Elsevier

Computers & Chemical Engineering

Volume 91, 4 August 2016, Pages 104-113
Computers & Chemical Engineering

Abnormal situation management: Challenges and opportunities in the big data era

https://doi.org/10.1016/j.compchemeng.2016.04.011Get rights and content

Highlights

  • Fault detection and diagnosis (FDD) techniques have not been mature since it was proposed 40 years ago.

  • Developments of chemical process FDD are reviewed from a perspective gained through a personal involvement in the research area.

  • The reason why process FDD has not been widely and successfully implemented in the chemical process industry is discussed.

  • A new framework is proposed based on big data from similar chemical processes for addressing the challenging issues in ASM.

Abstract

Although modern chemical processes are highly automatic, abnormal situation management (ASM) still heavily relies on human operators. Process fault detection and diagnosis (FDD) are one of the most important issues of ASM but few FDD systems have been satisfactorily applied in real chemical processes since the concept of FDD was proposed about 40 years ago. In this paper, developments of chemical process FDD are briefly reviewed. The reason why FDD has not been widely implemented in the chemical process industry is discussed. One of the insights gained is that some basic problems in FDD such as how to define faults and how many faults to diagnose have not even been addressed well while researchers tirelessly try to invent new methods to diagnose fault. A new framework is proposed based on the big data in a cloud computing environment of a big chemical corporation for addressing the challenging issues in ASM.

Introduction

Process control has made a major contribution to the production efficiency, product quality, process safety and environmental protection in the chemical process industry over the past decades. Although it makes chemical processes highly automatic nowadays, the automation of abnormal situation management (ASM) has not been realized yet. According to Nimmo’s estimation, abnormal situations brought an economic loss of 20 billion US dollars per year to petroleum and chemistry industries in USA (Nimmo, 1995). In UK, abnormal situations cause 27 billion US dollars per year (Venkatasubramanian, 2005). That is one of the main reasons why human operators are still needed in chemical plants. Industry 4.0 is regarded as the fourth industrial revolution. It was defined by Weyer and Schmitt (2015) as a synonym for the transformation of today’s factories into smart factories which are intended to address and overcome the current challenges of shorter product lifecycles, highly customized products, enhanced work safety and stiff global competition. According to the 5C architecture of the Cyber-Physical Systems (CPS) in a smart factory, cognition is a critical level or component because it creates a thorough knowledge about the monitored system (Lee et al., 2015). Making timely, reliable and automatic decision supports to the operators about the abnormal situations in the chemical processes should be an indispensable cognition function in smart factories.

ASM does not only mean early warning of abnormal situations, but also timely diagnosis of their causal origins and providing decision supports to operators to take actions to bring the process back to normal (Venkatasubramanian et al., 2003a). Adhitya et al. (2014) designed a generic experimental scheme to quantify the benefits of an early warning system which predicts the time of occurrence of critical alarms before they are actually triggered (Adhitya et al., 2014). Their study results showed that early warning was effective in improving diagnosis lag and subjectively found to be helpful by the experiment participants, but it did not improve diagnosis accuracy. Therefore, fault diagnosis is one of the most important components in ASM. It could be found out that some of the severe chemical accidents in the past were due to failure of timely fault diagnosis.

On March 23, 2005, an explosion occurred at the isomerization unit of the BP Texas City refinery, killed 15 people, injured 180 and caused an economic loss of about 1.8 billion US dollars including the income losses during the reconstruction from March 2005 to March 2006 (Manca and Brambilla, 2012). The accident occurred during the startup of the raffinate splitter column where the incoming raffinate feed was separated into light and heavy components. Before the accident, operators were feeding the splitter tower with raffinate. As the tower level raised, the high alarm of tower level was triggered to warn the operators but was ignored by them because a high tower level was good for operations in the downstream according to their operating experience. Due to a fault in the level transmitter, the tower level kept raising but the level reading was declining on the contrary. According to the accident investigation, the tower level finally reached 156 feet but the level transmitter only showed a reading of 7.90 feet. Therefore the operators still maintained the feed even after the tower was filled up. The raffinate filled up the tower and then flew to the blow down drum, but the high level alarm of the drum did not work. Finally the raffinate flew out of the plant, formed vapour cloud, and was ignited by a nearby vehicle, causing a catastrophic explosion. If the equipment faults and operation mistakes had been diagnosed timely and correctly, the accident could have been avoided. Therefore Saleh et al. (2014) argued that safety-diagnosability principle is an essential ingredient for improving operators’ situation awareness (Saleh et al., 2014).

