A conceptual framework for choosing a card-based pull production and
control system; a case study
M.R. Wagenaar, Dr. Ir. W. Klingenberg (Supervision), and Dr. M.J. Land (Co-assessor) Faculty of Economics and Business, University of Groningen,
Master thesis Technology Management
Abstract
This article discusses the key concepts of card-based pull systems and compares them with the original conditions for pure pull systems. Many studies have been performed on this topic, while none compares all card-based systems in order to present organizations a clear decision model. In order to get a clear comparison, nine essential conditions for pure pull, presented in the literature, are elaborated. Subsequently, the key concepts Kanban, CONWIP, POLCA and Cobacabana/WLC are elaborated and compared. The similarities and differences between the key concepts resulted in a conceptual framework for choosing the right card-based pull system. The framework contains two sections, one with entrance conditions and one based upon system features which can be compared with the features of an organization. This framework provides not just information about the feasibility of different key concepts, but delivers also accurate information about the current state of an organization. The functionality of the framework is tested at Schneider Electric Manufacturing, this company is looking for ways to improve its planning and control system.
Keywords: pull production, production planning and control, card-based, conceptual framework
1. Introduction
Less work in progress (WIP), shorter lead times, reliable due dates and less rescheduling is the aim of pulled Production Planning and Control (PPC) systems within organizations. This should result in a clear overview on the production floor and in clear decisions of order intake and release by planners. Companies which are searching for some sort of pull system most of the time opt for a system which can be implemented with minimal software support; especially small organizations (Land, 2009). This could be reached by introducing authorization cards. The pull authorization mechanism can be implemented physically without difficult software support, for example by introducing a limited amount of cards which can be attached to orders or by limiting the part buffers on workstations (Riezebos, 2006).
In the literature many systems are presented, but none compares all different card-based systems. Some articles are focusing on all planning and control key concepts and provide specific applicabilities, even though these do not present clear decision criteria (Zapfel et al., 1993; Stevenson et al., 2005). In these articles it becomes obvious that differences between the systems do exist, but how these relate in order to assist organizations in choosing, is not clear. Articles usually focus on the differences of two systems (Spearman, 1990; Germs et al., 2008), are working with a base system (Dogger, 2010), or creating a justified system for a special case (Bokhorst, 2010).
to present such a decision framework, whereby the different card-based systems are compared with the basic conditions of the traditional pull system. Key concepts of these card-based production planning and control systems are: Kanban, CONWIP, POLCA and Cobacabana/WLC. All these systems provide their authorization signals with help of cards and are, in some way, pulling orders through the production area. Each system has other characteristics and applicabilities to certain circumstances. The usability of this outcome, basic conditions and a decision framework, is tested at Schneider Electric Manufacturing (SEM). SEM produces MTS-MTO (30%-70%) cable management products for the market; this contains all products from wall trunking to cable trays. Schneider Electric has introduced a program to stimulate continuous improvements: the Schneider Production System (SPS). The SPS program is familiar to the widely known Toyota Production system; these are programs for continuously improving performance with the use of Lean manufacturing and Six Sigma. Pull systems are one of the priority areas, but are currently not implemented at SEM.
In order to define the systems on a good basis a theoretical framework will be elaborated. This will be done in section 2, starting with a discussion of the three different PPC approaches: push, pull and hybrid. These widely known terms are used in so many contexts that a definition for each term will be presented, in order to use it unambiguously. Hereafter, the required conditions for a successful implementation of a pull system will be elaborated. Finally, in this section, the different card-based systems will be described and compared with each other based on the needed conditions per system. In section 3, a conceptual framework will be proposed with a set of basic conditions and a framework for choosing the “right” card-based pull system. In section 4, the usability of the developed framework is tested at SEM. Finally, section 5 gives the conclusions and some remarks.
2. Theoretical framework
2.1 Push, Pull and Hybrid
Roughly we can distinguish three different planning approaches: push systems, pull systems and, a combination of the two, a hybrid push/pull production system (Spearman et al., 1992; Ghrayeb et al., 2008; Siha et al., 1994). The terms refer to the means for releasing jobs into the productions facility, push systems push their orders into the plant based on forecasts, pull systems pull the products through the plant based on customer demand. There is a lot discussion about what the right definition is for these widely used terms, descriptions of pull and push are used in different ways and not always in the same context.
2.1.1 Push
2.1.2 Pull
The pull production system was founded by the idea that by “a reduction of cost through eliminating waste” the lowest production costs can be achieved (Sugimori, 1977). A system which allows these features is needed, in order to reach this kind of production. This contains features like producing with the minimum amount of equipment, materials, and other resources. Toyota came up with the just-in-time philosophy, where production is balanced and waste is eliminated. Within the system only necessary parts are produced by using just the processes, equipment and materials needed. A few conditions are necessary for this just-in-time system, described for the first time by Sugimori (1977): signal approval by subsequent processes; one piece production and conveyance; and leveling of production. In order to strengthen these three conditions a pull system is needed, according to Sugimori (1977). Toyota practices the meanwhile well-known Kanban system, but more pull systems are developed.
