IIC Journal of Innovation 12th Edition | Page 124

Outcomes, Insights and Best Practices from IIC Testbeds: Smart Factory Machine Learning for Predictive Maintenance Tesbed customer engagements are currently being explored. limiting results to a proof-of-concept or applications in a controlled environment. An important issue is that, currently, there are many “expert systems” in production; however, those systems have problems with large false positive rates. In the testbed team’s experience, almost all expert systems of the type in focus experience shutdowns. In a facility such as the OEM testing this testbed solution, downtime costs USD 50,000 per hour, with unexpected failures potentially taking 40 hours to address. In this case, the testbed team’s approach is to test specific algorithms that take into account the dynamics of industrial systems, where degradation occurs and normal limits are understood (such as our cars where, five years later, they are not as new but work perfectly). These dynamic algorithms are also mathematically supported, where the designer (data scientist) is able to understand what the algorithm is doing internally (not black-box algorithms such as neural networks, deep learning, etc.). All these efforts help the predictive analytics to minimize the false positives rate and save significant costs.” E XPERIENCE Aingura IIoT has derived business value through the Smart Factory Machine Learning Testbed by connecting with high-quality vendors of specific technologies Aingura needs to develop products. Finding vendors with high-quality products to fulfill specific technology needs is very difficult, but the IIC ecosystem facilitates connections which would otherwise be unlikely to form outside the IIC. Not only has this helped Aingura create new customers, but the testbed itself has benefitted—for example, accessing a testbed partner’s products for use in its hardware platform. In addition, the testbed can act as a marketing tool for partners to showcase their products in real applications. The partnerships within the testbed expedited certain processes necessary to build the testbed. The relationship with iVeia and Xilinx, for example, enabled record-time building of the new hardware platform with the latest technology from each company. Such a complex platform was expected to take two or three years to create, but the relationships shortened this process to only six months. One key takeaway learned by the testbed team is that something is not finished when only tested at a home facility. The team may design the best machine-learning algorithms, best IoT hardware and best architecture, but many changes take place when going from a lab to a real industrial environment. If not tested during real production, there will be many false- positives. For other testbeds and companies considering an IIoT implementation, the testbed team strongly suggests conducting tests in real environments rather than IIC Journal of Innovation C LOSING The most significant surprise throughout the testbed’s journey has been the relationships formed between partners. In particular, it was unexpected how open to collaboration each partner was and how an open communication protocol was established. When examining the testbed’s progress, - 119 -