On November 13, 2005, an explosion occurred at the Jilin chemical plant, killing 8, injured 60 and polluted the Songhua River with about 100 tons of pollutants containing benzene and nitrobenzene (Zhang et al., 2010). The accident was caused by a wrong order of operations. Before the accident, the operators tried to stop the feed to the vacuum distillation tower but they did not stop the heating. When they resumed the feed, they started the heating steam before resuming the feed by mistake. When the cold nitrobenzene feed entered the heat exchanger, it was heated dramatically by the high temperature of the heat exchanger. The nitrobenzene then evaporated sharply and caused the flanges in the feeding pipe loosened. Afterwards the air was drawn in by the vacuum of the tower, and an explosive mixture was formed to lead to the explosion. No control interlocks were designed to prevent the misoperations or respond to the abnormal temperature rise which was as high as 60 °C before the explosion. There were two high temperature alarms before the explosion, which lasted for 80 min and 10 min respectively. But neither of them was recognized by the operators because alarm flooding was a common issue during the plant startup (Zhu et al., 2013).

The above two severe accidents showed that even though the process control systems have made modern chemical processes highly automatic accidents may still occur due to the mistakes of operators and malfunctions of equipment or instruments. Unfortunately, we have to acknowledge that there is no equipment or instrument that never fails and there is no human being that never makes any mistake. About 40 years ago, Prof. Himmelblau (1978) published the first book on chemical process fault detection and diagnosis (FDD) (Himmelblau, 1978). However, his dream to have a reliable early warning and diagnosis system that can widely be applied to different chemical processes has not been realized yet up to now. The situations in the development of the other automatic control systems are quite different. For example, the concept of programmable logic controller (PLC) was proposed by General Motors in 1968. One year later, Digital Equipment Corporation in the USA successfully released the first PLC. It took Honeywell only 5 years to launch the first distributed control system (DCS) TDC 2000 in 1975. In 1980, Cutler and Ramaker proposed the dynamic matrix control. In 1981, Garcia and Morari proposed the internal model control (IMC). Within 10 years, many advanced process control (APC) systems were successfully developed based on these concepts and widely applied in the refineries and petrochemical companies. However, process FDD is a 40 year dream that has not come true.

In fact, countless work has been done the area of FDD. In the 1980s, the IFAC Technical Committee on Fault Detection, Supervision and Safety for Technical Progresses (SAFEPROCESS) was founded. The IFAC Symposium on Fault Detection, Supervision and Safety for Technical Progresses is held every 3 years, aiming at strengthening contacts between academia and industry in the area of fault diagnosis, process supervision and safety monitoring. In 1994, the ASM Consortium was organized by Honeywell Company to address the chemical process industry concerns about the high cost caused by incidents and/or accidents, such as unplanned shutdowns, fires, explosions, emissions, etc. The ASM Consortium currently consists of companies including SASOL, BP, ExxonMobil, Honeywell, Shell, UOP and Human Centered Solutions, and universities including Mary Kay O’Connor Process Safety Center, University of California, Los Angeles, Pennsylvania State University, Nanyang Technology University and University of Alberta. The National Institute of Standards and Technology (NIST) funded the ASM Consortium with 8.5 million US dollars and the membership companies invested 8.5 million US dollars to support related research activities within the consortium. From 2008 to 2010, the ASM Consortium published three guidelines for (1) Effective Operator Display Design, (2) Effective Alarm Management Practices and (3) Effective Procedural Practices (Bullemer et al., 2011) respectively. Based on the first guidelines for effective operator display design, the graphical user interfaces of DCS systems have been changed to improve the operators’ ability to handle process abnormal situations that generate alarm floods (Laberg et al., 2014). Several alarm management systems such as Honeywell’s DynAmo, Yokogawa’s CAMS, PAS’s PlantState Suite, National University of Singapore’s IAMS (Srinivasan et al., 2004) and Tsinghua University’s iTAM were commercialized and the alarms flooding situations in chemical processes have been significantly mitigated by deploying these system. The suppression of nuisance alarms could help the operators take correct and quick action in response to serious alarms (Meel and Seider, 2008) and therefore prevent accidents.

