In complex dynamical systems, traditional rule-based fault detection algorithms remain highly static, and struggle to capture important patterns or trends. Worse, many traditional methods provide an overwhelming amount of poorly labeled information, making root cause analysis extremely difficult. In order to capture localized and system wide dynamic trends, an extension to rule-based fault detection algorithms is required. In this paper, we propose a spectral clustering method as a means to recognize dynamical patterns present in commercial HVAC system fault diagnostic signals. Through applications of the Koopman operator and spectral analysis, similar patterns in HVAC device faults are automatically detected and then described using principal component analysis. This method provides an advanced extension of traditional rule-based fault detection algorithms and shows effective root cause diagnosis. Finally, this method provides a comprehensive understanding of a complex system and increases the ability for automated fault detection. We display a proof of concept by applying the proposed method to a large commercial building with an extensive system of network sensors.