Evolvable Production Systems and Impacts on Production Planning Hakan Akillioglu, Mauro Onori Department of Production Engineering The Royal Institute of Technology Stockholm, Sweden [email protected], [email protected] Abstract— Production planning and control strategies have been changing in line with the constant change on product and customer requirements, under the light of technological and scientific advancements. Production systems which are based on mass production became obsolete in time hence companies, being profit oriented, are in need of new solutions towards mass customization to handle rapidly changing market conditions. To deal with this issue, production systems and production planning strategies have to be complementing each other. In this paper Evolvable Production Systems and its compatibility to Just in Time (JIT) Production compared to Material Requirement Planning (MRP) will be discussed. Keywords- Evolvable systems; process oriented planning; autonomous systems; production planning

I.

INTRODUCTION

Both Material Requirement Planning (MRP) and Just in Time (JIT) are basically aiming to organize the material planning and flow in a manufacturing environment. MRP has been used and benefited long time in an environment where products’ lifespan was longer, product varieties were less and volumes were higher. A system which is planned by MRP considers a manufacturing system and its stations as a box that requires input and extracts output. In fact the processes inside this box, its requirements and essentials are one of the determinant factors for the control of a production system. JIT on the other hand is essentially a philosophy which is composed of a statement of objectives that guides managing production system [1]. From the shop floor point of view, JIT used advantage of Single Minute Exchange of Dies (SMED) where the main purpose was decreasing the setup time which causes the manufacturing system to stop. This gave the opportunity to decrease lot sizes which in turn caused less work in process (WIP) inventories in the shop floor. Less WIP brings various advantages in a manufacturing environment which will be discussed. As in the SMED example, production system characteristics and processes have key effects on production planning and control. Evolvable Production Systems (EPS) [2] was first introduced in 2002 targeting highly adaptable mechanical and control solutions that can enhance reusability and interoperability of modules, enabling life time extension of the modules and short deployment times at the shop floor level. In this paper the basics of MRP and JIT will be explained and

how EPS affects production planning and control strategies will be addressed. II.

MATERIAL REQUIREMENT PLANNING

There are 2 essential philosophies in order to move material through the line which are push and pull production control systems. The main characteristic of a push system is that once the product is processed, it is pushed to the next station without considering if it is needed or not. Main disadvantages include large amount of inventories, high rate of scrap before diagnosing the error and the requirement of large and complex database maintenance. Besides, the production control issues are handled hierarchically and centrally, which can create difficulties in case of frequently changing system configurations in the way that a small change in the upper level may cause increasing effects on the lower level plans. Characteristically, MRP is a push system because the basic structure has a hierarchy starting from the top level long term planning and branches out until the lowest level. The inputs of a MRP system are master production schedule (MPS), on-hand inventory information, scheduled receipts and bill of material (BOM) as in Fig. 1 [3].

Figure 1. MRP inputs and outputs

The important point here is how MPS is created. There are various factors taken into consideration through the preparation of MPS which are customer orders, forecasts of future demand, safety stock requirements, seasonal plans and internal orders from other parts of the organization [4]. Although there are a variety of methods to be used for forecasting future demand, the important issue here is that the reliability of them is low considering the dynamicity of the market, product requirements and customer expectations. MPS stands at the top of a MRP system and it is where the unpredictable environment turns into a deterministic

