@JoergS5 @o_lampe
Thank you greatly for your advice! I have since replaced what I thought was a faulty harmonic gearbox, the backlash in that joint is now greatly reduced. For a university dissertation I had measured the stall torque and repeatability of the harmonic gearboxes, i found stall torque to be around 4.2Nm and precision to be around 0.28 arcmin. A MATLAB simulation model of my parallel SCARA revealed I only needed around 1.1Nm of torque to operate with the speeds and accelerations needed for thin steal.
I am still having some strange behaviours occurring with the machine, circles are still not fully circular, and occasionally joint positions are not achieved in a smooth manner. I am wondering if it's a calibration issue, however I shall now investigate the behaviour of bearings incase they're causing issues. I had previously tried using some sprung lanyards to apply tension on bearings and joints to ensure backlash and engagement is consistent. I will try and post some more pictures of the machines behaviours this evening.
Many thanks!