A prototype data-driven extended enterprise modeller and simulation builder has .... Model Builder. Supply. Chain. Database. Data entry/inquiry. Display supply.
18th International Conference on Production Research
Data-driven Simulation of the Extended Enterprise Bing Cao, Richard Farr, Mike Byrne and James Tannock Nottingham University Business School, Nottingham, NG8 1BB, United Kingdom
Abstract With the ever-increasing complexity and dynamics of supply chains in an extended enterprise, simulation has become a powerful tool to assess supply performance. In traditional simulation approaches, a model is constructed by a user, who defines it step-bystep via the user interface of the simulation tool. However, if the model configuration needs to be changed, this must be carried out by a simulation expert, thus greatly limiting the usefulness of simulation to non-expert users. To resolve this, the concept of data-driven modelling/simulation has been adopted. This has been made possible by the relatively limited domain nature of a supply chain. In a data-driven simulation, the model is constructed automatically by a software program, based on data existing in company IT systems. The model created this way can be reconfigured rapidly, by changing the external data. Hence, a user could explore the implications of radical changes to a simulated extended enterprise, despite having little knowledge of the simulation software itself. A prototype data-driven extended enterprise modeller and simulation builder has been developed by the authors. This is written in Visual Basic and includes codes to generate an Access database, a supply chain data model and an Arena simulation model. The paper outlines the approach, the performance metrics adopted, and presents a case study using semi-fictional data from the European civil aerospace sector. Key words: Data-driven, modelling, simulation, supply chain, extended enterprise, Arena, Visual Basic
1
INTRODUCTION
Over the last few decades, the nature of competition in many business environments has been changing, form competition between companies to competition between supply chains. A single company often cannot satisfy all customer requirements, including fast-developing technologies, a variety of product and service requirements, and shortened product life-cycles. Such developing business environments have made companies look to the supply chain as an ‘extended enterprise’, to meet the expectations of end customers. A number of definitions of the supply chain have been proposed. Christopher [1] defined it as, “a network of connected and interdependent organisations mutually and co-operatively working together to control, manage and improve the flow of material and information from suppliers to end users”. According to Johansson [2], one of the most common perceptions of the supply chain is, “a system whose constituent parts include material suppliers, production facilities, distribution services and customer linked together via the feed-forward flow of materials and the feedback flow of information”. Supply chains do not always behave as expected or desired. Excessive demand variability – due to information distortion in the supply chain, between one member and the next – can become a serious problem, and this led to some of the early studies of supply chain behaviour. Forrester [3] initiated the analysis of demand variability amplification and pointed out that it is a consequence of industrial dynamics, the time-varying behaviours of industrial organisations. In addition to demand variability and information distortion, other main issues in supply chain management relate to the uncertainties within the supply chain system. There are many sources of uncertainty in a supply chain. Davis [4] identifies three sources of uncertainties:
Supplier uncertainty measured in terms of suppliers’ on-time performance, average lateness and degree of inconsistency; Manufacturing uncertainty that arises due to process performance, machine breakdown etc; Demand or customer uncertainty arising from forecasting errors, irregular orders etc. Lee and Billington [5] claim that one of the potential pitfalls in managing supply chains is failing to understand the likelihood and the magnitude of impact of these uncertainties. Reiner and Trcka [6] argue that the main objective of problem-solving methods in supply chain management is to reduce uncertainties. Coupled with greater complexity and globalisation of supply chains, uncertainty brings ever greater supply chain risk. It is thus very important for organizations and supply chains to have the abilities, both at supply chain design and operation time, to be responsive to risks in order to achieve supply chain robustness and resilience. Simulation offers an ideal tool to analyse supply chain variability and risks. At the design stage, simulation would allow different supply chain configurations to be tested against different realisations of future supply chain scenarios. It would also allow limitations of the extended enterprise to be explored. When operated, it would allow supply chain management concepts, performance and risks to be quickly assessed. With traditional simulation approaches, the model is constructed by a user, who defines it step-by-step using the user interface of the simulation tool. This model is then run with different data scenarios and results analysed for these scenarios. However, if the model configuration needs to be changed, this must be carried out by a simulation expert, thus limiting the usefulness of simulation to non-expert users. This is true when a hypothetical collaborative network of businesses need to be constructed, based on a product’s
