University of Alberta Advanced Analysis and Redesign of Industrial Alarm Systems by Sandeep Reddy Kondaveeti A thesis submitted to the Faculty of Graduate Studies and Research in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Process Control Department of Chemical and Materials Engineering (cid:13)cSandeep Reddy Kondaveeti Spring 2013 Edmonton, Alberta PermissionisherebygrantedtotheUniversityofAlbertaLibrariestoreproducesinglecopiesofthisthesis andtolendorsellsuchcopiesforprivate,scholarlyorscientificresearchpurposesonly. Wherethethesisis convertedto,orotherwisemadeavailableindigitalform,theUniversityofAlbertawilladvisepotential usersofthethesisoftheseterms. Theauthor reservesallotherpublicationandother rightsinassociationwiththecopyrightinthethesis and,exceptashereinbeforeprovided,neitherthethesisnoranysubstantialportionthereofmaybeprinted orotherwisereproducedinanymaterialformwhatsoever withouttheauthor’spriorwrittenpermission. DEDICATED TO MY TEACHERS who taught me more by what they are... than what they said Abstract In the process industry, the alarm system acts as a layer of protection between theBasicProcessControlSystem(BPCS)andtheSafetyInstrumentedSystem (SIS). The BPCS is designed for automatic regulation of day to day process operation and SIS comes into picture when emergency shutdown is required. There are specific standards worldwide that define how BPCS and SIS system are to be designed and expected to work. However, not many well defined standards areavailable fordesign and management of industrial alarmsystems mainly due to the heavy involvement of human factors. Alarm management has gained complexity mainly due to increasing size of process plants and also due to the Distributed Control System (DCS) that presents little motivation for limiting the number of variables on which alarms can be configured. Alarm management lifecycle for maintaining an efficient alarm system as suggestedbytheInternationalSocietyofAutomationstandards(ISASP18.02) is discussed in this work with emphasis on monitoring and assessment and design stages. In this work, novel tools for assessment of alarm system based on routinely collected alarm event data are proposed and demonstrated. The primary focus of these tools is to identify nuisance alarms such as chattering and redundant alarms. Alarm event data is represented mathematically and indices are proposed to calculate the extent of similarity between two alarms and also to estimate the extent of chattering in an alarm. Two of the most commonly used techniques for reducing alarm chatter, delay timers and latches are discussed in detail. Effect of varying the size of on-delay, off-delay timers and latches on the accuracy of detection is discussed in the Receiver Operating Characteristic (ROC) framework from a theoretical view point by modeling them using Markov chains. Use of Return to Normal (RTN) information in addition to alarm events information in designing delay timers is also discussed. Finally, advantages of multivariate techniques such as Principal Components Analysis (PCA) based T2 and Q statistic as opposed to univariate monitoring are discussed in the same framework using simulation examples and an industrial case study. Acknowledgements It is a pleasant feeling to thank everyone who contributed to this study and made it a life changing experience for me. Firstly, I would like to express my deep gratitude to my supervisor Prof. Sirish Shah for his vision, attitude, patient guidance and constant encouragement. Dr. Shah has given me ample freedom, time and exposure to industrial setting which I believe were crucial for developing this thesis. I am very fortunate to have spent significant time with someone like Dr.Shahwithvast knowledge andexperience. Itrulyadmire and would like to emulate some of his great qualities such as being extremely patient, positive, respectful and determined. I would like to thank Prof. Shankar Narasimhan for exposing me at an earlystageinmycareer toindustrial dataanalysisandotherprojectsinvolving advanced process control. I am indebted to my classmate, friend and mentor Dr. Ramsharan Rangarajan for pushing me to pursue graduate studies when I was uncertain about what was in store. My sincere thanks to Dr. Arun Tangirala, Prof. Biao Huang, Prof. Sirish Shah, Dr. Amos Ben-Zvi, Dr. Jong Min Lee, Dr. Nilanjan Ray and Dr. Richard Sutton for their passion to teaching some of the advanced courses relevant to this thesis. I would like