How do you design a feedforward sigmoidal neural network in simulink in order to find a universal approximator

Question

I am trying to create a neural network in Simulink in order to find a universal apporximator for an input into a system of ODEs. I am not sure how to design this at all.

All i have is experimental data and knowledge of some peramiters. The key input in this is Cb, but I have a couple of other, but they are all constants.

I just need general help for creating a sigmoidal neural network in simulink using the default toolbox. If anyone can guide me, thank you in advance.

Neural network design. Cb is a perameter that changes over time with the experimental data. The output is the other variable I am estimating using the experimental data

I tried following a basic structure, by normalising my inputs and then applying random weights to each one. Then I feed them into a sigmoidal activation function I created and then form an output at the end. But what this gives me is just a random zig-zag as an output when I run the experiment in parameter estimator.

I think the main issues are with my inputs and just general design of the neural network. Apologies for not much info, I am very lost.