Learning from Productivity Learning Curves: A Log-Linear Model Can Help to Predict the Operating Performance of New Plants, Identify Under-Performing Processes and Benchmark Past Projects in Order to Identify Best Practices for Future Projects

New technology platforms evolve at USG Corporation through traditional discovery, development, pilot demonstration, engineering, and commercialization stages. Typical concept-to-commercialization (CTC) times are four to ten years. USG researchers have the leading role in the early stages and a strong support role during and after commercialization. A study of USG' s commercialization of new technology platforms was initiated to evaluate productivity gains through learning by doing (LBD) or experience-based learning. Specific goals of the work were to establish: (1) how do LBD concepts apply to USG production lines, (2) are there common rates of learning across panel technology platforms, (3) can historical results be used to predict future process performance, and (4) can LBD analysis be used as a tool to identify processes performing below their potential? This article addresses the productivity histories for one panel technology platform involving first- and second-generation process designs. Wright first proposed a log-linear relationship between worker productivity and accumulated experience as illustrated in Figure 1 (1). This model is widely used to analyze productivity learning curves and was selected as the model of choice for this work. Yelle suggested a tendency for productivity to plateau in more capital-intensive operations (2). Klenow's work indicated "evidence that learning is specific to each production technology and yields substantial and diminishing productivity gains" (3). And Argote and Epple raised the issue of "organizational forgetting" due to such factors as employee turnover and transfer of knowledge (4). The log-linear model along with concepts of "plateauing" and "organizational forgetting" were investigated in this study. Methodology Production records for new technology platform production lines were analyzed representing first- and later-generation process designs. Traditionally, productivity is measured as labor hours per unit output. For this study, productivity was defined as the net unit output per gross scheduled hour of operation. This metric accounts for factors such as production rates, process efficiency and process reliability. [FIGURE 1 OMITTED] P = 60 x [P.sub.rate] x [[P.sub.eff] x [P.sub.rel]], where: P = productivity, [ft.sup.2]/gross hour, [P.sub.rate] = production rate, [ft.sup.2]/min, [P.sub.eff] = process efficiency, percent, [P.sub.rel] = process reliability, percent. Process efficiency ([P.sub.eff]) is the ratio of net production to gross production in percent. Process reliability ([P.sub.rel]) is the ratio of actual production hours to scheduled production hours in percent. Overall productivity (P), process efficiency ([P.sub.eff]), and reliability ([P.sub.rel]) factors were analyzed versus cumulative net hours of operation. Net hours of operation was selected versus gross hours assuming that most learning occurs while the production lines are running. A process productivity progress ratio was defined as the ratio of productivity at a given time to productivity at 1,000 cumulative net hours of operation. The 1,000 net hours was selected as the baseline to avoid the highly volatile process performance experienced during initial production. Similarly, a process performance ratio was defined as the product of process efficiency and reliability [[P.sub.eff] x [P.sub.rel]] and normalized to performance at 1,000 cumulative net hours of operation. What We Learned Productivity performance at new USG operations was found to be consistent with the log-linear model. However, three distinct productivity periods were identified, as illustrated in Figure 2. [FIGURE 2 OMITTED] The first, or stagnant period occurs over the initial first 1,000 net hours of operation immediately following commencement of production. Typically, this period lasts less than six months. …