Looking at the three main conditions of Sugimori, we can denote that the final process, the last station, always gives authorization to preceding processes to start production. The last station knows the customer demand for final goods best. In other words, pull systems drive productions based upon customer demand (Ghrayeb et al., 2008). The final station pulls products downstream through the plant in order to finalize the part and make it ready for delivery to the customer. Each station in the process can be seen as an isolated station with its own supplier (the upstream station) and its own customer (the downstream station) (Ghrayeb et al., 2008). Eventually the conditions on the floor decide when a new order may enter the floor. Now it is becoming clear why more conditions are required, because by executing this kind production system, preceding processes must be able to produce flexible and demand has to be leveled in order to keep work load within boundaries. The number of cards or other signal types can restrict the amount of work in progress in order to level production (Spearman et al., 1992). We will define a pull system, in the most original way, as pure pull: a system characterized by the practice of downstream work centers which pulls limited levels of stock from previous operations, in order to meet customers demand without a planned master schedule.
2.1.3 Hybrid
In order to overcome the limitations and disadvantages of both systems, integration is suggested. The combination will result in a hybrid system, which makes use of the strong features of push and pull (Ghrayeb, 2008; Flapper et al., 1991; Huq et al., 1994; Benton, 1998). In push systems high inventory costs and WIP explosions can occur, but in return a low delivery lead time can be expected. Pull systems instead, can expect high delivery lead times in return of low inventory cost and better controllable processes. Applying a hybrid system will compromise the adverse features of both systems.
There are mainly two types of integration distinguished, vertical and horizontal integration (Ghrayeb, 2008). Vertical integrated systems consist of two levels; upper level production orderings are controlled by push and a lower level controlled by a pull production system (Ghrayeb, 2008). In this type MRP, upstream in the supply chain, is normally applied for production planning in a push way. A pull based system is applied on the “floor” to control and for execution in production. Horizontal integrated hybrid systems are systems where the production is pushed through the first stations and at a certain point will be decoupled and pulled further. This variant could be interesting when, for example, a plant produces a variety of products, which are built upon the same semi-products (platforms).
2.2 Pull conditions
There are several conditions needed for a plant to implement a pure pull production system successfully. When these conditions are not met, a pull system can be implemented, but it will not pay off the benefits it is expected to bring. Note that some of the conditions are important for every existing system, not specific for pull systems, but they are essential in making a pull system a success. Nicholas (1998) describes nine necessary conditions, Monden (1983) describes six conditions instead, to come to a successful implementation of pull. Both are displayed in table 1. Monden’s conditions are more general compared with the nine described by Nicholas. For example “autonomation” of Monden can be explained by two conditions of Nicholas, namely “planning and control responsibility” and “developing cooperative work attitudes and teamwork”. To be as complete as possible we will work out all nine conditions for pure pull. Based on these conditions it will be possible to compare the different card-based pull planning and control systems.
Nicholas (1998) Monden (1983)
1. Planning and control responsibility Autonomation 2. Production emphasis on producing to meet demand.
3. Reducing work in process inventories and unnecessary stock
4. Equipment preventive maintenance
5. Quality assurance efforts must be aimed at preventing defects from happening.
Standardization of work
6. Setup times must be small. Short setups 7. Plant lay-out must facilitate linking of all operations into the
process
Proper machine layout
8. Production plans and schedules must be somewhat uniform. “smooth” production involving a stable product mix
9. Developing cooperative work attitudes and teamwork Improvement activities
2.2.1 Planning and control responsibility
Shop supervisors and work teams know what is happening on the shop floor, therefore they are better in taking the right decisions at the right moment. For this reason they should be responsible for the planning and control according to Sugimori (1977). He describes this aspect as “self-display of workers’ ability”, where he emphasizes the attention for the respect to humanity of the workers in production environments. Making workers actively participate in running and improving their processes will make it possible for them to fully show their capabilities, which could be beneficial for motivation. Further, the responsibility of informing the workers of the priority of orders and the visibility of the state of these, the authority of job dispatching and overtime should lie by the foreman, without intervention of the management team (Sugimori, 1977).
2.2.2 Production emphasis on producing to meet demand between workstations
By focusing on customer demand, instead of forecasted production, overproduction will be ruled out and will not create that type of waste anymore. Waste like overproduction is one of the seven wastes described by Toyota (and Canon), and is seen by them as the worst waste. Overproduction raises the production costs and moreover it hides the causes of waste (Sugimori, 1977). Original pull systems pull products through the sequencing work stations, which makes the flow transparent. Wastes like unbalance between the workers and between processes, problems in processes, worker’s idle time, overcapacity and or insufficient maintenance will be noticed earlier. Producing to meet demand between workstations will rule out overproduction, stock and makes WIP controllable and is therefore a condition for a pure pull system. However, according to Darlington (2007) pull production is not in every sector feasible. For example, in the food industry where lead times are long and the demanded reaction times are short, is pull production not suitable, Darlington suggests for these industries a hybrid system. Besides, Buzacott (1994) emphasizes the need for safety stocks in a lot of industries, in order to overcome fluctuations in demand and other influences in production like frequent machine breakdowns. Also Ghrayeb et al. (2008) points out the danger of pull systems, where he suggests the dangers in exploding backlogs and, inherent too late deliveries.
2.2.3 Reducing work in process inventories and unnecessary stock
possibility of reducing WIP and inventory is a condition, although the other conditions are mainly responsible to achieve this. Nevertheless it will stay a condition in this article.
