Data-driven Uncertainty Analysis in Neural Networks with Applications to Manufacturing Process Monitoring

Date: 2022/08/02 - 2022/08/02

Academic Seminar: Data-driven Uncertainty Analysis in Neural Networks with Applications to Manufacturing Process Monitoring

Speaker: Bin Zhang, senior planning & control algorithm engineer at Baidu Inc

Time: 9:00 a.m.-10:00 a.m., August 2nd, 2022 (Beijing Time)

Location: via Feishu

Abstract

Uncertainty is ubiquitous in real-life data and must be considered methodically when building data-driven models. Unfortunately, uncertainty is often overlooked or underestimated in many studies, which hinders the robustness and reliability of data-driven models and limits their applications to practical problems. In this talk, a novel uncertainty analysis framework will be presented. The Gaussian mixture model with an adaptive refinement scheme, which can retain compactness and fidelity through the nonlinearity in neural networks, is proposed to characterize the probability distributions of uncertainties. A Bayesian Gaussian mixture filter is designed for state estimation of highly nonlinear systems and a probabilistic neural network algorithm with Gaussian-mixture-distributed parameters is developed for uncertainty learning and prediction from noisy data. The application of developed methodology to the condition monitoring of two manufacturing processes will be demonstrated, including a tool wear monitoring scheme for machining based on in-process signal processing, and a porosity monitoring scheme for additive manufacturing based on melt pool image processing.

Biography

Bin Zhang is currently working as a senior planning & control algorithm engineer at Baidu Inc. He received his Ph.D. degree in Mechanical Engineering from Purdue University, West Lafayette, IN, USA, in Aug 2021. Before that, he received the B.S. and M.S. degrees in mechanical engineering from Beihang University, Beijing, China and was a visiting student at the University of British Columbia, Vancouver, Canada. His current research interests include data-driven science and intelligent systems, with a focus on the uncertainty analysis methods and smart manufacturing applications. He has published 10 first-authored journal papers and 2 co-authored papers and served as reviewers for multiple journals including the Additive Manufacturing, Journal of Manufacturing Processes and Manufacturing Letters. He has also delivered a series of applied process monitoring projects to renowned industrial partners (e.g., GE, Boeing). His current work is exploring the data-driven decision making and planning for autonomous driving vehicles.