آموزش و بهبود فرآیند در صعود تولید

Rapid product lifecycles and high development costs pressure manufacturing firms to cut not only their development times (time-to-market), but also the time to reach full capacity utilization (time-to-volume). The period between completion of development and full capacity utilization is known as production ramp-up. During that time, the new production process is ill understood, which causes low yields and low production rates. This paper analyzes the interactions among capacity utilization, yields, and process improvement (learning). We model learning in the form of deliberate experiments, which reduce capacity in the short run. This creates a trade-off between experiments and production. High selling prices during ramp-up raise the opportunity cost of experiments, yet early learning is more valuable than later learning. We formalize the resulting intertemporal trade-off between the short-term opportunity cost of capacity and the long term value of learning as a dynamic program. The paper also examines the tradeoff between production speed and yield/quality, where faster production rates lead to more defects. Finally, we show what happens if managers misunderstand the sources of learning.

Many high-tech industries are characterized by shrinking product lifecycles and increasingly expensive production equipment and up-front costs. The market window for selling many products has shrunk to less than a year in industries such as disk-drives and telecommunications. These forces pressure organizations to cut not only their development times (time-to-market), but also the time it takes to reach full production volume (time-to-volume) in order to meet their financial goals for the product (time-to-payback). The period between the end of product development and full capacity production is known as production ramp-up. Two conflicting factors are characteristic of this period: low production capacity, and high demand. High demand arises because the product is still “relatively fresh” and might even be the first of its type. Thus, customers are ready to pay a premium price. Yet output is low due to low production rates and low yields. The production process is still poorly understood and, inevitably, much of what is made does not work properly the first time. Machines break down, setups are slow, special operations are needed to correct product and process oversights, and other factors impede output. Over time, with learning about the production process and equipment, yields and capacity utilization go up (although in many industries they never reach 100%). Due to the conflicts between low effective capacity and high demand, the company finds itself pressured from two sides, an effect referred to as the “nutcracker” [1]. A recent example of the importance of ramp-up can be found in AMD's efforts to compete with Intel in the microprocessor market. AMD had several generations of product that were slow to ramp, leading to limited market acceptance and financial difficulties for AMD. More recently, Intel experienced problems ramping up the yield of its 0.18 micron version of the Pentium. Industry observers speculate that an effective ramp-up of AMD's K7 processor will allow AMD to compete in the high end segment of the PC market (Electronic Buyers’ News, June 21, 1999). In this article, we analyze the interactions among capacity utilization, yields, and yield improvement (learning) during ramp-up. Traditional learning-curve models implicitly assume that manufacturing performance increases with cumulative output from the plant, more or less independent of managerial decisions. This is clearly an oversimplification, and there is much that managers can do to affect the rate of learning [2]. We concentrate on deliberate learning through experiments such as engineering trials, which are controlled experiments using the production process as a laboratory. Such trials are essential for diagnosing problems and testing proposed solutions and process improvements. But they also use scarce production capacity. This creates a paradoxical trade-off between regular production for revenue and experimentation for learning. We formalize this intertemporal trade-off between short-term revenues and long term learning benefits in form of a dynamic program, and derive solutions for the cost, value, and level of experimentation. The trade-off between short-term output and experiments, as well as more generally the phase of production ramp-up, is of substantial managerial importance. Launches of high-tech products are often either delayed or scaled back because of ramp-up problems. For example ramp-up problems in the production of video chips led to substantial losses during the launch of the Sega Dreamcast video console [3]. Similarly, pharmaceutical companies are struggling with ramping up the production of biotechnology-based drugs, leading to sales losses at the time when prices are at their premium [4]. This article models the complex dynamics of a new product's ramp-up, and assists decision making by providing concrete values for the cost and benefits of learning efforts. Specifically, we show that a misperception about the underlying drivers of learning can result in substantial financial losses over the lifecycle of a new product. The remainder of this article is organized as follows. Section 2 provides more background on the assumptions of our model, and discusses several strands of related literature. Section 3 describes the type of production environments our analysis is appropriate for and presents a simple model that captures the interaction among capacity utilization, process knowledge, and yields. The analysis of this (static) model will be the basis for our dynamic model of learning and process improvement during production ramp-up, presented in Section 4. Our results are illustrated by several numerical examples in Section 5, where we show that different cost and demand situations call for different ramp-up strategies. Section 6 provides a summary, managerial implications, and future research directions.

