I suppose this is due to your boundary conditions as some of your elements are not restrained but i cannot comment more as i am not sure what is the problem? and did you apply any surface boundary conditions?

This frequently happens in Abaqus explicit for several reasons

1. when you have a failure model. Due to the reduction in stiffness, the force on a single node can cause a huge acceleration

2. When you have strongly localized forces.

If you have a failure model and the elements have already failed, then this is nothing to worry about since these elements are not transmitting any forces anymore.

I can think of several possible remedies for this:

1) if you are using load-controlled boundary condition, probably changing to displacement-controlled would be better. I believe the experimentation was performed with displacement-controlled conditions.

2) Using element deletion for the elements that have undergone failure.

3) One of the heavily distorted element seems to be directly above your cylindrical roller support (on the right side of the figure). I assume you applied contact between the roller and the sample, and your roller is not round enough but with pointy edges, which may lead to severe surface penetration during contact modeling and convergence difficulties. Never save efforts on mesh discretization when it comes to correctly representing your geometries.

4) If you are using Abaqus/Standard, try switching to Abaqus/Explicit. It will be slower since you need to make sure your time increment/mass scaling/loading speed are sufficiently low, but the analysis usually go further than the implicit analysis.

Hope it helps.

Shiyao