OPERATOR JOB DESIGN
Work combination charts
The work combination chart is the key tool to design operator jobs that involve interactions with machines. While samples of these charts are in every book on lean manufacturing, most factories fail to use it in the initial stages of implementation. Once they start, however, they discover that they had left about half their productivity improvement potential untapped.
To evaluate the sequences of tasks we have assigned to each operator, we need to simulate their execution. When they hear the word "simulation," many engineers jump to the idea of using simulation software. The commercially available packages, however, are a poor fit for this application because (1) they are targeted at much more complex flows with random failures, and (2) a cell project can afford neither the money to buy them nor the time to learn them.
The work combination chart is a manual simulation tool, with which the analysis of several options for a cell can be completed start to finish within one afternoon. In the literature, work-combination charts are also called "standard work instructions" or "operator-machine balance charts." Even though "standard work instructions" is the Toyota term, we don't use it because it fails to convey the meaning that this is a tool for design and not only for training and communication.
When used as a design tool, work combination charts are best drawn by hand; as work instructions, they must be neat and easy to read. Line operators can be trained in reading these charts, but usually not in generating them. Among the tools used to design cells, work-combination charts are the most sophisticated and are understood and used only in a minority of the companies that are implementing lean manufacturing. People who design cells without this tool, however, typically leave on the table at least 20% of labor productivity increase. We have not seen a case where the extra time spent on generating these charts didn't pay off handsomely.
Breaking down operations into tasks
An operation is a change of state in a part. Generally, on the shop floor, a hole can be made in several machines, and even in different types of machines. Inside a cell, however, there is usually a one to one correspondence between operations and machines.
In Figure 1, we see the flow of materials through a counterclockwise cell that has one replicated machine, so that one part in two goes into each of the machines. While rare, this situation occurs occasionally. One machine may be duplicated, but a system with more than two machines for the same operation would no longer support the material flow pattern of a cell. Duplication also almost never happens with more than one type of machine in a cell, because if it needed it, the designers would most likely split it into two parallel cells, each with one machine of each type.
Figure 1. Breaking down an operation into tasks
This cell also has a machine with more than one station. A station is simply a location on a machine at which some form of operator intervention takes place. Machine suppliers wisely attempt to concentrate all the controls and supervisory displays on a machine at a single location but do not always succeed. On some machines, as the one shown here, the output location is not the same as the input location. On others, the operator must travel to a different location near the machine to monitor the process visually at some points during the operation.
For the operator, only the stations matter, and we can actually forget about the machines. While at a station, an operator carries out a task which may or may not be a complete operation.
The bottom of Figure 1 shows our graphic convention for work combination charts. The "wiggle" indicates manual work and the bar a machine cycle. When the wiggle is outside the bar, it shows the operator working on the machine while it is stopped, while the wiggle overlapping the bar represents the operator working on the machine while it is running. A prerequisite for drawing work combination charts is a time study that provides values for all these different times, as well as for walking times between stations.
Figure 2 illustrates the concept of the work-combination chart by applying it to a household activity. In the "before" chart, an operator making espresso from beans and water in an inefficient manner. The first three steps are purely manual, but the beans do not have to be returned to the freezer at this point. They can wait until the operator has time available while the machine is brewing.
Figure 2. Use of a work combination chart to improve an operation
Grinding the coffee counts as manual out time, because the machine is running while the operator is holding down the lid. It would be possible to free the operator during that time by using a coffee grinder that doesn't require the operator to hold down the lid. The operator also uses a measuring cup to dose 4 oz of water into the espresso maker, and a dosing device containing exactly 4 oz would eliminate the trials and errors associated with the measuring cup. The chart, however, draws attention to a much larger issue: the long brewing cycle. For over two and a half minutes, the operator has nothing to do but wait. The first part of the brewing cycle is spent heating water, and it seems that heating could be made concurrent with the manual activities.
The "after" chart shows how we can redesign the process based on these thoughts. The changes first go on the chart to simulate a new mode of operation. If a good half of the brewing time is spent heating up the water, then we should be able to start it at the very beginning of the cycle and still have time to grind the beans and load the grounds into the machine. In addition to preheating the espresso maker, we also replace the measuring cup with a ladle holding exactly 4 oz, and we delay returning the beans to the freezer until brewing starts. This should reduce the total brewing time to 3.91 minutes to 2.57. Then we run an experiment, which shows that the brewing time is cut even more than our calculations led us to expect.
Work combination charts as operator instructions
While Figure 2 was an example meant to illustrate the use of the tool in design mode, Figure 3 shows what a finished work combination chart for a cell looks like. This is now a communication tool, intended to tell an operator occupying any one of the positions what to do, and to enable a supervisor to monitor execution. The slanted lines between steps represent walking time and a vertical line marks the required takt time. The chart also includes performance summaries for both operators and machines.
