Monitoring large-scale systems is of fundamental importance in modern infrastructures. Many of these large-scale systems are complex interconnections of sub-components which interact by means of communication channels with limited bandwidth. Therefore the information must be encoded in order to be transmitted. Given a class of encoders, the problem of detecting faults affecting some of these sub-components starting from the encoded information is studied, and a precise characterization of a class of faults which can be detected is given.