Skip to content

Ansari Emaad Iftekar

Tue, 19 December, 2023
Ansari Emaad Iftekar (left) visiting TWI SEA office in Malaysia.
Ansari Emaad Iftekar (left) visiting TWI SEA office in Malaysia.

Student Name

Ansari Emaad Iftekar


Research Title

Development of an Improved Ensemble Learning Algorithm for Corrosion Erosion Prediction of Oil and Gas Pipelines



Corrosion Erosion Monitoring, Machine Learning, Ensemble Learning, Pipeline Corrosion, Corrosion Rate


Relevant Industries

Oil and Gas, Aerospace, Automotive



Lloyd's Register Foundation

Affiliated University

Universiti Sains Malaysia

Industry Supervisor(s)

Dr. Abbas Mohimi (TWI Ltd)


Academic Supervisor(s)

Dr. Elmi Abu Bakar and Dr. Mohammad Nishat Akhtar (Universiti Sains Malaysia)


Start Date

19 December 2023


Project Outline

Pipelines are constrained by their intricate applicability criteria and calculations. Due to their effectiveness and accuracy, data-driven models based on machine learning (ML) is quickly emerging as the new fashion. The internal corrosion leading to erosion (corrosion erosion) rate in oil and gas pipelines can be accurately predicted using robust ensemble learning models, which are proposed in this work as a practical application. Correctly estimating the rate of corrosion erosion in fluid-flowed oil and gas pipelines has a big impact on the system's safety and controllability. The proposed predictive data-driven models will use ensemble learning (EL) techniques in order to predict corrosion erosion rate.

A crucial first step in the implementation procedure is selecting the EL-models' parameters. In order to prevent overfitting problems and ensure better selection of the hyper-parameters for the ELmodels, the k-cross validation technique will be used during the training phase. This technique involves splitting the training data into k equal-sized subsets, using k-1 to train the EL-models with the pre-selected parameters, and then using the last fold to validate the model. The performance of the EL-models will be expressed as the average of the k fold outcomes. After creating the predictive models based on the training phase, the output (internal Crate) will be estimated using the test datasets. The developed predictive EL-models will be evaluated for performance using the obtained predictive results from both phases (training and testing), which will then be subjected to statistical and graphical analysis using a variety of criteria.