Modeling and optimization of ultrasonic cleaning process for hard disk drive arm using support vector regression

Abstract

The ultrasonic cleaning has been widely adopted in hard disk drive manufacturing industry due to its
effectiveness in removing contamination from the final products, sub-assemblies and components. This process is
suitable for delicate product with high complexity. The process efficiency depends on six factors including
ultrasonic frequency, type of liquid medium, time, temperature, power, and finally a number parts in cleaning
basket. These factors, if not carefully set, might result in failure in contamination removal indicated by high
level of liquid particle count (LPC), damages of hard disk drive and ultimately shorten product’s durability.
The ultrasonic cleaning process setting is usually determined by experience of operators, which might not
always result in optimum condition. Therefore, this study presents an application of the integration between
Taguchi Method and Computational Intelligence (CI) techniques, increasingly applied for modeling and
optimizing the performance of manufacturing industry, to identify the optimum setting of the cleaning process
parameters focusing on quality improvement of the finished hard disk drive arm that can deliver clean surface
of the product with no damage. The proposed method is as follows. Firstly, both Support Vector Regression
(SVR) and Artificial Neural Network (ANN) were trained with experimental data to model LPC level of the
cleaning process. The model with highest accuracy was selected to be a Suitable Computational Intelligence
Model (SCIM). Then, a Grid search was opted to the SCIM to find the optimum settings. Data from real
experiments of nigh hawk 1H hard disk drive arm were used to demonstrate the proposed method. The
experimental results suggested that SVR technique is capable of high accuracy modeling and results in much
smaller error and learning time in comparison with ANN.