Gas Turbine operator M&D centers currently spend a disproportionate amount of time responding to false alarms. Often, these false alarms are caused by errors in the instrumentation acquisition and recording chain. Sensors may fail outright, drift, or data may be corrupted during transmission or archival storage. Any one of these issues can lead to erroneous data which may lead to false alarms at the M&D center. Many advanced algorithms used for M&D are particularly susceptible to errors in input signals. Erroneous or missing data can cause issues during the training process and during diagnostic use. This work has constructed a physics-informed AI model which calculates a health indicator for faulty, mis-calibrated, or mis-recorded sensor readings. The process uses auto associative neural networks coupled with Historian data to identify not just a good/bad flag, but a likelihood score which indicates the probability that the reading is indeed erroneous. In the event the signal is anomalous, a digital twin model can be executed to determine if the issue is indeed instrumentation, or if the issue is hardware related. The method works for any physical asset and can be deployed using standard libraries available to gas turbine owners and operators.