Software Reliability Modeling with Fault Detection Data when Knowing Fault Severity

Software Reliability Growth Models (SRGMs) are commonly used to estimate the software quality in software engineering. Currently, most SRGMs employ the fault number data collected during software fault detection process and model the fault number data with corresponding detection time. In this process, fault severity is generally used as an unknown parameter to be solved by the modeling process. Few articles incorporate the fault severity as a known factor for the fault-detection-process modeling. In fact, each fault detected is classified into different severities during software testing process in a lot of software testing projects, that is, the fault severity can be treated as a known factor. Generally, the higher severity a fault has, the larger effect it may create. Therefore, besides the total fault count remained in software, the number of remained faults in different severity, especially the faults that may cause serious consequences, is more critical to the system operation. Incorporating the data information, we proposed one novel nonhomogeneous Poisson process software reliability growth model in this article, which involves both the failure time and the severity of each fault into modeling. In this article, we first discussed how to introduce the severity into modeling. In actual software development process, it has been observed that the fault in trivial severity is detected more easily and less influence by the learning effect than the fault in hard severity. Thus, we proposed a Severity Ratio Function (SRF) to describe the percentage of the fault detection rate in same severity out of the total fault detection rate changing in time. Then, based on the SRF, a new software reliability model is derived. Finally, this model are evaluated and validated on actual test data set collected from a nuclear power plant protection system. The results of numerical illustration demonstrate that the proposed MVF provide better estimation and fitting under comparisons.