Even though process safety and abnormal situation management have been enhanced worldwide during the past twenty years, there is no clearly visible overall decline in major accident events observed in either the USA or EU (Pitblado, 2011). While some companies may push safety culture and process safety management as the key drivers for enhancement of accident prevention, techniques such as real-time process fault detection and diagnosis will however be another viable technical key to avoiding major accidents.

Section snippets

Challenges and opportunities of chemical process fault diagnosis

The process FDD methods were categorized into three types: the qualitative model-based methods, the quantitative model-based methods, and the process history based methods (Venkatasubramanian et al., 2003a, Venkatasubramanian et al., 2003b, Venkatasubramanian et al., 2003c). Due to the increasing size and complexity of modern chemical processes and the prolife-rating quantity of available historical data, the process history based methods have shown great advantages in applicability over the

A new chemical process FDD framework

It is quite frustrating that process FDD technologies have not been as mature as other process control technologies such as model predictive control or advanced process control after 40 years of research and development. In our opinion, the main barrier that hinders the industrial application of FDD techniques is lack of historical fault samples. In Table 3 we listed the number of fault samples used to build process FDD models by some of the reported FDD technologies. It can be found a large

Conclusion

The concept of process Fault detection and diagnosis (FDD) was coined about 40 years ago. Its real industrial applications are still scarce. It is not because it is not a highly demanded technology nowadays when other process control technologies have widely been deployed in chemical plants. In contrast, it is still a dreamful and highly demanded technology that needs to be researched and developed in future. Many disasters such as the BP Texas city refinery explosion accident and the Jilin

Acknowledgements

The authors gratefully acknowledge support from the National High-Tech R&D Program of China (863 Program) (No. 2013AA040702) and the National Natural Science Foundation of China (No. 61433001).

References (86)

  • C.K. Lau et al.

    Fault diagnosis of Tennessee Eastman process with multi-scale PCA and ANFIS

    Chemom. Intell. Lab. Syst.

    (2013)
  • J.M. Lee et al.

    Fault detection of batch processes using multiway kernel principal component analysis

    Comput. Chem. Eng.

    (2004)
  • J. Lee et al.

    A cyber-physical systems architecture for industry 4.0-based manufacturing systems

    Manuf. Lett.

    (2015)
  • S.A. McCoy et al.

    HAZID a computer aid for hazard identification 1. the STOPHAZ package and the HAZID code: an overview, the issues and the structure

    Process Saf. Environ. Prot.

    (1999)
  • R.C. McFarlane et al.

    Dynamic simulator for a Model IV fluid catalytic cracking unit

    Comput. Chem. Eng.

    (1993)
  • A. Meel et al.

    Real-time risk analysis of safety systems

    Comput. Chem. Eng.

    (2008)
  • J. Mori et al.

    Quality relevant nonlinear batch process performance monitoring using a kernel based multiway non-Gaussian latent subspace projection approach

    J. Process Control

    (2014)
  • Y.S. Ng et al.

    An adjoined multi-model approach for monitoring batch and transient operations

    Comput. Chem. Eng.

    (2009)
  • R. Pitblado

    Global process industry initiative to reduce major accident hazards

    J. Loss Prev. Process Ind.

    (2011)
  • T.N. Pranatyasto et al.

