BRENNEN

Industries / Industrial equipment and services / Industrial services / Instrumentation and electrical

Testing safety trips

This work tests the trips and interlocks that shut a plant down safely, by forcing each one to act on a schedule and recording whether every part of it worked.

4

Milestones achieved, from 2012 to 2025.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2012

    Trip testing as a program

    A safety instrumented system is the set of sensors, logic and valves that takes a plant to a safe state when a reading passes a limit, apart from the system that runs the plant day to day; each of its trips is designed to one of four safety integrity levels (SIL), a grade of how much risk it must remove. A proof test is a periodic test that finds the failures a trip's own self-checks cannot see. ISA's technical report TR84.00.03, "Mechanical Integrity of Safety Instrumented Systems", approved on 28 August 2012, told plants to run inspections, calibrations, proof tests with pass and fail criteria, and controlled bypasses as one program that keeps the system "as good as new" and proves it with traceable records. It gives example demand logs, failure reports, a proof test template and worked proof tests for sensors, valves and bypasses, and its foreword says this edition sets out how to plan the program "rather than including specific test procedures as in the previous edition".

    webstore.ansi.orgcontrolglobal.com

    Credit

    ISA's ISA84 committee

  2. 2014

    Not every trip tested alike

    Jeff Wood of INEOS ChlorVinyls described how the company's Runcorn site in England sorted its 4,460 instrumented protective systems: 140 rated SIL 2, 1,120 rated SIL 1 and about 3,200 below SIL 1. After a visit, the UK Health and Safety Executive (HSE) asked the site to review how it tested the lower group. The site moved about 1,300 medium- and low-priority alarms out of that group, leaving about 1,900, and set lighter rules for the rest: tests using the instrument's built-in test functions, no formal test for systems that are called on more than twice a year, and credit for a real trip in the last year when records show the whole loop worked, from sensor to valve. The paper explains proof test coverage, the share of hidden failures a given test can reveal, and works the numbers for a tuning fork level switch whose test button reveals 52% of the sensor's hidden failures.

    Lessontesting every trip with the same rigor spreads effort away from the trips that matter most.

    icheme.org

    Credit

    Jeff Wood, INEOS ChlorVinyls, in a paper for IChemE's Hazards 24 symposium

  3. 2018

    A trip left switched off

    On 19 May 2018 an ethylene release caught fire at Kuraray America's plant in Pasadena, Texas, and injured 23 workers. During the startup, while troubleshooting why liquid would not flow out of a reactor, a board operator activated its high-high-pressure safety interlock by hand and then turned it to "off" with a panel switch; it stayed off. When the pressure reached the trip setting of 640 psi, the interlock did not open the cooling valve. The cause of the flow problem was a closed manual valve. The US Chemical Safety Board (CSB) concluded that Kuraray had no effective system to manage the disabling of safety interlocks and make sure they were back in service before startup continued, and OSHA cited the company for not using its own interlock bypass procedure.

    Lessona trip bypassed for a test or a fault is no protection until someone checks it is back on.

    csb.gov

    Credit

    investigated by the CSB

  4. 2025

    Failed tests counted offshore

    Norway's offshore safety regulator, Havtil (the Norwegian Ocean Industry Authority), asks every production operator on the Norwegian shelf to report the results of its routine periodic tests of selected safety barriers, and publishes the share of tests that failed, facility by facility, against industry norms, in its yearly Trends in Risk Level (RNNP) report. For 2025, 73 facilities reported fire detector tests; fire and gas detection stayed within their norms, while blowdown valves (0.012 against a norm of 0.010), deluge valves (0.014) and closure and leak tests of riser emergency shutdown valves were outside them. Havtil says barrier data have now been collected for 21 years for most barriers on production facilities.

    havtil.no

    Credit

    Havtil, RNNP

Watch for

  1. Real trips counted as tests

    A trip that happens in service, when the process really goes out of limits, exercises the same valves a proof test would. Yokogawa says release R3.60 of its Exaquantum/SFM software, of March 2026, lets a plant assign recorded trips of a safety function, and partial stroke tests of its valves, as proof tests, manually or automatically; the software records process data before, during and after each trip so that staff can check every device acted as intended, and compares real demands with those assumed in the safety design.

    Achieved whena plant or an outside assessor publishes how many scheduled proof tests were replaced by recorded trips over a period, and whether later full tests or failures showed that the credited trips had missed faults.

    yokogawa.com

    Claimed by

    Yokogawa, March 2026

Open

  1. 01

    Valve test signatures as data

    A public set of partial-stroke test signatures from bench shutdown valves with planted faults, and a model that names the fault. A partial-stroke test moves a shutdown valve a short way and back while the plant runs, and a smart positioner records the valve's position against actuator pressure, called its signature. Faults are planted one at a time: packing tightened to add friction, a weakened spring, low air supply, a leaking actuator seal, a partly blocked exhaust and a slow solenoid valve.

    The brief
    Credit · unclaimed
  2. 02

    Barrier test results as data

    An open dataset of the yearly share of failed tests for each barrier element Havtil tracks, such as gas detectors, fire detectors, emergency shutdown valves and blowdown valves, with the industry norms, from 2006 to 2025, and a chart of each against its norm.

    The brief
    Credit · unclaimed

Open 01 · the brief

Valve test signatures as data

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of partial-stroke test signatures from bench shutdown valves with planted faults, and a model that names the fault. A partial-stroke test moves a shutdown valve a short way and back while the plant runs, and a smart positioner records the valve's position against actuator pressure, called its signature. Faults are planted one at a time: packing tightened to add friction, a weakened spring, low air supply, a leaking actuator seal, a partly blocked exhaust and a slow solenoid valve.

Why it is still open

ISA-75.26.01, the standard for collecting and reporting valve signature data, says it "does not address interpretation of data or diagnosis of valve condition". The nearest public dataset covers other equipment: Virginia Tech's 2026 set from six industrial hydraulic actuators under faults and loads, where a model could not classify a fault type held out of training. Searches found no public set of shutdown-valve partial-stroke signatures with labeled faults.

isa.orgpapers.phmsociety.org

Built with what exists

Two small spring-return valves with pneumatic actuators, run on clean shop air with no process fluid; a HART smart positioner that records signatures, or a pressure transducer and position sensor logged by a LabJack or Raspberry Pi; Python with scikit-learn.

  1. First step

    Fit a positioner to one bench valve, record 20 healthy signatures, then tighten the packing in three steps and record 20 more at each.

  2. Done looks like

    600 labeled signatures covering six fault types at three severities plus healthy runs, released under an open license, and a classifier's accuracy on the second valve when trained only on the first.

Open 02 · the brief

Barrier test results as data

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open dataset of the yearly share of failed tests for each barrier element Havtil tracks, such as gas detectors, fire detectors, emergency shutdown valves and blowdown valves, with the industry norms, from 2006 to 2025, and a chart of each against its norm.

Why it is still open

Havtil publishes the barrier test results only as tables and charts inside PDF reports; the English summaries go back to 2006, the fuller main report is in Norwegian, and the report pages offer no data file.

havtil.nohavtil.no

Built with what exists

pdftotext or PyMuPDF to pull the tables, a language model to read jumbled tables and Norwegian text, Python and a spreadsheet; every value keeps its report and page number.

  1. First step

    Download the English summary reports for 2020 to 2025 and extract the table of failure shares by barrier element from each.

  2. Done looks like

    A CSV covering every barrier element and year in the English summaries from 2006 to 2025, each value traced to a page, with 10% of values checked by a second person and the error rate published.