 This paper proposes a novel unsupervised anomaly detection framework for natural gas-fired gensets in district heating networks. The framework uses machine learning algorithms to analyze SCOTA data and identify anomalous behavior. The model was tested on nine major failures of a genset in the Aosta DH plant in Italy, and it successfully detected anomalies before they resulted in unplanned downtime. This article was authored by Valerio Francesco Barnabe, Fabrizio Bonacina, Alessandro Corsini, and others.