One factor that contributes to successful long-term human-human interaction is that humans are able to appropriately express their emotions depending on situations. Unlike humans, robots often lack diversity in facial expressions and gestures and long-term human robot interaction (HRI) has consequently not been very successful thus far. In this paper, we propose a novel method to generate diverse and more realistic robot facial expressions to help long-term HRI. First, nine basic dynamics for robot facial expressions are determined based on the dynamics of human facial expressions and principles of animation in order to generate natural and diverse expression changes in a facial robot for identical emotions. In the second stage, facial actions are added to express more realistic expressions such as sniffling or wailing loudly corresponding to sadness, laughing aloud or smiling corresponding to happiness, etc. To evaluate the effectiveness of our approach, we compared the facial expressions of the developed robot with and without use of the proposed method. The results of the survey showed that the proposed method can help robots generate more realistic and diverse facial expressions.