S practical application in the Military training context. Dr Sondoss Elsawah is a senior lecturer at the University of New South Wales. Her research program focuses on understanding the behaviour of large complex problems and systemic risks that arise from the interactions between social, ecological, and technological systems. She is an expert on the application of systems thinking and system dynamics modelling methodologies. She has published widely including journal articles, conference papers, and book chapters. She has attracted research grants from Government and industrial agencies in Australia and overseas, including the Australian Research Council (ARC) and US National Scientific Foundation (NSF). Dr Mike Ryan is the Director of the Capability Systems Centre, University of New South Wales, Canberra. His research interests include project management, modelling for systems design, systems engineering, and requirements engineering. He is the author or co-author of 11 books, three book chapters, and over 160 technical papers.