Engineering systems are designed to serve society for decades and are exposed to multiple environmental and load conditions. This circumstance leads to the degradation of the system, causing different structural behaviours. In this regard, monitoring systems can assess structural health by recording time-dependent structural behaviour through a network of measurement instruments. However, monitoring systems can be affected by changing external conditions, leading to sensor degradation and faults that produce plausible data but incorrect information about the structural condition. Therefore, distinguishing between sensor faults and structural damage is critical to ensuring the reliability of SHM over the lifetime of the structure. Different techniques with varying effort and complexity could be used, such as linear regression, Mahalanobis distance, or multiple neural network architectures, to identify sensor faults. This study investigates the suitability of these techniques regarding the accuracy of sensor fault detection and computational effort on the example of a steel specimen equipped with nine strain gauges in a 3-point bending test with a cyclic load. In the test, one of the nine sensors showed a signal drift, which needs to be detected in a monitoring application using these sensor fault detection techniques. The results show that for this simple test, the Mahalanobis distance is the fastest and most accurate, while the linear regression technique is very fast but very imprecise, and the artificial neural network is relatively accurate but very slow. However, all the techniques used showed the potential to significantly increase the robustness of the monitoring system, which is important for more complex engineering structures.