(forecasted) situation. With the MPS on hand, resolving the planned order releases depends on stepwise procedure and calculations in MRP. To start with, netting is the step where net requirements are determined for each planning period by subtracting the on hand inventory and scheduled receipts from gross requirements that are specified in MPS. In lot sizing; according to the lot sizes of a specific product the periods are identified where new orders are required to be arrived. Lead time is required to be known for offsetting step in which the orders that are determined in the previous step are basically moved to previous periods according to the lead time to find out the planned order releases. Planned order releases are peculiar to a product. At this point the quantity and release dates of the product are determined and ready for applying to the components of the product. Bill of material includes the required information about the components of the product. MRP starts with collecting information for each end item to supply data for the preparation of MPS over the planned time period which is generally long. This data is not only used to prepare MPS but also to plan capacity and investment on the long term. Early investment on capacity can cause significant losses. Considering the current production systems capabilities and equipments, companies have to account for long term forecasts and investment strategies according to these forecasts. The 3 phases shown in Fig. 2 are narrowing the time period and application area from top to down where phase 1 includes long term decisions for the production line or company, phase 2 conclude information which has determined information for different periods and finally phase 3 has detailed planning information of shop floor. A problem arises when the plan in phase 1 is supposed to be revised and changed. In this case a small change in phase 1 may end up big modifications in the other phases. Other shortcomings of a MRP system are caused by uncertainty in the market and production environment, capacity infeasibility, fixed lead time and system nervousness, which are addressed below.

Uncertainty exists and cannot be predicted completely beforehand. External uncertainties are related with the market conditions, customer interests, technological shifts whereas internal uncertainties are various kinds of disturbance in the production system, change in the management strategies etc. To be exact, production planning systems and production systems are performing in an environment which is unpredictable, random, continuous and dynamic [5]. Taking precaution cannot be a reasonable strategy due to the reason that there might be some events whose likelihood of occurrence is very low but in case of occurrence the results could be fatal. Besides there might be critical parameters whose deviation is considerably high. Due to these conditions fixing some variables depending on the estimations before setting and running the production system includes substantial risks. Properties and abilities of the production system come into the picture again at this point. A production system which is not capable of reacting to reconfiguration needs, caused by design change of the product, capacity adjustment or frequent changeover, obliges the planner to foresee the production needs for a long period. Capacity infeasibility of MRP grounds on the fact that MRP does not consider capacity as a constraint whereas it assumes that capacity of the production system is able to meet the demand for each period. In the application of MRP, planners handle this issue by increasing working hours or number of shifts or alternately revising the MPS. What is more, capacity might seem to be sufficient at the high level of planning in MRP; however, it can possibly turn into a problem when MPS is translated into shop floor schedules. Fixed lead times are used at the offsetting step in MRP, which is in fact varying depending on the disturbances in the production system. To overcome this issue, either more safety stocks are kept on the shop floor or lead time is increased by some percentage. Increase of lead time results in early production of the orders which in turn causes high finished goods inventory. Both of the situations result in higher inventory levels which results in lack of responsiveness and higher inventory costs. System nervousness is another shortcoming of MRP which is explained as big effect on the low level shop floor schedule of a small revision in MPS. Revision in MPS might be required in some cases like updating forecasts, failure of equipment, problems with the suppliers, personnel problem etc. The number of shortcomings of MRP can be increased. Some of these issues are covered partially in MRP II however it is difficult for rigid MRP systems to cope with the uncertain nature of production environment. III.

Figure 2. MRP system

JUST IN TIME

On the other hand, in a pull system, the material flow starts with the introduction of a demand. The main difference between push and pull system is the order which is triggering the production. In push systems start time and quantity of production is determined beforehand and the system follows this schedule whereas in pull systems production starts with the information coming from the downstream station or machine. Just in time has been introduced by Toyota as a characteristic

