A general method to design optimal redundant sensor network even in the case of one sensor failure and able to estimate process key parameters within a required accuracy is proposed. This method is based on a linear model, which derived from a non-linear data validation model and the sensor network is optimised thanks to a genetic algorithm. To reduce the solution time, two parallelisation techniques, both using the Message Passing Inter-face (MPI) library, are compared: the global parallelisation and the distributed genetic algorithms. Both methods allow reducing the solution time, but the second one is more efficient. Results are presented for an ammonia synthesis loop. © 2005 Elsevier Ltd. All rights reserved.