Models for Estimation of Software Faults and Failures in Inspection and Test

Abstract: In software engineering, the quality of software and in particular the software reliability has to be controlled and improved. Hence there is a need for measuring and controlling the number and appearance of software faults and failures. This thesis presents models for estimation of residual faults after inspections and a model for usage specifications which enables estimation of operational failure behaviour and software reliability already in test. The fault estimation models are based on the capture-recapture technique. A filter approach and an experience-based approach are developed and evaluated in two experiments. The experimental results show that the new estimation models give less absolute error and more stable estimates than established models. However, replication of the experiments is needed to get sufficient confidence in the results. The failure estimation, or software reliability estimation models require test cases to be representative of the future usage of the system in order to make estimates in test. The future usage is specified in a usage specification. The state hierarchy usage specification model is developed to be suitable for modelling of the usage of real-time multi-user systems. The model is applied in early phases in the life cycle to dynamic analysis, as well as to test of components. Furthermore the implementation of the usage specification in an executable language for automatic generation of test cases is studied. The model is evaluated in minor case studies in different application domains, which have provided promising results. The models contribute to estimation of software faults and failures, hence providing means for control of software quality attributes.