TIME STANDARDS:ADJUSTMENTS TO TIME AND LEARNING

3. ADJUSTMENTS TO TIME: LEARNING

3.1. Learning

Failure to adjust standard time for learning is the primary cause of incorrect times. Learning occurs both in the individual and in the organization.

3.1.1. Individual Learning

Individual learning is improvement in time / unit even though neither the product design nor the tools and equipment change. The improvement is due to better eye–hand coordination, fewer mistakes, and reduced decision time.

3.1.2. Organization Learning (Manufacturing Progress)

This is improvement with changing product design, changing tools and equipment, and changing work methods; it also includes individual learning. Often it is called manufacturing progress.

Consider the server Maureen serving breakfast. During the individual learning period, she learned where the coffeepot and cups were, the prices of each product, and so on. The amount of time she took declined to a plateau. Then management set a policy to serve coffee in cups without saucers and furnish cream in sealed, one-serving containers so the container need not be carried upright. These changes in product design reduced time for the task. Other possible changes might include a coffeepot at each end of the counter. A different coffeepot might have a better handle so less care is needed to prevent burns. The organization might decide to have the server leave the bill when the last food item is served. Organization progress comes from three factors: operator learning with existing technology, new technology, and substitution of capital for labor.

Point 1 was just discussed. Examples of new technology are the subsurface bulblike nose on the front of tankers (which increased tanker speed at very low cost) and solid-state electronics. Moore’s law states that the number of transistors on a given chip size (roughly a gauge of chip performance) doubles every 1.5–2 years. Some example numbers are 3,500 transistors / chip in 1972, 134,000 in 1982, 3,100,000 in 1993, and 7,500,000 in 1997.

Use of two coffee pots by Maureen is an example of substituting capital for labor. Another example is the use of the computer in the office, permitting automation of many office functions. The ratio of capital / labor also can be improved by economies of scale. This occurs when equipment with twice the capacity costs less than twice as much. Then capital cost / unit is reduced and fewer work hours are needed / unit of output.

3.1.3. Quantifying Improvement

‘‘Practice makes perfect’’ has been known for a long time. Wright (1936) took a key step when he published manufacturing progress curves for the aircraft industry. Wright made two major contri- butions. First, he quantified the amount of manufacturing progress for a specific product. The equation was of the form Cost = a (number of airplanes)b; (see Figure 1). But the second step was probably even more important: he made the data a straight line (by putting the curve in the axis!) (see Figure 2). That is, the data is on a log–log scale.

On a log scale, the physical distance between doubled quantities is constant (i.e., 8 to 16 is the same distance as 16 to 32 or 25 to 50) (see Figure 4). Wright gave the new cost as a percent of the original cost when the production quantity had doubled. If cost at unit 10 was 100 hr and cost at unit 20 was 85 hr, then this was an 85 / 100 = 85% curve. Since the curve was a straight line, it was easy to calculate the cost of the 15th or the 50th unit. If you wish to solve the Y = axb equation instead of using a graph (calculators are much better now than in 1935), see Table 12.

For example, assume the time for a (i.e., cycle 1)= 10 min and there is a 90% curve (i.e., b = -0.152), then the time for the 50th unit is: Y = 10 (50)-0.152 = 10 / (50)0.152 = 5.52 min.

Table 13 shows how data might be obtained for fitting a curve. During the month of March, various people wrote down on charge slips a total of 410 hours against this project charge number.

The average work hours / unit during March then becomes 29.3. The average x-coordinate is (1 + 14) / 2 = 7.5. Because the curve shape is changing so rapidly in the early units, some authors recommend plotting the first lot at the 1 / 3 point[(1 + 14) / 3] and points for all subsequent lots at the midpoint.

