Evaluation of Hydrogen Embrittlement Effect on Mechanical Strength of Modern Martensitic High-Strength Steels - Experimental Analysis and Artificial Neural Network-based Prediction

Hydrogen embrittlement (HE) is a challenge affecting several engineering materials, with a significant impact on steels, targeting critical components in high-value industrial applications. To better understand the HE phenomenon and its impact on the mechanical performance of newly developed martensitic steels, this dissertation focuses on the following three key issues: (i) enhancement of experimental conditions for the accurate measurement of hydrogen concentration (CH) in metalic materials, (ii) assessment of the mechanical strength of steels under different hydrogenation and mechanical loading conditions, and (iii) a new artificial neural network (ANN)-based tool, applied initially to investigate the relationship between the hydrogen thermal desorption spectroscopy (TDS) and steel's susceptibility to H-induced damage, is further developed to predict the relation between CH and mechanical strength at failure, with reduced requisite experimental testing. A specimen cooling system was developed and integrated with the air-lock in a TDS apparatus with the objective of enhancing the accuracy of CH measurements in the studied steels. The application of the new specimen cooling system, enabling temperatures down to 213˚K, proved to provide a significant improvement in CH measurement, allowing more accurate analysis of obtained thermal desorption spectra, valuable for H-trapping analysis. The H-uptake and mechanical performance of four martensitic steel grades with varying prior austenite grain (PAG) morphology were studied by monotonic uniaxial tensile testing and TDS in different H-charging conditions. The results demonstrate that the H-uptake under the same Hcharging conditions, without external mechanical loading, does not vary between the steels. However, a variation in H-uptake was evident with external mechanical loading. This variation does not directly depend on, or correlate with the different PAG morphology. Furthermore, the results showed that CH in the steel has a greater influence on the H-enhanced creep rates leading to failure, than the magnitude of the applied loads, during constant load mechanical testing conditions. The developed dense ANN tools, encompassing three different models, were evaluated for the prediction of steel's susceptibility to HE and CH responsible for the H-induced failure with an approximate accuracy of about 98%. Enhancing the ANN models with training data may yield a powerful tool for the HE characterization of steels. This can be done with a reduced dependency on time-consuming experimental testing with hazardous and complex apparatus.