Skip to main content



Fujitsu Advances Smart Manufacturing Design with 3D CAD Model Search Solution

Fujitsu EMEIA

News facts: 
  • Fujitsu Laboratories of Europe has developed a new machine vision technology based on deep learning that enables fast searching and classification of CAD models in mechanical design
  • The solution significantly speeds up the identification and location of both models and components in large component databases, improving design quality and aiding creativity
  • Fujitsu’s new smart design technology advance is making its world-first appearance at the Fujitsu Innovation Gathering and Fujitsu World Tour event in Paris on June 9
London/Paris, June 09, 2016
Fujitsu Laboratories of Europe has today unveiled a new technology for smart design and manufacturing at the Fujitsu Innovation Gathering 2016, involving a new method for computing the similarity of 3D models. The new 3D model search solution forms part of Fujitsu’s research and development into machine learning-based solutions for smart design applications, and achieves over 96 percent accuracy for automatically locating elements in large component databases. The solution uses the power of deep learning to accelerate the process of mechanical design, enabling mechanical designers rapidly and efficiently to locate existing models, their components and associated manufacturing data – a process that currently accounts for more than 50 percent of the design process.

Developed to improve design productivity and quality, Fujitsu’s technology can benefit organizations in multiple ways. One example is in product data management, where large 3D model repositories can be rapidly labeled and augmented with meta-data. Model clustering tools based on Fujitsu’s 3D model search technology can reduce the time of labeling large collections of 3D models from days and even weeks to a few hours. The 3D model search solution can also significantly speed up the design process when integrated into a designer’s development environment. For example, it enables reviewers directly to locate elements of interest such as clips, bosses, ribs etc rather than having to search manually, enabling much faster design reviews. It can also assist with the rapid identification of design issues related to particular shapes before a prototype is made.

This new technology forms one component of Fujitsu’s machine learning-based solutions for smart design, combining its extensive existing library of proprietary mechanical design data with a variety of other technologies under development. Fujitsu Laboratories of Europe’s Chief Executive Officer, Dr Tsuneo Nakata, expands on the solution’s future potential : “The ultimate research and development objective is to create a design system in the future that automatically adds missing semantic information to 3D models, with complex product design rule checks capable of being defined and enforced. However, the applications of our new 3D model search technology are by no means limited to the field of manufacturing. It is an exciting development that has considerable potential – for example in healthcare and medicine, or any application where there is a need to compare 3D models extremely accurately, identify what they represent and ascertain whether they are normal or anomalous.“

The Technology Behind the 3D Model Search Solution At the heart of Fujitsu’s technology is a deep neural network that is trained on a combination of 2D image data, more than 1 million images from the ImageNet database , and 3D model data from Fujitsu’s own manufacturing designs – the latter being key to achieving high accuracy on 3D Computer Aided Design (CAD) models. The deep network is used to extract feature vectors from 3D models which, together with model dimensions, form a powerful shape descriptor that is independent of the orientation of the 3D model.

Taking only a few seconds, the search process generates a descriptor for the search item, which is then compared against tens of thousands of models in the database. The search then outputs an ordered list of similar models (see Figure 1). Current accuracy of this new method is over 96 percent for several classes of CAD models, with ongoing work to add more classes.

Figure 1: The first 10 results of a search for an LCD cover model. Results are ordered by similarity, with the first model being the most similar and the last one being the least similar.

Online resources

For information on related advances, plese see:

About Fujitsu

Fujitsu is the leading Japanese information and communication technology (ICT) company, offering a full range of technology products, solutions, and services. Approximately 156,000 Fujitsu people support customers in more than 100 countries. We use our experience and the power of ICT to shape the future of society with our customers. Fujitsu Limited (TSE: 6702) reported consolidated revenues of 4.7 trillion yen (US$41 billion) for the fiscal year ended March 31, 2016. For more information, please see

About Fujitsu Laboratories of Europe

Established in 2001 and with an active presence in Europe since 1990, Fujitsu Laboratories of Europe Limited represents Fujitsu Laboratories across EMEIA, focusing on regional initiatives that reflect the diverse mix of countries and ideologies. Fujitsu Laboratories of Europe is focused on the creation of cutting-edge solutions that benefit society, adopting a co-creation strategy and working with customers, collaboration partners and society as a whole to pioneer a new generation of user-centric applications and services underpinned by creative information analytics. As one of Fujitsu’s global centers of excellence for AI, its work encompasses security, social innovation, manufacturing, AI ethics, and high performance computing applications. For more information, please see

Georgina Garrett


Phone: Phone: +44 1903 854900
E-mail: E-mail:
Company:Garrett Axford Ltd (on behalf of Fujitsu Laboratories of Europe Ltd)

All other company or product names mentioned herein are trademarks or registered trademarks of their respective owners. Information provided in this press release is accurate at time of publication and is subject to change without advance notice.

Date: 09 June, 2016
City: London/Paris