The Moving Assembly Line

To achieve a high-volume mechanical assembly line requires reliable precision equipment and standardized shop practices (Heizer, 1998). In August 1908, while still producing the Model N, Henry Ford hired Walter Flanders who brought to Ford a much needed knowledge of machinery, layout, and production methods (Sorensen, 1956). The initial moving final assembly line proved so successful that three of them were built in the fall of 1913 (Heizer, 1998).

Advantages of the Moving Assembly Line By moving the work to the worker, the worker did not have to move all of his tools and materials to the work. This saved a great deal of time and made the assembly process much cheaper.

Disadvantages of the Moving Assembly Line Because the assembly line moved at a specific pace, the automobile chassis was in a given station for only a certain number of seconds. If any problem arose, that particular chassis could not have the operation completed before the chassis moved out of the area. This problem necessitated a "fix-it" station at the end of the assembly line, where automobiles with problems that occurred during assembly were completed.

Assembly Line Balancing

In designing an assembly line, the number of workers, and hence the direct labor cost, is minimized if every worker or station has an equal amount of work. If every station has an equal amount of work, the number of stations is minimized. Thus, a common field of study regarding assembly lines has long been assembly line balancing. Amen (2000) developed a list of heuristics for assembly line balancing. He later (2001) performed a study of the comparative performance of these methods. Becker and Scholl (2006) extend the discussion to include U-shaped lines, as are common in JIT facilities, and mixed model lines. U-shaped lines are used primarily to produce components for JIT operations, with material entering at one end of the U and exiting at the other. Workers usually perform multiple tasks with tasks often on each side of the U. The number of workers in the line varies by season to maintain a daily output consistent with daily sales.

Just-in-Time

The Toyota Production System (TPS) and Kanban System (Sugimori et al., 1977) were "developed by the Vice-President of Toyota Motor Company, Mr. Taiichi Ohno and it was under his guidance that these unique production systems have become deeply rooted in Toyota Motor Company...." Just-in-time is the successor of the TPS. The purpose of using JIT is to eliminate waste from processes (Hall, 1997). The name JIT is misleading because it suggests that the concept primarily involves materials arriving just in time for use. The major benefit of JIT techniques is the simplification of the processes themselves.

JIT implements a pull system of control often using cards or Kanbans to implement the pull system in which materials are replenished at approximately the same rate they are used.

The objective of JIT is to streamline a process-to change and improve the process itself, not to install a control pull system on a process undeveloped for it. Improvement is multidimensional: delivery (lead time and due date performance), cost, quality, customer satisfaction, and so on.

OPT-The Precursor to DBR

DBR gradually evolved out of Goldratt's experience with a shop floor scheduling software called OPT. In his article "Computerized Shop Floor Scheduling," Goldratt (1988) explains in detail how OPT evolved. The first version of the software was basically automated Kanban. Goldratt states that early versions of OPT were such that straightforward usage was restricted to repetitive environments.

Goldratt came to realize that not all machines need to be utilized 100 percent of the time-only constraints need this. OPT was reformulated to limit non-constraints to only the work necessary to keep constraints properly fed. This led to difficulty convincing supervisors of non-constraint resources to follow the schedules when these schedules called for less than 100 percent utilization. Goldratt realized that only the bottlenecks should be scheduled-other stations have excess capacity and can keep pace-and thus data accuracy was really needed only at the constraint.

The Nine OPT Rules We will now list the nine OPT rules (Goldratt and Fox, 1986, 179)4 and discuss them as special cases of mathematical programming and other methods: 1. Balance flow not capacity.

2. The level of utilization of a non-bottleneck is not determined by its own potential but by some other constraint in the system.

3. Utilization and activation of a resource are not synonomous [sic].

4. An hour lost at a bottleneck is an hour lost for the total system.

5. An hour saved at a non-bottleneck is just a mirage.

6. Bottlenecks govern both throughput and inventories.

7. The transfer batch may not and many times should not be equal to the process batch.

8. Process batches should be variable not fixed.

9. Schedules should be established by looking at all of the constraints simultaneously. Lead times are the result of a schedule and cannot be predetermined.

It is often counter-productive to attempt to balance capacity in order to get a flow-balanced plant. Because constraints determine system performance, constraints should have a buffer of material (represented as a time buffer) upstream of them to protect them from out-ages occurring upstream. This buffer will disappear as it is used to protect from outages. If the upstream workstations have the same capacity as the constraint, the buffer can never be rebuilt and the constraint utilization becomes a function of the vagaries of outages of upstream stations. To balance flow, the capacity upstream of the constraint needs to be bigger than constraint capacity to rebuild the material buffer. Likewise, when stations downstream of the constraint experience outages, the constraint will eventually run out of a place to store its output (the space buffer). Stations downstream of the constraint need more capacity than the constraint to empty the space buffer as needed.

