An Approach and Interface for Building Generic Manufacturing Kanban-systems Models

Simulation of manufacturing systems, historically the first major application area of discrete-event process simulation, is becoming a steadily more proactive and important strategy for achieving manufacturing efficiency. Concurrently, lean manufacturing has become a nearly essential corporate strategy to compete successfully in an increasingly austere and global business environment. Furthermore, industrial engineers responsible for supporting successfully competitive manufacturing operations have less and less time available for manipulating details deep within a simulation model in order to evaluate numerous complex alternatives. Convergence among these trends motivated the development of a generic manufacturing kanbansystems simulator that has Kanban inventory optimization capability, and an accompanying interface, described in this paper. 1 BACKGROUND AND MOTIVATION Improvement of manufacturing systems was one of the earliest significant applications of discrete-event process simulation analysis, and has consistently been one of the largest (Law and McComas 1999). Recently, improvements to manufacturing systems have stressed the importance of achieving lean production (Duggan 1998); manufacturing systems contributing to lean production reduce inventory via just-in-time techniques, reduce space requirements by shortening distances parts must travel, and reduce costs by elimination of non-value-added activities associated with inventory and material-handling (Heizer and Render 2001). Indeed, three of the eleven essential steps to eliminate waste, identified by (Cary 2002), are inventory reduction, motion reduction, and transportation reduction. Hence, there arises an “almost instantaneous demand to see its [simulation’s] benefits extended as far and as quickly as possible.” (Hartwig 2001). Selecting the kanban container size and the number of kanbans for each part type in a kanban system has been investigated in the past under deterministic and stochastic conditions (Monden and Aigbedo 2001; Askin and Goldberg 2002). Simulation is an effective tool in finding the proper values of these parameters in a stochastic environment. Timely and effective introduction of technology, such as process simulation, into a manufacturing organization, is one necessary step to achieving high competitive ability (Banerjee 2000). Increasing the efficiency and ease of use of input and output interfaces to simulation models, thereby saving time and reducing error risk on behalf of industrial and process engineers, contributes greatly to the successful integration of process simulation into ongoing process improvement (Krug 2001). An extensive survey of simulation usage versus non-usage in German industry identified complexity and difficulty of use as a significant barrier to the application of simulation technology (Hirschberg and Heitmann 1997). In this paper, we first describe the modeling context motivating the development of a generic kanban-system model and its data-input interface. We next describe the interface, the model, its built-in kanban optimization algorithm, typical user execution of the model, and representative model outputs. Last, we discuss plans for enhancement of the model and its outputs, and summarize the current status of this work. 2 OVERVIEW OF THE MODELING CONTEXT The generic model and its interface apply to a manufacturing or assembly system whose workstations are supplied by joint implementation of “Call” (“Electronic Kanban”) and “Card” (“Card Kanban”) systems. The Call system is conceptually responsible for sending signals (presumably electronic) to a warehouse or marketplace which supplies the manufacturing or assembly system. The Call system is responsible for sending these signals when inventory levels Williams, Ülgen, and DeWitt along the manufacturing or assembly line fall to a defined “trigger reorder” point (“Signal Kanban”). The Card system is responsible for the generation of kanban cards and the conceptual transformation of these cards into material delivery to line workstations from the warehouse or marketplace, in keeping with the usual definition of kanban cards as representing authorization to begin work (Hopp and Spearman 2001). The user of the model, typically an industrial, production, or process engineer, is presumably concerned with averting line stock-outs, reducing inventory levels within the constraint of averting line stock-outs, and avoiding congestion among the transport devices (e.g., tug trains) supplying the line via defined itineraries. The user will concurrently wish to lower the number of transport trains, relative to both the Call and Card systems, consistent with these objectives, due to both the capital costs and the operational costs of such material-handling equipment. 3 STRUCTURE OF THE MODEL INPUT INTERFACE The model input interface is an Excel© workbook containing seven worksheets. Within all worksheets, the user arranges information concerning one entity within rows, and hence places information of the same type within columns. The following list provides a summary of the seven worksheets and the purpose of each: 1. StationInfo, using four columns, specifies relationships between raw material usage points and raw material delivery locations. 2. StopInfo, using three columns, specifies aisle segments and relationships between those segments and origins of material. 3. KBInfo, using nine columns, specifies usage points, usage rates, packaging information, and delivery routes for raw material. 4. RouteDesc, using five columns, defines routes used by the delivery trains. 5. TrainSched, using five columns, specifies number, lengths, and schedules of delivery trains. 6. Misc, using fourteen columns, specifies miscellaneous logic flags and delay times (such as loading and unloading delays). 7. TransferInfo, using one column, specifies a list of aisle segments blocked to trains while skillets are transferred between assembly lines. Additionally, the workbook contains macros which check the input data for errors or internal inconsistencies and alert the user to any problems. For example, if the user inadvertently defines a raw material usage point but this usage point appears on no routes, the user will be warned of this discrepancy. Color coding, shading, and embedded comments within the worksheets increase the ease and reliability of their use. 4 STRUCTURE OF THE SIMULATION MODEL The simulation model is built in WITNESSTM (Thompson 1996), and thus supports run-time animation as well as simulation. The model contains two large arrays of WITNESSTM “elements.” The elements in one array are WITNESSTM “buffers” representing workstations which consume parts as raw material. The WITNESSTM “machines” in the other array represent stop-points along delivery routes where trains pick up kanban cards and/or deliver needed parts upon demand. Relative to the model, the inputs within the Excel workbook sheets specify interrelationships (e.g., which stop-points supply which workstations) by using the indices of the two part arrays. This model, significantly, uses neither WITNESSTM “tracks” nor WITNESSTM “vehicles.” The secondary advantage of this abstinence is greater model execution efficiency. This efficiency supports correct model usage, inasmuch as engineers are not tempted to run too few or too short replications (in the statistical sense) to shorten model execution time. More significant still is the ability afforded the user to modify the routes (either which workstations are supplied on which routes or which stop-points are on which routes) within the Excel workbook, with no need to revise the WITNESSTM model. 5 BUILT-IN KANBAN OPTIMIZATION ALGORITHMS This simulation model has an additional feature that creates kanbans (kanban cards and containers) whenever a station starves for a part type. The option for automatic generation of kanbans is triggered in the simulation based on user input at the beginning of a run. If the model is run under kanban-generation mode, the user, by starting with a low number of kanbans for each part type, can actually optimize the number of kanbans in the system. Since the kanbans are generated whenever needed, the lowest amount of kanban inventory will be achieved and thus identified by running the simulation in this mode. 6 USER EXECUTION OF THE MODEL After creating and checking the workbook described in Section 3, the engineer opens the WITNESSTM model, as described in Section 4. The user then assigns a warm-up period and a run length, and also chooses whether to run the model in “Advance” mode (animation provided) or “Batch” mode (faster execution without animation). During a run with animation, workstations (i.e., raw material usage points) appear as solid squares and stop points (i.e., raw material delivery points) appear as hollow squares. In Williams, Ülgen, and DeWitt “Advance” mode, typical animation niceties (such as identification of machine status by color and updating of variables in simulated time) are also provided.