We have presented an analytical model of pro- duction ramp-up, which combines a static trade-o !between yields and utilization with a dynamic trade-o ! of learning and process improvement. In today ' s rapidly changing environments, cutting de- velopment times (time-to-market) in itself is not su $ cient. Another key to achieving high pro " tis a rapid production ramp-up of a new product. This includes quickly achieving both high yields and a high level of utilization. Our " ndings have a number of managerial im- plications. Most basic is the need for managers to accept and deal with the inherent paradox of learn- ing during production ramp-up. At the beginning of a ramp when prices are at their highest, and yields and output at their lowest, it is nonetheless still the moment to further reduce output in order to run engineering trials and work on yield and speed improvements. This paradox often creates, in our experience, strong pressures to take shortcuts in learning, such as experiments with overly small sample sizes relative to the process noise level, or not running validation trials before implementing process changes. While this keeps up-time higher in the short run, it often leads to problems which reduce performance for the rest of the ramp-up period and beyond. We deal with this paradox by explicitly calculating the cost and value of experi- mentation as functions of time and processing ca- pability. Figs. 6 and 7 show the patterns that can result from optimal behavior. Second, we show the importance of understand- ing the sources of learning. It is incorrect to treat learning as an exogenous process beyond manage- rial control. Rather, there are three key high-level inputs which should be explicitly allocated and managed. These are normal production experience, capacity withdrawn from production for experi- ments of many kinds, and engineering time. Only the " rst of these happens automatically. Only en- gineering time (which we modeled as being lumped together with experimentation) appears explicitly in a cost accounting system. But the dollar costs of experimentation time, although not captured in accounting systems, can be a large investment as well, and are integral to success. Our analysis in Section 3 provides a " rst look at the important trade-o ! between yields and produc- tion speed. With di ! erent product economics, and at di ! erent times in ramp-up, the optimal levels ofcare and rework shift. It also serves as a strong reminder that in yield driven industries there is a large di ! erence between utilization and e ! ective utilization. Finally, this research illustrates the im- portance of time-to-volume compared with the still dominant paradigm of time-to-market. We show how di ! erent situations require di ! erent decisions during the ramp-up period. We have kept the model as simple as possible in order to focus on structural results. This approach clearly has limitations. Our assumptions that the processing capability can be represented as a single number, as well as the assumptions concerning the functional forms of learning rates and sub-additiv- ity are strong simpli " cation of real ramp-up situations. Re " nements of the model provide inter- esting avenues for future research. First, some of the assumptions could be relaxed. For example, prices and competitive behavior could be explicitly modeled. Spence [31] provided an in # uential analysis of the e ! ect of learning on strategic competition. He modeled a " rm investing in learn- ing early, in order to deter entry by potential competitors. In his model, learning was an inherent by-product of production experience, so that the form of ` investment a was to produce more . The " rm uses low prices both to encourage demand, and to serve as a signal to competitors that it has made an investment. This leads to the prescription to ` price ahead of the learning curve a . In our model, " rms can also invest in learning, but in the form of deliberate learning through more time for experiments (and less production). A combination of both models seems promising. Second, we see a strong need for more empirical research on this topic. Detailed case studies on the ramp-up period will help to reveal addi- tional variables [32]. Such case studies could try to develop a managerial check-list of items that need to be addressed before or during the ramp-up. Another empirical research opportunity lies in a detailed econometric analysis of yield and utiliz- ation curves over time, which tries to identify the most e ! ective variables that help increase the e ! ective capacity. Finally, the issues of production ramp-up should be linked to the existing " elds of product develop- ment and learning in manufacturing. Although inthe present paper we try to explicitly include " nd- ings from the manufacturing learning literature, we do not su $ ciently include aspects of product devel- opment. What happens during product develop- ment will have a strong impact on the initial processing capability as well as on the speed of ramp-up. Thus, linking the quality of the ramp- up to events during the product development process provides a third interesting avenue for future research.