AI, Process Hazards Analysis and Not Thinking the Unthinkable
We continue our discussion to do with AI (Artificial Intelligence) and its application in process safety.
One of the most important elements of a process safety is hazards analysis. After all, if we don’t know what the hazards are, we cannot remove or ameliorate them. There are many types of Process Hazards Analysis (PHA); they include Checklists, LOPA (Layers of Protection Analysis), HAZOP (Hazard and Operability Studies) and Fault Tree Analysis. Each has its place.
Most of these methods are somewhat formulaic; they rely on following a set of questions that tend to be repetitive in nature. For example,
Could we get reverse flow from B to A?
If so, what are the consequences (safety, environmental, economic)?
How likely is this event?
Is the risk high enough to call for corrective action?
Move on to the next set of guidewords.
However, an important part of a PHA discussion is providing an opportunity for creative or ‘off the wall’ thinking ― sometimes referred to as brainstorming. Such thinking may produce surprising and useful insights.
For example, an engineering company team was reviewing the layout of a standard operating unit in an oil refinery early in the design phase. One team member came up with a non-standard way of laying out equipment that saved space and helped improve safety.
Another example is to do with hydrogen safety (see the post Safety of Hydrogen/Methane Blends). Even if a process safety professional can demonstrate that the risk to do with hydrogen/methane mixtures is acceptable, an AI program may not be able to ‘create’ the emotional response from members of the public.
This type of thinking is not likely to come from an Artificial Intelligence program ― at least, not yet. These programs are not able to ‘Think the Unthinkable’. Because they rely on historical data they are not able to imagine ‘What Might Be’.
