Machine Learning Model Towards Evaluating Data gathering methods in Manufacturing and Mechanical Engineering

Abstract

To accurately simulate the behavior of intricate mechanical systems and processes, 
supervised machine learning (ML) models need a large amount of training data. 
Since real-world experiments or suitable simulations are costly, time-consuming, or 
incident-related, it is imperative that sample data be acquired efficiently. Data is 
often gathered in mechanical engineering and production using recognized Design 
of Experiments (DOE) techniques. At the same time, Active Learning (AL), which 
promises to reduce data, is becoming more and more popular in the academic 
community but is seldom used in business. In order to get better predictive 
outcomes with fewer samples on regression problems, we compare the most 
popular data sampling techniques with AL in this work. Regardless of their 
disparate criteria, we provide a unique assessment framework that enables the 
controlled and objective comparison of diverse sampling techniques. By examining 
the sample efficiency, stability, and prediction accuracy of the resultant machine 
learning models, we assess whether AL or DOE approaches should be used for the 
job of data creation using three illustrative use cases (UCs). For engineers and data 
scientists who need very effective data collecting for subsequent ML application, 
this article offers helpful advice