This study explores the shift from digital to analog computation using electronic circuits and investigates the use of surrogate machine learning models (ML) for designing analog circuits for numerical computations. The study focuses on the challenges of designing analog computers for the use case of Artificial Neural Networks (ANNs). Designing electronic circuits for specific mathematical models and ensuring stable and convergent parametrization of the model is complicated by Analog circuits which depends on environmental parameters like temperature and electronic component variations, which must be considered during the parametrization process. The study evaluates two methods for designing Analog ANN (AANN): simulation-in-the-loop ML training, which includes analog simulation of the target circuit in the training process and the loss function, and surrogate-models derived from simulation using digital ANN models, which provide much lower computational times. The integration of analog models in the training process increases the robustness of the final circuit against circuit and environmental variations and non-linearity of the analog components. A simple SHM example demonstrates the approach.