Applying DBR to Different Types of Facilities: VATI Analysis7

The TOCICO Dictionary (Sullivan et al., 2007, 51) defines "VATI analysis-The stratification of operations environments into four generic types referred to as: V, A, T, and I. Each environment has an inherent set of undesirable effects that, properly understood, make operations management easier. Each type is named for the letter that resembles a diagram of the logical flow (not the physical flow) of materials. Usage: A single plant may be a combination of more than one type." ( TOCICO 2007, used by permission, all rights reserved.) Umble and Umble (1999) discuss VAT analysis; that is, classifying plants as one of these three types and recognizing that certain characteristics are common to each type. They state that VAT classification was developed around 1980 by Goldratt.

Product flow diagrams for V-plants are characterized by divergence points (hence the V-shape). Three characteristics are typical of V-plants: 1. The number of end items is large compared to the number of raw materials.

2. All end items sold by the plant are processed in essentially the same way.

3. The equipment is generally capital-intensive, highly specialized, and typically requires lengthy setups.

A-plants are characterized by convergent assembly points throughout the process. In such plants, a large number of purchased or fabricated component parts and materials, generally produced in a job shop environment, are combined to form subassemblies that are used to build unique end products.

T-plants are dominated by a major divergent assembly point at final assembly, where many different end items are assembled from a relatively limited number of component parts.

Umble and Umble (1999) go on to discuss the specific placement of buffers in each type of plant.

I-Plant Research

Many simulation models used to study aspects of DBR are simple I-plants. This is because issues such as what constitutes adequate protective capacity or the buffer's impact on lead time can be studied in an I-environment without complicating factors that occur in A-, V-, or T-plants.

Fry et al. (1991) simulate an I-plant to show how having little WIP at non-constraints in DBR gives strong control of lead time.

Finch and Luebbe (1995) simulate a five-station system in which the constraint moves over time because of different learning curve rates at the five stations. Because of shifting work center times during much of the simulation, there is little or no protective capacity at non-constraints. The authors conclude that there are significant interactions between learning curve effects and constraint production and that there is need for further study of this issue.

Atwater and Chakravorty (1996) simulated simple 5- and 6-station serial lines (I-lines) with disruptions created by machine breakages using balanced, JIT, and TOC configurations. They found that TOC-based lines are less affected by variability than either balanced or JIT lines. Chakravorty and Atwater (1996) study line design for I-structures.

Kadipasaoglu et al. (2000) simulating an I-facility found that (1) when protective capacity increased from 0 to 12.5 percent, flow time decreased by about 40 percent; (2) there is a benefit to WIP level for having the constraint be the first station;8 and (3) non-constraint downtime and protective capacity tend to have opposite effects on flow-increasing non-constraint downtime decreases flow, which can be offset to an extent by increasing protective capacity. Betterton and Cox (2009) later studied this simulation and found that the methodology employed was not a correct implementation of DBR. First, Kadispasaoglu et al. (2000) had random arrivals released into the plant rather than using a rope to release material at the drum's pace. Second, using station 1 as the constraint, Kadispasaoglu et al. used infinite buffers at all downstream stations. Blocking can never occur with infinite buffers, so the constraint would never undergo blocking. Simulating the environment as a true DBR environment, Betterton and Cox (2009) found that some of Kadipasaoglu et al.'s findings were not correct.

Blackstone and Cox (2002, 419), using a simulated I-facility, define "protective capacity" as "the capacity needed at non-constraint workstations to restore WIP inventory to the location adjacent to and upstream of the constraint workstation to (create a time buffer to) support full utilization of the constraint workstation." It should be noted that the ability of downstream stations to empty the space buffer when it contains work is also protective capacity-protecting against blocking. Blackstone and Cox also show that the size of the time buffer required to adequately protect the drum is inversely related to protective capacity, a point that had been made previously by Atwater (1991).

Kim et al. (2003a) simulated a variety of flow control mechanisms within an I-line and found that, compared to output flow control and dynamic flow control, bottleneck flow control achieved greater output with less WIP while maintaining smaller lateness and tardiness of orders.

Real I-Lines I found no simulation studies or case studies of real I-lines. I think this is because even when the flow is straight line, real facilities tend to have multiple products that diverge into various configurations as they travel down the line. That is, they are V-plants, not I-plants.

