Perovskite lattice constant prediction framework using optimized artificial neural network and fuzzy logic models by metaheuristic algorithms

Perovskites have gained significant attention in recent years due to their unique and versatile material properties. The lattice parameters of the perovskite compounds play a crucial role in the engineering of layers and substrates for heteroepitaxial thin films. As an essential parameter in the cubic perovskite structure, the lattice constant, plays a significant role in the development of materials for specific technological applications and serves as a distinctive identifier of the crystal structure of the material. In the field of materials science, advanced Computational Intelligence (CI)-based techniques have become increasingly important for simulating the relationship between the physicochemical parameters of chemical elements and the physical properties of materials and compounds. Hence, this paper presents efficient techniques based on artificial neural network (ANN) and fuzzy logic to predict the lattice constants of pseudo-cubic and cubic perovskites. The identification of optimized parameters for the ANN and fuzzy logic models is accomplished using innovative metaheuristic algorithms such as, Particle Swarm Optimization (PSO), Invasive Weed Optimization (IWO) and Imperialist Competitive Algorithm (ICA). In the first part, the study assessed, the effectiveness of various metaheuristic algorithms (PSO-IWOICA) in tuning the parameters of the ANN prediction structure in order to get the optimal parameter of the ANN. Whereas in the second part, once we extracted the best optimization algorithm, we combined it with the fuzzy logic technique and then we compared the effectiveness of the two techniques, ANN and Fuzzy logic. On the basis of root mean square error (RMSE), mean absolute error (MAE) and the coefficient of determination (R2), the proposed PSO-ANN and PSO-Fuzzy based models are compared with existing and recent models such as Ubic, Sidey, and Owolabi. The proposed PSO-Fuzzy model performs better than our PSO-ANN model, the Ubic, Sidey, and Owolabi models, with performance improvement of 70.90%, 90.36%, 89.74% 84.46%, respectively on the basis of RMSE. Similarly, it attains performance improvement of 71.26%, 90.31%, 89.58%, and 85.02% on the basis of MAE. Furthermore, the developed PSO-ANN based model outperforms the Ubic, Sidey and Owolabi models with performance improvement of 66.86%, 64.74% and 46.60% respectively, on the basis of RMSE and percentage enhancement of 66.27%, 63.75%, and 47.90% when compared on the basis of MAE. Although the PSO-Fuzzy model has the best performance of all the compared models, the developed PSO-ANN based model possesses the advantage of easy implementation in addition to its moderate performance.