IIC Journal of Innovation 12th Edition | Page 120

Outcomes, Insights and Best Practices from IIC Testbeds: Smart Factory Machine Learning for Predictive Maintenance Tesbed To extend the usefulness of the published testbeds in the Testbed Program of the Industrial Internet Consortium (IIC), the Testbed Working Group has developed an initiative to interview the contributors of selected testbeds to showcase more insights about the testbed, including the lessons learned through the testbed development process. This initiative enables the IIC to share more insights and inspire more members to engage in the Testbed Program. This article highlights the Smart Factory Machine Learning for Predictive Maintenance Testbed. The information and insights described in the following article were captured through an interview conducted by Mr. Howard Kradjel, Vice President of Industry Programs at IIC, with Dr. Javier Díaz, Chief Technology Officer of Aingura IIoT. Javier is an active member in the IIC where he serves as a co-lead of the Smart Factory Machine Learning Testbed and is a key contributor to the Testbed Working Group. S MART F ACTORY M ACHINE L EARNING FOR P REDICTIVE M AINTENANCE T ESTBED – F ROM C ONCEPT TO R EALITY the quality of the data coming from the industrial environments (i.e., in terms of noise reduction) in order to get the right quality of data to perform the advanced analytics. The transportation of data from one place to another, the storage of that data and the implementation of these new machine learning algorithms must contribute to forming actionable insights for the end user. The actionable insights depend on the question or problem the end users are trying to solve. Related to predictive maintenance, the questions are related to the degradation level of a specific part of the machine, cell or line that could fail stopping the production, i.e., increasing downtime or long mean time between failures (MTBF), etc. Therefore, the output of the algorithms or the machine learning system is to tell the end user the remaining useful life (RUL) of that particular element. For example, a first use case in the testbed is working to predict the RUL of the frontal ball-bearing of a spindle head. Specifically, the actionable insights given to the end user is the % of RUL, Founded in 2017, the Smart Factory Machine Learning for Predictive Maintenance Testbed seeks to test algorithms and architecture solutions in the form of technologies: communication protocols, cloud platforms, cybersecurity, etc. to achieve predictive maintenance. Many existing machine learning technologies can be applied to predictive maintenance, but most are not fully developed to optimize the accuracy or performance of those elements in order to retrieve actionable insights. Many are able to detect a failure, but they do not have the ability to perform the specific activity in real, industrial environments. The testbed’s goal is to develop algorithms and test them in IIoT architectures and real industrial environments. Some experimentations lead to the development of dynamic algorithms— machine learning algorithms that are not necessarily well expressed in state-of-the-art industrial environments but that guarantee IIC Journal of Innovation - 115 -