In this article, we present a graph theoretic method to visualize and analyze the system behavior under different operating conditions. The system attributes (or variables) are the nodes in the graphs, and partial correlation between a pair of attributes defines the distance. between corresponding nodes, resulting in a fully connected graph. Then, the redundant links are reduced using Pathfinder Network Scaling technique to uncover the latent network structure. We use a simulated biological reactor dataset in normal and faulty operation to validate our method. The method is general and can be used to analyze several different systems.