This paper presents a comparative study of various neural network architectures for modeling the behavior of complex mechanical systems in the context of digital twins. The research is based on experimental bearing data provided by the Center for Intelligent Maintenance Systems (IMS) at the University of Cincinnati. Architectures such as convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory (LSTM), and transformers are examined and compared. The performance evaluation of these models was conducted based on criteria including bearing condition prediction accuracy, training time, computational complexity, and applicability for real-time systems. The research results show that transformer-based models demonstrate high accuracy, particularly for long-term forecasting when sufficient computational resources are available. CNNs offer significant computational efficiency with competitive accuracy, while standard RNNs provide a baseline with lower performance.