Advances in reducing operations and maintenance (O&M) costs are crucial to improving the viability of the nuclear energy industry. One of the important aspects to reduce the cost of maintenance activities in nuclear power plants is to automate equipment monitoring and fault diagnoses. As an inverse problem to fault diagnoses, finding a suitable population of sensors that enable a requisite degree of monitoring capability, preferably at low cost, is a prerequisite that ensures a successful monitoring and diagnosis capability. This work develops an optimization tool for the sensor assignment problem of thermal-hydraulic systems that minimizes the cost for a required diagnosing capability. The optimization is driven by a genetic algorithm (GA), with its parameters tuned by Bayesian optimization (BO). Compared to the conventional GA parameter-tuning approach based on experimental designs, the BO-tuned parameters show better performance for the test problem with various allocated computing resources. It is also verified that the BO-tuned parameters perform better for several problem variants based on the original test problem, which has practical values in meeting additional engineering goals in the sensor assignment process.