Adding visual search to our 3D printing proof-of-concept

Being able to search using images, not text, is very powerful when you’re searching for a specific part or product. If your text search does not exactly match the keywords or tags, you won’t find what you’re looking for. That’s where visual search can help.

Take the example of a vintage car owner who needs an obsolete spare part, which he requests by sending the OEM a photo of the part. The OEM would like to honor the customer’s request and 3D print the part but needs a way to search the digital warehouse of part models and has only the customer’s photo of the part.

To address this problem, we have added a visual search feature to our 3D printing procurement system proof-of-concept.

How visual search works

The OEM uploads the customer’s photo into the 3D printing procurement system (as shown in the figure below) and executes a visual search.

Uploading an image before executing a visual search. Click image to expand.


The search will return the results that best match the photo, showing the most accurate matches first (as shown in the figure below).

Executing a visual search for a part, which returns the best match first. Click image to expand.


The visual search uses TensorFlow, an open source search library. We needed to train it with the images in our digital warehouse so it could match an uploaded image against the warehouse images. We found TensorFlow’s accuracy to be quite good when compared to another open source computer vision library.

We have included the ability to share important information like whether the part has been 3D printed in the past, and the product ID and latest version number. This will help in handling requests for obsolete parts.

Next we are working on leveraging blockchain technology for the 3D printing proof of concept. We can use blockchain technology to have better traceability of transactions and to enforce security aspects like content hashing and restrictions on 3D printing a model, such as for a specified time frame (e.g., you have 24 hours to print the model) or based on the quantity to be printed or both.

ArunKumar Amarnath is a technologist in DXC Labs India, and is currently focused on 3D printing. Based in Bengaluru, his experience includes solution development, Agile, DevOps consulting and system implementation. @arun_amarnath



Baskar Venugopalan, associate director at DXC Labs India, is a technology enthusiast currently focused on 3D printing. As a lead technologist, Baskar has worked in the cloud and big data areas, with emphasis on technology adoption. He has led incubators, as well as teams of senior technologists and architects in the past. Based in Bengaluru, Baskar’s three decades of experience include working with multinationals like Microsoft and Oracle. @venugbx



  1. […] machine learning in identifying the 3D printability of the component(s), leveraging scanning and vision technologies in quality control, and using scanning technologies for model creation.  We will be working on […]

  2. […] we last talked about our 3D printing procurement system proof-of-concept, we promised to report back on how we are using blockchain features for access […]

Speak Your Mind


This site uses Akismet to reduce spam. Learn how your comment data is processed.