application of a pull system. JIT is described as the basis of Toyota production system where the right parts are needed in assembly line at the time they are needed and only in the amount needed [6]. It is important to emphasize that JIT reflects a management strategy rather than a physical system [7]. This necessitates new requirements from the operational and production system point of view. The underlying factors behind the applicability of JIT are Single Minute Exchange of Dies (SMED) and the information flow by means of kanban system. SMED, in essence, is developed for the purpose of reduction in changeover time between different product variants. Main idea is separating the setup time into two parts which are internal and external setup. External setup includes the setup operations which are possible to be conducted throughout the production system runtime whereas internal setup operations require the production system to cease. Hence, the more the setup operations are transformed into external setup, the shorter the production system stops during changeover. This critical improvement in the shop floor, namely reduction of setup time, created opportunities to the planning system by means of enabling to decrease lot sizes so work in process inventories. Low level of inventory makes some problems visible like long machine downtime, high scrap rates, vendor delinquencies and so on. The second factor is the contribution of kanban cards to the control of WIP. With the implementation of kanban cards the flow of information in the manufacturing system goes in reverse direction to the flow of material. By adjusting the number of cards, the WIP level can be kept under control. With these advantages JIT systems are aiming seven zeros [8], zero defects, zero lot size (lot size of 1 in reality), zero setups, zero breakdowns, zero handling, zero lead time between the stations and zero surging. These targets are extremely demanding and cannot be accomplished without improving the production system. Unless fundamental changes are implemented to the way that the production systems are built up, these targets cannot be realized. According to Polito et. al. [9] JIT has 5 shortcomings. First one is external obstacles which reveal the inability of JIT systems to respond to increasing rates in demand and high rate of variance in product. Second point is logistic issues. JIT is highly dependent on the coordination with the suppliers because a delay in one of the critical components can result in the stop of the whole system. Behavioral constraints are arising from the fact that JIT assumes employees are motivated and perform at best when entrusted with increasing responsibility and authority. Another point is intractable accounting system which depends on the fact that traditional accounting and financial measures generally tend to defeat JIT objectives. Finally, small suppliers resist to JIT because they do not benefit from it due to reason that they lack economies of scale that their high volume manufacturing customers have. In fig. 3 the production planning steps of a pull system is specified which is applicable for JIT [3]. At the top of the plan, forecasting feeds the strategic long term capacity and workforce planning. Subsequently MPS and WIP strategies are decided and applied to shop floor level. Throughout the capacity planning there are a number of critical factors which

are taken into account for the decisions. For example, the quantity and the type of equipment which has to be purchased should be decided. Another point is the question of how long will the products possibly be produced. Since the product lifecycles are shortening, the equipments which are specific to the product are in risk of being redundant. In fact, JIT distinguishes from MRP in the shop floor level, by the ability of managing WIP levels and efficient flow of material. For the long term plans, JIT is also dependent on forecasts and assumptions to be able to invest to production equipments and workforce.

Figure 3. Pull planning system

Although JIT and MRP today have different approaches and methods, the focus in this paper is how the production system characteristics affect the application of them at different levels of planning. IV.

EVOLVABLE PRODUCTION SYSTEM

As far as the limitations which are met by production planning and production systems are concerned, some similarities can be observed. The real problem was considered as uncertainty and solutions are improved on this focus. Since complexity and dynamicity of the environment is very high, foreseeing of future conditions may include considerable amount of restrictions and assumptions. This leaves substantial room for unexpected behaviours in the designed system therefore it is highly probable for such systems which are designed to meet a portion of future states to fail. Considering the current paradigms, flexible production systems [10] targeted to encapsulate the required abilities and functions of one or a few similar product families in one unit. It is a costly solution since end users are obliged to purchase extra process capabilities even though they will not need. In case of new product requirements which are beyond the available capabilities are introduced, configuration of the system became costly and/or time consuming or even impossible. Furthermore flexible production systems were

