There is at least one bug: to detect the Pareto front you have to use exact optimisation tools. GA is not going to provide you with guaranteed optimal solutions, hence the "front" that you receive is very likely not the front.

A good way to check the result is to plot all the feasible points that you have found to get a sense of the shape of the Pareto region. Then you can check if your code is able to find something that looks like the convex hull of your points.

Thanks Emil, I am trying to do similar things.

Thanks Michael, Can I request you to describe your idea ? Are you meaning what Emil recommends?

What I suggest is to skip GA, and instead utilise the best tools from mathematical optimization in order to at least approximate the Pareto front. No metaheuristic will be able to do that, but maybe perhaps through the use of scalarisation - or a combination thereof. I have not worked much on that myself, but I can bet that it will be better than GA. Perhaps explore (very) local moves could lead you to a better place.

I wanna compare between NSGA II VS weighted sum GA. The NSGA II is quite good but the problem with GA as there have not guaranteed front in a convex shape. something to do there as other literature have already done with weighted sum GA. I try to follow results to obtain convex shape.

You're certainly going about it in the right way. I would recommend varying weight values more, however. Try substituting w1>1 and see how the figure turns out from there. You need to exhaustively experiment with the changing weight factors until an acceptable Pareto front is achieved.

Another way is to get an approximate shape of the Pareto front with the NSAGA II method you're comparing with and perhaps direct your weight values accordingly to match it. My advice is to always use Pareto-based techniques, no matter what.