Magnetic sensors often experience faults such as no-response, noisy signal, and saturation. Yet, they have considerable object localization applications that require high precision, such as in medical operations. Conventionally, Dipole Magnetic (DM) position tracking is used for magnetic localization, even while a sensory fault occurs. But DM position tracking is not sufficiently accurate, and its computational cost is a matter of concern. Accordingly, the proposed approach here is in three folds. First, we propose to use a heuristic to detect faulty sensors and to stop the propagation of faulty reading by setting their readings to zero. Second is using a nonlinear modeling platform, Recurrent Neural Network (RNN) for the actual nonlinear mapping of the magnet sensory readings and placement due to its’ accurate outputs. And third is to prepare a sufficiently rich data set for training the network that is prepared under no sensory fault. The experimental study here confirms that the faulty sensory reading is successfully identified and set to zero by the proposed heuristic, and the nonlinear mapping of the neural network provides a good assessment of magnet localization even when the corresponding inputs from faulty sensors are set to zero. The experimental setup here consists of a network of eight magnetic sensors, one of which becomes faulty during the experimentation process. More specifically, results show that the accuracy of our method has improved up to 444.3% to DM method and its robustness enhanced to 105.3% to an RNN which is trained without our rich data set.