Industrial components made with fiber-reinforced composites undergo changes in material properties during the manufacturing process. Characterizing these constituent material properties poses many challenges in modelling microstructural variation in fiber distribution, fiber orientation, volume content and complex topology. Complimenting experiments with virtual material characterization helps save development time. The virtual workflows using traditional FFT homogenization techniques for scale bridging are slow. So using reduced order methods to replace traditional FFT simulations for scale bridging serves as an established approach for realizing multiscale simulations. When these models are used for predictions outside of the training data set, the accuracy of the prediction tends to be low. Moreover these models need to be re-identified in case the material model is changed . In recent years, Deep Material Networks (DMNs) have shown promise in providing a robust workflow to deal with a large number of different material models. Most importantly, the key idea of learning a complex microstructure with these material networks using linear elastic homogenization as a training data isolates the computational bottleneck encountered at the microscale. Once these networks are identified using machine learning methods, the network topology is considered fixed, and an inelastic analysis is enabled via the DMN. Thus during inference the same material model and balance equations are solved for, ensuring thermodynamic consistency of the predictions. The state-of-the-art (SotA) work proposes novel laminate blocks to ensure faster model identification and a two step sampling strategy for the training data. The current work starts with the implementation of the SotA , applies it to fiber reinforced composites, discussing challenges in industrial scaling of these models. It also aims to discuss applied industry grade case studies to complement experiments with virtual material characterization in the context of high cycle fatigue.