Post by Mennatullah Khaled

DEPI-series [GenAI] - Capstone Project | PetroMind 🛢️ PetroMind: Predictive maintenance system for industrial facilities. Industrial facilities rely on reactive or schedule-based maintenance; leading to: -> unplanned downtime -> excessive costs -> loss of operational knowledge Engineers lack a unified system combining real-time sensor monitoring, intelligent failure prediction, and instant access to technical documentation. PetroMind combines: • SCADA/PLC sensor ingestion • LSTM failure prediction (F1 ~0.98) -> predicts binary failure risk from 30-cycle sensor windows • RUL estimation (RMSE ~15.6) -> estimates remaining useful life in cycles • Hybrid RAG layer over TM 5-692-1 -> dense (BGE) + sparse (TF) retrieval over TM 5-692-1 with cross-encoder reranking • ML-to-RAG Bridge -> auto-translates predictions into maintenance queries • Multi-Agent System -> 19 tools, served via FastAPI Tech Stack: PyTorch · NASA C-MAPSS & N-CMAPSS · HDF5 · FastAPI · Pinecone · Groq Llama 3.3 70B · SentenceTransformers Huge appreciation for my teammates for all their hardwork :> - Eng. Engy Abdelmalak - Eng. Christine Zakaria - Eng. Karim Zaki - Eng. Mostafa Mohamed - Eng. Mohammed Ragab And our supervisors: - Eng. Ahmed Hekal - Dr. George Samuel, MSc. Deeply grateful for Eng. Lobna Hamdy Basmsm for the great insightful discussion of our project; she gave us such a great feedback and mentored us towards helpful tasks for our future work💖 Attached down-below a demo for the implemented system functionalities: • Chatbot • Agent trace • Live monitoring #DEPI #PredictiveMaintenance #MachineLearning #LSTM #RAG #PyTorch #IndustrialAI #GenAI

Video Content