bill of materials and lead times, etc. An extended enterprise is also likely to change over time; partners come and go as different levels of inputs are required at different stages of a project, and different processes and logistic concepts may become viable at different volumes of manufacture. To improve the utility of simulation when applied to the extended enterprise, the concept of data-driven modelling and simulation has been adopted. In a data-driven simulation, the model is constructed automatically by a software program based on user input data. The advantage of a data-driven simulation approach is that the model can be reconfigured rapidly, by changing the external data. Hence, a user could explore the implications of radical changes to a simulated extended enterprise, with little knowledge of the simulation software itself. Data-driven simulation would mean that a database of supplier information would be interrogated, and a corresponding set of nodes created within a network model of a potential supply chain configuration, each node with appropriate properties, representing the relevant capabilities of a supplier. A prototype data-driven extended enterprise modelling and simulation builder has been developed by the authors. This paper outlines the approach, the performance metrics adopted, and presents a case study using semi-fictional data from the European civil aerospace sector. 2
DATA-DRIVEN SIMULATION
Data-driven modelling and simulation has been used in financial analysis, engineering, management and supply chain management, and is gaining popularity, especially in domain specific modelling. However, in this relatively new area, authors have expressed different concepts of the nature of data-driven modelling and simulation. Solomatine [7] describes data-driven modelling approaches to the management and control of water resources. Datadriven modelling is based upon the analysis of all the data characterising the system under study. A model can then be defined on the basis of connections between the system state variables (input, output and internal), with only limited knowledge about the “physical” behaviour of the system. D'Souza and Allaway [8] describe a data-driven modeling approach to product level decision support. They define data-driven modeling as a process of model-building wherein models are created that fit the dynamics of the data rather than assuming a priori relationships among brands and their marketing mix elements. Based on a combination of time-series and econometric modeling methods, these models can significantly improve a modeller's ability to capture marketplace structure and dynamics. Although more complex than their predecessors, the capabilities of these new data-driven decision-support models make them potentially very powerful tools, improving intuition and managerial understanding of marketplace dynamics while suggesting improved decision alternatives. A somewhat different concept is described by Darema [9], who introduces the concept of ‘dynamic’ data-driven simulation systems. He claims that this approach represents a new paradigm for application simulations, where the applications can accept and respond dynamically to new data injected at execution time, and suggests measurement methods, where the application systems will have the ability to dynamically control the measurement processes. Darema
considers that the ‘synergistic and symbiotic feedback control-loop’ between simulations and measurements can open new domains in the capabilities of simulations with high potential pay-off, and will create applications with new and enhanced analysis and prediction capabilities. Clark & Cash [10] consider that that simulation modellers need to develop domain-specific models (as opposed to generic products) that are data driven and only require minimal simulation knowledge so long as the users apply these models in a specific domain of application. They present a data-driven simulation of a network using manufacturing blocking to control work-in-process. The model is object oriented and constructed in C++ using an object-oriented simulation environment called HOSE. An important feature of the model is a sequential procedure for determining run length to achieve statistical precision in an output performance measure. The principal conclusion is that the data-driven network simulation can be applied by users with minimal simulation knowledge as a rapid modeling tool. 3
STRUCTURE OF THE DATA-DRIVEN SIMULATION SYSTEM
A data-driven supply chain simulation would mean that a database of the bill of materials (BOM) and supplier information would be interrogated, a simulation model be created, the simulation would be run and the results exported and analysed. Figure 1 shows the data-driven modelling and simulation architecture for the prototype system. This has been written in Visual Basic, and includes codes to generate an Access supply chain database, a supply chain data model and an Arena simulation model. Data entry/inquiry
Display supply chain
Supply Chain Model Builder
Arena Simulation of Supply Chain operations
Supply Chain Performance Reporting Supplier Data (Collaborat -ion Hub)
Supply Chain Database
Figure 1. The data-driven modelling/simulation architecture At the centre of the architecture is the Supply Chain Model Builder (SCMB), an application which integrates all the other components. Figure 2 is a snapshot of the SCMB with example data. At the top is the menu bar where all other activities are controlled and accessed. The left pane is the graphical representation of the data model, for a selected subset of the whole product supply chain (to be explained later on in the section) based on the user’s requirement. On the right, the top pane lists BOM component details for the selected product, whilst the bottom pane shows the supplier performance table for the components of the selected product. A data entry/enquiry form (not shown) allows the user to import data from various sources, to manually add new data to the system or to modify existing data. It also allows the user to make various queries on the database.