2.2.4 Equipment preventive maintenance
In general three kinds of maintenance can be distinguished: reactive, preventive and predictive based maintenance (Smith, 2001). Reactive maintenance is, as its name suggests, only carried out when equipment fails. This maintenance strategy is expensive, because it can hold up production for a long time when key equipment is broken down. Moreover, costs are higher than they normally by planned maintenance will be, because there has to be given priority to this kind of maintenance. Preventive maintenance schedules on regular basis maintenance on critical equipment. It will be done on a calendar basis or on a production time/amount of the machines, this will reduce the risk of unexpected breakdowns and increase the equipment availability. In addition, when breakdowns or wear of machines can be traced with the help of computers these could be predicted. Predictive maintenance can replace the preventive one and reduces waste in lowering the costs of parts and labour, because (now) only the needed parts are replaced. Therefore, preventive and predictive maintenance can reduce costs and improve equipment reliability and availability (Smith, 2001). It will be obvious that when organizations begin to work more with pull their reliability also has to increase, a preventive or predictive maintenance strategy should be implemented.
2.2.5 Quality assurance efforts must be aimed at preventing defects from happening
Since pull systems carry minimal buffers and allow no defective items to proceed through the process, defects and quality issues will halt production flows. Therefore, there are basically two topics covered here; defect detection will be done in order to make sure that the products delivered to the customers are perfect and defect monitoring has the aim to improve quality and prevent quality issues from happening (Nicholas, 1998). These two topics, with the emphasis on defect preventing, can be subscribed to the methodology of Total Quality Management (TQM). Within the TQM philosophy the integrated approach is important, quality should be a top management issue and continuous improvement efforts, together with the zero error objective, should be companywide and also beyond the supply chain boundaries. This includes self-assessments through internal quality audits, focus on administrative routines for quality improvements, as well as quality improvements in design and manufacturing operations (Soderquist, 1999).
2.2.6 Setup times must be small
2.2.7 Plant lay-out must facilitate linking of all operations into the process
Every work station must be seen as an element of a linked, synchronized system, in order to support the flow within the plant. For a smooth flow every station or process needs the same capacity and they have to be able to produce at the rate of the final station. For the applicability of a pull system, the shop floor configuration is a major determining factor (Nicholas, 1998; Oosterman, 2000).
2.2.8 Production plans and schedules must be somewhat uniform
A pure pull system, like Kanban, needs uniform schedules because large variations and seasonal patterns have to be anticipated and are leveled in advance to minimize the impact on upstream processes. This seems contradicting with the aim to hold as less inventory as possible, but is necessary because a Kanban system with a given number of cards can only effectively deal with demand variations in the range of only plus or minus 10% (Nicholas, 1998). Other pull systems cope better with these kinds of fluctuations, like Paired Overlapping Loops of Cards with Authorization (POLCA) and Workload Control (Stevenson et al., 2005). These systems release work with more freedom in the release opportunities.
2.2.9 Developing cooperative work attitudes and teamwork
Some of the production planning and control will be done on the floor by the workers teams, as described in the first condition; therefore, it is important that they can cooperate. By making the workers more cooperative and give them more authorization, they will also feel more responsible. This results in more improvements suggested by the workers and also certain actions are better understood (Sugimori, 1977). Management’s role regarding workers is important too; managers must provide workers with the opportunity to develop and expand their skills and utilize their capabilities (Sugimori, 1977; Nicholas, 1998).
2.3 Different kinds of card-based pull systems
In order to point out the differences between the different card-based pull systems, each is elaborated. Resulting in a set characteristics which is compared to the traditional conditions for pure pull systems. The following ones are investigated:
Kanban (Sugimori et al., 1977)
CONWIP (Spearman et al., 1990)
POLCA (Suri, 2003)
Cobacabana/WLC (Land, 2009)
2.3.1 Kanban
From the first descriptions by Sugimori and Ohno of Kanban in its purest form, are currently many variations known in order to make it more applicable in various organizations. In fact, researchers agreed that pure pull systems, Kanban according to Sugimori (1977), are not feasible in most real life organizations. Pure Kanban needs a continuous flow, can only handle a limited amount of different parts, can only deal with few setups and needs low demand variability in order to become effective, as described in the conditions for pure pull systems. Hence, this system is for MTO and high variety organizations, which have high routing variability and less repetitive processes, difficult to implement. Besides, Spearman et al. (1992) proves that the limited work in progress is contributing to the benefits of such systems, not the pulling through the whole system. After these recognitions systems as CONWIP are further developed.
2.3.2 Constant work in progress (CONWIP)
Constant work in progress is a release method for continuous work on the floor. Cards are regulating the flow of work, but in this system they are not part specific as in Kanban, instead they are job number specific. These cards are traveling with the products or batch the whole length of a process, instead of being pulled between each process, like Kanban (Spearman et al., 1990; Stevenson et al., 2005). Spearman et al. (1992) concludes that CONWIP can generate more throughput than Kanban, because it is naturally focused more on WIP. Consequently, CONWIP provides a more stable production environment by constantly controlling WIP over the line, instead of controlling WIP at each station.
CONWIP is less restricted in the essential conditions compared to pure pull. By using CONWIP a stable product mix is less necessary, because this system is controlling WIP only for a whole line or a certain routing. Any part/product that uses that certain routing can be brought into the system and can receive a signal for authorization. Different products and parts could be in the system, as long as they have the same routing. Instead of Kanban which is using WIP for each station, effective control would be impossible with a whole range of different parts. Controlling the line is easier with CONWIP, as the line is only limited in WIP, instead of WIP amount decisions for each station by Kanban. Furthermore, predicting the performance of the line is easier, because the queuing models needed for these lines are better and simpler than queuing models for each station (Spearman et al. 1992). Modeling and optimization is therefore easier with CONWIP than for Kanban.
