How intel/predictive-maintenance-pipeline works
PACE is a proof-of-concept predictive-maintenance system for pipeline defects. Its main workflow runs image, video, or sensor inference, stores results in SQLite, then supports agent-assisted analysis and chat. The README also documents visualization, policy filtering, evidence, ticketing, and optional text-to-SQL/LLM modes. Some links below rely on README statements or the source index; unsampled implementations and their internal call paths are not inferred.
Main entry points: inference_runner.py Inference runner · meta_agent.py Meta-agent · app.py Web application · sqlite_client.py SQLite detections
Components 24
Inference & ingestion
- Pipeline runner
run_complete_pipeline.py - Inference runner
src/inference/inference_runner.py - Handler dispatch
src/inference/dispatcher.py - Inference handlers
src/inference/handlers - Visualization
src/inference/visualization.py - Output writer
src/inference/output_writer.py - Database cleanup
scripts/clean_slate.py
Storage & query
- SQLite detections
src/utility/sqlite_client.py - Text-to-SQL executor
src/agents/utility/sql_query_executor.py - Agent database backend
src/agents/utility/database_backend.py
Agent analysis
- Agent graph builder
src/agents/graph_builder.py - Meta-agent
src/agents/meta_agent.py - Policy agent
src/agents/policy_agent.py - Policy filtering
src/agents/utility/policy_filter.py - Analysis agent
src/agents/analysis_agent.py - Evidence agent
src/agents/evidence_agent.py - Corrosion agent
src/agents/corrosion_agent.py - Ticketing agent
src/agents/ticketing_agent.py - LLM client
src/agents/utility/openvino_llm.py - Response cache
src/agents/utility/response_cache.py
User interfaces
- Web application
web_app/app.py - Interactive chat
interactive_chat.py - Chat intent routing
src/utility/chat_intent.py
Other components
- Operator
Connections 26
- Operator → Pipeline runner: starts pipeline
README.md - Pipeline runner → Inference runner: runs inference
run_complete_pipeline.py - Inference runner → Handler dispatch: dispatches handlers
src/inference/inference_runner.py - Handler dispatch → Inference handlers: selects handler
- Inference runner → Output writer: writes outputs
src/inference/inference_runner.py - Inference runner → Visualization: generates visualizations
src/inference/inference_runner.py - Inference runner → Database cleanup: cleans database
src/inference/inference_runner.py - Inference runner → SQLite detections: stores detections
src/inference/inference_runner.py - Output writer → SQLite detections: writes detections
- SQLite detections → Text-to-SQL executor: serves SQL queries
src/agents/utility/sql_query_executor.py - Agent database backend → SQLite detections: uses client
src/agents/utility/database_backend.py - Operator → Web application: uses interface
README.md - Web application → Pipeline runner: launches pipeline
web_app/app.py - Web application → Chat intent routing: routes questions
web_app/app.py - Web application → Ticketing agent: creates tickets
web_app/app.py - Operator → Interactive chat: asks questions
interactive_chat.py - Interactive chat → Chat intent routing: classifies questions
interactive_chat.py - Agent graph builder → Meta-agent: includes coordinator
src/agents/graph_builder.py - Meta-agent → Policy agent: coordinates agents
README.md - Meta-agent → Analysis agent: coordinates agents
README.md - Meta-agent → Evidence agent: coordinates agents
README.md - Meta-agent → Corrosion agent: coordinates agents
- Analysis agent → Policy filtering: filters detections
src/agents/analysis_agent.py - Evidence agent → Policy filtering: filters detections
src/agents/evidence_agent.py - LLM client → Response cache: checks and stores
src/agents/utility/openvino_llm.py - Text-to-SQL executor → LLM client: generates SQL
src/agents/utility/sql_query_executor.py