Recently, continuous power reactor core surveillance, which is based on fixed in-core detector readings, has exhibited a trend of growing significance. The fixed in-core sensors can only be replaced during shutdown periods; therefore, it is important to have reliable information on the quality of each detector in advance to be able to create an appropriate detector replacement schedule. During the operating cycle, the continuous core surveillance system should rely only on reliable measurements, and only an effective detector failure diagnosis can ensure avoiding falsified information. At the same time, most published signal validation methods are not well suited for an extensive set of fixed in-core detectors. A relatively simple, though powerful and robust, method is proposed that can be applied for both signal validation and early failure detection. The basic idea of the method is that inevitably there exist such process noise components in the detector signals that are characteristically correlated within some well-determined groups of sensors. The lack of such correlation most probably occurs due to some detector failure. When a smaller, localized subgroup of the detectors shows a decreased level of correlation with the majority, that is typically caused by some abnormal event in the technological process. In such cases the results of this method can be utilized as a target identification tool for the more sophisticated noise diagnostics methods. The method has been thoroughly tested with an extensive data set, including rhodium self-powered neutron detectors and assembly outlet thermocouple signals, which was collected throughout a complete operational cycle of a VVER-440/213-type pressurized water reactor.