to thank Dr. Iman Izadi for his guidance, encouragement and like-mindedness. We had numerous discussions that were instrumental in developing someimportantideasinthisthesis. ManythankstoProf. Tongwen Chen for leading the study and providing constructive feedback on some of the work. Special thanks to Dr. Dave Shook and Dr. Ramesh Kadali for theirsupport andinvaluableguidanceespecially onhumanfactorsinindustrial operations due to which the dissertation has significant practical application. I must also thank Trevor Hrycay, Rajeev Varma, Jamie Dang, Tim Black, Tim Burton and David Joe for taking time in supporting this study. Ifeelextremelyfortunatetobepartofapowerhouse, theComputerProcess Control (CPC) group. The knowledge gained through technical presenta- tions/discussions and the friendships developed during my stint with members of this association will last forever. I also enjoyed being a member of the Advances in Alarm Management and Design group and wish the group a huge success moving forward. I would like to specially thank the following group members for their support and friendship: Fan Yang, Prakash, Harigopal, Sankar, Phanindra, Mridul, Enayet, Venkat, Karteek, Saneej, Pabbi, Abhijit, Barath, Fei Qi, Tim, Salim, Shoukat, Hailei, Mohammad Iqbal, Rumana, Anuj, Yashaswi, Samidh, Aditya, Yue, Ping, Naseeb, Amin, Kabir, Elham and Sridhar. I would like to thank my friends in Edmonton, Hari, Varghese, John, Ananth, Anand, Ravi and Sriram for the wonderful memories I had at Garneau towers and during several outings that made me feel at home. I would like to take this opportunity to also thank my friends Suman, Mirza and Dhanno for tolerating my idiosyncrasies. I would like to acknowledge the Department of Chemical and Materials Engineering, University of Alberta for providing me the opportunity to pursue my doctoral degree. I also gratefully acknowledge the financial support from the Natural Sciences and Engineering Research Council of Canada (NSERC), Matrikon Inc., Suncor Energy Inc. and the Alberta Innovates Technology Futures (AITF). I would like to thank Lily Laser, Marion Pritchard and Betty Daniel for their help with various administrative requests. I would like to express my earnest gratitude to my parents and both sisters Supriya (brother in-law Praveen) and Saveena (brother in-law Raghu) who made many sacrifices to raise me and give me this education. I am eternally grateful to my wife, Nishu for her caring love, support and patience without which the thesis wouldn’t have been possible. Contents 1 Introduction 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1.1 Importance of safe process operation . . . . . . . . . . 2 1.1.2 Independent layers of protection . . . . . . . . . . . . . 2 1.2 Role of alarms in the control room . . . . . . . . . . . . . . . 3 1.2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.2.2 Major incidents . . . . . . . . . . . . . . . . . . . . . . 5 1.3 Research objectives and scope of this work . . . . . . . . . . . 8 1.4 Organization of the thesis . . . . . . . . . . . . . . . . . . . . 8 2 Introduction to alarm management 12 2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.2.1 Alarm . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.2.2 Alarm design considerations . . . . . . . . . . . . . . . 13 2.2.3 Types of alarms, their attributes and states . . . . . . 15 2.2.4 Alarm system . . . . . . . . . . . . . . . . . . . . . . . 19 2.3 History of alarm management . . . . . . . . . . . . . . . . . . 20 2.4 Benefits of alarm improvement . . . . . . . . . . . . . . . . . . 24 2.5 Standards, guides and references . . . . . . . . . . . . . . . . . 24 2.6 Alarm management lifecycle . . . . . . . . . . . . . . . . . . . 25 2.6.1 Brief description of the stages . . . . . . . . . . . . . . 25 2.6.2 Importance of monitoring and assessment stage . . . . 27 2.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 28 3 Graphical Tools for Routine Assessment of Industrial Alarm Systems 30 3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.3 Proposed methodology . . . . . . . . . . . . . . . . . . . . . . 33 3.3.1 Down sampling to seconds . . . . . . . . . . . . . . . . 33 3.3.2 Binary sequence representation of alarms . . . . . . . . 33 3.3.3 Graphical tools . . . . . . . . . . . . . . . . . . . . . . 35 3.4 High Density Alarm Plot . . . . . . . . . . . . . . . . . . . . . 35 3.4.1 Each row corresponds to an alarm . . . . . . . . . . . . 