Any time there is a process upset in the plant, an investigation is undertaken to determine the root cause.  Many times, it is found that the upset is due to either a misguided decision/oversight by a member of production or a random failure of equipment that caused an issue immediately. These can be improved by better maintenance practices and by training/communication amongst production staff, but cannot be avoided altogether. On the other hand, there are process upsets that could have been prevented with data-driven decisions. Whether it is due to scale building up on the inside of a heat exchanger, a slowly drifting measurement causing the operation to go outside of normal ranges, or a general shift in operation over an extended period that went unnoticed, potential breakdowns in the process could have been caught in advance. These are instances where if someone had been diligently looking at the PI data trends, the process upset would have been caught and avoided.

It is not feasible to dedicate the time it would take to manually monitor all the different things that can go wrong in the process, and it can become tedious and difficult to create event frame notifications that are accurate and impactful for every single detail in a complex facility. That is where machine learning can be beneficial to be the watchful eye looking out for deviations in the normal flow of the process.  With SAMSON’s process monitoring software called SAMGUARD, many process upsets have been identified before they have happened. This advanced notice allows the necessary time to investigate the situation in more detail, make decisions to alter course or schedule downtime for maintenance, and avoid serious issues and unplanned downtime.

In this presentation, it will be shown how the software integrates and relies on the quality data provided by the PI system for its machine learning model, how alerts are received, and then explore a couple of examples in depth where great outcomes were realized through data-driven decision making.