V-Plant Research

Simulations of Real V-Plants Vaidyanathan et al. (1998) describe the simulation of a coffee production facility having moving CCRs. The simulation model was used to develop a schedule for this V-plant. The simulation showed that output could be increased by approximately 40 percent by using the simulation model to develop the schedule.

Hasgul and Kartal (2007) used the Wagner-Whitin algorithm, a very sophisticated technique used to attempt an optimal sequence of jobs over a lengthy planning horizon, to schedule a simulated refrigerator plant. The portion of the company they were simulating corresponded to a V-plant. They reported achieving an average cycle time decrease from 12 days to 7 days when DBR was applied.

Case Studies of V-Plants Chakravorty (1996) reports a case study at Robert Bowden, Inc., a $40 million sales supplier of residential and light commercial building products whose manufacturing facility is a V-plant. After implementation of DBR (which is described in the article), the average number of orders processed increased by 20 percent with no increase in staff, and expediting of orders was significantly reduced.

Rerick (1997) presents a study of semiconductor wafer manufacture at Harris Corporation, which reduced cycle time by approximately 50 percent while almost doubling output. Wafers were made for automotive, telecommunications, and computer markets. A control point was selected to implement a DBR system.

Huang and Sha (1998) use a hybrid DBR/Kanban system to model a wafer fabrication facility through a simulation model. Kanbans, which pull material forward station-by-station, somewhat override the purely informal DBR approach to non-constraint dispatching. Huang and Sha also attempt to determine the optimal size of Kanbans in such a system.

Hurley and Whybark (1999) studying a simulated V-plant correctly point out that variance reduction can reduce the need for protective capacity and protective inventory.

Chakravorty (2000) presents a second case study of DBR at Robert Bowden, Inc., emphasizing the fact that it is a V-plant. V-plants running DBR have not received a great deal of attention in the literature. The plant used two buffers-constraint and shipping. Between 1996 and 1999, annual sales in units increased from approximately 58,000 to over 80,000 while the number of workers only increased from 12 to 16. During the same period, the stock of finished goods was reduced from 3800 to 1325, while late orders decreased from 19 to 7 percent.

Frazier and Reyes (2000) present a detailed description of how DBR was applied to the Dallas, Texas, plant of a company manufacturing cable and telecommunication equipment in a V-plant. After three months, WIP decreased to one-third of its previous level, raw materials inventory value decreased by approximately 30 percent, and percent on-time completion of jobs increased by more than 30 percent.

Schaefers et al. (2004) report the implementation of DBR in a facility that buys large rolls of metallic sheets and cuts them into smaller coils with less width and length. This appears to be a V-plant. The firm is a make-to-order (MTO) operation with no internal constraint so it used the shipping schedule as the drum (S-DBR). Before implementation, lead time varied from 21 to 182 days. After implementation, it was a stable 10 days. Customer service level increased from 34 to 87 percent. The exact change in profitability was not reported, but the authors did say that the facility changed from losing money to making money.

Belvedere and Grando (2005) report on a DBR implementation at an Italian chemical company producing dyes and pigments. Because the main raw materials are natural products, it was difficult to obtain the desired color precisely. The solution would be diluted and color-tested repeatedly, causing a dilution and the sample-testing department to be the constraint. In two years, the DBR led to a decrease in raw materials and finished goods inventory and to an increase in the number of stock turns, which almost doubled between 1999 and 2001.

Umble, Umble, and Murakami (2006), noting the lack of case studies from Asian implementations, report a case involving Hitachi Tool Engineering, a Japanese tool-engineering firm employing approximately 1100 people. They describe the plant they studied as a V plant. In addition to implementing DBR, the firm implemented some TOC thinking processes. A simple DBR system was set up using three shelves at the bottleneck with each shelf containing a day's work for the bottleneck. The authors report that this was adequate to buffer the bottleneck and to subordinate other resources to the bottleneck's schedule.

A-Plant Research

Simulation of Hypothetical A-Plants In this section, I discuss simulations of hypothetical lines that appear to me to be A-plants. Unless specifically mentioned, the authors did not specify the plant type using the VATI breakdown.

