A Novel Energy Performance-Based Diagnostic Model for Centrifugal Compressor using Hybrid ML Model
Mukhtiar Ali Shar,Masdi B Muhammad,Ainul Akmar B Mokhtar,Mahnoor Soomro
DOI: https://doi.org/10.1007/s13369-024-08793-1
IF: 2.807
2024-03-12
Arabian Journal for Science and Engineering
Abstract:Faulty compressors must be timely detected to prevent excessive energy consumption, maintenance, and energy costs. Existing diagnostics models lack addressing the energy performance indicators and do not provide effective hybrid machine learning (ML) model for advanced fault diagnosis to prevent compressors from becoming energy hogs. Therefore, this study proposed a novel approach in the form of an energy-based diagnostic model for integrating energy performance indicators to detect the healthy and faulty behavior of the compressor using a hybrid ML model. The time series analysis and Isolation Forest techniques have been used to detect faulty and healthy behavior of centrifugal gas compressor. To obtain more insight and prevent false alarms the hybrid ML model was introduced. The Ridge regression was used as the meta-classifier in the suggested hybrid model, which receives the input from the base classifiers Decision Tree (DT), k -Nearest Neighbors (kNN), and Gradient Boosting (GB) to optimize the performance and accuracy of hybrid model. This study was conducted on a two-stage centrifugal compressor that compresses production gas for export powered by a gas turbine at Malaysia's PETRONAS Angsi oil and gas field. According to the findings, the energy efficiency predicted for the first stage was 84.2% for healthy behavior and 69.7% for faulty behavior, while for the second stage, it was 83.2% for healthy behavior and 68.1% for faulty behavior, indicating high energy efficiency during the healthy operation of a centrifugal compressor in comparison with faulty behavior. The slight difference between the proposed diagnostic model training, testing, and prediction performance accuracy 0.98, 0.97 and 0.99 proposes a model is efficient neither overfitting nor underfitting according to the value of co-efficient of determination ( R 2 ). The R 2 values for training, testing, and prediction performance accuracy for the GB model were 0.95, 0.93, and 0.94; for kNN, 0.89, 0.87, and 0.86, and for Tree, 0.95, 0.94, and 0.93 respectively. According to the results, the proposed hybrid model performs more eloquently and efficiently than other single models DT, kNN, and GB. This study empowers operators to take critical measures to increase energy efficiency, reduce downtime, and schedule maintenance to improve the reliability of centrifugal gas compressor.
multidisciplinary sciences