I have an airfoil (S3010, 9% thickness) that gives the resulting Cp distribution at 2 degrees and Re=500k.
I suspect that the abrupt increase in pressure before the suction peak is a separation bubble. I've read somewhere that this behaviour is common for thin airfoils.
How can I modify the airfoil to prevent leading edge separation? The end goal is to increase the L/D
The velocity-jump at $\simeq 0.5c$ is due to a laminar separation bubble.
But you ask two different things here:
A leading edge separation may happen as the laminar separation bubble progresses forward for higher angles of attack and is more common in thin airfoils. A way to suppress this is by the use of roughness elements, a trip or vortex generators. You can experiment with this using Xfoil and fixing the transition location at the point (or in a distance before this point) that you assumed that you have a roughness element for example. Otherwise try modifying the geometry with Xfoil's inverse design routines and if you're seriously into this read these Low Reynolds Number Airfoil Design Lecture Notes provided by Selig.
Increasing $L/D$ for a specified angle of attack for these kind of airfoils and for the low-Reynolds regime, means in general to reduce the velocity-jump of the laminar separation bubble. If I am correct, Selig's airfoils are already an improved version of similar NACA airfoils for low-Re. To do this you should again consult the aforementioned notes.
Are you sure that the spike on $c_p$ plot before the suction peak is not a paneling issue? It seems unphysical. Have you tried refining your airfoil? I do not capture this with the same analysis using Xfoil with 200 panels.
A rule of thumb is that the shape parameter ($H_k$) of the boundary layer at a laminar separation bubble is around 4. You can see from the following plot that you have a laminar separation bubble around $0.5c$ on the upper surface. Not in the leading edge though.
This site lists the S3010 as an airfoil for low Reynolds numbers. This book is by Michael Selig, the person that the S3010 airfoil is named after. The book contains $C_P$ data at different Reynolds numbers, the highest being 300k. So it Looks like the OP measurement is at a Reynolds number that was higher than the airfoil was designed for, leading to turbulent boundary layer behaviour at the nose and early separation.
What is apparent from the profile drawing, is the relatively sharp nose when compared to NACA profiles of the same thickness. The obvious remedy would be to increase nose radius, round out the nose - but that would remove a design feature that all the low Reynolds number airfoils in the book have. The other option is to use an airfoil that is designed for higher Reynolds numbers.
The NACA 6-series profile listed in this question uses a leading edge radius of 1.561% c.