During April, 9 units passed final inspection and 191 hours were charged against the project. Cumulative hours of 601 divided by cumulative completed output of 23 gives average hr / unit of 26.1. The 26.1 is plotted at (15 + 23)/2 = 19. As you can see from the example data, there are many possible errors in the data, so a curve more complex than a straight line on log–log paper is not justified. Figure 4 shows the resulting curve.

Although average cost / unit is what is usually used, you may wish to calculate cost at a specific unit. Conversely, the data may be for specific units and you may want average cost. Table 12 gives the multiplier for various slopes. The multipliers are based on the fact that the average cost curve and the unit cost curve are parallel after an initial transient. Initial transient usually is 20 units, although it could be as few as 3. The multiplier for a 79% slope is (0.641 + 0.676) / 2 = 0.658. Thus, if we wish to estimate the cost of the 20th unit, it is (24.9 hr)(0.658) = 16.4 hr.

Cost / unit is especially useful in scheduling. For example, if 50 units are scheduled for September, then work-hr / unit (for a 79% curve) at unit 127 = (13.4)(0.656) = 8.8 and at unit 177 = 7.8. Therefore between 390 and 440 hours should be scheduled.

Looking at Figure 4, you can see that the extrapolated line predicts cost / unit at 200 to be 11.4 hr, at 500 to be 8.3, and at 1,000 to be 6.6. If we add more cycles on the paper, the line eventually reaches a cost of zero at cumulative production of 200,000 units. Can cost go to zero? Can a tree grow to the sky? No.

The log–log plot increases understanding of improvement, but it also deceives. Note that cost / unit for unit 20 was 24.9 hr. When output was doubled to 40 units, cost dropped to 19.7; doubling to 80 dropped cost to 15.5; doubling to 160 dropped cost to 12.1; doubling to 320 dropped cost to 9.6; doubling to 640 dropped cost to 7.6. Now consider the improvement for each doubling. For the first doubling from 20 to 40 units, cost dropped 5.20 hr or 0.260 hr / unit of extra experience. For the next doubling from 40 to 80, cost dropped 4.2 hr or 0.105 hr / unit of extra experience. For the doubling from 320 to 640, cost dropped 2.0 hr or 0.006 hr / unit of extra experience. In summary, the more experience, the more difficult it is to show additional improvement.

Yet the figures would predict zero cost at 200,000 units, and products just aren’t made in zero time. One explanation is that total output of the product, in its present design, is stopped before 200,000 units are produced. In other words, if we no longer produce Model T’s and start to produce Model A’s, we start on a new improvement curve at zero experience. A second explanation is that the effect of improvement in hours is masked by changes in labor wages / hr. The Model T Ford had a manufacturing progress rate of 86%. In 1910, when 12,300 Model T Fords had been built, the price was $950. When it went out of production in 1926 after a cumulative output of 15,000,000, the price was $270; $200 in constant prices plus inflation of $70.

The third explanation is that straight lines on log–log paper are not perfect fits over large ranges of cycles. If output is going to go to 1,000,000 cumulative units over a 10-year period, you really shouldn’t expect to predict the cost of the 1,000,000th unit (which will be built 10 years from the start) from the data of the first 6 months. There is too much change in economic conditions, managers, unions, technology and other factors. Anyone who expects the future to be perfectly predicted by a formula has not yet lost money in the stock market.

4.1.4. Typical Values for Organization Progress

The rate of improvement depends on the amount that can be learned. The more that can be learned, the more will be learned. The amount that can be learned depends upon two factors: (1) amount of previous experience with the product and (2) extent of mechanization. Table 14 gives manufacturing progress as a function of the manual / machine ratio. Allemang (1977) estimates percent progress from

product design stability, a product characteristics table (complexity, accessability, close tolerances, test specifications, and delicate parts), parts shortage, and operator learning.

Konz and Johnson (2000) have two detailed tables giving about 75 manufacturing progress rates reported in the literature.

