Ulku G. Oktem

Near‐Miss Incident Management in the Chemical Process Industry

This article provides a systematic framework for the analysis and improvement of near-miss programs in the chemical process industries. Near-miss programs improve corporate environmental, health, and safety (EHS) performance through the identification and management of near misses. Based on more than 100 interviews at 20 chemical and pharmaceutical facilities, a seven-stage framework has been developed and is presented herein. The framework enables sites to analyze their own near-miss programs, identify weak management links, and implement systemwide improvements.

Assessment of Catastrophe Risk and Potential Losses in Industry

This paper describes the potential contribution of near-miss management systems to improving company profitability and reducing the frequency and severity of major industrial accidents. The near-miss concept has long been understood in different industries, as examples in this paper illustrate. However, what has been largely missing is the integration of near-miss management into the culture and day to day operations in a manner that underlines the critical connections between near-misses and behavior. Often, near-miss management has played an ex post forensic role in risk management rather than an alerting one, summarizing leading indicators and precursors of hazardous conditions. This paper describes several strands of recent research that aim to correct this and to make near-miss management an organic element of Enterprise Risk Management. In this respect, a new concept, “potential safety profit loss”, is introduced to calculate the potential monetary losses due to unexpected shutdowns and accidents.

Improving Process Safety and Product Quality using Large Databases

Abstract This paper introduces a novel modeling and statistical framework (based on Bayesian theory) that utilizes extensive distributed control system and emergency shutdown databases, to perform thorough risk and vulnerability assessment of chemical/petrochemical plants. Quality variables are utilized, in addition to safety (or process) variables, to enhance both process safety and product quality . To effectively achieve these objectives, new concepts of abnormal events and upset states are defined, which permit the identification of near-miss events from the databases. The databases for a fluid catalytic cracking unit at a major petroleum refinery are used to demonstrate the application and performance of the techniques introduced herein. The results show that with the novel utilization of near-miss data, one can perform robust risk calculations using both product-quality and safety data.

Analysis of management actions, human behavior, and process reliability in chemical plants. I. impact of management actions

While management and engineering actions have a significant impact on process reliability, these factors have received too little attention in calculating plant risks. In this work, the focus is on understanding and modeling the influence of human behavior patterns on plant safety in two settings. The first, introduced in Part I, involves a framework to estimate the impacts of management and engineering decisions, process operator performance, and processing equipment operations on the failure state of chemical plants. As examples, the impacts of poor training, maintenance problems, operator inabilities, control system failures, and excessive feed quantities, on failure states are studied. The management and engineering team and the operators are found to have significant impacts on process reliability. While the theoretical framework introduced herein is illustrated using hypothetical plant data, it should provide a basis for more quantitative safety analyses. Attempts to obtain operating data in industrial plants for validation of the framework were unsuccessful because of confidentiality and liability issues associated with industrial data.

Analysis of management actions, human behavior, and process reliability in chemical plants. II. Near-miss management system selection

To understand the behavior patterns of managers, engineers, and operators, in Part II, a game‐theoretic decision model is developed for a specific plant to balance the advantages and disadvantages of having a Near‐miss Management System (NMMS) with different sophistication levels. Assuming that management and engineering preferences differ from those of the process operators, the tradeoffs between them are balanced. As anticipated, it is shown that the choices of the management and engineering team, and the operators, for the selection of a NMMS, are sensitive to the contributing factors. This article introduces a theoretical approach, illustrated using hypothetical data, which should be effective in industrial operations. Attempts to obtain data for validation of the framework were unsuccessful because of confidentiality and liability issues associated with industrial data.

Design for Process Safety – A Perspective

Abstract Key safety-related elements in process design are discussed: inherent safety, Hazard and Operability (HAZOP) analysis, process integration with improved safety, Model- Predictive Control (MPC) for improved control and risk reduction, identification of special causes, and multi-objective optimization. Recent developments in data mining to extract risk estimates from extensive near-miss records (e.g., alarm databases) are reviewed. Building upon these techniques, the development and application of statistical methods to both predict abnormal events (generally resulting in alarm activations) with leading indicators and to diagnose their special (e.g., root) causes follows. In our view, the goal is to develop and implement techniques and safety systems to better provide plant operators actionable (e.g., valid, timely, non-redundant) information on emerging abnormal situations. These improved techniques should account for today’s dynamic operation of process plants, which are more integrated and therefore have increased multivariable interactions that must be taken into account.

On Deriving Indicators from Texts

This paper presents and explores the idea of deriving numerical indicators from texts, that is, converting text data to numerical data that has predictive or diagnostic value. One application of such a general capability is to the provisional identification of networks, or rather, of associations within networks. Conversely, given a network structure among entities that are associated with various texts, the network structure can itself contribute usefully to construction of indicators derived from texts. The focus of the paper is on basic concepts and methods for deriving indicators from texts. Much research remains to be done.

Understanding rare safety and reliability events using transition path sampling

Abstract In the chemical and process industries, processes and their control systems are typically well-designed to mitigate abnormal events having potential adverse consequences to human health, environment, and/or property. Strong motivation exists to understand how these events develop and propagate. These events occur so rarely that statistical analyses of their occurrences alone are incapable of describing and characterizing them − especially when they have not yet occurred. Moreover, the use of process models to understand such rare events is hampered by the orders of magnitude separating the frequencies with which reliability and safety events (years to decades) occur and the duration over which they occur (minutes to hours). To address these challenges, we adapt a Monte-Carlo based, rare-event sampling technique, Transition Path Sampling (TPS), which was developed by the molecular simulation community. Important modifications to the TPS technique are needed to apply it to process dynamics, and are discussed herein.