Not necessarily. You only need bounds if you think your model is overfitting the data (e.g. due to not enough data points and/or imperfect system excitation) and you know some solutions are wrong due to prior knowledge about the true system.
But before starting to add bounds, make sure you have done everything else correct! Including:
- removing normal operating points from your data
- choosing how the initial condition should be handled (e.g. zero or estimate)
- comparing the estimated model output with the data and seeing if it matches what you are expecting
- ...etc.