In a simple line, it is easy to see that constraints determine non-bottleneck performance. If there are two or more bottlenecks in a line, the constraint will be the station with the least capacity. Stations downstream of the constraint can process no faster than the constraint because material must pass through the constraint to get to them. Stations upstream of the constraint could work faster than the constraint, but this will build inventory at the constraint and eventually the futility of having upstream non-constraints working faster than the constraint will be recognized and the practice will be stopped.

To activate (a non-constraint producing more work that the constraint can process) a resource when the resulting output cannot get through the constraint is the meaning of Rule 3. Activating a non-bottleneck resource to produce more than can be processed by the constraint does not add any value to the company.

A bottleneck is a bottleneck only if it cannot keep up with market demand working 24/7. Thus, there is no reservoir of time from which an hour lost at the constraint can be replaced. It is simply lost to the system.

It has long been known that the slowest station in a line determines output. OPT extends this principle to job-shop type flows. In a job shop, the constraint may shift around somewhat as the mix of orders varies from season to season but there is generally one machine that is the heart of the plant and the reason most of the orders are obtained. This machine or work center tends to be needed on almost every job and becomes a long-term constraint on the system. Thus, even beyond simple lines, the constraint determines the output of the system.

By having the transfer batch (the number transferred between two stations) be less than the process batch (the number processed between setups), it is possible to have several stations working on an order simultaneously. This gets the order through the facility very quickly. It could be done to expedite the order. Alternatively, it could be done simply to use a short lead time as a competitive weapon in the marketplace.

Process batches should be variable, not fixed. If a product is seasonal and a shop always makes one week's worth of demand as a process batch, then the process batch will vary naturally over the course of the year. This approach allows little inventory to accumulate. If a fixed process batch large enough to cover a week's demand during the peak season was to be used, inventory covering several weeks demand would be created during the off-peak periods. Having a variable process batch makes more sense. Traditional plants may use the Economic Batch Quantity (EBQ) (the number of units processed at a time to minimize setup and carrying costs) formula to determine a fixed process batch size, but the EBQ formula assumes a fixed demand so its use really is not appropriate in this situation.

Derivation of DBR Using the Five Focusing Steps

TOC says that constraints (anything that limits a system from achieving a higher performance versus its goal) determine the performance of a system and TOC provides methods for efficiently and effectively utilizing these constraints. Since it is not the main topic of this chapter, here I will present only a key definition and the Five Focusing Steps (5FS) without elaboration or delving into ramifications. There is expanded coverage of 5FS in Chapter 8 and elsewhere in the book.

The 5FS are as follows: 1. Identify the system constraints.

2. Decide how to exploit the system constraints.

3. Subordinate everything else to the above decision.

4. Elevate the system constraints.

5. If, in the previous steps, the constraints have been broken, go back to Step 1, but don't let inertia become the system constraint (Goldratt, 1988).

In Step 1, a company defines its drum. In Step 2, it develops buffers at shipping and the internal resource constraint if it exists. At Step 3, the rope is tied between the buffer and material release to maintain the constant buffer.

A number of articles have discussed 5FS. These include Mabin and Davies (1999), Ronen and Spector (1992), Jackson and Low (1993), Politou and Georgiadis (undated), Mabin and Davies (2003), and Trietsch (2005). In addition, Gupta et al. (2002) introduced a series of simulation models that were run with each successive model introducing another step.

Jackson and Low (1993) note that an important contribution of constraints management is the focus it provides the entire organization. When everyone understands the vital role the constraint plays in the organization, everyone measures their actions according to the effect on the constraint and thus the total productivity of the system.

Scheduling the Resource Constraint

In TOC, all workstations work to maintain the schedule set at the constraint resource. Goldratt (1990) describes how this schedule is derived in The Haystack Syndrome. For each order, we have the due date of the order. We also have an estimate of the time it will take for the order to move from the constraint resource to the shipping dock-the shipping buffer. Scheduling the resource constraint involves loading each job onto the constraint, a shipping buffer time before its due date, and resolving any timing conflicts. The Avraham Y. Goldratt Institute produced a set of production simulators (a Windows version is provided in Goldratt, 2003b) to teach potential users constraint-scheduling concepts.

