AGENTS
Software Test Case Generation Agent (ISO 26262 Functional Safety)
Auto-generates ISO 26262 Part 6 software test cases with joint SSR/SAD/SUD multi-document input.
Core Pain Points
- Multiple safety documents need unified understanding: Test inputs include the software SSR, software SAD, software SUD, and test case template; humans must extract constraints across documents.
- Complex authoring of functional-safety cases: ISO 26262 Part 6 has explicit requirements for software testing activities; authoring test conditions, expected results, and coverage relationships is labor-intensive.
- High requirements for deliverable format and traceability: Generated test cases must conform to the project template and be traceable back to requirements and design evidence.
Solution & Technical Implementation
- Structured document parsing: A parsing engine reads the SSR, SAD, SUD, and template, extracting requirements, design constraints, and test fields.
- Workflow + knowledge-base enhancement: Workflow orchestrates the generation steps; a test-knowledge-base RAG retrieves functional-safety testing knowledge.
- LLM generation + template output: The LLM (DeepSeek-V3-670B) combines with private configurations to generate test cases; a template-report generator then outputs the files.
Highlights
- Combined multi-document input: Leverages SSR, SAD, and SUD simultaneously, reducing isolated interpretation.
- Knowledge-base RAG: Incorporates functional-safety testing knowledge into the generation context.
- Template-constrained deliverables: Outputs directly target project test-case files, reducing format rework.
- Traceable generation pipeline: Preserves the provenance relationship among documents, knowledge, and cases.
Business Value
- Automates document understanding and initial case drafting, reducing expert time spent writing cases item by item.
- Template and knowledge-base constraints help unify case structure.
- The parsing engine, workflow, and test knowledge base can be applied across different software modules, vehicle models, and safety-level projects.