Heat Exchanger Performance Prediction using Machine Learning

Project Overview

In the petroleum refining industry, sensor data is often noisy, making it difficult to accurately predict equipment performance. This project aims to build a robust Machine Learning model to predict the Heat Load of a heat exchanger using noisy input data (Temperature & Flow Rate), simulating real-world refinery conditions.

Methodology

1. Baseline Model: Trained a Linear Regression model on clean simulation data to establish theoretical physics.
2. Real-world Simulation: Tested the model on noisy data (simulating sensor errors) after applying IQR outlier removal.
3. Advanced Modeling: Implemented a Random Forest Regressor to capture non-linear relationships and improve resistance to noise.

Key Findings

• The theoretical maximum R² score for this noisy dataset is approximately 0.57 (due to the high aleatoric uncertainty/noise ratio).
• The Linear Regression model underperformed on noisy data.
• The Random Forest model achieved an R² score of ~0.50, successfully capturing the physical patterns despite the heavy noise, proving its robustness for industrial applications.

Tools Used

• Python (Pandas, NumPy, Scikit-Learn)
• Data Visualization (Matplotlib, Seaborn)

Dataset Source

The dataset used in this project is publicly available on Kaggle: Heat Exchanger Dataset : Parametric Study(KAGGLE_DATASET_LINK)