Industrial Internet of Things (IIoT) requires reliable, low-latency, and secure wireless links to support rate/delay/security-intensive applications. In this letter, we model the IIoT network as a geometric graph with a virtual coordinate system (VCS). One major problem with VCS is that it is sensitive to the placement of reference nodes. This letter first proposes a general formulation for selecting a specific set of reference nodes that minimizes the network error. Then, we derive the upper bound for the optimization problem and present a novel iterative method that selects additional reference nodes based on the network inference. Simulation results compare the proposed method with the random selection of reference nodes and show that the proposed method performs better regarding packet delivery rate, network connectivity, and total energy consumption.