inadequate to meet the requirements with its centralized and hierarchical control system [11]. Modular production systems aim to overcome the challenges of the market by building flexible production systems from standardized modular machine elements. The module categories are composed of four classes, namely process machine primitives, motion units, modular fixturing and configurable control systems [12]. Once the manufacturing system reconfigurability is targeted, the control architecture has to be given priority since the reprogramming takes considerable fraction of reconfiguration. Unless the modular elements are combined with distributed control systems, which can be realized by embedded computational power to the modules, the real reconfigurability cannot be reached. Reconfigurable production systems went one step further and developed modular approach to the production equipment for the purpose of increasing reusability. This enabled the system to have flexible capacity and functionality together with reusability [13]. The key characteristics of a reconfigurable system are modularity, integrability, convertability, diagnosability and customization. The reasons why reconfigurable systems could not achieve true reconfigurability are; firstly, intelligent and distributed control systems were not benefited in the production environment, secondly reconfigurable systems were focusing only on production environment and did not evaluate the whole enterprise with product design, organization, operation etc. and finally modularity of the equipments could not enable short deployment time without supplying autonomous components. Holonic manufacturing systems introduced the concept of holon which stands for any unit that encloses an information processing part and optionally a physical processing part. These units are classified as resource holons, product holons and order holons [14][15]. There are 2 fundamental characteristics of a holon which are autonomy and cooperative behavior. The control structure of holonic systems lies between fully hierarchical and heterarchical structures to compensate a weak point of fully heterarchical control systems, which is the inability to guarantee a certain global behaviour and performance [16]. Comparing to the current paradigms the aim of EPS has not been to develop limited flexibility or barely physical reconfigurability with central control mechanisms [17]. The focus of EPS paradigm is to achieve overall system adaptability by modules which are dedicated to specific processes with the capability of short deployment time at shop floor without reprogramming effort. Modules are the basic building blocks with the characteristics of being process oriented and embracing embedded computational power in order to enable autonomous behaviour based on multi agent system architecture. In EPS approach intelligent modules form a dynamic network for the purpose of dealing with production system requirements by means of collaborative behaviour. An Evolvable Production System envisages three key aspects: process-oriented mechatronic modules, fine granularity and intelligence at component level.

A. Process oriented modularity Process oriented modular structure enables the modules to be re-used effectively at different configurations since the product characteristics change very often however the processes needed are quite stable [18]. Hence with a detailed ontology it is possible to translate product requirements into processes. In traditional approaches there is one way flow from product design to production system which means that product requirements are imposed to production system by leaving freedom to product designers and assigning complexity to production systems. Product oriented focus results in product dedicated equipment which becomes obsolete after lifecycle of the products comes to end. Changing market conditions forces traditional system to repetitive equipment investment for the upcoming product requirements. Hence, same processes are invested more than one time in respect of new product needs. Investment to product specific equipments might be misleading in short term because very low re-usage of the equipment results in redundant investment in long term (Fig. 4).

Figure 4. Traditional system design

Process oriented approach is a fundamental cornerstone of Evolvable Systems, being supported by skill-based modules (Fig. 5).

Figure 5. Process oriented system design

There are 2 types of skills which are atomic skills and composite skills. Composite skills are composed of atomic skills to create coalitions for more complex tasks. Modules can be used lifelong by being configured according to changing

product requirements. Process dedicated modules bring the opportunity for the designers to be aware of the available processes which are the ground of two way information flow between product design and production system [19]. Comparing to traditional equipment investment, EPS modules have long term advantage considering their reusability for different products. B. Fine granularity The EAS approach is distinctive comparing to other paradigms on achieving fine granularity at component level with agentified components (Fig. 6). Granularity can be described as the level of complexity of the component which is building the system [20]. The higher the properties of modularity are exhibited in assembly system levels, the coarser the granularity is. Extending or shrinking a system by reconfiguring cells is a case of granularity on cell level. The feasibility of device level granularity has been tested on an EAS during the EUPASS project [21], resulting in the development of a MASMEC cell

Figure 6. Production system levels

Even though modularity at the device level enables physical adaptability; control system has to be even more adaptive in order to achieve full system adaptability. C. Intelligence at component level To achieve autonomy in the system, modules must have computational power together with standardized interfaces. Computational ability inside the modules allows them to selfconfigure and to eliminate reprogramming effort whereas highly developed interfaces enable modules to communicate and negotiate effectively to be able to create coalitions to perform operations which are not possible to be handled by the skill of a single module. Emergent behaviour, which can be described as the properties that emerge from the coalition of modules, is playing essential role in Evolvable Systems. This means that addition of a module to another module or to a coalition may contribute with more than its basic skill. The EAS control structure is based on Multi Agent System (MAS) characteristics. There are two main reasons: MAS already holds the principle and structure of interaction within a society of individual agents, and emergence is best handled with multi-agent approach [22]. Each agent has three abilities; (1) being aware of its environment with the help of communication capability enabled by interfaces, (2) evaluating

the information obtained by its embedded computational ability, and (3) acting appropriately. V.

