A Practical Approach for Reliability Prediction of Safety Critical Software Using Multi-Model Ensemble Techniques

This paper introduces a novel and comprehensive approach for estimating the reliability of safety critical software components in autonomous vehicle motion systems. The proposed approach in this paper presents a combination software reliability model (CSRM) that integrates multiple non-homogeneous Poisson process (NHPP) software reliability growth models (SRGMs) to achieve a reasonable compromise between accuracy, the trade-off between the goodness of fit and the simplicity, and stability. By using machine learning techniques, the CSRM effectively combines the strengths of individual SRGMs while mitigating their weaknesses through suitable evaluation and calibration techniques. The developed CSRM has been successfully applied and validated to facilitate a smoother and more efficient evaluation of reliability targets for software components in autonomous robotic wheelchair (ARW). Based on the validation results, the new CSRM has significantly enhanced the efficiency of the process for evaluating whether the defined reliability goals were being achieved. Furthermore, has facilitated a more accurate assessment of the need for further test executions and better planning of the required verification and validation session. This new approach provides valuable insights into the reliability of the developed software, particularly for software developers lacking extensive experience in identifying and applying appropriate SGRM.

I. Introduction

To mitigate software system malfunctions and prevent the violation of the defined functional safety goals for the intended functionality, it is crucial to ensure reliability throughout all development life cycle phases of software systems. To achieve this target, the intended software reliability level shall be reached. In this context an adequate planning and execution of the verification and validation measures hold a central role.