The article by Schragenheim and Ronen (1990) is the most often cited description of how DBR scheduling works. They list three steps: (1) schedule the constraint, (2) determine the buffer sizes, and (3) derive the materials release schedule according to steps (1) and (2). Schragenheim and Dettmer (2001) and Schragenheim, Dettmer, and Patterson (2009) provide perhaps the most in-depth discussion of DBR, including a special case called simplified DBR, and such issues as multiple constraints, moving bottlenecks, multiple operations occurring at the bottleneck, and other complications. Simplified DBR (S-DBR) assumes that the market is the constraint and therefore uses only one buffer-the shipping buffer (frequently called the production time buffer). Of course, if there is an internal constraint, material will naturally accumulate upstream of the constraint establishing a de facto constraint buffer.

Scheduling Non-Constraints

The pure DBR methodology does not develop a formal schedule for non-constraints. Rather, the rope determines when material is to be released to the first station on a routing and material is allowed to flow naturally between workstations. If decisions made by workstation supervisors result in a hole deep in the buffer, then expediting by using small transfer batches to achieve overlapped operations at a few stations may be needed to get material into the buffer in time to avoid the hole reaching the buffer origin (starving the constraint).

The individual work center (non-constraint) supervisor is advised that when a hole deep in the buffer appears, he or she should schedule the missing job first. If there are no significant holes in the buffer, he or she is free to run any job next. The supervisor might choose a job because of a short, sequence-dependent setup time, for example. Many academics are uncomfortable with this informal, ad hoc, logic for dispatching at non-constraints. Some researchers have developed alternative mechanisms for scheduling non-constraints.

Protective Capacity

TOC breaks capacity at non-constraints into three categories5: (1) productive capacity, (2) protective capacity, and (3) excess capacity. Productive capacity is that capacity equal to the constraint's capacity-the ability to produce the number of units that the constraint canproduce. Protective capacity is capacity needed to restore buffers to their ideal state after a disruption-to refill the time buffer that has become depleted or to empty the space buffer that now has material awaiting processing downstream of the constraint. This restoration of the ideal state of the buffer needs to be done quickly, before another disruption occurs. Excess capacity is capacity over and above productive and protective capacity.

Protective capacity is one of the most vital aspects of DBR because if there is insufficient protective capacity, then the buffer cannot be refilled quickly enough when the buffer is low and thus the drum is vulnerable to possible starvation6 by upstream stations or blocking by downstream stations. Since an hour lost at the drum is an hour of lost output if the drum is a resource constraint, downtime at the drum can be extremely expensive. Protective capacity is idle when the buffer is in an ideal state and needs no restoration. The non-constraint station uses only enough capacity to produce at the drum's pace. However, once a buffer leaves its ideal state, all affected non-constraints must use their protective capacity to restore the buffer to an ideal state before some other problem threatens to idle the drum.

Of course, in a deterministic environment there would be no need for protective capacity because a constant amount of inventory would be held in the time buffer. An issue related to the establishment of a DBR system is "How much protective capacity is needed and how should it be arranged?" There have been only a few studies of this issue. This issue is especially important if there are capacity constrained resources (CCRs) in the system. Recall, a CCR is defined by the TOCICO Dictionary (Sullivan et al., 2007, 7) as "any resource that, if its capacity in not carefully managed, is likely to compromise the throughput of the organization." ( TOCICO 2007, used by permission, all rights reserved.)

Literature on DBR Scheduling

In discussing literature on DBR, I first present some overview articles that principally discuss the 5FS or the nine OPT rules. Then I move to simulation models and case studies divided by VAT classification. V, A, and T represent types of plants with V-plants dominated by divergences in flows, A-plants dominated by assembly operations, and T-plants experiencing a huge increase in variety in the final operations. After the sections on V-, A-, and T-plants, I present those simulations and cases that could not be assigned a specific VAT class.

Overviews

When TOC first appeared, total quality management and JIT were also gaining popularity. Because Goldratt was doing most of his information transfer via workshops and his books The Goal and The Race, many people lacked a true sense of what TOC entailed. A number of people sought to fill this relative void by introducing articles covering the 5FS or the nine OPT rules, and especially DBR and Buffer Management. Some also reported on DBR implementations. Because the articles are broader than case studies of a single implementation, this section of the chapter was developed to gather these broad overviews.

Cox and Spencer (1998) devote a chapter of The Constraints Management Handbook to the DBR scheduling method. Throughout the chapter, they give a detailed three-product, five-work-center example, showing how to develop a schedule for the drum and for shipping. They also present a section on Buffer Management and a section on how DBR works within a material requirements planning (MRP) system. Overall, this is an excellent short summary of DBR.

