ARC Advisory Group’s Paula Hollywood moderated a session, IT/OT/ET Convergence at the 2018 ARC Industry Forum. Here’s the description from this session:
The convergence of IT (Information Technology at the enterprise level) and OT (Operations Technology, the information and automation technologies employed in the plant), is not new, but the relationship is intensifying. What is changing is the addition of ET (Engineering Technology) to this convergence. ET consists of newer technologies such as Ethernet/Wi-Fi, virtualization, cloud, SaaS, analytics, Big Data, mobile, social, modeling and simulation, augmented reality, machine learning, remote monitoring, and digital twin that create virtual models. Convergence of all three disciplines enables digital enterprises to operate in innovative and collaborative ways to improve performance.
Paula opened the session describing the Digital Twin, a virtual product, piece of equipment, etc. that replicates the physical one, where data is linked between the virtual and physical to optimize the physical representation. Real time condition monitoring can be performed. Predictive maintenance alerts and what if scenarios can be performed. Paula and the ARC team expects the usage of digital twin technology to grow rapidly. A digital twin fosters collaboration, compresses time to value, optimizes production and maintenance, provides early discovery of issues, and allows continuous refinement of models.One of the speakers was Blair Fraser with Emerson local business partner, Lakeside Controls. Blair is a maintenance and reliability leader and has over 20 years’ experience in designing, commissioning, maintaining and improving manufacturing equipment and processes for the manufacturing industry.
Blair described a project with a pharmaceutical manufacturer who was experiencing reliability issues around an autoclave. It was important to start with the fundamentals. It’s important to understand all the possible failure modes. This not only help understand what the failures were, but got a cross functional team talking about the issues and bringing them together on a path to a solution.
The first approach was to perform manual data collection. Because the autoclave operated in so many different states, it didn’t work to compare data taken at different times because the operating conditions were so different. The second approach also failed. It was based on rules for online condition monitoring. The problem was that it would require 32 factorial—a number so large it could not be implemented.
The approach that did work was to use ultrasonic sensors to listen to each of the valves and use machine learning for pattern recognition of the condition of each valve. With the valves operating normally, the patterns would remain consistent when viewed together over a long-time interval, measured in months.
Blair and the project team built the model—they trained it, they chose the influencing factors, and they re-trained the model. The model runs continuously and provides an anomaly score from 0 to 100—0 being what’s expects and 100 being the complete opposite of what is expected. An alert is generated based on the degree of anomaly and for how long it is present. This was tuned by the plant staff.
Blair shared the story of the model showing the anomalies being high on several valves before the start of the batch. The plant started the batch and it failed from a compressed air shortage during the batch. The lesson was to trust what the model was indicating and determining the root cause before starting the batch.