ForgeShield is a research-oriented AI framework designed to integrate predictive maintenance, anomaly detection, explainable AI, and Generative AI into a unified industrial safety-intelligence pipeline. The system combines supervised machine-learning models for failure prediction with unsupervised anomaly detection, predictive-health modelling, and Remaining Useful Life (RUL) analysis. Using the AI4I 2020 Predictive Maintenance Dataset, the framework evaluates models including Gradient Boosting, Random Forest, XGBoost, SVM, and other machine-learning approaches, while an Isolation Forest model provides a continuous anomaly signal.
These outputs are combined through a unified risk score defined as Risk = 0.60 × Failure Probability + 0.40 × Anomaly Score, producing LOW, MEDIUM, HIGH, and CRITICAL risk levels. The framework further applies SHAP-based Explainable AI to identify the contribution of important operational variables such as tool wear, rotational speed, torque, mechanical power, and temperature differential. A separate temporal modelling pipeline using NASA C-MAPSS data explores LSTM, GRU, CNN-LSTM, and LSTM Autoencoder approaches for machine-health and RUL analysis.Beyond prediction, ForgeShield investigates how AI can connect machine intelligence with evidence-based safety reasoning. Its Retrieval-Augmented Generation (RAG) layer uses embeddings and ChromaDB to retrieve relevant incident records and safety procedures, which are then provided to a locally hosted Qwen3:8B model through Ollama. The system generates structured incident intelligence covering abnormal behaviour, possible contributing factors, hypotheses, corrective actions, preventive actions, and supporting evidence.
An evidence-aware Safety Copilot further separates facts, possible hypotheses, recommended actions, evidence sources, and limitations, helping reduce unsupported LLM-generated claims. The project integrates these capabilities into a Streamlit-based ForgeShield Command Center covering live telemetry simulation, machine intelligence, predictive health, anomaly detection, explainability, incident intelligence, and safety assistance. As a research Proof of Concept, ForgeShield does not replace certified industrial safety systems or provide direct machine control; instead, it demonstrates a technical architecture for moving from isolated AI predictions toward explainable, evidence-grounded industrial safety intelligence.Project created by Aashi.
Project Github Repo : https://github.com/aashi2709/ForgeShield
Please check the project and explore the technical implementation. For any questions, doubts, or further information, feel free to write to info@macroedtech.com.
