Intelligent digital twin to make robot learn the assembly process by Deep Learning
Loading...
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UMT, Lahore
Abstract
The fundamental objective of this thesis is to make an intelligent digital twin through
deep learning for operational support of a human-robot assembly station. Digital
twin, as a virtual portrayal referring to a product or system, is used for the purpose of
designing, simulating and optimizing the complexity of the system under observation.
For testing purposes, convolutional neural network (CNN) are integrated with a digital
twin. It is used for application of a collaborative robot for an assembly application.
Collaborative robots are a new form of industrial robots that are safe for humans and
can work alongside humans and have received ample attraction in the recent past years
for automation of simple to complex tasks.
Artificial intelligence can enable us to couple the effects of machine learning and data
analytics to establish a digital twin covering the whole of the lifecycle of a
manufacturing system; and probe, sense and respond to its behavior. One solution that
can come out of this is an adaptive behavior of robot when interacting with humans and
intelligently generating robot program to perform a task. However, in a factory
environment there can be a large variety of components that robot may need to handle.
For the experimentation, it is exemplified with LEGO elements.
The data is collected from Kaggle, where data in the imagery structure is presented. A
corresponding model is developed for the purpose of classification of LEGO
construction elements and attain a recognition accuracy of at least 90 percent.
Significant computing power is required in the case of LEGO elements to achieve a
satisfactory proficient of distinguishing in a neural network. An optimal solution is
achieved by restraining the distinguishable classes up to 16.
Then in Tecnomatix Process Simulate software, which is a simulation software, a digital
twin of the collaborative robot was generated and was programed for assembly tasks
after classification model. The data synchronization with a digital twin and an
automated generation of robot program is the core developments of this thesis.
This thesis aims to:
ix
a. Develop a digital twin (DT) of a physical robotic assembly station
b. Generate robot actions in the digital twin to assemble LEGO elements
c. Develop a deep learning algorithm (DLA) to learn LEGO elements
d. Synchronize DLA with the DT to automatically generate assembly sequence