Variants as m-CONWIP were developed for different routing possibilities (Hopp et al., 1996). However, with a higher variety of products within this system, the number of cards can get quite high, which makes it difficult to control. The cards control the WIP, therefore they should have a similar work load, which means that products still should be more or less the same (Stevenson, 2005). CONLOAD was developed to overcome the load drawback, although it (still) cannot handle high routing variation and needs accurate lead time information (Rose, 1999). Another disadvantage of CONWIP is the free flow in the line, meaning that inventory within the lines is not controlled. Apparently, it is possible to build up WIP for slower machines or machines with frequent breakdowns (Dogger, 2010). At last it is obvious that production plants with variable routings cannot or very difficultly implement a system like CONWIP. The basic CONWIP system uses a single loop covering all resources in the production line.
control is per station, therefore this condition less important. Besides, the condition of pulling between work stations is less applicable. In this system the line is WIP controlled, not controlled at every work station. But the controlled WIP can be brought in several times in a line, therefore this one remains a necessary condition for CONWIP.
2.3.3 Paired cell overlapping loops of cards with authorization (POLCA)
POLCA is a hybrid push-pull material control system that combines the features of card-based pull (Kanban) systems and push (MPR) systems, with emphasis on reduction of lead times, reducing product costs and shorter feedback loops in rework (Suri, 2003). This hybrid system is developed to overcome the limitations, which are known for push and pull driven systems, as described in previous sections.
Within POLCA the flow of orders through different cells is controlled by a combination of release authorization and production control cards (Suri, 2003). POLCA controls the flow of work between cells (Riezebos, 2010). Problems in normal systems can arise in insufficient synchronization of the processes between cells, which results in waiting times between the completion of a job in one cell and its release in the other. Also some cells can face a lack of work which can be done; known as starvation. The combination results in an unbalanced factory. By the capacity signal of POLCA cards, the card signals available capacity downstream and starvation will be avoided (Riezebos, 2010; Suri et al., 2003).
POLCA works as follows: a company first need to create cells, in these cells the focus must lie in similar processes for parts. Orders will be released through a MRP system with a certain routing and these will flow, if capacity is available, over the floor. Orders can be different in size and routing, with paired cells forms this not a problem.
At the first sight POLCA may seem familiar with the releasing of jobs within Kanban, but there are some key differences. POLCA is only used to control movements between cells, not within these cells, and POLCA cards are assigned to a pair of cells. The last aspect makes different routings possible, instead of Kanban at which cards are assigned to every specific product (Suri et al., 2003). However, this system entails also some disadvantages; for example orders with large differences in work load could disrupt the flow of the system. Therefore, as well as for CONWIP also for POLCA load-based variants are developed; General-POLCA (Fernandes, 2006) and Load Based-POLCA (Vandaele, 2008).
2.3.4 Cobacabana/WLC
Workload Control (WLC) is a system specifically designed for the MTO industry, it uses a pre-shop pool of orders to reduce waiting times on the floor and hence makes it more controllable with shorter queues (Stevenson, 2005). The jobs must be selected from the pre-shop pool for release, based upon available capacity. Once a job is released, it stays on the floor until it is finished (Oosterman et al., 2000). By using the pre-shop pool the performance of the plant is stabilized, because it will be less dependent on the variability of the incoming orders and jobs will be only released if enough capacity is available. These pre-shops make WLC a hybrid system, the floor pulls new orders, but where this pre-shop is situated is dependent of the type of organization (e.g. before or after a platform). Jobs are released when the critical work stations have capacity. Moreover, in WLC the capacity of (critical) work stations is calculated and on the basis of throughput time of orders are these released in to the plant, making sure that capacity of the critical stations will be respected.
Cobacabana (Control of Balance by Card-Based Navigation) is developed special for the small job shops, which have limited software control and consequently demanding a card-based system. Cobacabana is developed on the fundamentals of WLC, as described above. The Cobacabana system uses card loops between critical work stations and the planner, who is performing the releases on all critical work stations (Land, 2009). The release cards authorize the planner to send new orders into the plant, each order gets the right amount of cards for each work station before it enters the floor. An order can receive more than one card of a work station, since it is representing the capacity of that work center; a big order takes more capacity than a little one and takes more cards. Capacity of each work center is expressed in percentages and summed up to 100%. The task of the planner is better manageable, through a simple overview generated by the released capacity and the available capacity. Such overviews make it possible for the planner to decide which order to release and also to identify easily bottlenecks (Land, 2009).
The WLC sounds obvious in theory, although there are some problems in practice. Several authors briefing longer lead times when a pre-shop pool is used (Stevenson et al., 2005). Causes of these differences are probably due to the limited autonomy of the foremen, limited standardization and constraints by labour unions. Furthermore, feedback delays can play a role here, because WLC has no direct short feedback loops, which does not warn previous stations in time. These drawbacks are concerning WLC, improvements are found in WLC-based system SLAR (superfluous load avoiding release) (Land, 2004). The card-based variant of WLC, Cobacabana, is not (yet) empirical documented, so further research is needed.
2.4 Differences
Based on the traditional pure pull system conditions, the different systems from the previous systems are validated. Table 2 shows the results of the needed conditions for each system.
