پديد آورنده :
مناني، زهرا
عنوان :
پيش پردازش و الگوكاوي داده هاي آلارم صنعتي
مقطع تحصيلي :
كارشناسي ارشد
محل تحصيل :
اصفهان : دانشگاه صنعتي اصفهان
صفحه شمار :
سيزده، 91ص، مصور، جدول، نمودار
استاد راهنما :
ايمان ايزدي
توصيفگر ها :
پيش پردازش داده هاي آلارم , آلارم هاي نوساني , جاي گذاري داده هاي آلارم , الگوكاوي زماني , داده كاوي
استاد داور :
مريم ذكري، فريد شيخ الاسلام
تاريخ ورود اطلاعات :
1398/03/13
دانشكده :
مهندسي برق و كامپيوتر
تاريخ ويرايش اطلاعات :
1398/03/13
چكيده انگليسي :
Preprocessing and Pattern Mining of Industrial Alarm Data Zahra Mannani z mannani@ec iut ac ir April 2019 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Assist Prof Iman Izadi iman izadi@cc iut ac ir Advisor Assist Prof Nasser Ghadiri nghadiri@cc iut ac ir Abstract Today alarm system is an important part of the industry and processes By using advanced technologies defining alarmsin each section of the process is both easy and cost effective Thus the increased number of alarms in any process maycause confusion for the operator and deviate his her judgement to identify the root cause of the problem So improving alarmmanagement systems is very crucial in today s industrial processes Analysing historical alarm data can identify problemsin alarm systems Data mining algorithms and techniques are utilized to analyze alarm data Data mining is the scienceof finding patterns in large data Various frameworks are available for data mining One of the most important steps indata mining is the preprocessing of data Therefore by considering both data approaches point based and interval based a structure is proposed for preprocessing of alarm data Removing chattering and fleeting alarms and imputing missingalarm messages are the most important parts of preprocessing of alarm data Several algorithms are introduced to removechattering alarms and the effects of process variable types are investigated Then we have described the concept of missingmessages in the interval based alarms and two methods for imputation are presented ON OFF RLD median index andtemporal patterns are used to reconstruct missing alarms and then place them into the dataset The first one is based onthe information of the same unique alarms in the dataset and the latter on the information of the other unique alarms forestimating missing message time The goal of the modeling step is to find the related alarms and their relationship as temporalpatterns By using developed TPMiner algorithm a new method to find suffix sequences and alarm domain concepts theATPMiner algorithm is introduced The occurrence and duration probability of temporal patterns provide more informationto the analyst about the existing alarms in the pattern By adopting the developed P TPMiner algorithm for calculating theprobability of occurrence and the duration of a specific pattern a new algorithm AP TPMiner is introduced Finally agraphical user interface is designed based on this algorithm to display the association of the existing alarms in the temporalpattern and related probabilities This user interface is provided as an analytical tool for the expert analyst A case studydemonstrates the effectiveness of the proposed preprocessing and modeling methods on real industrial alarm data Key Words 1 Preprocessing of Alarm Data 2 Chattering Alarms 3 Alarm Data Impu tation 4 Temporal Pattern Mining 5 Data Mining
استاد راهنما :
ايمان ايزدي
استاد داور :
مريم ذكري، فريد شيخ الاسلام
