Every microprocessor-based device is capable of providing some form of self-diagnostics—whether those diagnostics are used is a separate question. Fortunately, according to recent information I read, they're being used as much as 40% of the time when the device is able to communicate on a network (HART, one of the fieldbuses, or wireless).
Having the information to understand the current condition of your facilities and how the individual components work together to make your product is one step toward an effective asset management program.
Assuming you’ve determined that applying asset management to your intelligent devices is a good place to start, your critical list will probably include your safety system, valves and any other devices that, like valves, have moving parts susceptible to wear and tear and, in the event of failure, will affect operations.
Then comes the challenge of how to manage the orders-of-magnitude of added data available from each intelligent device—not only determining which information is important, but also how to route and manage this information, so it doesn’t negatively impact the operation of your control system.
One approach proposed by the NAMUR working group and implemented in many of today’s intelligent devices is to pre-process diagnostics in the device itself (basic edge computing) to provide a predetermined notification (NE 107–5 states) or a common trouble alarm (NE43–2 states) with one state being good or normal operations. Then, knowing the type of alarm and routing/using it properly, the device can be interrogated for more detailed information, providing the added advantage of less data on the network.
Another decision that need to be made is, since it doesn’t make sense to run a process using bad data or a failed signal, how you are going to integrate the newly available device status information with your control logic? If you use the device status information, who are you going to notify that you've taken this action, and how will you do so, especially as we have competing efforts in alarm management to contend with?
Many of these questions are being discussed across several standards organizations, including ISA-108/SC65E WG10, which is working on intelligent device management documents to provide guidance on how to develop the systems and programs to manage and use the information available from these devices to improve plant reliability at reduced costs.
The alternate approach to a dedicated asset management system is to gather all the data, and use the same algorithms for cloud-based data analysis. This will identify patterns leading to greater insight and perhaps a more holistic view of the broader implications of how your facility is functioning overall based on diagnostics data, process information, system demands, mechanical equipment status, etc.
In his keynote presentation, "AI–New Business Imperative,” at the ISA Calgary 2019 conference, Dariusz Piotrowski, director, global AI solutions development, IBM Natural Resources Industry Platforms, said, based on experience including Woodside energy applications in Australia, “AI/data mining exposes dark data, which is 80-90% of all data, resulting in a 20-30% increase in workforce productivity, and 5-15% increase in efficiency/cost reduction.” However, being able to achieve this means putting your data in the cloud because that’s where there are processors capable of churning through all the bits and bytes to reach the kind of conclusions leading to these gains. Local computing platforms just don’t have the necessary horsepower. These cloud solutions aren't yet using instrument diagnostic information, and they present an opportunity for experienced practitioners to share their knowledge by verifying the models AI algorithms generate, which can make algorithms stronger.
In both cases, the good news is that it’s the knowledge in our heads as control practitioners that still makes these systems work.
About the author: Ian Verhappen
