Refinery system, a typical example of process systems, is presented as complex network in this paper. The topology of this system is described by task-resource network and modeled as directed and weighted graph, in which nodes represent various tasks and edges denote the resources exchanged among tasks. Using the properties of node degree distribution, strength distribution and other weighted quantities, we demonstrate the heterogeneity of the network and point out the relation between structural characters of vertices and the functionality of corresponding tasks. The above phenomena indicate that the design requirements and principles of production process contribute to the heterogeneous features of the network. Besides, betweenness centrality of nodes can be used as an importance indicator to provide additional information for decision making. The correlations between structure and weighted properties are investigated to further address the influence brought by production schemes in system connectivity patterns. Cascading failures model is employed to analyze the robustness of the network when targeted attack happens. Two capacity assignment strategies are compared in order to improve the robustness of the network at certain cost. The refinery system displays more reliable behavior when the protecting strategy considers heterogeneous properties. This phenomenon further implies the structure-activity relationship of the refinery system and provides insightful suggestions for process system design. The results also indicate that robustness analysis is a promising application of methodologies from complex networks to process system engineering.