Abstract

This work explores the use of Machine Learning techniques for advanced level control and fault prediction in a coupled tank system. The main objective is to design a predictive model capable of improving liquid level regulation and enabling early fault detection, addressing the limitations of conventional Proportional-Integral-Derivative (PID) controllers in dynamic industrial environments. Two Machine Learning techniques, such as Multilayer Perceptron (MLP) and Random Forest, were used for level control and fault prediction, respectively. The results showed that MLP models with four hidden layers significantly reduced prediction errors, while Random Forests provided superior fault prediction accuracy due to their robustness against overfitting. Compared to traditional methods, the proposed Machine Learning-based control system demonstrates enhanced adaptability, precision, and stability in scenarios with high variability and noise. In addition, the system’s fault prediction capabilities improve predictive maintenance and operational efficiency, which are crucial for industrial safety and reliability. These facts highlight the potential of Machine Learning techniques to transform process control by addressing complex challenges and enabling smarter, more reliable industrial systems.