Needed conditions Conditions
Kanban CONWIP POLCA Cobacabana
1. PPC responsibility
2. Pull between work stations
3. Pos. reducing WIP/inventory
4. Prev. maintenance
5. Quality
6. Small setup times
7. Lay-out
8. Uniform planning
9. Teamwork/ autonomous
TABLE 2. CONDITIONS PER SYS TEM
3. Conceptual framework
3.1 Features
The key concepts have four conditions in common, according to the comparison of the different systems (table 2); we can call these the four basic conditions for pull. These conditions need a certain level of attention, performance and have to be continuously improved before any pull system could be implemented successfully.
The first basic condition is represented by reducing work in progress and reducing inventories, this seems logical due to the pull character of all systems. Within the more hybrid approaches, organizations should be able to respond fast. Most important result, with reductions in WIP, can be expected in flow times, which make organizations more flexible. The next essential condition is preventive equipment maintenance. This can be explained by the aim for smaller buffers and inventories; hence machines and production processes should be reliable. The maintenance condition is more or less applicable to every system, and should be a concern to every organization. Monitoring and improving quality of work is also a condition which is not only applicable to pull systems. Although by using a pull system these improvements can be done more easily, as delaying processes will be easier traced. Finally short setups, these are functional in order to reduce the lead times of products. By pull production setups become more relevant, because smaller batches or order driven production could be applied.
The other conditions are dependent on the desired pull system, starting with the condition of a stable product mix. This condition is relevant for the repetitive production environments, which is logical because these systems, as Kanban, are developed in this sector. When more variants of a product are made, WIP’s of the specific parts could become too high to let the system function as it originally was developed. CONWIP can handle different orders/products better; when product families can be produced on lines, not every station has to keep a certain amount of WIP, but a total WIP for the line should be determined.
plants produce a high variety of products. M-CONWIP was developed to overcome this shortcoming partially, but this variant is limited in its application, as a lot of organizations produce a lot of different products. POLCA has no problems with high variety, because of its paired cells. By making cells with more or less the same tasks, routings could be split up in pairs of cells. Therefore, the condition of sequential layouts becomes less relevant.
POLCA is less dependent of the upstream and downstream stations and release of orders takes place with help of a MRP system, the link with the downstream stations and demand is obviously thinner. If a work cell is obligated, a connection with another paired cell, which is available, could possibly be made. This makes the pull condition for POLCA less relevant than in CONWIP or Kanban. The system becomes in this way more hybrid, which enables the system to deploy it with a push system for order releases and supply, because orders can arrive with (more) time variability.
The described systems all rely on the ability to lay responsibility on the work floor and let operators, instead of the (master) planners, decide how orders have to flow on the work floor. Overview is limited at local stations or work centers only, this is reached by the authorization through the cards; a planner has less to do in fact. However, it could be possible in some situations, for example in situations with high variety in workloads or a lot of priority orders, that planners have to stay responsible for the authorization and need global overview of the production floor. Less responsibility on the work floor and authorization by the planners is given by the Cobacabana system. This system is easier to apply without some authority on the production floor. Consequently, the workers only have to focus on their main job, instead of thinking about fast releases and improvements in throughput. Hence, these conditions (improvement/teamwork/autonomous) become for Cobacabana less relevant, although improvements are still important, it is not an essential condition to implement Cobacabana successfully.
3.2 Decision framework
Translating the above mentioned similarities and differences into a methodological framework, results in the following conceptual framework:
Basic conditions Additional conditions System
1. Kanban
2. CONWIP
3. POLCA
4. Cobacabana/WLC
TABLE 3. CONC EPTUAL FR AMEWOR K FOR C HOOSING A SYSTE M
choice of an appropriate planning and control system and eventually in one key concept. Once one of these is selected, it would be necessary to investigate these systems in depth. As mentioned are for each system many variations developed.
Basic condition Quantifier Desired state
Reducing work in process inventories and unnecessary stock
Amount of WIP and total value of inventory
Should be possible; e.g. safety stock requirements should not be too high Equipment preventive maintenance Variability of breakdowns Must be low or moderate Quality assurance efforts at preventing
defects
Plant return rates, internal rejection rates
Rates must be close to zero
Setup times must be small Variability of setups; average time of the setups
Must be low or moderate; could be multiplied with the average times for an improvement comparison TABLE 4. BASIC CONDITIONS C ONCEPTUAL FR AM EWOR K
Features Description Overview Sugested
system/subsystems
Card
1. 1. Produces the company a stable product mix? 2. Do orders arrive at a
stable interval?
3. Is there a sort of repetitive production?
Local Kanban Part/batch
orientated Station restricted
2. 4. Is pure pull between work stations not feasible? (stable demand?)
5. Can production lines handle several products? (product families?) Local CONWIP M-CONWIP CONLOAD Order orientated Line restricted
3. 6. Does the company produce high-variety products?
7. Clear go-through of different work cells? 8. Desired overview on the
floor? Local POLCA G-POLCA LB-POLCA Order orientated Paired cell restricted
4. 9. Is it necessary to let the planner stay responsible for every flow and release?
10. Should it be possible to adjust orders lately?
Global Cobacabana/WLC SLAR Order and capacity orientated Critical workstation restricted
4. Case study
This article presents a new conceptual framework based upon the required conditions and characteristics of each card-based pull system, to give organizations a clear overview of the different systems. To evaluate this new conceptual framework a case study has been performed at SEM, where demand for a card-based pull system is growing.
4.1 Current situation
SEM produces wall trunking and cable tray products, which are completely produced in the factory. SEM is executing since 2004 a continuous improvement program, based on the principles of the Toyota Production System. In this program tight improvement targets are set, consequently there are always projects running to discover the value-added activities and to eliminate waste at SEM. Better overview of the released orders, overview of unused capacity, transparent planning activities and more pulled control was recognized as one of the improvement areas.
