## TL;DR
AI is the brain of modern manufacturing — predicting equipment failures before they happen, detecting microscopic product defects at production line speed, and simulating entire factory operations in real-time digital twins. Industry 4.0 plus AI ("Industry 5.0") reduces downtime by 30-45%, defect rates by 35%, and energy consumption by 25%.

## Core Explanation
Manufacturing AI stack: (1) Data acquisition — IoT sensors (vibration, temperature, acoustic, current draw), cameras (visible, IR, hyperspectral), PLC/SCADA logs, quality inspection records; (2) Predictive maintenance — anomaly detection on sensor streams (autoencoders learning normal operation → flag deviations), remaining useful life (RUL) estimation via LSTMs/Transformers predicting time-to-failure from degradation patterns; (3) Quality control — computer vision defect detection (CNNs, Vision Transformers) running at 60+ fps on production lines, classifying surface defects, dimensional deviations, and assembly errors; (4) Digital twins — real-time virtual replica of factory/line/machine synchronized with physical sensors, enabling what-if simulation; (5) Process optimization — reinforcement learning for optimal machine parameters (feed rate, temperature, pressure) balancing quality, throughput, and energy.

## Detailed Analysis
Predictive maintenance evolution: Level 1 (reactive — fix when broken) → Level 2 (preventive — scheduled maintenance) → Level 3 (predictive — AI forecasts failures). Modern PdM systems (Springer 2025 literature review) combine: vibration FFT analysis (bearing degradation), acoustic monitoring (air leaks, grinding), motor current signature analysis (MCSA for rotor bar faults), and thermography (hot spots). Federated learning across factory sites preserves IP while improving models. Digital twin manufacturing (Frontiers 2025): the AI-DT system continuously compares physical sensor data against digital twin predictions — discrepancies trigger anomaly alerts. Generative AI creates synthetic failure scenarios to stress-test manufacturing lines that rarely experience downtime. Augmented reality overlays digital twin data onto physical equipment for maintenance technicians. Industry adoption: Siemens (MindSphere, closed-loop digital twins), GE (Predix for aircraft engines), Fanuc (Zero Down Time for CNC machines), TSMC (AI process control reducing wafer defects). Key challenges: (1) Data scarcity for rare failure modes — one factory may never experience certain failures, making supervised learning impossible; synthetic data and transfer learning partially address this; (2) Legacy equipment without IoT sensors — retrofitting is expensive; (3) The "last mile" — AI recommendations must be actionable for operators on the factory floor, requiring intuitive UIs and trust-building.

## Further Reading
- Siemens MindSphere & Xcelerator Digital Twin
- Google Cloud Visual Inspection AI
- Augury: Machine Health AI (PdM)