Context
- Goal of Numerical Simulation and Machine Learning : Predict the behaviour/performance of a system/product/process
- Data can be rather expensive
- Specific challenge of small but smart datasets
→ Frugal learning - Combine simulation and ML towards hybrid modeling approaches & Digital Twins
- Vast amount of domain prior knowledge is available
→ Multi-source data
What's the Objective ?
- Develop predicitive models for physical fields
- Develop hybrid modeling methods for more reliable and explainable models → Guide critical decisions and better anticipate
- For manufacturing and process industries
What's at stake ?
- AM has revolutionized the manufacturing sector - Freedom of design & potential of (nearly) zero-waste
- Not yet widespread because numerous and complex parameters to be monitored and controlled → Part defects are still a major issue
- AI could help to better understand and master the AM process and tend towards in-process control, certified quality assurance and real-time closed-feedback loop
What are the challenges ?
- Complex and still young process
- Improve the quality and reduce the cost of the simulations
- Enhance industrial process mastering to avoid systematic post-production control
- Take better decisions in engineering processes and reduce production cycles
→ Towards Process Quality Control & First Time Right
How to use AI to better master and understand manufacturing processes by jointly exploiting numerical simulations, experimental data and additional prior knowledge ?
What are the possible solutions ?
- "Real-time" physical fields prediction
- Merge heterogeneous data (experimental data, simulated data of different levels of fidelity)
- Hybridize expert knowledge with ML via Informed Neural Networks
→ Better results and more physical solutions with less data - Use explainable AI to expose the correlations between process, properties and performances
- Combine reinforcement learning with human feedback for process control
→ Synergy / close cooperation needed between computer science, material science and process know-how
Key AI topics
- Physical fields prediction
- Physics-informed Neural Network
- Fusion of data
- Frugal Learning
- Reinforcement learning