Page 21 - eBook_Proceedings of the International Conference on Digital Manufacturing V2

Collaborative Study on Weight Optimisation of Lubricant Bottles under
Stacking Condition using Finite Element Analysis and Machine Learning

(DiRaddo & Garcia‐Rejon (1994); Yousefi, Collins, Chang &
DiRaddo (2007)). The profile configuration of desired product is
divided into several points, depending on the specification of
parison thickness controlling system. Each point corresponds to
the required parison thickness and the value of desired parison
wall thickness is maintained through the movement of angular die
gap between the extruder die and the core of inner mandrel (Saito,
Akimoto, Oizumi & Ogura, 1995). Thus, the cultivation of parison
controller programming for the manufacturing of HDPE
containers can substantially improve the uniformity of bottle
thickness and help in the reduction of material consumption.

The process of structural optimisation of complex materials
like rubber and high-density polyethylene through FEA modelling
and experimental technique is considered as computationally
expensive and time-consuming Nguyen, Huynh, Nguyen,
Nguyen & Truong (2021). Therefore, the scope of Artificial
Intelligence, especially machine learning and deep learning, is
considered as the most effective tools to predict mechanical
properties, the discoveries of advanced material compositions or
structures, and the transformation of computational approach
while maintaining precision (Guo, Yang, Yu & Buehler, 2021).
For example, Anush, Yashwanth, Shashank, Reddy & Kumar
(2021) addressed the common challenges faced by bottle
industries related to error in filling, caused by mechanical
inaccuracies or human intervention and implementation of
Multilayer Perceptron (MLP) classifier, an advanced algorithm of
Artificial Neural Network to classify the bottles into correctly
filled, underfilled or overfilled. Demircioğlu, Bakır & Çakır
(2024) evaluated several ML algorithms to predict the placement
of optimal mass location based on cutout positions and natural
frequencies of perforated cantilever beam. Liang, Liu, Martin &
Sun (2018) introduced deep neural network (DNN) model to
predict the thoracic aortic wall stress distribution by training the
model with a large dataset (729 aorta geometries as input). The
trained DL model generated stress distribution in under 1 second,
compared to 30 minutes, taken by Abaqus FEA software. De
Iuliis, Miceli & Castaldo (2024) utilised 20 different ML models
to identify a set of highly predictive seismic parameter for

5