SEM is a production plant with 10 automated machine lines, a paint shop and some manual machines. High variety products (>2500) are produced in different ways, some (45%) are produced on only one machine while others have a routing of more than four work stations. In the current situation planning is performed on a push basis (for mainly the “runners”), where a queuing tool is used to prevent too much work in progress. A lot rescheduling has to be performed due to frequent breakdowns in the plant. The scheduling is done by a single person, others are not involved in this process, which makes the planning process not transparent. Control loops run via the production chef and the operators, the production chef keeps in touch with the planner at the office. Therefore, in order to increase the accurate information, better global overview, transparent planning, lower inventory, and less WIP the feasibility of a card-based pull system was investigated.
4.2 Assessments of the characteristics
According to our developed conceptual framework, we have to judge the four basic conditions and subsequently the sequel questions. This will lead to a card-based pull system feasible for the organization now or in the future, in case the basic conditions are not met (yet).
4.2.1 Assessment of the basic conditions (Table 4)
Reducing work in progress and keeping less inventory makes the organization more flexible and enables them to use their capital in other ways. Looking at the current situation of SEM, inventory is kept for every running product, in order to deliver customers as soon as possible. This inventory is currently needed because of the, sometimes, large orders by customers and the narrow due dates in combination with the setup and breakdown variability. In total the inventory is about a half million euro (see appendix A). Work in progress is about six days; this is not very long but could be shorter by a tighter planning. For example, most rush orders can be accomplished in two days. By making planning and control more transparent and easy, WIP will decrease and the whole process would become more reliable. In the end it will be possible to drop the total inventories.
throughput times, which is negative for the productivity of the plant. In a situation where the plant wants to go to a more pull system, machines should become more reliable.
In order to prevent producing defects, some quality efforts are made. For every workstation measurement/check lists are available, operators can easily check the tolerances of the output with these lists. The first and last product of a batch is checked and depending on the kind of product, and the amount, also interval checks are done. This result in only setting defects and almost no defects in the end products, the plant return rate is even close to zero.
To asses the setup times of the machines, we have summed up the setup times of the machines and divided them by the amount of setups. The result is an average setup time for each machine (appendix D). Besides, also the variability of the setups is calculated, which indicates if a machine performs sufficient on this subject (see for data appendix C). It turns out that SEM has two machines with high variability and long setup times and also two machines with only long setup times and moderate variability.
Consequently, not every condition is performing so well that it convinces directly for the implementation of a pull system. Although two of the four conditions are sufficient, the other two are not. The first condition tested, reducing inventory, is not expected to form a big barrier. Reducing inventory is possible and will happen automatically when the other conditions are met. Furthermore, the condition quality is sufficient, as we have seen, there are no problems within this aspect; the plant return rate is even close to zero. The other two conditions instead, need both more attention; together they cause high variability in the processes (see appendix C). Preventive maintenance and also the setup times are not on the right level to perform sufficient in a pull system. As indicated, too many breakdowns cause an unreliable production and unstable processes and require therefore a lot of rescheduling. Further, also the setup times of some of the machines are too long, this includes four of the ten automated machines. These two conditions need to be improved, before a pull system can be introduced at SEM.
4.2.2 Assessment of the features (Table 5)
We have assessed the features of SEM to make the required data complete in order to succeed the conceptual framework. Table 6 gives an overview. In this table each question is finally answered with a yes, a no or when it is doubtful to give a clear answer a yes/no.
4.3 Desired system
TABLE 6. FEATUR E ELABOR ATION AT SEM
Feature Description (SEM) Answer
1.1 Produces the company a stable product mix? The plant produces an unstable product mix, SEM has to cope with a high variety; in total are about 2500 different products possible. In 2011 were about 1794 different products produced, in 2012 are already 686 different products ordered/produced. So there is not a stable product mix, although we can distinct a group of products which are produced most; the runners. There are in total 201 runners, calculated by a Pareto analysis, see also appendix E .
NO
1.2 Do orders arrive at a stable interval? Orders of the same family arrive not at a stable interval, because of the high variety it is difficult to predict which products have to be produced. For this reason and the poor basic conditions stock is held currently.
NO
1.3 Is there a sort of repetitive production? A repetitive production line is a work cell or line which is always performing the same task. In this plant there are some machine lines which are doing the same tasks in some way, because of the high variety in products and (thus) routings it cannot be seen as repetitive.
YES/NO
2.4 Is pure pull between work stations not feasible? (stable demand?)
In Kanban systems buffers of WIP are held between each work station, but with a variety in routings because of different products this is impossible at SEM. Controlling the amount of WIP in one line would be a solution, but SEM has too much variety in routings.
NO
2.5 Can production lines handle several products? (product families?)
Only when a family of products can be made on a line CONtrolled WIP will work. Here, the variety of routings is too high to make it work, for the 200 runners (most produced products) already 33 different routings are needed (see appendix F).
YES/NO
3.6 Does the company produce high-variety products? This question is in some way the consequence of the first question, when there is not a stable product mix it implies a high variety of products. As said, there are produced up to 2500 different products, thus high variety at a small plant.
YES
3.7 Clear go-through of different work cells? In this plant some of the products only go through one work station/cell (see for a partition appendix F), for the runners is this about 45% of the produced products (see appendix E). Sometimes the work stations are independent work cells and sometimes dependent work cells. This means that not al products have a routing, but are independently produced and SEM has not clear routes between work cells.
