Innovative human-centric AI solution: AI solver - design engineering ecosystem

• a deep learning framework that learns the response of a system
• data-driven learning technique from physics-based simulation

Use case: Flow Dynamic based on Fujitsu AI-Solver technique.

Leveraging an Intel python environment including MKL as well as TensorFlow, the AI solver works on the principle that a neural network can be trained to produce results that are close to those derived from direct numerical simulation. AI solver uses a data-driven technique that learns from physics-based simulations to instantly predict the principal field distribution within a 3D space. With training, results are accurate enough for data-driven methods to become viable in the design engineering ecosystem.

Non-destructive testing solution - 3D volumetric analysis

• faster inspection and higher throughput
• coverage of new designs through unsupervised learning
• consistency and traceability

Use case: Quality control in composite manufacturing - accelerating sign-off for complex aviation parts.

Non-destructive testing is performed on all carbon composite aviation parts before shipment. AI models locate and characterize discrete and non-discrete defects from raw pulse-echo ultrasound data using the original 1D signal (A-scan) in a 3D context.

Through-transmission interpretation delivers a complete Assisted Defect Recognition (ADR) solution that links quality assurance to the production process.

3D quality control solution - finding defects with 3D scanning

• 100% defect-free manufacturing
• 100% damage and fatigue found during maintenance
• production cycle adaptation

Use case: Improving the non-destructive testing of large single carbon aircraft structures.

Single carbon composite aircraft parts manufactured today are very large (±30m x ± 10m) and made as one piece with highly complex structures – shape, attachments and cutouts. Ordinarily, quality control of these parts requires 24 hours of ultrasound scanning and 18 man-hours of analysis on 2D image projections. Supervised 2D CNNs are impossible to generalize as defects are rare. With Fujitsu’s unsupervised 3D deep learning methodology, 100% coverage is achieved with complete scalability and higher precision from training on the entire wing data.

2D image analysis solution - automating quality assurance

• 100% coverage for repair and recovery
• faster automated quality checks to lower scanning costs
• lower risk and zero wastage

Use case: Automating quality assurance for wind turbine blades.

Composite wind turbine blades fabrication is highly manual with common defects. All blades must be scanned with ultrasound to locate defects for repair which is a long and variable process. Working with a leading blade manufacturer, we have developed and deployed an industrialized AI solution using pre-trained deep learning algorithms for 2D image analysis that delivers direct efficiency gains and process improvements.