Multi-Fault Detection in Fluid Catalytic Cracking Unit Using Hybrid CNN-LSTM Deep Learning Model

This paper introduces an innovative blend of deep learning models that combines Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks for multi-fault detection in Fluid Catalytic Cracking Units (FCCUs). Using data from an open-source FCCU, we generated datasets for normal operations and five fault scenarios. The hybrid model uses CNNs for spatial characteristic extraction and LSTMs for temporal dependency modeling across 46 process variables. Custom feature engineering and sequence-based inputs enhance fault detection sensitivity. Results demonstrate high accuracy in distinguishing between normal operations and various fault conditions, with success in identifying subtle, evolving faults. Feature importance analysis provides insights into critical variables for fault detection. This approach offers a promising solution for real-time fault diagnosis in FCCUs, potentially reducing downtime and improving system reliability in oil refineries.