IIC Journal of Innovation 12th Edition | Page 123

Outcomes, Insights and Best Practices from IIC Testbeds: Smart Factory Machine Learning for Predictive Maintenance Testbed IISF from the cybersecurity point of view. published a book, Industrial Applications of Machine Learning, in late 2018 which features two chapters that discuss the testbed’s results in the Aingura lab. Time Sensitive Network (TSN) standards, as related to the IIC TSN Testbed, play a role in the implementation of the Smart Factory Machine Learning Testbed. Between decision-making, processing and pre- processing, there is communication—the transportation of data from one place to another. The deterministic approach of TSN increases the security of the data during communication, helping to avoid the loss of that data. In the Smart Factory Machine Learning Testbed, the gathered data needs to feed into the machine learning algorithm—TSN protects the data during this communication. The deployment of the testbed’s technologies in the controlled lab environment is fairly straightforward, but deploying in the real industrial environment involves several challenges. In a real production facility, the window of time available to deploy the technology is drastically limited due to the production schedule of the facility. As a machine is producing, it cannot be stopped or tampered with for the sake of running tests. When that short window of time opens, the system is connected and deployed, but to check if it is working properly may take several months and is thus not feasible. This differs from deploying a brownfield system which can be deployed in existing production lines. In a scenario where production machines are built in the Aingura lab and sent to the end user with the technology already installed, proper validation tests can be implemented. However, it is not common that a machine is installed in a new production line—so the opportunities to deploy the testbed’s technology are limited. T ESTBED R ESULTS The testbed’s first deliverable was published in the IEEE Internet of Things Journal in May 2018, covering the results of one of the testbed’s first dynamic machine learning algorithms which works with the data stream coming from a machine. At this time, the testbed team is working on a new article for the Engineering Applications of Artificial Intelligence which is aimed to be published by early 2020. This paper discusses machine learning algorithms oriented to predictive maintenance. Next, the testbed team was granted a U.S. patent for their architecture that increases computing power by connecting several aspects of Aingura’s hardware platform to deploy the testbed’s solutions. The hardware integration gives more computing power at the edge and allows for more complex machine learning algorithms to be deployed in real industrial environments. Finally, the testbed team To bypass this challenge, Aingura’s parent company Etxe-Tar, which has developed strong relationships with OEMs for the last 20 years, is able to provide a channel for the testbed’s technology to find its way into production facilities. Without a strong relationship with an end customer’s production department, this is far more difficult. The testbed team seeks to find other ways to implement the testbed in real production environments, and some - 118 - November 2019