Real-time Production Performance Analysis Using Machine Degradation Signals: a Two-Machine Case Yunyi Kang, Hao Yan and Feng Ju Abstract— Machine degradation has significant impact on the production system performance. Its variation might lead to large deviation from the steady state performance of the whole system. In this work, we build up a model to estimate the long-term production performance of the two-machine-and-one-buffer production systems, given the real-time machine degradation signals. A phase-type distri- bution is generated to mimic the remaining life distribution of each machine given the degradation signal. Then a continuous Markovian model is formulated to predict long- term system throughput rate for a two-machine-and-one- buffer system. With the fluctuation of machine degradation signals, such a model can effectively estimate the expected system performance in real-time. I. I NTRODUCTION Performing real-time system output estimation and control of modern manufacturing systems is impor- tant while still challenging for many practitioners. The fluctuations of machine conditions, random disruption events, and maintenance activities can influence instant system performance and long-term predictions [1]. Espe- cially as modern manufacturing systems are becoming more and more complex, system performance predic- tion based on real-time system information is highly demanded. Moreover, with the development of sensor technology, massive sensors have been deployed to collect machine condition information, such as machine degradation sig- nals, to monitor real-time system performance. However, a significant gap still exists between data collection and system level decision making. Therefore, a novel approach by integrating available sensor information and system physical properties is needed to evaluate system performance and conduct production control in real- time. In this work, machine degradation, a gradual and accumulating process that deteriorates the machine op- erating conditions, is considered. It can influence the performance of a production system, such as production rate, product quality, energy consumption and produc- tion cost, which are discussed in many research works *Y. Kang H. Yan, and F. Ju are with School of Computing, Informatics, and Decision Systems Engi- neering, Arizona State University, Tempe, AZ 85281, USA. ykang37@asu.edu,HaoYan@asu.edu, Feng.Ju@asu.edu [2]. Typically, a degradation signal is defined as a quantity computed from sensor information that captures the current state of the machine and provides information on how that condition is likely to evolve in the future [3]. Machines will fail when the degradation signal reaches a well-defined threshold. To model the production systems with degradating machines, earlier research works develop queuing mod- els to estimate system long-term performances [4], [5], [6]. In addition, Markovian analysis is also investigated based on the Bernoulli models, geometric models, multi- state models, and exponential models [7], [8], [9]. How- ever, these models are difficult to incorporate real-time degradation information due to the strong requirement on operating distributions. Moreover, such a requirement is not widely met in the production practice, according to the empirical and analytical studies [10], [11]. In recent years the newly developed analytical models provide higher flexibilty, with adjustable paramters on machine and thus less requirement on machine operating distri- bution [12], [13], [14], [15]. Nevertheless, these models do not provide clear guidance on incorporating real- time degradation signals to update system parameters and to fit for long-term production system performance estimation. On the other hand, the reliability and prognostics research typically investigates real-time sensing data and degradation signals for degradating machines to predict performance based on machine health status from a single sensor [3], [16] and multiple sensors [17], [18]. However, such research mainly focuses on individual machine, seldom discussing how to integrate the study results to complex manufacturing systems and improve the system performance. In complex manu- facturing systems, it is difficult to directly extend the study on individual machine to larger systems without a comprehensive study on system properties. Therefore, a model on system performance estimation using real-time degradation signals is pressingly needed. To fill the research gap, we develop an analytical model to estimate the long-term performance of a two- machine-and-one-buffer production system given the degradation signal of each individual machine. To the best of our knowledge, this is the first work on system throughput modeling utilizing the real-time degradation 2018 IEEE 14th International Conference on Automation Science and Engineering (CASE) Munich, Germany, August 20-24, 2018 978-1-5386-3593-3/18/$31.00 ©2018 IEEE 1501 Authorized licensed use limited to: ASU Library. Downloaded on May 22,2020 at 06:07:22 UTC from IEEE Xplore. Restrictions apply.