LAUSR

Analytics applications are becoming indispensable in today’s business landscape. Greater data availability from self-monitoring production equipment allows firms to empower individual workers on the shop-floor with powerful decision support solutions. To explore the potential of such solutions, we replicate an important manual leak detection process from high-tech composite manufacturing and augment the system with highly sensitive sensors. Based on this setup we illustrate the main steps and major challenges in developing and instantiating a predictive decision support system. By establishing a scalable and generic feature generation approach as well as leveraging techniques from statistical learning, we are able to improve the forecasts of the leak position by almost 90%. Recognizing that mere forecast information cannot be evaluated with respect to business value, we subsequently embed the problem in an analysis of the underlying searcher path problem. We compare predictive and prescriptive search policies against simple benchmark rules. The data-supported policies dramatically reduce the median as well as the variability of the search time. Based on these findings we posit that prescriptive analytics can and should play a greater role in assisting manual labor in manufacturing environments.