Prediction the extent of incomplete contact zones in two-dimensional static contact problems using artificial neural network

This research investigates the performance of artificial neural network in solving contact problems, which are important due to their inherent complexity and prevalence in engineering systems. While analytical and numerical methods can solve some contact problems, they often struggle with complicated loading and geometries. This study specifically examines the neural network's capability to determine the extent of incomplete contact zones in symmetric geometries involving single and multiple contacts between frictionless surfaces. To achieve this, a multi-layer perceptron network is implemented using the Levenberg-Marquardt algorithm within MATLAB. The optimal network model for each problem is identified through a trial-and-error approach. Furthermore, the performance of this algorithm across all case studies is evaluated and compared with other methods. The findings demonstrate that the optimal neural network model can predict outcomes with over ninety percent accuracy when compared to analytical, semi-analytical, and numerical solutions. This accuracy is achieved for random inputs within the training data range and for one to three outputs. Such results underscore the potential of neural networks to deliver high precision across various applications.