Industries / Industrial equipment and services / Industrial services / Process safety consulting
Safety systems
Safety systems are the automatic trips, relief valves and emergency shutdown valves that act when people and ordinary controls fail. Consultants specify them, check their reliability figures and plan how they are tested.
5
Milestones achieved, from 1989 to 2016.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1989
A valve on every riser
A riser is the pipe that brings a pipeline up onto an offshore platform. The UK Offshore Installations (Emergency Pipe-line Valve) Regulations, in force from 12 July 1989, require a valve on every riser that is held open by a signal and closes by itself if the signal is lost. People next to it can close it, and so can the platform's emergency shutdown system, and it must be shielded from fire and blast. Each valve is inspected at least every 3 months and partly closed in a test at least every 6 months, and two to four months after each partial test it must be fully closed through the shutdown system, with every test recorded. In March 1989 the energy secretary told Parliament the rules followed his department's technical investigation of the Piper Alpha accident.
Credit
UK Department of Energy
1996
A US standard for trips
ANSI/ISA-S84.01-1996 set requirements for safety instrumented systems (SIS) in process plants, from design through installation, maintenance and testing. An SIS is an automatic chain of sensors, logic and valves that takes a plant to a safe state. In March 2000 the US Occupational Safety and Health Administration (OSHA) told ISA it treats the standard as "recognized and generally accepted good engineering practice", so a plant that documents and meets it complies with the SIS requirements of OSHA's process safety rule. In November 2005 OSHA said the same of ANSI/ISA-84.00.01-2004, the US version of IEC 61511. Both times OSHA said no single standard is the only benchmark.
Credit
ISA; OSHA
2004
An interlock bypassed by hose
On 23 April 2004 at Formosa Plastics in Illiopolis, Illinois, an operator cleaning one reactor probably opened the bottom valve of a full, heated, pressurized reactor nearby. An interlock blocked that valve above 10 psi, but operators could get around it by connecting an emergency air hose. Vinyl chloride escaped and exploded, killing five workers and seriously injuring two. The CSB found that a 1992 hazard study had recommended changing the bypass, while a 1999 study accepted the interlock, backed by procedures and training. The same bypass had been used without permission about 60 days earlier, and nobody could see when it was used.
LessonA bypass controlled only by a written procedure leaves one person as the last safeguard, so bypasses need hardware and authorization that leave a trace.
Credit
CSB, final report March 2007
2014
Relief valves measured in use
A spring-operated relief valve opens when pressure overcomes its spring. Researchers from Villanova University, exida and the Savannah River Site analysed 195 proof tests of one group of these valves at the US Savannah River Site, done from 2003 to September 2012, and 13 of the tests found a valve that failed to open. They showed that a site's maintenance and pre-installation testing change the valves' average probability of failure on demand (PFDavg, the chance a device does not work when needed), and concluded that a manufacturer's failure rate must be adjusted to the site's own practices.
LessonPart of a safeguard's reliability depends on the plant that maintains it.
Credit
Julia Bukowski, William Goble, Robert Gross and Stephen Harris, ASME Pressure Vessels and Piping Conference 2014
2016
Security enters the safety lifecycle
Edition 2 of IEC 61511-1, published on 24 February 2016, replaced the 2003 edition. It added a requirement for a security risk assessment of the SIS, expanded the rules for counting the basic process control system as a protection layer, and changed the rules for hardware fault tolerance, meaning how many spare channels a safety function needs.
Credit
IEC subcommittee 65A
Watch for
Agents that act before trips
OperAItor is an open-source framework for testing AI agents that handle faults in a simulated reactor and distillation plant with 22 alarms and 11 independent safety instrumented functions. Without an agent, 18 of 20 fault scenarios ended in a plant stop. With checks on the agents' actions, GPT-5 nano, Qwen3 8B and Mistral 7B cut the stops to between 5 and 8, and recovered $265,000 of a $399,000 loss, using the authors' assumed costs.
Achieved whenanother group reproduces the result on the open benchmark, or a plant trial reported by outsiders shows fewer trips with no safety function bypassed.
Claimed by
Javal Vyas and Mehmet Mercangöz, November 2026, in a preprint
Open
01
Find gaps in safety specifications
A tool that reads a safety requirements specification (SRS) and lists, quoting the line, each item IEC 61511 requires that is missing, vague or contradicted. An SRS is the document saying what each safety function must do, how fast, how reliably and how it is tested.
The briefCredit · unclaimed02
Grade a plant's proof tests
A tool that compares a plant's written proof-test procedure for a device with the example procedure and failure-mode breakdown in the device's published FMEDA report (failure modes, effects and diagnostic analysis, a part-by-part study of how a device can fail). It lists the dangerous failures the plant's procedure would miss and gives a cautious proof-test coverage figure for PFDavg calculations.
The briefCredit · unclaimed
Open 01 · the brief
Find gaps in safety specifications
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A tool that reads a safety requirements specification (SRS) and lists, quoting the line, each item IEC 61511 requires that is missing, vague or contradicted. An SRS is the document saying what each safety function must do, how fast, how reliably and how it is tested.
Why it is still open
HSE traced 44% of 34 control system failures it studied to the specification. A 2025 paper by Hamid Jahanian says a good SRS is still hard to write and that some requirements are never written down. The nearest AI work, SAFER (2026), checks requirements for gaps and contradictions in a drone model, not a process plant, and exida's exSILentia writes an SRS inside its own tool but does not grade documents written elsewhere.
Built with what exists
Claude or Gemini for long documents; a paraphrased checklist of what IEC 61511-1 requires in an SRS; Python for scoring; each finding citing its line (Note: "Findings that cite notes").
First step
Write one complete SRS for a single high-pressure trip, then make five copies, each missing different required items.
Done looks like
30 SRS documents for textbook safety functions, each with 3 seeded omissions or contradictions. The tool finds at least 80 of the 90 and averages no more than one false flag per document, checked by two certified functional safety engineers.
Open 02 · the brief
Grade a plant's proof tests
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A tool that compares a plant's written proof-test procedure for a device with the example procedure and failure-mode breakdown in the device's published FMEDA report (failure modes, effects and diagnostic analysis, a part-by-part study of how a device can fail). It lists the dangerous failures the plant's procedure would miss and gives a cautious proof-test coverage figure for PFDavg calculations.
Why it is still open
exida's William Goble writes that working out coverage means examining every part of a product, which is "beyond the scope of anyone not intimately familiar with a product", so exida publishes coverage only for example proof tests. exSILentia's Proof Test Generator writes new procedures; it does not grade a plant's existing ones.
Built with what exists
FMEDA reports that instrument and valve makers publish for certified products; Claude or Gemini to match procedure steps to failure modes; Python to recalculate PFDavg.
First step
Download five FMEDA reports for common pressure transmitters and turn each example proof test into a step-by-step table.
Done looks like
20 devices with public FMEDA reports and 3 procedures each: the published example and two versions with steps removed. On the 20 examples the tool's coverage comes within 5 percentage points of the published figure, and on at least 36 of the 40 edited versions it names every failure left unchecked.