Developing Machine Learning Methods for Industrial Big Data Analysis and Traceability of Product Quality Abnormality

Project: Research project

Project Details

Description

"Fault detection and traceability of product quality abnormality (FDT-PQA) play a significant role in maintaining product quality in industrial manufacturing processes such as those in the petrochemical, mineral processing, and steel manufacturing industries. Recently, several data-driven-based FDT-PQA techniques have been proposed and investigated in power distribution networks, the Tennessee Eastman process, and other domains, but most of them focus on small-scale and static fault location tasks. They are generally ineffective or may even fail when applied to real-world complex industrial processes. In general, industrial processes characterized by large-scale multiple processes with unclear latent mechanisms and operations in a dynamic environment result in three key issues for FDT-PQA: (1) Multiple processes and complex reactions generate high-dimensional heterogeneous data with categorical and numerical attributes (i.e. mixed-type attributes). Selecting discriminative features from such data is a challenge; (2) Cross-coupled industrial processes often feature complex mutual reactions with multiple variables, making the traceability of PQA challenging; (3) Dynamic manufacturing environments will occasionally cause the occurrence of concept drift, i.e. the patterns that a model have learned are not held any more. Furthermore, it is often expensive or even impossible to obtain the labels of the generated data from the manufacturing process. Another challenging issue is therefore how to quickly update the FDT-PQA model under concept drift with few labeled data.

This project will develop machine learning methods to address the three above-mentioned challenges. First, we will design feature selection methods for high-dimensional heterogeneous data with mixed-type attributes. Then, a generative adversarial network-based one-classification model and a hierarchical representation learning method will be proposed for FDT-PQA. Furthermore, we will design a local similarity measure and a contrastive learning-based method to address the problem of concept drift in dynamic environments with insufficient labeled samples. Finally, we will design and build a semi-physical simulation platform to evaluate the effectiveness of the proposed method.

Through this project, we will gain a thorough understanding of the underlying mechanism of FDT-PQA. The project’s output will be an FDT-PQA solution for the industrial manufacturing process. Furthermore, the research findings will provide the basis for the development of FDT-PQA. The techniques developed in this project will also be essentially applicable to a variety of applications in machine learning and data science."
StatusActive
Effective start/end date4/11/214/11/25

UN Sustainable Development Goals

In 2015, UN member states agreed to 17 global Sustainable Development Goals (SDGs) to end poverty, protect the planet and ensure prosperity for all. This project contributes towards the following SDG(s):

  • SDG 9 - Industry, Innovation, and Infrastructure

Fingerprint

Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.