Learning the manufacturing capabilities of machining and finishing processes using a deep neural network model

In this work, we present a deep neural network model to automatically learn the capabilities of discrete manufacturing processes such as machining and finishing from design and manufacturing data. By concatenating a 3D Convolutional Neural Network (3D CNN) with a simple Multilayer Perceptron (MLP),...

Full description

Saved in:
Bibliographic Details
Published in:Journal of intelligent manufacturing 2024-04, Vol.35 (4), p.1845-1865
Main Authors: Zhao, Changxuan, Melkote, Shreyes N.
Format: Article
Language:English
Subjects:
Artificial neural networks
Business and Management
Computer aided operations
Control
Decision trees
Finishing
Machine learning
Machines
Machining
Manufacturing
Manufacturing industry
Material properties
Mechatronics
Multilayer perceptrons
Neural networks
Process planning
Process selection
Processes
Production
Robotics
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this work, we present a deep neural network model to automatically learn the capabilities of discrete manufacturing processes such as machining and finishing from design and manufacturing data. By concatenating a 3D Convolutional Neural Network (3D CNN) with a simple Multilayer Perceptron (MLP), we show that the model can learn the capabilities of a manufacturing process described in terms of the part features and quality it can generate, and the materials it can process. Specifically, the proposed method takes the part feature geometry, material properties, and quality information contained in a part design as inputs and trains the deep neural network model to predict the manufacturing process label as output. We present an example implementation of the proposed method using a synthesized dataset to illustrate automatic manufacturing process selection. The performance of the proposed model is compared with the performance of interpretable data-driven classification methods such as decision trees and random forests. By comparing the performance with different combinations of input information to be included during training, it is evident that part quality information is necessary for characterizing the capabilities of finishing processes while material information further improves the model’s ability to discriminate between the different process capabilities. The superior prediction accuracy of the proposed deep neural network model demonstrates its potential for use in future data-driven Computer Aided Process Planning (CAPP) systems.
ISSN:0956-5515
1572-8145
DOI:10.1007/s10845-023-02134-z