Open AccessMachine Learning in AWS for IoT-based Oil Pipeline Monitoring System
Author(s) -
Sabreen J. Siwan,
Waleed F. Shareef,
Ahmed R. Nasser
Publication year - 2022
Publication title -
webology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.259
H-Index - 18
ISSN - 1735-188X
DOI - 10.14704/web/v19i1/web19209
Subject(s) - cloud computing , pipeline transport , computer science , python (programming language) , internet of things , random forest , pipeline (software) , support vector machine , decision tree , real time computing , classifier (uml) , embedded system , artificial intelligence , operating system , engineering , environmental engineering
The world's economy is dominate by the oil export business, which is heavily reliant on oil pipelines. Due to the length of the pipes and the harsh environment through which they pass, continuous structural health monitoring of pipelines using normal methods is difficult and expensive. In this paper, an IoT system integrated with cloud services is propose for oil pipeline structure monitoring. The system is based on collecting data from sensor nodes attached to the pipeline structure, which collectively form a network of IoT devices connected to the AWS cloud. Measurements from sensor nodes are collect, stored, and filtered in AWS cloud. Measurements are also make accessible to users through the internet in real-time using Python web framework, Flask, and sending alarms via email in real-time. The performance of the system is evaluate by applying damaging events (hard knocking) on the oil pipeline at several distances. Analysis of IoT data by machine learning classification algorithms, apply and comparison between SVM, Random Forest Classifier, and Decision Tree to determine the best one, and then built in EC2 Linux in AWS to analyses the measurements and classify new events according to their distances from the sensor nodes. The proposed system is test on field measurements that were collect in Al-Mussaib Gas Turbine Power Station in Baghdad. Among the three classifiers, Random Forest achieved 90% classification rate.