Figure 3. Work combination chart for a two-operator cell
This example shows both the power and the limitations of the technique. It helps you evaluate such job designs as the patterns of operator movements in Figure 1, and find ways to improve them, but it does not generate these patterns from scratch. It is also predicated on the assumption that all operators can do all the tasks that may be assigned to them. It is a reasonable assumption in a design situation, because (1) having multiskilled operators is a goal that lean manufacturing organizations pursue, and (2) the design of an operator job cannot be based on the specific capabilities of the individual who holds it today.
Lack of software support for work combination charts
Technically, there is nothing about the generation of work combination charts that could not be embedded in a software product that would be similar in of complexity to those available for, say, project management on PCs. The only reason none is available that we know of is that the tool has too few users with collectively not enough money to support a supplier.
One tempting approach is pressing into service software that has been designed for other applications, such as production simulation, production scheduling, or project management. Inevitably, however, the application mismatch causes the user to spend more time struggling to control the behavior of the software than it takes to generate and maintain charts manually.
Baton-touch, caravans, and bucket brigades
The three following approaches are used to design operator jobs in U-shaped cells:
Baton-touch. This is the most common approach, and is the one shown in Figures 1 and 3. In this mode, a job makes one operator move through a loop of tasks in the same direction as the work pieces, and in such a way that movement is either to the next station on the same side of the cell, or across to the other side. As shown in Figure 1, there are at most two crossings between sides. There are exceptions to this with, for examples, machine-tools with long cycles that require operators to go back and forth to the same machine several times during one cycle.
Figure 4. Baton-touch task assignments
The baton-touch approach lends itself to work combination chart analysis, and to the design of a job pattern such as is shown in Figure 5, in which all operators except the one in charge of the first and last operation are fully occupied. The one who is not acts as cell leader, feeding work into the cell, relieving others as needed, filling out the paperwork, etc.
Figure 5. Desired operator loading
Caravan, or rabbit-chase. This approach is popular with operators, but limited to cells with at most two operators and loads both equally. In it, each operator works on all the jobs in the cell in succession, picking up at each operation the last part processed through it and loading it into the next machine.
Figure 6. Caravan/rabbit-chase style
Bucket brigades. The bucket brigade, also known as "bump-back system," is the latest addition to the list of approaches, and the only one that does not lend itself to a work combination chart analysis, because the job assignment are not fixed. It is mostly useful in mass customization settings, such as sandwich making at Subway or custom bag assembly at Timbuk2 designs. Operators are arranged slowest to fastest along the line. Every time a unit is finished out of the cell, each operator takes over the next unit from his or her predecessor, and the first operator starts a new work piece, as shown in Figure 7. This system has been shown to be self balancing, with all operators tending to take on work that requires the same amount of time. Click here (http://www.isye.gatech.edu/faculty/John_Bartholdi/bucket-brigades.html) for J.Bartholdi's write-up and animation explaining bucket-brigades.
Figure 7. Bucket brigades operations
DAILY TASK ASSIGNMENTS
Daily challenges on the shop floor
An actual shop floor, even the leanest one, never has a full slate of multiskilled operators. In any cell, we have operators who may be in the process of learning all the tasks in the cell, but, as of today, are fully proficient in some, partially in others, and not at all in the rest. Productivity improvements result in promoting the best operators out of the cell, thereby requiring a build-up in the skill level of the remaining team. In addition, operators are occasionally absent. These situations occur routinely in practice, and force cell teams to dynamically revise job definitions and assignments. Today, teams muddle through, using such resources as the cell leaders, floaters, or water spiders.
Towards a systematic approach
Let us examine what it would take to have a systematic approach and what tools could, if not automatically solve the problem, at least support decision making. In a typical, realistic situation, a production supervisor has to run a cell short-handed, because one operator has the flu, and with operators who are less than fully multiskilled.
This is not a situation where we need to design the cell and can set aside "quality time" to do it. It needs to be done quickly, in the heat of the action at shift start. We need a way to assign tasks to operators within the cell to maximize its output - that is, to get it as close as we can to what it would be with a full team of fully multiskilled operators. The output of the cell is determined by the operator who needs the longest time to complete his or her tasks. If we designate this time as the operator's "workload," we need to find the task assignment that minimizes the maximum of the operators' workloads, within the constraints of equipment characteristics and layout.
