An automated tool for efficiently generating a massive number of random test cases

Certain software systems, such as process control and avionics systems, require an extremely large amount of testing to measure their reliability according to the ultra-high reliability requirements imposed on them. Therefore, it is essential to speed up the test generation process to reduce the certification time. We develop techniques to enhance the effectiveness of an automated program for generating random test data. Our approach consists of three major phases, viz. (1) a predicate decomposition phase, (2) a test data generation program creation phase and (3) a random data generation phase. During the phase 1, the predicate is decomposed into independent subpredicates resulting in a partition of the input variables. Test data can then be independently generated for every subset of variables subject to satisfying the corresponding subpredicate. During phase 2, the source code of the test data generation program is created according to the results of predicate decomposition of phase 1. The actual test data points are generated in phase 3 by compiling and running the program generated in phase 2. A preliminary performance evaluation is presented