The Kenexis Lecture Series
Kenexis CEO and Chief Scientist, and technical safety luminary, Ed Marszal, will be providing an exciting day of lectures on cutting edge developments in technical safety for the chemical process industries. The lecture series will be presented in Kuwait City, on 20 Feb 2025 at the Ahmadi Mega Complex (Auditorium F-115 – Zone ‘F’ ground floor).
The lecture series will include four separate lectures throughout the day. Please feel free to attend select sessions in your area of interest or stay for the entire day.
7:00 AM – Registration Opens
8:00 – 9:30 AM
Using Large Language Models (AI) for HAZOP Automation and Augmentation Tool
Hazards and operability studies (HAZOP) are a critical first step in ensuring safe chemical process facilities. Unfortunately, they are time and resource intensive. Even good HAZOP studies can miss a large number of very significant scenarios or come to insufficient conclusions. This problem is being exacerbated by the loss of corporate memory that results from experienced personnel leaving the workforce. Research is currently being performed to employ artificial intelligence tools like Generative Pretrained Transformers (e.g., Chat GPT) to assist, or potentially complete automate the HAZOP process. In order to be effective, the GPT will have to be applied to curated datasets that only contain trusted engineering knowledge that is strictly proprietary – removing the “junk” general knowledge that hampers their performance.
This lecture begins with a discussion of the fundamentals of AI from its inception, starting with expert systems that were abandoned for fuzzy logic and neural networks that underpin modern large language models such as GPT. The lecture will then discuss how HAZOP data such as HAZOP and LOPA worksheets can be fed into a large language model and how a GPT will create a model for predicting information in a HAZOP based on training data.
9:30 – 10:00 AM Break
10:00 – 11:30 AM
Use Fault Tree Analysis When LOPA Fails
Layer of Protection Analysis (LOPA) is a ubiquitous tool for performing risk analysis that is more detailed than a qualitative hazards and operability study (HAZOP) but less cumbersome than a full quantitative risk analysis. This is especially true when the analysis must result in performance targets for safeguarding equipment such as Safety Integrity Levels (SIL) of Safety Instrumented Systems (SIS). The ease of use of LOPA relies on conservative assumptions about system design that are not always true or appropriate, specifically, the independence between initiating events and protection layers. Use of LOPA in these cases can result in either inappropriate or impracticable. In these cases, use of fault tree analysis, which is more sophisticated and better suited to handling the complex logic of shared components, allows a more accurate risk analysis and better. This paper explains some situations where LOPA is expected to result in inaccurate risk analysis and presents who they analysis could be better performed using fault tree analysis. The concepts are presented through an example of a butane sphere batch filling operation where single pieces of equipment are commonly used as parts of initiating events and multiple protection layers that turn out to be not so independent.
11:30 to 1:00 PM – Lunch and Prayers
1:00 – 2:30 PM
LOPA is Obsolete! Move to Quantitative Bowtie Analysis
Layer of Protection Analysis (LOPA) is a ubiquitous part of the workflow for process hazards analysis and engineered safeguard design. Even so, from its inception LOPA has suffered from limitations that are rooted in the “one-cause, one-consequence” paradigm to the risk analysis the limits the analysis scope. This limitation can result in design errors where engineered safeguards that protect against a consequence with multiple causes can be under-designed by looking at one cause at a time. Also, safeguards that reduce the magnitude of consequences (i.e., mitigate) instead of preventing loss of containment cannot be addressed at all without ignoring the residual consequence that exists even if the safeguard activates successfully. While some extensions of LOPA that address the issue of multiple causes are commonly used, but mitigative safeguards are rarely appropriately addressed and designed. Finally, the textual nature of the process also makes results hard to communicate to non-practitioners.
Recent research into Unified Hazard Assessment that combines HAZOP, LOPA, and bowtie analysis has yielded techniques that elegantly address all the limitations of LOPA while also providing a graphical presentation that facilitates result communication. This paper will provide background on how Unified Hazard Assessment yielded the techniques of Quantitative Bowtie analysis. The paper will also describe in detail how to implement quantitative bowtie analysis along with the mathematical concepts used for quantification of risk for multiple causes and multiple consequences inside a single scenario. The concepts will be presented using example studies that include mitigative safeguards and multiple cause scenarios.
2:30 – 3:00 PM Break
3:00 – 4:30 PM
Gas Detector Scenario Coverage Calculation by Employing Dispersion Model Results
Gas detector placement is moving from a imperfect heuristic art to a systematic engineering task that employs quantitative risk analysis tools. This presentation will provide a discussion of how gas detector scenario coverage, as defined in the ISA 84.00.07 technical report Guidance on the Evaluation of Fire and Gas System Effectiveness is being quantitatively calculated. The presentation discusses the mechanics of calculating scenario coverage from the results of dispersion modeling and highlights its accuracy and effectiveness with respect to more simplified methods such as geographic coverage and heuristic techniques.
The presentation will begin with a discussion of the creation of the most common dispersion models, Gaussian, from a sufficient set of scenarios defining factors, such as process operating conditions and ambient weather. These modeling results will then be manipulated to consider multiple release orientations and their relative frequency, along with consideration of wind directions by modifying the relative directional release frequency based on wind direction frequency. Next a discussion of integration of the risk is presented, discussing accumulation release frequency at all geographic positions-based release scenarios present in those locations. Scenario coverage calculation is then described as the determination of which scenarios are detectable given the detector locations relative to locations of released gas clouds.