4.1.5. Typical Values for Learning

Assume learning has two components: (1) cognitive learning and (2) motor learning (Dar-El et al. 1995a, b). Cognitive learning has a greater improvement (say 70% curve), while motor learning is slower (say 90% curve). For a task with both types, initially the cognitive dominates and then the motor learning dominates. Use values of 70% for ‘‘pure cognitive,’’ 72.5% for ‘‘high cognitive,’’ 77.5% for ‘‘more cognitive than motor,’’ 82.5% for ‘‘more motor than cognitive,’’ and 90% for ‘‘pure motor.’’ Konz and Johnson (2000) give a table of 43 tasks for which learning curves have been reported.

The improvement takes place through reduction of fumbles and delays rather than greater move- ment speed. Stationary motions such as position and grasp improve the most while reach and move improve little. It is reduced information-processing time rather than faster hand speed that affects the reduction.

The range of times and the minimum time of elements show little change with practice. The reduction is due to a shift in the distribution of times; the shorter times are achieved more often and the slower times less often—‘‘going slowly less often’’ (Salvendy and Seymour 1973).

The initial time for a cognitive task might be 13–15 times the standard time; the initial time for a manual task might be 2.5 times the standard time.

4.1.6. Example Applications of Learning

Table 15 shows the effect of learning / manufacturing progress on time standards. The fact that labor hr / unit declines as output increases makes computations using the applications of standard time more complicated. Ah, for the simple life!

4.1.6.1. Cost Allocation Knowing what your costs are is especially important if you have a make–buy decision or are bidding on new contracts. If a component is used on more than one product (standardization), it can progress much faster on the curve since its sales come from multiple sources. Manufacturing progress also means that standard costs quickly become obsolete.

Note that small lots (say due to a customer emergency) can have very high costs. For example, if a standard lot size is 100 and labor cost is 1 hr / unit and there is a 95% curve, a lot of 6 would have a labor cost about 23% higher (1.23 hr / unit). Consider charging more for special orders!

4.1.6.2. Scheduling Knowing how many people are needed and when is obviously an important decision. Also, learning / manufacturing progress calculations will emphasize the penalties of small lots.

4.1.6.3. Evaluation of Alternatives When alternatives are being compared, a pilot project might be run. Data might be obtained for 50–100 cycles. Note that the times after the pilot study should be substantially shorter due to learning. In addition, the learning / manufacturing progress rate for alternatives A and B might differ so that what initially is best will not be best in the long run.

Even a small learning rate can have a major effect on performance. X = experience level of the operator when time study was taken, for example, 50, 100, or 500 cycles. The table gives time / unit based on a time standard of 1.0 min / unit; therefore, if actual time standard were 5.0 min / unit, then time / unit at 98% and 2X would be 0.98 (5.0) = 4.9.

4.1.6.4. Acceptable Day’s Work Assume a time standard y is set at an experience level x. Assume further, for ease of understanding, that y = 1.0 min and x = 100 units. A time study technician, Bill, made a time study of the first 100 units produced by Sally, calculated the average time, and then left. Sally continued working. Let’s assume a 95% rate is appropriate. Sally completes the 200th piece shortly before lunch and the 400th by the end of the shift. The average time / unit for the first day is about 0.9 min (111% of standard). The second day, Sally completes the 800th unit early in the afternoon. She will complete the 3200th unit by the end of the week. The average time of 0.77 min / unit during the week yields 129% of standard!

Point. If none of the operators in your plant ever turns in a time that improves as they gain experience, do you have stupid operators or stupid supervisors?

Next point. The magnitude of the learning effect dwarfs potential errors in rating. You might be off 5% or 10% in rating but this effect is trivial compared versus the errors in using a time standard without considering learning / mfg. progress.

Third point. Any time standard that does not consider learning / manufacturing progress will be- come less and less accurate with the passage of time.

Final point. Since learning and manufacturing progress are occurring, output and number of work hours should not both be constant. Either the same number of people should produce more or fewer people can produce a constant output.