Industry 4.0 enabled electrochemical process lines and data analysis for quality control

Abstract

Online machine learning or stream learning is a machine learning algorithm that updates its model with each new data instead of the standard batch learning that need the whole dataset at the time of training. One particular trait of the stream learning algorithm is its robustness against concept drift, which is the change in the input and output relationship within the dataset. In this paper, multiple stream learning algorithms are compared against batch learning algorithms across various datasets. Among these datasets is the parameter optimization of black nickel coating process to test the applicability of stream learning algorithm in electrochemical process. The result of the comparison has shown that batch learning performs better on dataset with minimal concept drifts, but the stream learning algorithm outperform it when the dataset contains significant concept drift. When applied on the black nickel coating dataset, the stream learning algorithm does not perform well due to the small amount of data points. The number of data significantly affect the performance of stream learning algorithm, thus when the data collection process is slow and the expected collected data does not amount to much, a batch learning algorithm is preferred.