18th International Conference on Production Research
codes using the Arena VBA object model. The data-driven Arena model creation mainly concerns the supply chain network structure and the number of suppliers. The Arena model thus created is a hierarchical model of three levels. Figures 3, 4, and 5 show the three levels of a prototype Arena model without control mechanisms. The top level represents the selected supply chain (Figure 3) The second level represents individual suppliers and transport between suppliers (Figure 4) The third level represents assemblies and routing between assemblies (Figure 5) Figure 2. A snapshot of the SCMD main screen display The product BOM, as shown in Figure 2, represents the hierarchy of components, sub-assemblies and assemblies that make up the final product. This is the primary data framework for the supply chain modeling process. The BOM data is currently obtained from company ERP systems. ERP systems (such as SAP) allow the export of data in ASCII text formats, which can be directly imported into the SCMB application. Such an approach could be implemented by a single company that wished to simulate the behaviour of its supply chains, for different products. However, this research focuses particularly on the extended enterprise scenario, where BOM and supplier data will be obtained from a collaboration hub (an internet/intranet portal for document, data, application exchange and integration). The data requirement from this source will be an extension of the traditional BOM, in the sense that it should also contain supplier names, capability and performance data for each component. Supplier data should include product lists, material and product Inventory levels, effective capacities, production rates and costs, interconnections between suppliers, transportation lead times and costs with associated variability etc. All the data and results for the data-driven system are stored in a staging database such as Microsoft Access (the supply chain database box in Figure 1). This stage database can be created and populated using functionality provided by the SCMB. The tables and queries are created by the execution of SQL commands in external text files; hence the database design is independent of the SCMB code. Once data has been imported or entered and validated, a subset of the data can be chosen, for generation of a data model of the required supply chain. This data model is a representation of a supply chain (or part of a supply chain) using appropriate data queries generated by the SCMB application from user selections. It is graphically represented by the TreeView control in Visual Basic (see Figure 2). There are two basic types of supply chain data models which may be selected; product supply chain models and supplier supply chain models. The former is an extended BOM with representations of alternative suppliers and optional configurations (see Figure 2). The latter displays supplier relations for material flows, also in a TreeView format. After the user has selected the supply chain or part of the supply chain of interest and created the corresponding data models, a simulation model based on user specified parameters is created. This is accomplished by Visual Basic
Figure 3. The model of a data-driven supply chain in Arena
Figure 4. A supplier sub-model In this prototype model, each supplier is modelled as a set of assembly or processing operations. Transport is modelled as delays with costs for each product and pair of suppliers. Each sub-model is built with a facility to handle incorrect arrivals, so as to help ensure that the simulation can run to completion. This is one of the mechanisms required in a data-driven simulation system, where the issues of model validation and robustness must be carefully considered. There is danger of inconsistencies such as ‘orphan’ components arriving at a sub-model, and having no defined handling mechanism.
Figure 5. An assembly process sub-model After model creation, control is handed over to Arena, where the model can be run just as if it was created manually. At the end of the run, the results and statistics are exported back to the database, to be viewed and analysed using the SCMB application. 4
PERFORMANCE EVALUATION
Having created a model and run simulations of a potential or actual supply chain, it is necessary to have a means of assessing its performance. This requires a strategic understanding of the relative merits of cost reduction, leadtime reduction, reliability (on-time delivery performance), robustness (time to resume a steady state after disruption), etc. Since some of these goals may be conflicting, and their importance varies in different business scenarios, it was necessary to establish a system of weighting, allowing a user to specify the goals for a multi-criteria optimisation process. There is considerable literature and many proposed models of supply chain performance measurement. The performance evaluation system was designed taking account of the performance attributes and metrics of the SCOR (Supply Chain Operations Reference) model [11] and uses several categories of metric. 4. 1
Metric calculation and aggregation
Supply chain performance is calculated and aggregated from the simulation results, using the following sequence: Collection of relevant results for multiple replications of the simulation process Calculation of statistics for individual metrics in relevant periods (e.g. means, confidence intervals) Setting of performance targets Application of weightings to each metric category, as appropriate Normalisation of metrics in each category Aggregation of metrics in the same category, e.g. lead times, costs Calculating the overall supply chain performance using the metrics scorecard The business requirement in each category may be different. For example, lead time should be short and component cost low, but the reliability and robustness metrics should be high. Similarly, the range of possible performance will vary between metrics.
