Anomaly detection is key to satellite equipment as it can lead to the detection of potential faults that can lead to great damages. So far that are various methods that have been developed ranging from simple thresholds to machine learning methods. The latter receives recently great attention due to its efficiency and capabilities in treating complex situations. However, still there are many challenges in the application of these methods due to several reasons related to the learning complexities as well as the characteristics of the processes. During this work, we evaluate some machine learning (with different architectures) methods on common NASA datasets (SMAP and MSCL) based on different parameters and criteria to have a clear vision of the methods impact. Results show that the complexities of ML methods learning still poses a challenge for the filed. Moreover, it is advised to use lightweight ML methods to get better performance in terms of time and learning complexities.