On goodness-of-fit tests for accelerated life models