With the acute shrinking of product life cycles as well as the increasing pace of technological and organi-zational innovation, in most situations facilities should not anymore be located and laid out assuming a steady state perspective as was generally done in the past. Layout and location dynamics, explicitly con-sidering the time-phased evolution of facilities and networks, is thus also becoming a key issue for the engineer (Rosenblatt 1986, Montreuil 2001). He has to recognize that as the current design is about to be transformed into the proposed design, this proposed design will have a finite existence. It will also have to be transformed into a subsequent design at a later time. The same will occur to this subsequent design and all subsequent others, in a repeating cycle over the entire life of the facility in layout cases or the network of facilities in location cases.
Only when relaying out or redeploying facilities involves insignificant efforts can the engineer opti-mize the next design strictly for the near future expectations, as (i) there will be negligible costs in transforming the current design into the next design and (ii) it will later be easy to reshuffle this next design into subsequent designs as needed. In most cases, however, there are significant costs involved in dynamically altering designs. Thus, the engineer has to explicitly deal with the dynamic evolution of his designs. This implies for him developing a dynamic plan as illustrated in Figure 5.13, which shows a four-year layout plan for a facility. Figure 5.13 uses gray shadings to distinguish processors in terms of expected moving cost.
In this age of high market turbulence, the complexity of dealing with the dynamic nature of the design task is confounded by the fact that all demand, process, flow, and space requirements are estimates based on forecasts and that these forecasts intrinsically are known to be ever more prone to error as one looks further into the future. For example, what will be the demand for a product family tomorrow, next month, next quarter, next year, in three years?
ABCD.1 ABCD.2 ABCD.3 ABCD.4 E.1 E.2
FGJ.2
FGJ.1 HI.1 HI.2
HI.3 HI.4
HI.5 HI.6 HI.7 HI.8
HI.9 HI10
HI11 HI12 HI13 HI14
HI15 HI16
HI17 HI19 HI20 MP/PF
MP / PF
MP / PF
HI18
ABCD.1 ABCD.2 ABCD.3 ABCD.4 E.1 E.2
FGJ.2
FGJ.1 HI.1 HI.2
HI.3 HI.4
HI.5 HI.6 HI.7 HI.8
HI.9 HI10
HI11 HI12 HI13 HI14
HI15 HI16
HI17 HI19 HI20 MP/PF
MP / PF
MP / PF
HI18
ABCD.5 ABCD.6 E.3
ABCD.1 ABCD.2 ABCD.3
ABCD.4 E.1 E.2
FGJ.2
FGJ.1 HI.1 HI.2
HI.3 HI.4
HI.5 HI.6 HI.7 HI.8
HI.9 HI10
HI11 HI12 HI13 HI14
HI15 HI16
HI17 HI19 HI20 MP/PF
MP / PF
MP / PF
HI18
ABCD.5 ABCD.6 E.3
E.4 E.5 ABCD.10 ABCD.11
ABCD.7 ABCD.8 ABCD.9
MP/PF
ABCD.1 ABCD.2 ABCD.3
ABCD.4 E.1 E.2
FGJ.2
FGJ.1 HI.1 HI.2
HI.3 HI.4
HI.5 HI.6 HI.7 HI.8
HI.9 HI10
HI11 HI12 HI13 HI14
HI15 HI16
HI17 MP/PF
MP / PF
MP / PF
HI18
ABCD.5 ABCD.6 E.3
E.4 E.5 ABCD.10 ABCD.11
FGJ.2 HI19 HI20 HI21 HI22
E.6 E.7
MP/PF MP/PF
AB CD 7
AB CD 8
AB CD 9
AB CD 12
AB CD 13
AB CD 14
AB CD 15
Year 1
Year 2
Year 3
Year 4
FIGuRE 5�13 Myopically generated dynamic layout plan.
Tables 5.10 and 5.11 illustrate this phenomenon for the products of Table 5.4. In these tables, the demand stated in Table 5.4 becomes the expected average daily demand in the fourth future year. There are also forecasts for the first three years preceding this fourth year. Table 5.10 shows that the demand for some products is forecasted to be expanding while the demand for others is forecasted to be shrinking. For each forecast of Table 5.10, Table 5.11 provides the estimated standard deviation over the forecasted mean.
