The HAZOP method is pretty good, if done right. IF you apply it to continuous mode correctly using nodes and all applicable deviations, and adding loss of containment deviation for damage mechanisms; and then if you apply it separately to analyze deviations from steps in startup, shutdown, and online maintenance; and IF you have the fully competent and experienced PHA/HAZOP leader and support scribe, and IF you have a expert team members from the units under review; and IF you have reasonable up-to-date P&IDs, procedures for startup, etc., and good understanding of the process technology, then you will find most of the scenarios ahead of the accident, and nearly all of the major accident scenarios. To date, though we have led and documented many thousands of unit/plant-sized HAZOPs, when the IFs above are met well, the accidents that do occur can be found in the HAZOP report. When the HAZOPs have missed scenarios, it is usually because the client did not agree to do a HAZOP of startup, shutdown, and online maintenance modes of operation by using HAZOP (normally 2 guideword) and/or What-If analysis of these procedures. 70-80% of major accidents occur during non-routine cmodes of operation; so the PHA/HAZOP needs to cover these modes well.
All of the these points are made well in the various chapters of “Guidelines for Hazard Evaluation Procedures”, 3rd edition, CCPS/AIChE, 2008. You can also find most of these points in the free papers on our website at: www.piii.com/resources/.
With all of that said, no group of team members and process knowledge is perfect. And if there are any deficiencies in the quality items listed earlier (if there are inaccurate procedures, poor P&IDs, lack of expert team member, etc.) then the chances of missing accident scenarios greatly increase. And yes, there are many examples where lack of insight or expertise in what certain chemical reactions can occur have caused some specular accidents. But, in such cases we have investigated, these gaps in knowledge by the “site” can be traced to deficiencies in the management systems for deriving process safety information.
With respect to probabilities of missing scenarios given that all of the factors for the HAZOP are good: we have run experiments at several sites where we did two PHA/HAZOPs of the same unit with different, but equivalent expert teams. There was an 80% overlap of findings and 5% of the major scenarios found by one team were not found by the other team. (Each team missed 5% of the major accident scenarios found by the other team.) So far, this has not led to the accident occurring though (not yet, anyway).
Bottomline: Yes, HAZOP teams will miss scenarios due to variations in expertise, quality of data, quality of team leadership, personalities, etc. BUT, the biggest reason and recurring reason for accident scenarios to be missed is that the HAZOP team is not allowed to or chooses not to due a PHA/HAZOP of non-continuous modes of operation; this has indeed led to specular accidents; and this gap continues to exist in +80% of the thousands of HAZOPs we have audited around the world. I suggest we focus on closing that gap before worrying about any other limitations (since these other limitations are small in comparison).