Legacy Textual Requirement Software System Modeling and Test Generation

The cost of poor quality software to the US economy is estimated to be in the trillions. In safety-critical software, like that used in commercial airliners, the cost of software verification can exceed 75% of the total software development cost. Automated test tools can help to accelerate test development to "shift-left" testing to be run earlier in the software life cycle. While the cost of qualifying such automated tools is very high, independent human oversight can replace the need for qualification. Many of these safety-critical systems still rely on difficult to parse text-based requirements. To address these challenges, this paper presents research into parsing textual requirements to enable automated analysis of the quality of the requirements and generate tests. The research begins with parsing of structured requirements into logical models, enabling an investigation of the benefits of system analysis and test generation. The requirement analysis ranges from analysis of signal propagation and requirement conflicts preventing output values from being set or inputs being masked, to full systems containing thousands of requirements. Test generation includes methods for creating full Multiple Condition/Decision Coverage tests. Next, Machine Learning approaches are investigated to broaden the requirement parsing for better handling of natural language requirements. Comparisons of the automated requirement analysis and test generation with manual approaches demonstrate a 45x speedup of system analysis time and suggest potential savings that could exceed tens of thousands of hours in Verification and Validation time.