Structural health monitoring (SHM) systems rely on a network of sensors to assess the health of engineering structures that are designed to last for decades. Over time, engineering structures such as wind turbine towers can experience degradation-related damage, while the monitoring systems themselves can age and degrade, becoming less reliable. This aging of SHM systems can result in sensor faults that produce plausible but incorrect data, leading to misinterpretations of structural integrity and potentially catastrophic failures. Therefore, distinguishing between sensor faults and structural damage is critical to ensuring the reliability of SHM over the lifetime of the structure. This study presents a probabilistic sensor fault detection approach to address this issue. The sensor correlation within the sensor network is analyzed and sensor faults are detected using Mahalanobis distance. The sensor fault detection is validated on a real support structure in the field, which is realized as a 9-m-high lattice mast under real environmental conditions. The results show that different sensor faults such as bias and drift can be accurately distinguished from structural damage, whereas gain and noise increase are not detectable. The advantage of this approach is that a generalized threshold can be defined based on the probabilistically based Mahalanobis distance, which enables automated sensor fault detection. Overall, increasing the robustness of SHM systems will significantly improve the reliability of data-based assessments, a task that is becoming increasingly important for long-lived structures.