EPS AND PRODUCTION PLANNING

Fundamental changes in production systems which are carried with architecture of Evolvable Production Systems have the potential of yielding a shift in the production planning and capacity planning strategies since constraints coming from the production system insufficiencies are relaxed to a large extend. From the capacity planning point of view, the investments is done under the light of long term capacity predictions for both push and pull systems. The trade off for companies at this point is the choose of either early investments, which bring opportunity costs together or the risk of late reaction in case of sudden changes in the product requirements, which causes loss of market share. For the sake of company future and competitiveness it is more preferred to invest for long term. As far as production system is concerned the main reasons standing behind are long system integration time, the replanning effort and production system infeasibility to changes. Optimality desire is a misleading element causing to narrow down the focus on short term product oriented decisions. With the intention of shifting the focus from products to processes EPS is targeting to a new philosophy which is; optimize over time, not for any given state. System approach behind this purpose comprises 3 stakeholders which are system integrators, module developers and end users. From the shop floor level production planning point of view one of the major contributions of EPS paradigm is minimizing, even eliminating in some cases, the changeover time by means of modular structure which have embedded intelligence and ability to adapt to system without reprogramming effort. This creates extensive opportunities for shop floor planning. To be clear, setup time elimination in a system gives the freedom to achieve lot size of one which in turn enables to have no WIP in the production flow. WIP is a burden to the system to respond to frequent changing product requirements even for a fully reconfigurable system. EPS architecture enables the production systems to have considerably shorter system integration/setup time and minimized/eliminated changeover time for different product requirements with the condition that the necessary modules (processes) are available in the repository. For both push and pull systems, the availability of short system integration and changeover time at the shop floor will enable to shorten the planning span and increase fast reaction without taking high investment risks. Besides the concern will shift from forecast dependent long term plans to short term and more precise planning. As well as, the early investment of the end products can be modified to shorter term more frequent investments according to the need which bring a range of advantages such as elimination of the opportunity cost which is arising from early use of money (Fig. 7). Although this kind of incremental investments is a typical outcome of modularity, the real challenge is the system integration and setup time to make the system start instantly.

REFERENCES [1] [2]

[3] [4] [5]

Figure 7. Investment strategies of EPS and traditional systems

[6]

Multi agent system based control architecture in EPS enables agents to collaborate in order to fulfill the requirements of product by communicating and negotiating. Routing of the products is decided as a result of communication between the product agent and resource agent. If the resource agent which has demanded service is available the product goes there by means of another collaboration with the transportation agent which can be a conveyor, AGV or similar. This enables the flow of the product to dynamically alter according to the availability of resources. However, transfer of products by agents might not be as effective as a traditional conveyor system considering the transfer time and resource utilization. This issue is possible to overcome under the condition that the resource planning is performed properly. On the other hand, no WIP inventory, unlike a traditional system, in EPS compensates this by shortening the throughput time.

[7]

VI.

FUTURE RESEARCH

The production systems which are designed with the aim of performing at optimum for a specific production state have to be configured and planned every time a change in the system is required, which brings about a considerable amount of cost, time and effort. Although optimality is possible to achieve, in the dynamicity of the environment and market, for most products it is rarely advantageous to delay the production for the sake of running the system in optimum. The trade off at this point is between system responsiveness and production optimality for certain states. EPS serves to the system responsiveness more than system optimality, with the philosophy of optimization over time, not for any given state. Research will follow by developing a production planning methodology which is feasible with the structural essentials of EPS. Process capabilities and characteristics will be at the centre of this method instead of product specific equipments and machines characteristics. The shop floor level planning will be performed by the multi agent system structure whereas the capacity planning will be conducted by the help of virtual process repositories where the process specifications and capabilities with detailed operational information will be available.

