A Collection of Exploratory Data Analysis Techniques for Process Fault Detection in the Tennessee Eastman Dataset

The Tennessee Eastman process (TEP) is a widely studied industrial benchmark for testing fault detection and diagnosis techniques, given its complexity and the need for reliable fault detection systems in chemical plants. This paper proposes a comprehensive application of exploratory data analysis (EDA) techniques to the TEP dataset to enhance process fault detection and improve understanding of the dataset's features. The proposed methodology involves plotting time series data to capture temporal dynamics and applying Shewhart control charts to monitor process variability. Additionally, deviation atlases are used to detect anomalies across all features, dendrograms are employed to identify hierarchical clusters, and correlation heatmaps visualize interdependencies between process variables. By applying these EDA techniques, patterns and relationships that were less apparent in the raw dataset became identifiable, contributing to a more thorough fault diagnosis. The findings highlight the value of EDA in uncovering hidden insights, which could aid in the development of more robust fault detection algorithms for real-world industrial applications.