Intelligent Inspection Framework: 12 Week POC

Synthetic data helps train AI models to make real-world decisions.

Inspecting complex manufacturing or facilities equipment is often expensive and challenging, which means these processes need to become more intelligent to maximize the use of the equipment while ensuring employee safety. Artificial intelligence, computer vision, and drones or cameras can quickly identify issues and take action to boost facility revenue, reliability, and utilization. But a lack of visuals for defect detection or other critical scenarios makes it hard to develop reliable AI models. For manufacturing and production facilities, the need for inspection, critical scenario detection, health and safety, emission monitoring, and quality control is now. Capgemini has built in partnership with Microsoft a computer vision framework solution that will help to inspect assets in factories and refineries using AI ML computer vision-enabled techniques.

Synthetic data from simulations Capgemini’s computer-vision framework can receive visual data from drones or other devices to detect defects or scenarios. It applies AI models that are trained for this purpose. When there is a lack of visual information available, synthetic data is generated using metaverse simulations for AI model training; this delivers a wide range of scenarios and data to drive machine learning and better AI models.

3D models are generated using simple visual or asset specifications, with computer graphics to simulate realistic use of operating assets and different potential defects over time. Simulations can generate thousands of images for multiple scenarios per minute to train the computer-vision model.

Computer-vision framework The plug-and-play framework receives real-world visuals, applies scenario detection using computer vision AI models, and sends the results to systems to manage the work order or inspection.

The registry in the framework can host and manage its own custom or third-party pre-built AI models in one place. The framework organizes training and detection visuals in an easy-to-use structure and allows traceability of results. And users see the results visualized in an intuitive manner. 3D models are generated using simple visual or asset specifications, with computer graphics to simulate realistic use of operating assets and different potential defects over time. Simulations can generate thousands of images for multiple scenarios per minute to train the computer-vision model. Capgemini’s solution delivers: • Synthetic data to train AI models when real-world visuals are unavailable • Proactive asset maintenance to allow operators to act on potential issues, preventing costly unscheduled downtime • Regulatory compliance, so operators meet all commitments • Scalability with the flexibility to extend beyond the visual spectrum and to incorporate sensor data from assets and the environment • Quality assurance and improved product quality with visual AI • Improved worker safety, with visual AI and drone inspection handling hazardous environments. The cost-effective, plug-and-play framework on the cloud provides the flexibility of adding models from new or existing applications and minimizing integration points in the IT landscape.

Capgemini’s strong industry capabilities combined with AI, computer graphics, and metaverse experience deliver a flexible and scalable framework. Now manufacturers and facilities can create a safer and more productive environment with Capgemini’s computer-vision framework delivering the future they want.

1. The Proof of Value Approach and Business Model
2. Our Proof of Value approach will validate that AI/ML can be deployed to perform automated inspections with long term goal of eliminating the need for manual analysis. The high-level objective of the project was to develop a generic framework that receives inspection requests for checking safety and working condition of assets, receive visual image/video data, perform automated inspection using computer vision (CV) artificial intelligence (AI) models to detect defects, and send the results back to the inspection request system to create workorders based on the identified defect(s).
3. In approximately 12 weeks the Proof of Value/Pilot will: a. Provide discovery analytical services in support of imagery and video provided by the client for specific list of anomalies defined by the client via Azure Machine Learning and Artificial Intelligence algorithms developed by Capgemini. b. Detect other anomalies that are not regularly detected visually c. Considering the short duration and complexity of visual analytics, the report should explain how the POC/ Pilot model demonstrated the feasibility of detecting the anomalies from visuals.
4. Components and Azure Services used include: ADF, Azure ML, ADF Azure Function, WebApp, Blob Storage, Synapse, ADLS (Azure Data Lake Storage Gen2)