Doctorant IA

CIFRE PhD – Multimodal Foundation Models for 3D Defect Perception in Industrial Inspection

Location: Orsay, FR

Position Summary

Westinghouse is seeking a PhD candidate to develop the next generation of Artificial Intelligence technologies for industrial inspection and defect detection in the nuclear industry. The objective of the project is to create AI models capable of combining multiple inspection data sources to build a coherent three‑dimensional perception of inspected components and support defect identification, characterization and operator decision‑making. The research focuses on explainable AI, multimodal data fusion and industrial deployment in safety‑critical environments.

Research Challenge

Industrial inspections generate large quantities of heterogeneous information. Visual inspection data, inspection history, operator observations and advanced non‑destructive examination techniques each provide only a partial view of the inspected component. The challenge is therefore not to reconstruct a component in 3D, but to enable an AI system to develop a meaningful three‑dimensional understanding of the inspection context and potential defects by combining complementary sources of information. This approach aims to provide operators with richer and more actionable information while maintaining transparency, traceability and explainability.

Main Objectives

• Multimodal Data Fusion: Develop methods to combine heterogeneous inspection data; Create unified representations from complementary information sources; Investigate relationships between different inspection modalities.
• AI‑Based 3D Perception: Develop models capable of understanding spatial relationships within inspected components; Provide a three‑dimensional interpretation of indications and defect signatures; Improve defect detection and characterization capabilities.
• Explainable and Physics‑Informed AI: Design interpretable models suitable for industrial and nuclear applications; Associate AI outputs with physically meaningful features; Develop confidence and uncertainty indicators to support expert decision‑making.

Stage 1 – Visual Testing (VT)

• Creation of labeled datasets.
• Development of AI architectures and fusion methodologies.
• Validation of representation‑learning approaches on visual inspection data.

Stage 2 – Advanced Ultrasonic Testing (UT)

• Transfer and adaptation of developed models.
• Integration of FMC, TFM and advanced UT information.
• Fusion of VT and UT information within a common representation framework.
• Development of enhanced defect characterization capabilities.

Candidate Profile

We are looking for candidates passionate about Artificial Intelligence applied to complex industrial challenges.

Preferred Backgrounds

• Artificial Intelligence
• Machine Learning
• Computer Vision
• Data Science
• Applied Mathematics
• Signal Processing & Computer Engineering

Experience With

• Deep Learning
• Foundation Models
• Representation Learning
• Autoencoders
• Generative Models
• Multimodal AI
• 3D Computer Vision
• NDE (Non‑Destructive Examination) would be considered a strong advantage.

Benefits

• Attractive remuneration
• Great benefits for your convenience and safety:
  • Peer‑to‑peer recognition program.
  • Learning and development opportunities.
  • Employee Assistance Program: confidential counseling and resources for employees, eligible dependents, and household members.
• We encourage our employees to participate in community service events and other team‑building activities to strengthen their bonds and inspire each other.

Westinghouse is an Equal Opportunity Employer including Veterans and Individuals with Disabilities.