A machine learning-based inverse analysis procedure for concrete softening law prediction using non-experimental datasets

This paper studies the mechanical behaviour of concrete as one of the most widely used quasi-brittle construction materials emphasizing on the importance of knowing its mechanical parameters and their evolution during the inelastic stage, i.e., the softening law. The softening curve, which describes the response of the material under damage or cracking, is critical for predicting the behaviour of concrete structures subjected to extreme loads. Experimental tests are commonly employed to obtain this information either directly or indirectly. Some of the indirect methods are based on inverse analysis and/or artificial intelligence techniques, both of which capable of predicting the mechanical parameters of concrete from the experimental results of one test, e.g., a notched beam subjected to vertical loads. However, an important drawback of these procedures is that they require a large dataset constructed from data gathered in multiple experiments in order to be developed. Consequently, most existing methods are tailored to specific types of experiments and even limited to certain specimen dimensions. Additionally, these procedures primarily focus on predicting mechanical parameters rather than determining the softening law. To address these limitations, this paper proposes a machine learning-based algorithm for the inverse analysis of an experimental test capable of predicting both the softening law and the mechanical parameters of concrete. By generating a non-experimental dataset through the Sequentially Linear Analysis (SLA) procedure, the proposed algorithm can be applied to other experimental setups suitable for analysis with SLA. The results of the application example demonstrate the effectiveness of the proposed approach.