Reliable and timely fault detection and isolation are necessary tasks to guarantee continuous performance in complex industrial systems, because early fault detection can avoid failure propagation in the system and help to minimize downtime. Model-based diagnosis provides different techniques to fulfill those requirements, and has the advantage of model reusability. However, reusing existing complex non-linear models for diagnosis in large industrial systems is not straightforward. Most of the times the models have been created for other purposes different from diagnosis, and many times the required analytical redundancy is small. In this work we propose to use Possible Conflicts, which is a model decomposition technique. Possible Conflicts provide the structure (equations, inputs, outputs, and state variables) of the minimal model able to perform fault detection and isolation. Then we propose to use such structural model to design a gray box model by means of a state space neural network. We demonstrate the feasibility of the approach in an evaporator for a beet sugar factory using real data.