Independent Real-Time Verification of Manual Valve Position as an Additional Protection Layer Against Misrouting During Product Transfers: A Monte Carlo Barrier-Effectiveness Assessment
Year: 2026
Type: preprint
Abstract: Manuscript prepared for the Journal of Loss Prevention in the Process Industries (Elsevier), version 1.1.0, with supplementary information, highlights, title page and cover letter (docx and pdf). Supersedes version 1.0.0. Abstract. Manual valve line-ups before product transfers rely on rounds, radio confirmation and checklists, none of which observes valve position independently. We ask when an additional layer of temporary clamp-on position sensors on existing manual valves reduces transfer risk, and when alarm fatigue, bypass, common-cause failures and incomplete coverage remove the benefit. A Monte Carlo model of the line-up task, verified against closed-form results, an independent analytic model and automated tests, gave with assumed inputs for a 14-valve route an undetected-error probability per transfer of 4.03% with existing checks and 0.29% with direct sensing and NFC identity (14-fold; sampling interval 13.6–14.3), assuming that operators always act on alarms. Varying 13 inputs gave a median reduction of 9.7 (5th–95th percentile 4.4–28); adding uncertain operator response, alarm fatigue and common-cause failures of the sensing layer gave 2.5 (0.9–4.8), with response the strongest driver. A mean response time of 10 min against 20 min available gave a factor of about 2. A factor of 10 required a response of at least 97%, bypass below 53% and common-cause probability below 2.3%. An illustrative event tree gave a 14-fold reduction in expected loss. Clustering of validations by valve cut the power of a 1000-validation pilot from 95% to 50%. Inputs, consequence probabilities and losses are assumed: the results are conditional, not credited risk reduction. Code and data: https://github.com/sandlerleon/hydrasense-valve-verification-mc (archived at 10.5281/zenodo.24046773). The study uses no field data. A generative AI tool (Claude, Anthropic) assisted with drafting, code and figures; the author reviewed the content and takes full responsibility.
Author: Sandler, Leon
Language: English
Topic: Risk and Safety Analysis
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225568296