Taylor (1999) simulated a traditional push (MRP) system versus a pull (JIT) and "hybrid" (DBR) system regarding their impact on financial measures. His simulation model appears to be an A-plant. It contained 29 stations. Independent variables included buffer size and location. He found that the DBR system had higher profit, return on investment (ROI), and cash flow while using considerably less inventory. The pull system placed second in financial results with the push system placing last. Taylor (2000) studied this same plant for impact on TOC operational measures such as Throughput, Inventory, and Operating Expense.

Atwater and Chakravorty (2002) found that mean flow time through a simulated system that has a jumbled flow and appears to be an A-plant decreased as protective capacity increased but at a diminishing rate as protective capacity reached 7 percent. Mean tardiness decreased in the same fashion. In their study, they varied constraint utilization from 94 to 98.5 percent. They compared releasing jobs immediately upon arrival in the system to releasing jobs according to the DBR schedule and found that while DBR had a smaller mean flow time through the system, the immediate release approach resulted in fewer tardy jobs.

Simulations of Real A-Plants Wu, Morris, and Gordon (1994) show how DBR improves makespan when compared via simulation to a traditional production control system. Make-span is the time from the start of processing until the final unit clears the system. The Wu et al. simulation is an A-plant based on a furniture manufacturer. They demonstrated that a Taiwanese furniture manufacturer would benefit significantly from makespan by implementing DBR. In their simulation, makespan decreased approximately 50 percent when DBR was added to the environment.

Guide (1995) presents a simulation model used to estimate ideal buffer sizes in a DBR implementation at a naval repair depot. A naval or air force repair depot completely disassembles a plane (while this may resemble a V-plant, in contrast to a V-plant parts flow down each path instead of one or the other paths), repairs or replaces components as needed (probably A-plant), and reassembles the plane (A-plant). This process is known as remanufacturing.

Steele et al. (2005) simulate a shop using both DBR and MRP. They found that DBR has much better performance and suggested the use of DBR within MRP systems. They based their simulation on a bearing manufacturer. This involved an assembly that sat atop two V-lines.

Case Studies of A-Plants Andrews and Becker (1992) present a case study of Alkco Lighting, noting "Buffer Management" as a keyword. This A-plant involves several assembly operations. Alkco changed its primary measurement from efficiency to Throughput. As a result, WIP inventory improved significantly and there was an accompanying improvement in cash flow. Prior to implementation, the company was promising delivery in 60 to 90 days, had an on-time rate of only 65 percent with 16 percent of deliveries being more than one week late. Thirty-two percent of Inventory was in finished goods. The DBR system as managed by Alkco freed up 40 percent of its total floor space. Five years into the implementation, lead time was reduced to one week, while on-time delivery increased to 98 percent, sales volume increased 20 percent, and before-tax profit increased 42 percent.

Spencer (1994) reports on improvement from Trane Co. of Macon, Georgia, where output changed from an average of three units per day to six units per day with the same work-force when DBR was implemented. At this location, Trane assembles large air conditioners designed to cool commercial facilities.

Guide (1996; 1997) and Guide and Ghiselli (1995) present three discussions of the application of DBR in remanufacturing applications such as a military repair depot. As in Guide's (1995) simulation discussed above, this facility appears to be an A-plant (reassembly) sitting atop a disassembly operation. Disassembly is somewhat akin to a V-plant in that a single plane diverges into many components to be evaluated and repaired, replaced, or reused. However, the consensus is that a disassembly operation is different from a V-plant where a part flows to one product or another.

Luck (2004) presents an Ashridge Business School (UK) study of a supply chain centered on a manufacturing company called Remploy, which makes military garments. Remploy had two plants, a V-plant that cut material and an A-plant where sewing was accomplished. Five months into a standard DBR implementation, Throughput had increased 19 percent, output per employee was up 13.4 percent, WIP was reduced more than 50 percent, and absenteeism was down an average of 7 percent. There was some increase in transportation cost, but it was small compared to the increased profitability.

T-Plant Research

I was unable to find any simulations or studies that dealt with T-plants.

Research That Could Not Be Classified as V, A, or T

Sometimes research cannot be classified as V, A, or T. For one thing, the research may include more than one plant with the types of plant being different. Even if a single plant is described, in many instances, there may not be enough information provided to make a reasonable conclusion on whether the plant is V, A, T, or I.

