APPLICATION AND REFINEMENTS OF THE REPS THEORY FOR SAFETY CRITICAL SOFTWARE

With the replacement of old analog control systems with software-based digital control systems, there is an urgent need for developing a method to quantitatively and accurately assess the reliability of safety critical software systems. This research focuses on proposing a systematic software metric-based reliability prediction method. The method starts with the measurement of a metric. Measurement results are then either directly linked to software defects through inspections and peer reviews or indirectly linked to software defects through empirical software engineering models. Three types of defect characteristics can be obtained, namely, 1) the number of defects remaining, 2) the number and the exact location of the defects found, and 3) the number and the exact location of defects found in an earlier version. Three models, Musa's exponential model, the PIE model and a mixed Musa-PIE model, are then used to link each of the three categories of defect characteristics with reliability respectively. In addition, the use of the PIE model requires mapping defects identified to an Extended Finite State Machine (EFSM) model. A procedure that can assist in the construction of the EFSM model and increase its repeatability is also provided. This metric-based software reliability prediction method is then applied to a safety-critical software used in the nuclear industry using eleven software metrics. Reliability prediction results are compared with the real reliability assessed by using operational failure data. Experiences and lessons learned from the application are discussed. Based on the results and findings, four software metrics are recommended. This dissertation then focuses on one of the four recommended metrics, Test Coverage. A reliability prediction model based on Test Coverage is discussed in detail and this model is further refined to be able to take into consideration more realistic conditions, such as imperfect debugging and the use of multiple testing phases