Advanced & Sustainable Manufacturing

The Advanced & Sustainable Manufacturing (ASM) division integrates I2R’s capabilities in Sensory, Connectivity, Cybersecurity, Artificial Intelligence, Robotics, Data Analytics & Computer Vision R&D to provide Industry 4.0 testbed capabilities for Manufacturing, Engineering and Aerospace MRO industries to enhance productivity and improve operational efficiency. We co-develop with our industry collaborators on cutting-edge Industry 4.0 technology solutions including real-time process optimization & control, next generation logistics and supply chain, large-scale automated inspection and predictive maintenance. 


Our Smart Maintenance and Inspection (SMI) unit focus on the use of AI to automate industry processes such as asset identification with computer vision, visual inspection, and AI decision-making for workflow planning. 

Most deep learning methods are difficult to deploy in real world applications as they require a lot of training data. During deployment, AI expertise and GPU resource are also required to re-train neural networks when new or more data becomes available. To overcome above problems, our team focus on using low resource deep learning to enable easy and efficient AI solution deployment. 

Our capabilities include:

  • Meta-learning based few-shot incremental learning for image classification;
    Image segmentation for object detection and measurement using few training examples;

  • Anomaly detection in images using one or few defective images for neural network training;

  • Semi-supervised progressive learning to minimize the efforts for data annotation and improve accuracy continuously;

  • 3D defect detection and segmentation using 3D deep learning.

    In the age of Industry 4.0, our Digital Factory (DF) unit work with industry collaborators to help them embark on digital transformation for their factory and manufacturing process. We help our collaborators evaluate their current manufacturing practices and identify areas for improvement using customised Industry 4.0 technologies. We provide a one-stop solution that transfer cutting-edge research outcomes from advanced machine learning, data analytics, sensory, connectivity, and cybersecurity to specifically help each of our collaborators tackles their own important challenges.

    Our key strength lies in our leading use-inspired academic research, and our ability to translate academic research into real-world problem-solving technologies. With this approach,  we have worked with various multinational and local collaborators in the aerospace, semiconductor, precision engineering and other sectors to solve their specific challenges that couldn’t be easily solved using off-the-shelf commercial solutions including Yield Optimization, Predictive Maintenance, Advanced Anomaly Detection and Model Compression  Many of our solutions, such as the Predictive Maintenance solution developed for our aerospace collaborator, are first of its kind in the industry.   


    Our Smart Automated Aircraft Visual Inspection System (SAAVIS) programme investigates the use of automation for the visual inspect of large structure, such as an aircraft. It is estimated that 80% of all inspections for aircraft are carried out visually (Source: FAA). Currently, in MRO, trained aircraft inspectors are required to climb onto boom lifts/high platforms to inspect the top surface; this raises safety concerns for the worker.As a human’s attention and field-of-view is limited, defects may also be missed at times. The severity of some defects is also subject to human biasness.

    Automating the inspection process through cameras and artificial intelligence (AI) can improve outcome by reducing human errors and biases. The programme investigates modes of imaging large structures through moving camera systems carried by autonomous robots, drones, mobile devices or PTZ cameras, localization and AI-power defect detection capabilities. AI is employed to automatically detect and classify the different types of defects (e.g. missing parts, cracks, dents, unsecured panel etc.) to automatically identify the repairs required, while working alongside human inspection. A proof-of-concept prototype has been developed. Trials are ongoing in actual locations to scale up this technology to operational levels.