[8] [9]

[10]

[11]

[12]

[13]

[14]

[15] [16] [17]

[18]

[19]

[20]

[21]

[22]

O’Grady P.J., “Putting just in time philosophy into practice: A strategy for production managers”, London: Kogan page, 1988. Onori M., “Evolvable Assembly System – A new paradigm?”, Proceedings of the 33rd International Symposium on Robotics, 2002, Stockholm, Sweden. Hopp J.W., Spearman L.M., “Factory physics”, Mcgraw hill, 3rd ed., 2008. Nahmias S., “Production and operation analysis”, Mcgraw hill, 5th ed., 2005. J.M. Wing, “Modeling unpredictable or random environments,” AAAI 2001 Fall Symposium, Technical Report FS-01-04, AAAI Press, North Falmouth, Cape Cod, MA, November 2001, pp. 144-145. Ohno T., “Toyota production system: Beyond large scale production”, Productivity press, 1988. Benton W.C., Shin H., ”Manufacturing planning and control: The evolution of MRP and JIT integration”, European Hournal of Operational Research 110, pp. 411-440, 1998. Edwards J.N., “MRP and Kanban-American style”, APICS 26th Conference Proceedings, pp. 586-603, 1983. Polito T., Watson K., “Just in Time Under Fire: The Five Major Constraints Upon JIT Practices”, The Journal of American Academy of Business, Cambridge. 9(1), pp. 8-13, 2006. Elmaraghy, H.A., “Flexible and Reconfigurable Manufacturing Systems Paradigm”, International Journal of Flexible Manufacturing Systems 17, pp. 261-267, 2006. Onori M., Semere D., Lindberg B.; "Evolvable Systems : an approach to Self-X Production"; Proceedings of the DET/CIRP 2009 conference on Digital Enterprise Technology, Hong Kong, Dec. 2009. Rogers G.G., Botacci L., “Modular Production Systems: A New Manufacturing Paradigm”, International Journal of Intelligent Manufacturing, Volume 8, Issue 2, pp. 147-156, 1997. Mehrabi M.G., Ulsoy A.G., Koren Y., “Reconfigurable Manufacturing Systems: Key to future manufacturing”, Journal of Intelligent Manufacturing 11: pp. 403-419, 2000. Bussman S., “An Agent-Oriented Architecture for Holonic Manufacturing Control”, First Open Workshop IMS Europe, Lausanne, Switzerland, 1998. Valckenaers P., Brussel V. H., “Holonic Manufacturing Execution Systems”, Annals of the CIRP 54/1: pp. 427-430, 2005. Scholz-Reiter B., Freitag M., “Autonomous Processes in Assembly Systems”, Annals of the CIRP 56: pp. 712-729, 2007. Akillioglu H., Neves P., Onori M., “Evolvable Assembly Systems: mechatronic architecture implications and future research”, 3rd CIRP Conference on Assembly Technologies and Systems, Trondheim, Norway , 2010. M.Onori M, J. Barata, “Evolvable Production Systems: mechatronic production equipment with process-based distributed control”, Proceedings of the IFAC/SYROCO Conference, Gifu, Japan, 2009. Gawel. A., "XPRES: Production integration in conceptual design", Study thesis, XPRES project, School of Industrial Engineering and Management, Dept. of Machine Design, KTH, Stockholm, Sweden, january 2011. Maffei A., Onori M., Neves P., Barata J., “Evolvable Production Systems: Mechatronic Production Equipment with Evolutionary Control”, Emerging Trends in Technological Innovation 314: pp. 133142, 2010. EUPASS-Evolvable Ultra Precision Assembly, EUPASS Adaptive Assembly Roadmap 2015 deliverable 1.5f, Project Report-Public Document 1.5f, NMP-2-CT-2004-507978, 2008. Barata J., Matos C.L., Onori M., “A Multiagent Based Control Approach for Evolvable Assembly Systems”, 3rd IEEE International Conference on Industrial Informatics (INDIN), pp. 478-483, 2005.