If you don't know the severity (for example, the amount each muscle contracts when it's stretched), then it seems like you could end up with a problem that cannot be solved. An analogy would be knowing the displacement of a rope in a tug-of-war but not knowing how many people are pulling each end. How are you going to ensure the problem has a unique solution?
We were thinking about making an iteration between the real subject movement and forces(we are using a robot to get this data) and the model in OpenSim, so we can try different spasticity gains in the software in order to get the one that better fits the real subject values. It's similar to the Ashworth scale technique that physiotherapists use in their diagnosis:
https://scireproject.com/outcome-measur ... fb7a3-f550
This solution may not be appropriate for your study, but it depends on how the problem is posed, how you have modified the muscle model, etc.
You will need to build either a model or a solver that takes this into account.
That's exactly why I'm building a new model class based on Thelen2003Muscle, so we'll be able to modify the excitations the spastic muscle is generating. Furthermore, the project is meant to work only with patients with spasticity, so we won't need to differentiate between healthy subjects and those with spasticity. And this is why I'm looking for a method in Thelen2003Muscle or relate classes that will be called during a CMC and can modify these excitations for each state. By the way, as I've seen in previous researches, this excitation is proportional to muscle fiber velocities.
I'm really sorry if I didn't explain it clear in the first time.