Avionics Requirements Validation AI for Completeness Checking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual review of software requirements in avionics systems is prone to oversights and inconsistencies, especially in large and complex projects, leading to potential safety hazards and financial burdens due to missed or incomplete requirements.
Innovation Solution
A generative AI-assisted system that integrates a large language model fine-tuned with regulatory guidelines and historical data to automatically analyze software requirements, providing comprehensive and accurate validation, including compliance checks and predictive analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual review of requirements is used, then domain expertise and contextual understanding are provided, but oversights and errors increase in large and complex projects
Solution Approach 1:
The requirement review process is segmented into multiple validation layers: automated AI-based initial screening, followed by targeted manual review of high-risk items. The system divides the review checklist into specific validation criteria (completeness, consistency, traceability, safety) and applies AI validation to each segment independently, improving overall reliability while maintaining human expertise for complex judgments.
Solution Approach 2:
An AI-based validation engine serves as an intermediary between requirement authors and human reviewers. The system automatically validates requirements against regulatory guidelines (DO-178C) and historical defect data, generating structured feedback that guides human reviewers to focus on high-risk areas, thereby improving both accuracy and efficiency.
2Measurement precision
If manual review is performed thoroughly, then validation completeness improves, but review time and cost increase
Solution Approach 1:
The system performs preliminary automated validation of all requirements against the review checklist and regulatory guidelines before human review begins. This preliminary action identifies complete and potentially problematic requirements in advance, allowing human reviewers to focus only on items that require their expertise, thereby maintaining validation completeness while reducing overall review time.
Solution Approach 2:
The AI validation engine provides continuous feedback to reviewers by highlighting requirements that fail specific validation criteria and suggesting corrections. This feedback mechanism ensures that reviewers can quickly identify and address completeness issues without re-reading entire requirement documents, significantly reducing review duration while maintaining thorough validation.
3Productivity
If AI validation is introduced, then review efficiency and consistency improve, but system complexity increases
Solution Approach 1:
The AI validation engine is designed as a multi-functional system that performs multiple validation tasks simultaneously: checking completeness against the review checklist, verifying consistency with regulatory guidelines (DO-178C), detecting traceability issues, and identifying historical defect patterns. This universal approach improves review efficiency across multiple dimensions while consolidating complexity into a single integrated system rather than multiple separate tools.
4Reliability
If comprehensive validation is performed, then error detection improves, but development cycle time increases
Solution Approach 1:
The AI validation engine operates continuously throughout the software development lifecycle, providing real-time validation feedback as requirements are created and modified. This continuous validation detects errors immediately rather than requiring separate validation phases, enabling early error detection that prevents costly rework later and ultimately reduces overall development cycle time while maintaining high reliability.
Data Source
AI summary
A system comprises a set of databases in communication with a processor, and a requirements validation AI engine hosted by the processor. The databases comprise software requirements, regulatory guidelines, historical requirements defects, requirements review checklist, and prompts. The AI engine communicates with the databases, and a fine-tuned LLM communicates with the AI engine. A user interface communicates with the AI engine, and a configuration management repository communicates with the user interface. The user interface sends new requirements, from the configuration management repository, with corresponding check instructions to the AI engine, which selects prompts for review, based on types of new requirements and the check instructions. The AI engine sends the prompts and the new requirements to the LLM to review for any defects in the new requirements. The LLM sends a response to the user interface, to report on validity of and any defects in the new requirements.