We propose a stochastic model of fragmentation, developed by taking into account the fragment lifetime as a function of fragment size based on the Gibrat process. If the lifetime is determined as a power function of the fragment size, numerical results indicate that size distributions at different times can be collapsed into a single time-invariant curve by scaling the size by the average fragment size (i.e., the distribution obeys the dynamical scaling law). If lifetime is determined as a logarithmic function of the fragment size, the distribution does not obey the scaling law. The necessary and sufficient condition under the scaling law is obeyed is obtained by a scaling analysis of the master equation.