Today, low frequency disc-brake noises are commonly explained as self-sustained stick-slip oscillations. Although, at a first glance this explanation seems reasonable, there are indices that cast doubt on it. For instance, the basic frequency of the observed oscillations does not scale with the disc-speed as it is with stick-slip oscillations and the classical model does not explain the observed ending of the vibrations beyond a certain speed. Indeed, our experimental studies on groaning noises reveal two different vibration patterns: stick-slip vibrations at almost vanishing relative speed and a second, differing vibration pattern at low to moderate relative speeds. Yet, these two patterns produce a very similar acoustic impression. While the experiment provides a vast amount of data, the dimension and structure of the underlying oscillation is not known a priori – hence, constructing phenomenological minimal models usually must rely on assumptions, e.g. about the number of DOF, etc. Due to noise and complexity, the measured raw data did only allow for a first straight forward insight, rendering further analysis necessary. Hence, time-delay embedding methods together with a principle component analysis were used to reconstruct a pseudo-phase space together with the embedded attractor to analyse for the system's dimension and to separate signal from noise.