Automatic identification of mechanical parts for robotic disassembly using the PointNet deep neural network
by Senjing Zheng; Feiying Lan; Luca Baronti; Duc Truong Pham; Marco Castellani
International Journal of Manufacturing Research (IJMR), Vol. 17, No. 1, 2022

Abstract: Identification is the first step towards the manipulation of parts for robotic disassembly and remanufacturing. PointNet is a recently developed deep neural network capable of identifying objects from 3D scenes (point clouds) irrespective of their position and orientation. PointNet was used to recognise 12 instances of components of turbochargers for automotive engines. These instances included different mechanical parts, as well as different models of the same part. Point clouds of partial views of the parts were created from CAD models using a purpose-developed depth-camera simulator, reproducing various levels of sensor imprecision. Experimental evidence indicated PointNet can be consistently trained to recognise with accuracy the objects. In the presence of sensor imprecision, the accuracy in the recall phase can be increased adding stochastic error to the training examples. Training 12 independent classifiers, one for each part, did not yield significant improvements in accuracy compared to using one classifier for all the parts. [Submitted 13 September 2019; Accepted 27 March 2020]

Online publication date: Mon, 21-Mar-2022

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Manufacturing Research (IJMR):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com