The normalisation method is based on a linear 0–10 scale, as this is intuitive and enhances readability and interpretation of metrics. When normalising scores, category targets are assigned appropriately between 0 and 10, and results are then scaled according to the target position and range of performance for each category. Consistency is recommended when using normalised scales, and hence higher values will always mean superior supply chain performance, regardless of the metric under study. Aggregation typically requires calculating an average of the normalized scores for each category. So, for example, various costs associated with materials, manufacture, transportation, etc., can be aggregated into a single cost metric. This may or may not be a weighted average. Agreement on the relative importance (weightings) of the individual metrics for the aggregation process will be developed in conjunction with industrial partners during this research. 4. 2
The supply chain metric scorecard
A supply chain scorecard, based on the Balanced Scorecard (BSC) concept, will be created to display the aggregated and normalised metrics. The BSC (Balanced Scorecard Institute [12], Rohm [13]) has attracted a lot of attention in recent years, as a means of broadening performance measurement systems to enhance understanding of business performance. The BSC approach takes account of metrics associated with issues like operational excellence, integration, customer perspectives, internal business processes, organisational growth, learning, and innovation, as well as the traditional financial metrics. The BSC approach is widely used to link strategy with operational activity measures, balancing the financial and non-financial perspectives and driving accountability down though the organisation. It also emphasizes that to develop meaningful performance measures, one has to understand the desired outcomes and the processes that are used to produce these outcomes. The BSC was originally focused and used at corporate level but is increasingly used for multi-enterprise evaluation of the extended enterprise, which can also benefit from this approach. It is proposed to adapt and simplify the approaches as described by Rohm [13] using a two layer structure. 5
CASE STUDY: AEROSPACE LOGISTICS
The aerospace industry offers particular opportunities to test the usefulness of a data-driven simulation approach, due to the complexity of the product, the high cost of holding inventory, and the complex interrelationships between businesses that can operate at several levels in the supply network. With the increasing emphasis on the extended enterprise, complexity can be expected at every stage in the supply chain, and analysis should yield benefits in terms of competitiveness for the enterprise as a whole. It is commonly assumed that upstream processes are somehow simpler, or less deserving of study than those nearer the customer, but these processes nevertheless have a great bearing on the overall performance of the extended enterprise. Because of the confidential and proprietary nature of much of the information in this area, a partially fictional version of
18th International Conference on Production Research
the supply chain for a civil turbofan engine manufactured by a leading UK company, has been developed to provide a dataset for system testing and verification. This dataset has been used to provide the diagrams used in the paper. A considerable amount of information is already in the public domain, in the form of news releases relating to formulation of partnerships and the award of contracts, etc. Some BOM data were drawn from company websites and publicly available sources, whilst other data were invented by the authors for testing purposes. A three-level prototype Arena model based on this dataset, showing only material flows, has been shown in Figures 3, 4 and 5. Supplier performance data are estimated from data obtained on previous research projects with partners from the Aerospace industry, and from industry general knowledge. In the prototype model, only processing and distribution lead times and costs are considered. These use the above performance data as most likely values in triangular distributions. Material arrivals use exponential distributions with mean values depending on demand. The model-building process operates successfully. Whilst operating, the SCMB software will generate error messages to alert the user to data conditions requiring a decision – for example where more than one supplier exists for a single component. Messages are also generated where data are duplicated, or missing, or where redundant elements (suppliers, etc.) are found in the database. Thus the model building activity forms part of the model validation process. Preliminary testing of Arena simulation models generated automatically using this system has been conducted by the authors, using the test dataset described above. Running the generated model in the Arena environment produces a large volume of individual performance metrics. At the time of writing, these have not been fully formulated in the way described in Section 4. However, an example of raw output is illustrated in Figure 6, for the end product engine, where the x-axis is the month and the y-axis represents the number of engines produced. The first few months show a warm-up period for the simulation, and thus show an atypical level of output. There are numerous limitations and simplifications to the data used at present, including a lack of resource limitation data at companies in the supply chain, efficiencies, quality issues (scrap and rework), demand, calendar and order schedule information, etc.