NO
3.8 Desired overview on the floor? This question is the result of the desired autonomy of the workers on the production floor within Lean-Six Sigma programs. Most systems are only controlled on the floor by the operators themselves, but in some production environments this is not desirable, because of the frequent rush orders or the fluctuations in work load at the different work stations. SEM is such a plant: regularly rush orders are produced and also workload of the different workstations is difficult to plan. Various kinds of orders are produced, some are very big others are very small and work load is always different, so capacity needs to be planned carefully. Moreover because of the independent produced products, the planning task is extra complex.
NO
4.9 Is it necessary to let the planner stay responsible for every flow and release?
The foregoing point, added with the supply argument in the next question, makes clear that the planner has to stay responsible for the releases of the orders on the floor and also of the workload of every station.
YES
4.10 Should it be possible to adjust orders lately? Orders are dependent on raw materials, which is steel in this case. Steel is a complicated material and, because of the small size, SEM is dependent of their suppliers. So it is not uncommon that orders are rescheduled because of a lack of raw materials.
5. Conclusion and remarks
The aim of this article is to present company managers a useful decision framework for choosing a card-based pull system. Many key concepts are presented in the literature, but none makes it possible to select in a clear way, based upon real facts/features, an appropriate system. In this article it became clear that every key concept of a card-based pull system contains four common basic conditions as presented in the first section. Also that based upon the plant features a decision could be made for one of the key concepts. Therefore the concept decision framework has two sections. The first section contains basic conditions, which functions as entrance conditions. The second section aims to select in a clear way one of the card-based systems based upon (plant) features. The use of a case study shows the applicability of the developed decision framework. It turns out that SEM is not yet ready for implementing a card-based pull system, because it does not meet the minimal requirements of the four basic conditions. However, based upon the second section we could already select a system for SEM; most suitable would be Cobacabana. The framework provides not just information about the feasibility of the different systems, but delivers also accurate information about the current state of the organization. This is helpful for management in identifying the priorities.
The role of suppliers and the just in time aspect is not taken into account in the original conditions for pure pull in the beginning of this article. Once an organization is producing with a pull approach, the supplier relationships become interesting and just in time deliveries of materials could become necessary. Especially small organizations have less power to enforce supplier agreements. The feasibility of the supply chain within (small) companies to serve pull systems would be an interesting point for further research. Additionally, it will be interesting to improve the currently developed conceptual decision framework by elaborating each variant of the key concept systems presented. The original WLC concept showed negative influences on lead time and due date performance, although in general positive impacts are noticed by introducing WLC. These positive influences could be subscribed to the more transparent work floor situation, where fewer things go wrong (Land, 2004). Implementation at SEM would present new data how this will work with capacity cards in reality. For Cobacabana/WLC was also a variant developed, as table 5 suggest. SLAR is a WLC variant without norms and a control loop between the different machines based upon due dates and priorities. These conditions results in better due date accomplishments. Hence, further research would be needed to show the relation between Cobacabana and SLAR. And if these two could work together, it would be interesting to research if it will fit at SEM and what the benefits are after implementation.
6. References
Benton, W.C., Shin, H., 1998, Manufacturing planning and control: The evolution of MRP and JIT integration, European Journal of Operations Research, vol. 110, 411-440
Bokhorst, J.A.C., Slomp, J., 2010, Lean production control at high-variety, low-volume parts manufacturer, Interfaces, vol. 40, 303-312
Buzacott, J.A., Shanthikumar, J.G., 1994, Safety stock versus safety time in MRP controlled production systems, Management Science, vol. 40(12), 1678-1689
Dogger, B., Land, M. and van Foreest, N., 2010. Making CONWIP Work in High-Variety Manufacturing, 16th International Working Seminar on Production Economics, Innsbruck. Preprints 2, 133-144
Dar El, E. M., 1999, CONWIP-based production lines with multiple bottlenecks: Performance and design implications, IIE transactions, vol. 31(2), 99
Darlington, R., Rahimifard, S., 2007, Hybrid two-stage planning for food industry overproduction waste minimization, International Journal of Production Research, vol. 45(18-19), 4273-4288
Diaby, M., 2000, Integrated batch size and setup reduction decisions in multi-product, dynamic manufacturing, International journal of production economics, vol. 67(3), 217
Flapper, S.D.P., Miltenburg, G.J., Wijngaard, J., 1991, Embedding JIT into MRP, International Journal of Production Research, vol. 29(2), 329-341
Fernandes, N.O. & do Carmo-Silva, S., 2006, Generic POLCA—A production and materials flow control mechanism for quick response manufacturing, International Journal of Production Economics, vol. 104(1), 74-84
Germs, R., Riezebos, J., 2008, Workload balancing capability of pull systems in MTO production, Preprints 15th working seminar on production economics, Innsbruck, Austria.
