This paper presents a graphical approach to assess the detectability of multiple simultaneous faults in mechanical systems such as Air Handling Units (AHUs). Symptoms of multiple simultaneous faults can cancel each other out, resulting in no indication of abnormality and rendering these faults undetectable when the measurements collected from AHUs are processed. Still, such undetectable faults can lead to significant energy inefficiency and economic loss. We present a method to assess the detectability of a given set of faults provided the causal relations among the variables and the sensor configuration in the system. Fault detectability assessment informs decision makers about possible risk scenarios the system is exposed to and enables them to design/modify the system to reduce this exposure, e.g. by adding extra sensors where necessary. Our approach first constructs a graphical representation of a given system so that the causal relations among the variables and whether they are sensed can be visually observed, which can help predict until where a fault symptom can propagate. Then, our approach examines a given set of faults to check whether it is possible for that set to be detected by the given set of available sensors. We illustrate the application of our approach to an operational AHU and classify which sets of faults are detectable and which are not. We compare our approach with alternative methods and show that one of its benefits is the capability of handling systems with unpredictable input variables. (C) 2018 Elsevier B.V. All rights reserved.