AI-Assisted TPS Development and Sustainment

Abstract:  Test Program Set (TPS) development is both difficult and time consuming.  It is also error prone.  The combination of these factors increases both the calendar time and the cost to produce a good working TPS.  Artificial intelligence (AI) is typically data driven but because TPSs vary so greatly, it is difficult to obtain sufficient data points for training AI. Almost every unit under test (UUT) has its own unique test program and depending on large volumes of data to train the TPS development process may be futile.  Also, test development is a function of the way the UUT is described. In this project the product descriptions will be based on SysML models.  The extent to which these models are detailed will influence AI’s success, but presently little or no such rules exist.  In this project we will study the performance of AI in developing tests for various SysML descriptions.  We will also use various AI generators and vary the way the problem is presented to them. Our results will be tabulated, analyzed and conclusions reamed from them.