In industry Statistical Process Control (SPC) techniques are typically used to predict whether a process is out of control or not. More recently, Machine Learning, in particular, Artificial Neural Networks, has been a great enabler of data-driven techniques for improving quality control in industrial applications, particularly for predicting whether control limits will be violated in advance so that appropriate mitigation actions can be taken.
However, our recent projects with industrialists demonstrated that there is a lack of understanding on how AI-based techniques could be applied. Given a time series data and a monitoring goal, it is not clear how to formulate machine learning problem. Should it be a classification problem, clustering, or prediction of mean/variation? There are also issues relating to the quality and completeness of the dataset which needs to be addressed. Datasets tend to be small, imbalanced, and have discontinuities.
In this project, we aim to tackle these issues by creating novel machine learning frameworks that can be used for quality prediction, and the optimisation of quality control parameters.
Funded by: Pitch In, Research England
Collaborators: Federal-Mogul, University of Sheffield
Researchers: Bang Xiang Yong