This paper presents a genetic-algorithm-based optimization method to design a sensor network for pipeline fracture detection in petrochemical plants. To avoid catastrophic consequences, it is desirable to detect any pipeline fractures as early as possible, while minimizing false alarms. Sensor network design is the first step in structural health monitoring for early detection of pipeline fractures. Sensor network design is conducted in three steps. First, a hydraulic simulation model of the pipeline system is built to calculate pressure drops before and after fracture occurrences. Inherent randomness in model parameters, such as pipeline roughness, is incorporated in the simulation model as physical uncertainty, while uncertainty in pressure measurements is considered as statistical uncertainty. Statistical distributions of pressure heads at different locations are calculated with the simulation model. Second, for a particular set of fracture scenarios, the statistical distributions of pressure heads at individual nodes are computed with the probabilistic simulation model. Then, true positives and false negatives are quantified using a detectability metric. Finally, a mixed integer nonlinear programming (MINLP) optimization problem is formulated to find the optimal sensor locations by maximizing detectability and the optimization problem is solved using a genetic algorithm. The optimal locations computed by three different MNNLP algorithms are compared in terms of accuracy and computational cost. In the future, the optimal design of sensor networks suggested in this work will be compared with experimental results.