PNRR is expected to boost Italian wind energy production by repowering old Wind Farms (WF). Unfortunately, this process is being performed by replacing old Wind Turbines (WT) with used WTs from northern Europe. Although replacements improve the power, the WTs remain old products subjected to a greater risk of faults. The project intends to develop a tool for the identification of WT faults, even in real-time, providing an early warning. The methodology is based on monitoring the emitted sound and analyzing it with Machine Learning (ML) techniques. The expected project result is a simple, economical and automated tool that finds its perfect use in old generation WTs, devoid of onboard fault diagnosis control systems. The tool improves maintenance by reducing WT downtime, as well as reducing direct inspections on the machine, which are usually complex and expensive given the remote location of the WTs. To achieve the purposes, the project starts with the classification of the WT failure modes by drafting the Failure Mode and Effect Analysis document followed by the definition of the subset of failures that are expected to influence the emitted sound. These deliverables, together with the definition of the measurement protocols, represents the kick-off of an extended experimental study where multiple noise and weather data measurement campaigns are performed at different WFs in Campania, Sicily and Tuscany, whose owners have already declared interest in collaborating. The first measurements are performed at WTs that have known damages and lead to the characterization of the acoustic footprint that the faults produce. The automatic fault recognition procedure is then developed through ML algorithms applied to the acquired signals. An automatic removal of spurious events from measured sound is developed, allowing to feed the Artificial Neural Network with sound signals with the best signal-to-noise ratio. Measuring the WT immission is notoriously difficult and requires specific techniques. In fact, the immission and residual signals can overlap in time and frequency within the measured noise and they should be separated. At each measurement point the immission level is modeled as a function of blades rotational speed, and the residual as a function of wind speed at ground height following the ISPRA Guidelines for WT noise, co-authored by the PI. This, combined with the operating and weather data provided by the operators, allows the definition of the healthy WT. Further measurement campaigns are performed for the validation phase. Measurements are repeated over the year in order to include the seasonal variation of the wind-induced vegetation noise that could produce false positives if not considered. The environmental sustainability of the solutions proposed in the project are also investigated by means of the Life Cycle Assessment methodology according to ISO 14040.