Simulations of DBR Systems

A number of individuals have simulated DBR systems, sometimes to estimate DBR's parameters, such as the time buffer, and other times to compare DBR's effectiveness with systems such as Lean or CONWIP. Guide (1995) experimented with different buffer sizes and Buffer Management techniques at a naval air station. Kosturiak and Gregor (1998) simulated a flexible manufacturing system (FMS) using MRP, Load-Oriented Control (LOC), DBR, and Kanban and found that LOC and DBR had the best performance, while DBR was easier to implement. Hasgal and Kartal (2007) combined DBR with the Wagner-Whitin lot-sizing algorithm and reduced cycle times and WIP in a simulation model compared to DBR by itself.

Kayton et al. (1997) simulated a wafer factory running DBR to better understand the impact of preventive maintenance in such a facility. They found that downtime at non-constraints can become problematic in facilities using DBR even when significant protective capacity exists.

Lea and Min (2003) simulated a seven-station, three-product line using both JIT and DBR and found that JIT had slightly higher profits and service levels. They also found that activity-based costing systems slightly outperformed traditional costing and Throughput Accounting systems.

Case Studies

Several articles present case studies of successful implementation of DBR and constraint management. These case studies could not be classified as V, A, or T. Often the reports include multiple plants.

Gupta (1997) discusses DBR benefits to a supply chain. In 1998, Gupta discussed the need for software to implement DBR, describing some situations that are too complex for manual implementation. Koziol (1988), a manager at the Valmont plant in Brenham, Texas, discusses the successful implementation of DBR at that facility.

Spencer and Cox (1995) report a study of nine repetitive-manufacturing companies, three of which were pure JIT, three added MRP to JIT, and three added TOC (OPT or DBR) to JIT. No specific improvement numbers were reported; however, they found that the existence of repetitive manufacturing does not preclude the application of any of the three production planning and control systems.

As mentioned earlier, Wolffarth (1998) presents practical lessons learned from an implementation of DBR within an ERP system. Umble and Umble (2006) describe how Buffer Management was used in two accident and emergency facilities in Oxfordshire, UK.

Guide and Ghiselli (1995) report on the implementation of DBR at Alameda Naval Air Depot. This disassembly / repair facility implemented preventive maintenance, added small transfer batches, eliminated local efficiency measures, and took other DBR-related steps. Results achieved included increasing Throughput while reducing WIP, reducing airplane turnaround times, and increasing the turns ratio. Further refinements of DBR at the facility were reported to have been planned.

Umble et al. (2001) report a case study of DBR used within an ERP system. The case is Oregon Freeze Dry (OFD), which processes products by removing water at low temperatures and pressures. A branch of OFD implemented DBR in 1997, identifying a resource constraint that was designated as the drum. ERP was implemented at about the same time. The authors report that an ERP system makes DBR more effective. Once the drum schedule is determined, the ERP system was used to tie the rope. They state that the integration of TOC/DBR may be the key to ERP success.

Corbett and Csillag (2001) report on seven DBR implementations in Brazil. Five of the companies were multi-nationals, while two were Brazilian. Six used MRP and one used Kanban before using DBR scheduling. Average time to implement DBR was 3.6 months with the longest being 7 months. All seven companies started showing beneficial results during the implementation period. Six of the seven reported that they were satisfied with DBR. Even the one reporting dissatisfaction experienced a 50 percent drop in WIP and lead time and an increase in revenue per employee from US$56,000 to $64,000.

Lindsay (2005) reported on the implementation of DBR in Intel distribution centers (DCs) in an attempt to reduce order cycle time and reduce Inventory. Five DCs located in five countries have implemented DBR with an average cycle time reduction of more than 60 percent and a standard deviation reduction of more than 70 percent.

Vermaak and Ventner (undated) report the use of TOC in conjunction with computer simulation of a conveyor system in a coalmine, which resulted in an 8 percent increase in output.

Mabin and Balderstone (2003) report on an analysis of over 80 successful TOC implementations taken from a search of available literature. A portion of one of their tables reporting percentage improvements in various measures is shown below.

Huff (2001) reports that Bal Seal Engineering used DBR to increase Throughput, reduce Inventories, improve due date performance, reduce Operating Expense, and double net profit. Boeing and Rockland Manufacturing also achieved dramatic improvements relative to Throughput, Inventory, and profit.

Special Cases

The TOC literature contains a number of articles describing research that does not fall neatly into the previous categories but that are significant in their contributions to the body of knowledge. I have classified this research into the topics given in the next sections.