The power can be easily estimated using simulations. The example bellow was based on: http://egap.org/content/power-analysis-simulations-r#Power-analysis-for-multiple-treatments

I will do it using just the Shapiro-Wilk test -- but the logic is straightforward and you can apply it to whatever test you want.

The power of a test varies according to two things

1 - The sample size

2 - The significance level you adopt

I'll keep the significance level constant at 0.05 and will vary just the sample size.

#### CODE ####

#This is a vector of sample sizes we will use:

possible.ns

what do you need the test for? In case you want to test your data is normal to apply a staistical test, better go for simple QQ-plots. They will tell you fairly well if the distribution of the data fits your chosen model. Especially with small sample sizes (< 30) the tests can be misleading if you are only guided by their significance. QQ-plots and visual exploration tells you far more about your data.

@Christian Michel ..QQ plot is not enough for testing normality ,numeric tests should be applied besides graphs to confirm the answers...