Abstract: Methods, systems, and assemblies for detecting faults in an electrolyser having a plurality of electrolysis cells are described. The method comprises obtaining voltage measurements of the electrolysis cells during operation of the electrolyser, generating synthetic cell voltages for the electrolysis cells using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters, comparing the voltage measurements to the synthetic cell voltages for corresponding ones of the electrolysis cells to obtain voltage differences, and detecting a fault in the electrolyser when at least one of the voltage differences reaches a threshold.