Ghrayeb, O., Phojanamongkolkij, N., Tan, B.A., 2008, A hybrid push/pull system in assemble-to-order manufacturing environment, Journal Intell. Manuf., vol. 20, 379-387
Hopp, W. and Spearman, M., 1996, Factory Physics: Foundations of Factory Management, Irwin/McGrawHill, Chicago, IL
Huq, Z., Huq, F., 1994, Embedding JIT in MRP: The case of job shops, Journal of Manufacturing Systems, vol. 13(3), 153-164
Land, M.J., 2009, Cobacabana (control of balance by card-based navigation): A card-based system for job shop control, Int. J. Production Economics, vol. 117, 97–103
Land, M.J., 2004, Workload control in job shops, grasping the tap, PhD Thesis University of Groningen, Labyrint publications, Ridderkerk
Oosterman, B., Land, M.J. & Gaalman, G., 2000, The influence of shop characteristics on workload control, International Journal of Production Economics, vol. 68(1), 107-119
Monden, Y., 1983, Toyota production system: Practical approach to production management Nicholas, J.M., 1998, Competitive manufacturing management, Chicago, McGraw-Hill
Riezebos, J., 2006, Polca simulation of a unidirectional flow system, in: Riezebos, J., Slomp, J., (eds), 2006, Proceedings of the Third International Conference on Group Technology / Cellular Manufacturing, University of Groningen, The Netherlands, July 2006
Riezebos, J., 2010, Design of POLCA material control systems, International Journal of Production Research, vol. 48(5), 1455-1477
Rose, O., 1999, CONLOAD - A new lot release rule for semiconductor wafer fabs, Proceedings of the 1999 Winter Simulation Conference
Ryan, S.M., Choobineh, F.F., 2003, Total WIP and WIP mix for a CONWIP controlled job shop, IIE Transactions, vol. 35(5), 405-418
Siha, S., David, H.T., 1994, A Finch-Foster Duality between 'Pull' and 'Push' Production Systems, The Journal of the Operational Research Society, vol. 45(2), 179-186
Smith, D., 2001, Predictive and preventive maintenance pays dividends, Power Engineering, vol. 105(2), 39
Soderquist, K., 1999, Quality issues in lean production implementation: a case study of a French automotive supplier, Total quality management, vol. 10(8), 1107
Spearman, M.L., 1990, CONWIP: an alternative to Kanban, Operations Research, vol. 28(5), 879-894 Spearman, M.L., Zazanis, M.A., 1992, Push and pull production systems: Issues and comparisons, Operations Researc, vol. 40(3), 521-532
Stevenson, M., Hendry, L.C., Kingsman, B.G., 2005. Are view of new and established production planning and control (PPC) methods and their applicability to make to order (MTO) companies. International Journal of Production Research, vol. 43(5), 869–898
Suri, R., Krishnamurthy, A., 2003, How to plan and implement POLCA: A material control system for high-variety or custom-engineerd products, Technical report, Center for Quick response manufacturing, university of Wisconsin-Madison, WI
Vandaele, N., Van Nieuwenhuyse, I., Claerhout, D. & Cremmery, R., 2008, Load-Based POLCA: An Integrated Material Control System for Multiproduct, Multimachine Job Shops, Manufacturing Service Operations Management, vol. 10(2), 181-197
Appendices
Appendix A - Current inventory state
Here was the relation and value showed between Make-to-order and Make-to-stock products at SEM.
Appendix B – Machine breakdown times
In this table were the percentages given of the time a machine is broken down against the total time in use.
[table of machines and percentages]
Appendix C – Variability
In this appendix the variabilities of the different conditions (maintenance and setups) were assessed. This is done with help of the classes of variability from Hopp and Spearman (1996), showed below. Variability of process times per machine according to Hopp and Spearman(1996)
Natural process times (excluded random downtimes, setups and any other influences like coil changes)
For a good basis we have calculated the natural process times of the machines, with these data we could assess then the variabilities of maintenance and setups.
These natural process times could already be classified with the classes of variability of Hopp and Spearman (1996). This indicates if a machine is producing stable or not.
Natural process time= total process time – (Unclear + Planned + Breakdown + Coilchange + Setup + Unplanned)
Coefficient of variation natural process time = standard deviation natural process time / mean natural process time; ) ( ) ( X E X Cv
[table with variabilities at SEM]
Classes of Variability (Hopp and Spearman, 1996)
Variability class Coefficient of Variation Typical situation
Low (LV) c < 0,75 Process times without outages Moderate (MV) 0,75 < c > 1,33
Process times with short adjustments (e.g. setups) High (HV) c > 1,33
Preemptive outages (separate): breakdowns
In this section the variability of the breakdowns is calculated. This is done with the main time to failure (MTTF), the main time to repair (MTTR) and the data of the foregoing, the natural process times. These variabilities can also be classified according to the classification of Hopp and Spearman (1996)
Effective process time = mean process time/availability; A t te
0 Where Availability is defined as;
r f f m m m A
Squared coefficient of variation preemptive outages;
0 2 2 0 21
1
t
m
A
A
r r ec
c
c
[table with preemtive outages: breakdowns]
Non- preemptive outages (separate): setups
In this section the variability of the setups is calculated. This is done with the formula beneath. Again are the variabilities classified with Hopp and Spearman (1996).
Effective process time = mean process time + mean duration setups/average parts between setups;
s s e N t t t 0
Squared coefficient of variation non- preemptive outages; 2 2 2 e e e
t
c
Whereby; 2 2 2 2 0 21
s s s s s et
N
N
N
[table with Non-preemptive outages: setups]
Preemtive and non-preemptive both
At last we have taken both variabilities and calculated the combined effects. Also this resulted in a classification of the machines.
Effective process time = mean process time + mean duration setups/average parts between setups;
s s preemptive e e N t t t
Appendix D – Machine setup times
In this table a percentage was given of the time a machine in changeover against the total time in use. [table of machines and percentages]
Appendix E – Pareto analysis
This appendix showed the Pareto analysis of the products produced at SEM.
Appendix F - Routings
