To facilitate efficient control and monitoring, massive wireless sensors and measurement devices are deployed in Oil and Gas Refineries. These sensors are deployed along the pipes and data measured are correlated. In-network processing with enough data along a pipe allows abnormal events to be detected early by an analyzer deployed on-site. The data collection structure should then be carefully developed to facilitate each pipe to be monitored efficiently. On the other hand, as the sensors are deployed in harsh environment and subject to hardware damages, they may fail. Sensor failures may disrupt the on-site monitoring of pipes by analyzers. In this paper, we study the resilience issue in refinery sensor networks to facilitate the robustness of fast data collection and abnormal event monitoring even some devices fail. We present a quorum scheme to ensure every analyzer would get enough relevant data for analysis. We apply a multi-tree data collection structure to achieve fast data collection. To tolerate node failures, we present a novel distributed Refinery Resilient Protocol (RRP), which enables the affected nodes to discover an alternative path to relay data. The simulation results show that the RRP maintains efficient data collection and event monitoring even a significant portion of sensors have failed.