Software reliability analysis for safety‐critical and control systems

Abstract

The transition from analog to digital safety-critical instrumentation and control (I&C) systems has introduced new challenges for software experts to deliver increased software reliability. Since the 1970s, researchers are continuing to propose software reliability models for reliability estimation of software. However, these approaches rely on the failure history for the assessment of reliability. Due to insufficient failure data, these models fail to predict the reliability of safety critical systems. This paper utilizes the Bayesian update methodology and proposes a framework for the reliability assessment of the safety-critical systems (SCSs). The proposed methodology is validated using experiments performed on real data of 12 safety-critical control systems of nuclear power plants.We believe that SRM for SCS must have some special desirable characteristics:1. There should be acceptable documentation of the method and its applications that can be understood and evaluated. 2. There should be a strong justification of the assumptions made. 3. It should be possible to consider the specific operating conditions of the software. 4. It should be able to address all the uncertainties for better accuracy. 5. There should be sufficiency of model parameters. 6. The models should have proper verification and validation ways.Several models do not have description of intended uses of the method and all assumptions. 6,7 In general, the existing SRM assumes unrealistic assumptions and use failure history to estimate the parameters of the models in order to calculate the reliability. As of our knowledge, we have found very few SRM which consider the specific operating conditions of the software. 8 Software reliability growth model (SRGM) estimates the software reliability with high accuracy, if there is sufficiency of failure data and therefore are not suitable to SCS.Software errors are generally due to unstated, ambiguous, and unrealistic requirements or because of design errors. From this, we believe that if it is possible to formulate any model that can accommodate errors in each phase of software development life cycle (SDLC) and also can propagate them in next phase, prior to the completion of software development, it will give more accurate reliability prediction results. A perfect designed software requirement and specification (SRS) not only depends on the requirements and analysis but also on the expert's opinion, the domain knowledge, and the experience of the SRS designing team, the resources available, and many other constraints. Taking these as the inputs for the requirements phase, a perfect SRS is designed. These inputs may be considered as "safety cases" or "safety arguments" as these are the collection of evidence responsible for the proper development of final product. Evidence for safety cases or safety arguments for each phase exists. Using these evidences, we can create BBN model for each phase of life cycle model and calculate the inferred probability using the safety argument values. A proper conditio...

Feature selection methods for root‐cause analysis among top‐level product attributes

“…Pattern recognition methods can also be used for reliability estimation and failure prediction during software development. Kumar et al employed a Bayesian update methodology to evaluate the reliability of safety‐critical control systems in a nuclear power plant 17 …”

“…Kumar et al employed a Bayesian update methodology to evaluate the reliability of safety-critical control systems in a nuclear power plant. 17 Blumenstock et al introduced interactive decision trees to conduct a stepwise exploration of automotive warranty data and to visualize product attributes relevant to a particular problem. Their interactive approach allowed experts' domain knowledge to be combined with an automated statistical analysis of the data.…”

Modelling reliability growth for multi‐version open source software considering varied testing and debugging factors

“…With the help of NHPP, many critical decisions regarding reliability prediction, resource allocation, cost analysis, testing time etc. can be determined 1,2,35–36,38,39,41,43 . To improve software quality and achieve prosperity in the software industry, software industries are switching from closed source data to open‐source data.…”

“…can be determined. 1,2,[35][36]38,39,41,43 To improve software quality and achieve prosperity in the software industry, software industries are switching from closed source data to open-source data. Few SRGMs have been proposed for OSS which is the expanding field of software development.…”