Figure 8. Minimizing the maximum workload
Assume the operators have a skills matrix of the type shown in Figure 9. We use numbers rather than the usual graphic symbols to represent proficiency, because we want to translate it to speed. This is reasonable, because, presumably, the more proficient operators perform the tasks faster. When we translate that in times required, we get infinity for the jobs the operator is not qualified to do. We can use these numbers to assign tasks to operators, but we must be aware that they does not take into account any layout dependent information, such as walk time, or the need to coordinate people and machines, which would require a work combination chart.
Figure 9. An actual skills matrix
A common approach is to allocate tasks first to the least flexible members of the team, and it would be correct in the absence of volume pressure. If an operator only has one skill, then the choice is easily made, and then we reserve the most versatile operators to pinch-hit wherever needed. This approach, however, does not maximize production, which is what we need if the team is short of its normal size. So, instead, we assign each task to the operator who requires the least amount of time on it, and use balancing considerations to break ties.
As shown in Figure 10, not only do this throw the workloads grossly out of balance, but the resulting operator paths are hopelessly convoluted. Then we look in the table for opportunities to reduce the maximum operator workload by switching the task assignments. While the resulting operator paths, in Figure 11, are less convoluted, they still require William and Averell to cross paths. Since Averell has at least 18 seconds of slack time, this may be workable, at least for this shift.
Figure 10. First cut at task assignments
Figure 11. Improved task assignments
Requirements for decision support software
This is all done manually, which begs the question of whether it could be done better with software. The task assignment problems in a cell are complex but small in size, and, because of the small size, brute force enumeration of candidate solutions is feasible, not by hand, but with appropriate software on a desktop PC.
With the assignments in Figure 11, the cell can produce at least 10% more than with those in Figure 10, but could we do any better? As indicated in Figure 9, if we ignore all layout considerations, there are 1,536 possible assignments in this cell. Using Access and Excel, it is possible to list all the corresponding 15,361 individual assignments, as shown in Figure 12, and use sorts and crosstabs or pivot tables to identify the assignment that minimizes the maximum workload. The result, shown in Figure 13, is that assignment 1087, the same as shown in Figure 11, is optimal.
Figure 12. Table of possible assignments
Figure 13. The assignment that minimizes the maximum workload
This is not to suggest that these database manipulations are the solution, only to show that any supervisor's desktop PC has enough computing power to host software that could, in a few minutes at shift start, enumerate all possible assignments in four to six cells, evaluate them by a work combination simulation, and recommend the assignment best suited to a variety of goals. In the above example, the goal was to maximize production, but it could as well be to minimize the number of operators needed to sustain a given level of production. Again, the only reason such software is not available off the shelf today is that too few supervisors have been confronted with the problem to date.
ABOUT THE AUTHOR
Since 1987, Michel Baudin has consulted for such clients as Honda of America, Dell Computer, Canon Virginia, Boeing, Raytheon, Unilever, MetalEurop, the CIADEA automotive group, Hoechst, and others on lean manufacturing implementation, and for high-technology companies like Hewlett Packard, Intel, Motorola. Winbond, and National Semiconductor on production scheduling, process transfer from R&D to production, and computer system architecture for manufacturing applications. He also designed the MS/X OnTime production scheduler marketed by Tyecin Systems and led the EU-funded INRECA research project.
Since 1995, he has taught short courses on the details of lean manufacturing, the management of lean manufacturing implementation, the lean approach to quality, and lean manufacturing for small and medium-size companies, as well as customized in-house seminars for consulting clients. These courses have been offered to the public through UC Berkeley extension, the University of Dayton's Center for Competitive Change, and the Hong Kong Productivity Center, and have been used in house by Honda, Boeing, Canon, Raytheon, Applied Materials, VDO, Siemens, and others.
His prior experience includes being a director of the Menlo Park Technology Center of Teknekron Corporation, leading a group at Schlumberger/Fairchild that designed, tested, and supported maintenance management, production scheduling, and quality control software that is in use in semiconductor factories; giving technical support for CIM installations in Japan on behalf of Consilium corporation; and implementing the OPT scheduling system in two General Motors factories.
Mr. Baudin is author of two books, Lean Assembly, from Productivity Press (2002) and Manufacturing Systems Analysis, from Prentice Hall (1990) and 22 articles and papers in various journals since 1977. His academic background includes a Master's Degree in Engineering from the Ecole des Mines, Paris; work at the Hahn-Meitner Institute of Berlin; and research at the University of Tokyo. He is a senior fellow of the University of Dayton's Center for Competitive Change, and a member of the IMSE External Advisory Board of Ohio University. Michel Baudin is fluent in French, Japanese, and German, and is learning Spanish.
MMTI - Manufacturing Management & Technology Institute