Mabin and Balderstone (2000) present a book containing over 300 abstracts for books and papers on TOC published prior to 2000. They present a tree showing all aspects of TOC, with DBR belonging to a branch on production management. Only a few of the abstracts relate to DBR, including Atwater and Chakravorty (1994) who present a simulation study of the importance of protective capacity. Betz (1996) presents a study of an implementation at Lucent Technology. Coman et al. (1996) discuss the successful implementation of DBR at an Israeli electronics firm. Conway (1997) presents a concern over DBR scheduling the constraint carefully and non-constraints loosely as previously described by Simons and Simpson (1997).

Danos (1996) discusses how an implementation of DBR software increased profits by 300 percent at one company. Demmy and Demmy (1994) present a novel use of DBR by a photographer (treating himself as the constraint) in scheduling students to have their pictures taken for their yearbook. Demmy and Petrini (1992) describe the successful implementation of DBR to control aircraft maintenance within the Air Force Material Command. Duclos and Spencer (1995) use a simulation model of three different environments to show how DBR produces significantly better results than MRP in a hypothetical company. Fawcett and Peterson (1991) include DBR in a discussion of manufacturing-related aspects of TOC. Fry (1990) discusses an important aspect of buffers-the impact of work-in-progress (WIP) inventory on lead times. Because most of the time that a part spends in a facility is waiting for service rather than being serviced, there is a strong correlation between WIP and lead time. In a follow-up article, Fry et al. (1991) discuss the implementation of DBR to control lead time. Gardiner et al. (1992) provide a comprehensive overview of DBR and Buffer Management. Gardiner et al. (1994) present a brief discussion of DBR and Buffer Management in discussing the evolution of TOC.

Grosfeld-Nir and Ronen (1992) discuss the application of OPT to the single-bottleneck problem. Lambrecht and Alain (1990) present the results of a simulation comparison of JIT and DBR. In an earlier paper, Lambecht and Decaluwe (1988) show that DBR is more robust than JIT in managing bottlenecks. Pinedo (1997) provides a second commentary on Simons and Simpson (1997) praising the overall article but raising an issue of lack of comparison with other software. Radovilsky (1994) uses queuing theory to estimate the size of time buffers in DBR (Goldratt and others suggest using an amount equivalent to a portion of the existing lead time). Radovilsky (1998) presents a follow-up to the initial article, also estimating initial time buffer size using queuing theory. It should be noted that Buffer Management would be used to adjust this initial estimate based on whether too much or too little material is present in the time buffer.

Reimer (1991) outlines DBR and discusses it with a modified MRP framework. Schragenheim and Ronen (1990; 1991) are discussed; these articles were discussed at length earlier in the chapter. Russell and Fry (1997) discuss order review/release mechanisms that could be used to fill the function of the rope and discuss lot splitting into several transfer batches as an expediting methodology. Schragenheim et al. (1994) discuss modifications of DBR for use in process industries. Simons and Simpson (1997) present a concise history of the evolution of DBR and the algorithm in detail, and relate the algorithm to alternative methods.

Spearman (1997) gives both positive and negative comments on TOC and the Goal System software. Spencer (1991) discusses the basic theory behind DBR and how to marry DBR and MRP II. Spencer and Cox (1994) discuss the distinctions between OPT and TOC. Spencer and Wathen (1994) present a case study of service functions at Stanley Furniture including an implementation of DBR. Stein (1996) includes a discussion of the advantages of DBR and dynamic buffering in a generalized manufacturing situation. Umble and Srikanth (1995) include a thorough discussion of DBR in their pioneering book, Synchronous Manufacturing.

Wolffarth (1998) presents practical lessons learned from an implementation of DBR within an Enterprise Resources Planning (ERP) system. Yenradee (1994) presents a case study from a battery factory using a manual DBR system in conjunction with the nine OPT rules. Mabin and Balderstone (2000) also present a list of 34 books that had been published on TOC by 2000.

There are several overviews of TOC including some discussion of DBR published since the Mabin and Balderstone (2000) book appeared. Rahman (1998, 337) states that TOC contains two major components-the logistics paradigm, including DBR, and the Thinking Processes, which he calls a "generic approach for investigating, analyzing, and solving complex problems." He includes the 5FS, the nine OPT rules, and the definitions of the three operational measures (Throughput, Inventory and Operating Expense). He also includes a table of 139 articles and conference proceedings broken down by year and journal. Gupta (2003) provides an overview that relies heavily on both Rahman and Mabin and Balderstone as an introduction to a special issue of International Journal of Production Research. Watson et al. (2007) update the comprehensive discussion of the evolution of TOC previously discussed in Gardiner et al. (1994).

Boyd and Gupta (2004) give an excellent overview of TOC, comparing its philosophy to several somewhat similar philosophies but giving only a rudimentary overview of DBR.