In modern engineering, structures are characterized by the rapid increase in their complexity due to demanding criteria related to durability and overall efficiency, while maintaining low costs. Therefore, the application of appropriate materials with optimal properties has an enormous impact on structural integrity and the extension of the lifespan of structures. As known, engineering structures are subjected to various forms of loading. Numerous degradation mechanisms occur in the materials, compromising their integrity and shortening their lifespan. Numerical simulations are widely used as an alternative to expensive experiments. Unfortunately, numerical methodologies based on traditional constitutive models often struggle to capture governing microscale behaviours, necessitating the development of novel modelling techniques. Multiscale methods provide a numerical framework for analysing material behaviour across multiple scales. However, applying multiscale analysis to practical problems results in high computational costs. Due to advances in computing power, machine learning has recently emerged as a complementary approach, by collecting and processing large amounts of data to determine both known and unknown physical laws. Fracture is one of the most commonly encountered failure modes in modern engineering structures. Accurate prediction of damage accumulation is therefore a key factor in the design of reliable products. The Phase-Field fracture formulation has proven itself capable of achieving this with its ability to simulate complicated fracture processes, including crack initiation, propagation, merging, and branching, without the need for additional ad-hoc criteria. In this research, the aim is to develop a methodology for modelling the constitutive behaviour of heterogeneous materials based on machine learning techniques. The methodology developed in the project relies on the machine learning algorithms for modelling heterogeneous materials and assessing the integrity of structures based on phase-field method. All numerical algorithms will be verified through comparison with standardized test samples and real structural components.