I have tried to change the time-step method from automatic into user specified and also change the length scale from conservative to aggressive but both of them give me the diverged solution. In my case, I use automatic time step with timescale factor 0.1 and it can be run but the problem is the residuals keeps fluctuating even though I have run the simulation for almost 2 weeks in 175k iterations.

Do you have a better recommendation according to my problem? I would be happy to hear that and try to implement it directly to see whether it's suitable or not 

Thank you,

Best Regards,