    Sensor validation and process fault diagnosis for FCC units under MPC feedback

    Control Eng. Pract.

    (2001)
  • Y. Qian et al.

    Implementation of knowledge maintenance modules in an expert system for fault diagnosis of chemical process operation

    Expert Syst. Appl.

    (2005)
  • K. Renganathan et al.

    Observer based on-line fault diagnosis of continuous systems modeled as Petri nets

    ISA Trans.

    (2010)
  • J.H. Saleh et al.

    Texas City refinery accident: case study in breakdown of defense-in-depth and violation of the safety-diagnosability principle in design

    Eng. Fail. Anal.

    (2014)
  • Y. Shu et al.

    Data-driven causal inference based on a modified transfer entropy

    Comput. Chem. Eng.

    (2013)
  • R. Srinivasan et al.

    Online fault diagnosis and state identification during process transitions using dynamic locus analysis

    Chem. Eng. Sci.

    (2006)
  • V. Venkatasubramanian et al.

    Intelligent systems for HAZOP analysis of complex process plants

    Comput. Chem. Eng.

    (2000)
  • V. Venkatasubramanian et al.

    A review of process fault detection and diagnosis part I: quantitative model-based methods

    Comput. Chem. Eng.

    (2003)
  • V. Venkatasubramanian et al.

    A review of process fault detection and diagnosis: part II: qualitative models and search strategies

    Comput. Chem. Eng.

    (2003)
  • V. Venkatasubramanian et al.

    A review of process fault detection and diagnosis part III: process history based methods

    Comput. Chem. Eng.

    (2003)
  • V. Venkatasubramanian

    Prognostic and diagnostic monitoring of complex systems for product lifecycle management: challenges and opportunities

    Comput. Chem. Eng.

    (2005)
  • S. Verron et al.

    Fault detection and identification with a new feature selection based on mutual information

    J. Process Control

    (2008)
  • J. Wang et al.

    An online method to remove chattering and repeating alarms based on alarm durations and intervals

    Comput. Chem. Eng.

    (2014)
  • X. Wang et al.

    Recursive partial least squares algorithms for monitoring complex industrial processes

    Control Eng. Pract.

    (2003)
  • S.H. Yang et al.

    Neural network based fault diagnosisusing unmeasurable inputs

    Eng. Appl. Artif. Intell.

    (2000)
  • J. Yu

    A nonlinear kernel Gaussian mixture model based inferential monitoring approach for fault detection and diagnosis of chemical processes

    Compu. Eng. Sci.

    (2012)
  • W. Zhang et al.

    Transport and fate modeling of nitrobenzene in groundwater after the Songhua River pollution accident

    J. Environ. Manage.

    (2010)
  • J. Zhang et al.

    Modeling fluid catalytic cracking risers with special pseudo-components

    Chem. Eng. Sci.

    (2013)
  • J. Zhang

    Improved on-line process fault diagnosis through information fusion in multiple neural networks

    Comput. Chem. Eng.

    (2006)
  • C. Zhao et al.

    Reconstruction based fault diagnosis using concurrent phase partition and analysis of relative changes for multiphase batch processes with limited fault batches

    Chemom. Intell. Lab. Syst.

    (2014)
  • J. Zhao et al.

    Multidimensional non-orthogonal wavelet-sigmoid basis function neural networks for dynamic fault diagnosis

    Comput. Chem. Eng.

    (1998)
  • C. Zhao et al.

    PHASuite: an automated HAZOP analysis tool for chemical process: part I: knowledge engineering framework

    Process Saf. Environ. Prot.

    (2005)
  • J. Zhao et al.

    Learning HAZOP expert system by case-based reasoning and ontology

    Comput. Chem. Eng.

    (2009)
  • D. Beebe et al.

    The connection of peak alarm rates to plant incidents and what you can do to minimize

    Process Saf. Prog.

    (2013)
  • Cited by (0)

    View full text