Column generation is often used to solve problems involving set partitioning constraints, such as vehicle routing and crew scheduling problems. When these constraints are in large numbers and the columns have on average more than eight to twelve nonzero elements, column generation often becomes inefficient because solving the master problem requires very long solution times at each iteration due to high degeneracy. To overcome this difficulty, we introduce a dynamic constraint aggregation method that reduces the number of set partitioning constraints in the master problem by aggregating some of them according to an equivalence relation. To guarantee optimality, this equivalence relation is updated dynamically throughout the solution process. Tests on the linear relaxation of the simultaneous vehicle and crew scheduling problem in urban mass transit show that this method significantly reduces the size of the master problem, degeneracy and solution times, especially for larger problems. In fact, for an instance involving 1600 set partitioning constraints, the master problem solution time is reduced by a factor of 8.