The primary goal of Spacecraft Failure Detection, Isolation, and Recovery (FDIR) is to ensure the reliability, availability, maintainability, and operational autonomy of missions, thus securing their success even in the face of potential failures. Traditional FDIR approaches mandate the identification of all potential failure scenarios during the spacecraft’s design phase, which often leads to substantial development and operational costs associated with resolving unanticipated in-orbit anomalies. Therefore, it can be more cost-effective to employ an on-board system capable of learning from telemetry data, enabling it to perform monitoring tasks with minimal prior knowledge of expected failures. While numerous strategies for detecting failures and anomalies in time series data have been developed and utilized in various missions, the increasing complexity of modern spacecraft presents ongoing challenges for both ground-based and on-board smart anomaly detection. A significant constraint is the limited hardware and computational resources, with processors like LEON IV and space-qualified FPGAs offering far less computing power compared to contemporary GPUs. Consequently, it becomes essential to adapt these techniques. This study offers an initial evaluation of the performance of diverse machine learning methods in identifying different failure scenarios. It also highlights the specific intricacies and obstacles involved in implementing these techniques on board a spacecraft.