37 3.4.2 Color coding of alarm count in time intervals . . . . . . 37 3.4.3 Rank ordering of alarms . . . . . . . . . . . . . . . . . 38 3.5 Alarm Similarity Color Map (ASCM) . . . . . . . . . . . . . . 39 3.5.1 Padding each binary sequence with extra 1’s . . . . . . 41 3.5.2 Alarm similarity index . . . . . . . . . . . . . . . . . . 41 3.5.3 Re-ordering of the similarity matrix . . . . . . . . . . . 43 3.5.4 Color coding . . . . . . . . . . . . . . . . . . . . . . . . 44 3.6 Second case study . . . . . . . . . . . . . . . . . . . . . . . . . 46 3.6.1 HDAP for the second case study . . . . . . . . . . . . . 46 3.6.2 ASCM for the second case study . . . . . . . . . . . . 49 3.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.8 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 51 4 Quantification of alarm chatter based on run length distribu- tions 54 4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.3 What is a chattering alarm? . . . . . . . . . . . . . . . . . . . 57 4.4 Runs and run lengths . . . . . . . . . . . . . . . . . . . . . . . 58 4.4.1 Brief history of the use of runs and run lengths . . . . 58 4.4.2 Run lengths in the alarm context . . . . . . . . . . . . 59 4.4.3 Run Length Distribution . . . . . . . . . . . . . . . . . 61 4.5 Chatter index based on the run length distribution . . . . . . 63 4.5.1 Differences between chattering and non-chattering alarms 63 4.5.2 An index to measure alarm chatter . . . . . . . . . . . 65 4.5.3 Chatter index based on inverse weighting of the DPF . 66 4.5.4 Properties of the proposed chatter index, Ψ . . . . . . 66 4.5.5 Scope for a modified chatter index . . . . . . . . . . . 67 4.6 Industrial case study . . . . . . . . . . . . . . . . . . . . . . . 68 4.6.1 High Density Alarm Plot . . . . . . . . . . . . . . . . . 70 4.6.2 RLDs . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 4.6.3 Chatter indices Ψ and Ψ . . . . . . . . . . . . . . . . 75 τ 4.7 Improvement in chatter index after redesign of alarms . . . . . 76 4.7.1 Flow tag - tag.id5 . . . . . . . . . . . . . . . . . . . . . 77 4.7.2 Density tag - tag.id6 . . . . . . . . . . . . . . . . . . . 77 4.7.3 Results and Discussion . . . . . . . . . . . . . . . . . . 77 4.8 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 80 5 On the use of delay timers and latches for efficient alarm design 84 5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 5.3 Delay timers and latches . . . . . . . . . . . . . . . . . . . . . 86 5.3.1 Alarm delay timers . . . . . . . . . . . . . . . . . . . . 86 5.3.2 Alarm latches . . . . . . . . . . . . . . . . . . . . . . . 87 5.4 Markov chains . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 5.5 Modeling delay timers and latches using Markov chains . . . . 91 5.5.1 Receiver Operating Characteristic curve . . . . . . . . 93 5.5.2 ROC analysis for delay timers and latches . . . . . . . 94 5.5.3 Case study and discussion . . . . . . . . . . . . . . . . 97 5.6 Design of delay timers using historical alarm and return to normal data in an industrial case study . . . . . . . . . . . . . 98 5.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 100 6 Application of Multivariate Statistics for Efficient Alarm Gen- eration 104 6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 6.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 6.2.1 Fault detection and process monitoring . . . . . . . . . 105 6.2.2 Alarms and alarm systems . . . . . . . . . . . . . . . . 106 6.3 Multivariate Statistical Process Control . . . . . . . . . . . . . 107 6.4 PCA based Statistical Process Monitoring (SPM) . . . . . . . 109 6.5 Receiver Operating Characteristics (ROC) curves . . . . . . . 111 6.6 Effectiveness of multivariate techniques . . . . . . . . . . . . 113 6.6.1 Linear System . . . . . . . . . . . . . . . . . . . . . . . 113 6.6.2 Tennessee Eastman Process (TEP) . . . . . . . . . . . 116 6.6.3 Industrial case study (Pump example) . . . . . . . . . 124 6.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 129 7 Concluding remarks 131 7.1 Main contributions of this thesis . . . . . . . . . . . . . . . . . 131 7.2 On going work and recommendations for future work . . . . . 134 A Quantification of alarm chatter 137 A.1 Theoretical bounds on Ψ . . . . . . . . . . . . . . . . . . . . . 137 A.2 Fictitious example of a chattering alarm . . . . . . . . . . . . 138 A.3 Proof that Ψ ≥ Ψ ∀ τ < τ . . . . . . . . . . . . . . . . . 139 τ1 τ2 1 2 A.4 Non-uniqueness of Ψ . . . . . . . . . . . . . . . . . . . . . . . 139