Page 30 - eBook_Proceedings of the International Conference on Digital Manufacturing V2

Proceedings of the International Conference on Digital Manufacturing –
Volume 2

For case (d), a thin surface bottle divided into 100 sections was
imported to Ansys from SolidWorks and unstructured tetrahedral
mesh was generated. A grid independency test was executed at
initial thickness of 1.17 mm, and same element size was then used
to conduct thickness variation analysis through parametric study.

Methodology for Machine Learning

In this section, the application of ML is integrated to predict the
mechanical behaviour of HDPE lubricant bottle under the stacking
condition. This study focuses on two structural parameters’ i.e.
bottle’s deformation and von Mises stress resulting from
variations in bottle thickness across 100 points.
Therefore, Regression, a supervised machine learning technique
is utilised because of label input and output data. By employing
parametric study in finite element analysis, a dataset was extracted
which consists of 100 input features and two output features, as
detailed in Table 3, exploring the influence of thickness variation
on the overall weight of bottle.

As illustrated previously in Figure 2, the workflow began with
importing the dataset into Jupiter Notebook, followed by (80:20)
split into training and testing dataset, as previously done by
Demircioğlu, Bakır & Çakır (2024). Based on prior studies in the
field of non-linear finite element analysis (Nath, Ankit, Neog &
Gautam, 2024; Hussain, Sakhaei & Shafiee, 2024; Pranckevičius
& Marcinkevičius, 2017; Kadiyala & Kumar, 2018), different ML
algorithms, such as support vector regressor (SVR), bagging
regression, decision tree, forest tree, and boost gradient, were
individually used to develop the hypothesis using the identical
training dataset. The test set was then used to predict the output
features and evaluate model accuracy by comparing the predicted
and actual values using standards performance parameters.
Models’ performance was assessed using Mean Square Error
(MSE) and R-square (R2) score. MSE indicates how far the
model’s predictions are from the true values. Lower MSE means
better predictions. R2 score is a statistical error that measures how
well a model fits. The value of R2 score varies between 0 and 1;
however, the closer the value of it to 1, the better the model fits

14