This thesis proposes a neural network-based approach for gas leakage detection in accumulators of offshore wind turbines, aiming to improve maintenance strategies and minimize downtime. The research begins with the development and validation of an accurate accumulator model using data that captures various operational scenarios. The model is validated using experimental data from the Hydraulics Laboratory at AAU. Subsequently, a Fully Convolutional Network (FCN) model is developed for gas leakage detection. It is designed to classify a group of input signals and determine the corresponding gas pre-charge pressure. The FCN model is trained using time series data and evaluated for its performance. The study explores the impact of input variables, sliding window size, hyperparameters, and sensor utilization on the performance of the neural network. Experimental results show that incorporating oil pressure, along with oil and ambient temperature signals in the neural network model achieves an accuracy of 95% when classifying the pre-charge pressure. Adding thermocouples to the accumulator's surface significantly enhances the neural network performance, reaching 100% accuracy.