Engines completed
Supply Chain Output 35 30 25 20 15 10 5 0 1
2
3
4
5
6
7
8
9
10
11
12
Month
Figure 6. End product output for the prototype model 6
FINDINGS AND RECOMMENDATIONS
Although at an early stage of the project, some findings and recommendations can be shared here:
7
Data-driven simulation can be a useful tool given that supply chains are generally similar but their behaviours and impact differ widely. Given the significant effort required to develop such a simulation approach in a new domain, it is sensible to first determine the necessity of datadriven simulation. Data-driven simulation is a not a replacement for general-purpose simulation tools. It is only worthwhile if the system developed will be repeatedly used and its structure is relatively stable. Pre-determination of project scope. Should the whole supply chain or some sub-section of the supply chain be considered? This is dependent on the viewpoint of the user (systems integrator, component supplier, etc) and also security and confidentiality issues, such as access to other suppliers’ data, perhaps via a suppler portal in an extended enterprise collaboration hub. Pre-processing of data for errors (e.g. redundancy trapping) may be important, because of the potential model validity issues. So far as possible, inconsistencies within the database should be handled without demanding excessive human intervention, since even automated model construction process may be a lengthy process. It is vital to select a simulation tool that has a model-building macro language capable of dynamically creating and configuring complex models. The tool should also be able to integrate seamlessly with other database and reporting tools. A modular or object based approach is very useful in creating the simulation model using a macro language.
FORTHCOMING WORK
No control or planning and scheduling mechanisms have been included in the data-driven supply chain models created at the time of writing. The system currently builds models composed only of companies that use a simple push scheduling method, based on MRP philosophy. A JIT (just in time) philosophy has not been applied, but is needed to model the behaviour of some companies in the supply chain. In addition, the concurrent commitments of each manufacturing system are not yet taken into account, and each company in the supply chain starts with an empty factory. These issues will be addressed by the authors as the research proceeds. For enhanced model detail and accuracy, it will be necessary to add more detail for suppliers, especially where their capacity is tightly constrained. Safety stocks may also be introduced for participants in the supply networks, as appropriate. After incorporating such control mechanism and details, animation will be added to the Arena model. One of the key issues in data-driven simulation of the supply chain relates to model validity. Model validation is an important issue which takes on new dimensions when using a data-driven simulation approach. The authors intend to address this area in future publications.
ACKNOWLEDGEMENTS The work presented is part of the EU framework 6 VIVACE project. The authors acknowledge the funding from the EU and collaboration from our industrial partners Volvo Aero and MTU. The supply chain model builder was developed by a former MSc student Chang-Seop Kim of the University of Nottingham. REFERENCES [1] [2]
[3] [4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Christopher, M. (1998), Logistics and Supply Chain Management, Financial Times. Johansson, M. (2002), The impact of supply integration and information flow on supply chain performance, University of Nottingham thesis. Forrester, J. W. (1961), Industrial Dynamics. Waltham, Mass., Pegasus Communication. Davis, T. (1993), Effective Supply Chain Management, Sloan Management Review, Vol. 34, No. 4, pp. 35 – 46. Lee, H. L. and Billington, C. (1992), Managing Supply Chain Inventory: Pitfalls and Opportunities, Sloan Management Review, Vol. 33, No. 3, pp. 65 – 73. Reiner, G. and Trcka, M. (2004), Customized supply chain design: problems and alternatives for a production company in the food industry. A simulation based analysis, International Journal of Production Economics, Vol. 89, pp. 217 – 229. Solomatine, D. P. (2002), Data-driven modelling: th paradigm, methods, experiences, Proc. 5 International Conference on Hydroinformatics, Cardiff, UK. D'Souza, G. and Allaway A. (1997), A data-driven modelling approach to product level decision support, The Journal of Product and Brand Management, Vol.6, Issue. 2, pp. 130. Darema, F. (2000), Dynamic Data Driven Application Systems (Symbiotic Measurement & Simulation Systems), “A new paradigm for application simulations and a new paradigm for measurement systems”. NSF sponsored Workshop. Clark, G. M. and Cash C. R. (1993), Data-driven simulation of networks with manufacturing blocking, Proceedings of the 1993 Winter Simulation Conference, G. W. Evans, M. Mollaghasemi, E.C. Russell, W.E. Biles (eds.). Supply Chain Council (2004) SCOR Overview Version 6.1, retrieved from http:// www.supplychain.org/SCOR. The Balanced Scorecard Institute (2004), What is the Balanced Scorecard? http://www.balancedscorecard .org. Rohm, H. (2002), A Balancing Act, Perform, Vol. 2, Issue 2.