| Abstrakt: |
Mathematical programming can express competency concepts in a well-defined mathematical model for a particular Any system that runs is always be expected to experience faults in different ways. Any change in the physical state of numerous components, control machinery, as well as environmental factors, might result in these problems. In process industries, where prompt detection is crucial in maintaining high product quality, dependability, and safety under various operating situations, finding these flaws is one of the most difficult tasks. The goal of this project is to implement several machine learning techniques for fault identification and classification in a binary distillation column. A pilot binary distillation unit (UOP3CC) is utilized for this purpose. The set up is run under normal operating conditions and the real time data is collected. Three common faults namely reboiler fault, feed pump fault and sensor fault are introduced one at a time and the faulty data is collected. These data are then introduced in to different machine learning algorithms like Logistic Regression, KNN, Naive Bayes, Decision Tree, Gradient Boosting, X Gradient Boosting, SVC and Light Gradient Boosting for model development. 70% of the data samples used for training and 30% of data samples are used for testing. It is found the Decision tree algorithm gives the best accuracy possible with 99.9%. Using decision tree algorithm, fault classification is performed for different datasets and is found that the algorithm was able to classify accurately even for new untrained datasets. [ABSTRACT FROM AUTHOR] |