For example, in year 1 the average daily demand for product 3 is forecasted to be 6,000 units with a stan-TABLE 5�10 Multi-Year Demand Forecasts
Product Number
Expected Daily Demand per Year
Y1 Y2 Y3 Y4
1 1000 2000 3000 2500
2 0 0 270 560
3 6000 5750 5500 5250
4 300 275 250 225
5 0 0 0 120
6 0 1000 2000 3000
7 0 0 0 50
8 0 0 0 30
9 0 0 0 30
10 1000 1000 950 875
11 800 750 700 650
12 25 25 25 25
13 0 0 0 25
14 0 0 0 75
15 200 300 450 625
16 0 50 100 175
17 200 500 1000 1500
18 125 110 100 125
TABLE 5�11 Multi-Year Uncertainty of Average Daily Demand Forecasts
Product Number
Standard Deviation of Expected Average Daily Demand
Y1 Y2 Y3 Y4
1 150 360 675 750
2 0 0 30 83
3 300 345 413 525
4 20 22 25 30
5 0 0 0 25
6 0 200 500 1000
7 0 0 0 12
8 5 0 0 8
9 2 0 0 3
10 50 60 71 88
11 200 225 263 325
12 5 6 8 10
13 0 0 0 5
14 0 0 0 20
15 40 72 135 250
16 0 15 38 88
17 140 420 1050 2100
18 6 6 7 12
dard deviation of 300 units, while in year 3 the average daily forecast is down to 5,500 units, yet with a high standard deviation of 413. Such information may come from analyzing historical forecast performance in forecasting demand for such a family, respectively, one year and three years ahead (Montgomery et al.
1990). This means that within two standard deviations (two-sigma) or 98% probability using normal dis-tribution estimation, in current year zero the average daily demand for P3 is expected to be between 5,400 and 6,600 units in year 1, and between 4,674 and 6,326 units in year 3.
Using the process requirements of Table 5.4 and assuming the flexible processors introduced in the lower part of Figure 5.9, these forecasts permit computation of estimates for the average expected number of processors of each type, provided in Table 5.12. Also, they allow robust estimations for processor require-ments, such as the two-sigma robust estimates provided in Table 5.13. In year 3, for processor type ABCD, the average estimate is 9 units, while the two-sigma robust estimate is 11 units. Overall the robust estimate adds up to a total of 28 processors in year 1 to a total of 47 in year 4.
Figure 5.13 provides a four-year layout plan generated in year 0 by an engineer. To help understand the compromises involved, the engineer was asked to first generate a design for year 1 based on the estimates for year 1. He had to then transform this year-1 design into a year-2 design, taking into consideration the expected flows for year 2 and the cost of transforming the year-1 design into the year-2 design. He had to repeat this process for years 3 and 4. Clearly, this is a rather myopic approach because in no time was he considering the overall forecasted flows and processor requirements for the entire four-year planning horizon. Analyzing the plan, it is clear that the engineer’s decision in year 1 to lay out the two FGJ proces-sors adjacent to each other has defined a developmental pattern that has had repercussions on the designs he has produced for year 2 to year 4. Even though possible, he has not planned to move any of the processors E and FGJ once laid out in their original location, which has created a complex flow pattern in year 4, as contrasted with the elegant simplicity of the lower layout of Figure 5.9. Formally evaluating the dynamic plan requires evaluation of each design statically as described in the previous sections, and computation of the expected transformational costs from year to year. The evaluation requires the gen-eration of demand scenarios probabilistically in line with the forecast estimates of Tables 5.10 to 5.13.
Due to space constraints, the results of such an evaluation for the plan of Figure 5.13, and the generation and evaluation of alternative plans that take a more global perspective, are left as an exercise to the reader.
TABLE 5�12 Multi-Year Expected Average Processor Requirements
Processor Type
Expected Average Processor Requirements
Y1 Y2 Y3 Y4
ABCD 4 5 9 13
E 2 2 3 4
FGJ 2 2 2 2
HI 19 19 19 20
Total 27 28 33 39
TABLE 5�13 Multi-Year Two-Sigma Robust Processor Requirements
Processor Type
2-Sigma Robust Processor Requirements
Y1 Y2 Y3 Y4
ABCD 4 6 11 15
E 2 3 5 7
FGJ 2 2 2 3
HI 20 20 20 22
Total 28 31 38 47
In practice, there are two main strategies to deal with dynamics. The first is to select processors and facilities that enable easy design transformation, and to try to dynamically alter the design so as to always be as near to optimal as possible for the forthcoming operations. Figure 5.13 can be seen as an example of this strategy. The second is to develop a design that is as robust as possible, as immune to change as pos-sible, a design that requires minimal changes to accommodate in a satisfactory manner a wide spectrum of scenarios (Montreuil and Venkatadri 1991, Montreuil 2001, Benjaafar et al. 2002). Figure 5.14 provides an example of this strategy by simply expanding the robust design of Figure 5.11 to be able to deal with the estimated requirements for year 4. It is left as an exercise to assess how to subtract processors from Figure 5.10 to deal with the lower expected requirements for years 1 to 3.
In the above examples, a yearly periodicity has been used for illustrative purposes. In practice, the rhythm of dynamic design reassessment and transformation should be in line with the clock speed of the enterprise, in synchronization with the advent of additional knowledge about the future and the lead time required for processor and facility acquisitions and moves. Even decades ago, some companies were already reconfiguring their shop floor layouts on a monthly basis, for example, in light assembly factories dedicated to introducing new products on the market, assembling them until demand justified mass production.
In the illustrative example of Figure 5.13, the planning horizon has been set to four years. Again, this depends on the specific enterprise situation. It can range from a few days in highly flexible, easy-to-alter designs to decades in rigid designs in industries with low clock speeds.