Algorithms based on neural network for segmentation of defects on metal sheet images

A manufacturing system is governed by its various processes upon which its efficiency is dependent. Since failure results in considerable losses, many manufacturing systems have certain redundancies for some processes. These redundancies cause the system to work under different efficiency states called multi-state elements. In this paper, various processes of metal sheet manufacturing unit have been categorized as subsystems to determine the multi-state probabilities of its different efficiency states. Artificial Neural Network Technique (ANN) has been used to estimate the change in these multi-state probabilities over time. The ANN has also been used to estimate variation in upstate and downstate probabilities of the system for a particular-time period. The results have been used to determine variation in profit over time for the system.

Download Full-text

Parameter selection by an artificial neural network for a laser bending process

Proceedings of the Institution of Mechanical Engineers Part B Journal of Engineering Manufacture ◽

10.1243/095440502320783350 ◽

2002 ◽

Vol 216 (11) ◽

pp. 1517-1520 ◽

Cited By ~ 15

Author(s):

G Casalino ◽

A D Ludovico

Keyword(s):

Neural Network ◽

Model Simulation ◽

Numerical Models ◽

Back Propagation ◽

Metal Sheet ◽

Analytical Models ◽

Processing Unit ◽

Feed Forward Neural Network ◽

Laser Bending ◽

Central Processing

Based on thermally induced plastic deformations produced by laser irradiation, metal sheet laser bending can be a valid alternative to dies for rapid prototyping and manufacturing. Some numerical models have been built in order to improve the understanding and prediction of mechanisms. Drawbacks entailed with those models have been found. Finite element model simulation has proved to be time and CPU (central processing unit) memory consuming. The analytical models have been cumbersome and unsatisfactory. Nowadays, it is possible to build a neural network model for process modelling directly from data collected during the experiments. In this paper a feed-forward neural network with a back-propagation learning function has been designed and its performances have been evaluated for metal sheet laser bending. This technique has proved to be effective and efficient, providing the process parameters that are necessary to achieve a desired bending angle.

Download Full-text

Improved knowledge-based neural network (KBNN) model for predicting spring-back angles in metal sheet bending

International Journal of Modeling Simulation and Scientific Computing ◽

10.1142/s1793962313500268 ◽

2014 ◽

Vol 05 (02) ◽

pp. 1350026 ◽

Cited By ~ 6

Author(s):

Tinh Quoc Bui ◽

Anh Viet Tran ◽

Abid Ali Shah

Keyword(s):

Neural Network ◽

Empirical Model ◽

Metal Sheet ◽

Spring Back ◽

Element Analysis ◽

Knowledge Based ◽

Surface Method ◽

Sheet Bending ◽

Artificial Neural Network Ann ◽

The Relationship

We develop an efficiently improved knowledge-based neural network (KBNN) associated with optimization algorithms and finite element analysis (FEA) to accurately predict spring-back angles in metal sheet bending. The well-known V and U prevalent processes of bending are considered. The KBNN predictive results are based on the empirical model and artificial neural network (ANN) modeling. The empirical model is constructed from the FEA results using response surface method, while the multilayer perceptron is employed to create the ANN. The trained KBNN can accurately model the relationship between the spring-back angles and process parameters. The obtained results are validated against other existing methods showing a high accuracy.

Download Full-text