2024年5月24日学术报告

报告一：Data Analytics and Machine Learning for Feature Extraction.

报告人：Prof. Biao Huang, University of Alberta

报告时间：2024年5月24日09:00

报告地点：物联网工程学院B222室

简介: Modern industries are awash with a large amount of data. The extraction of information and knowledge discovery from data for process design, control, and optimization, especially from day-by-day routine process operating data, is interesting but challenging. Big data analytics is an emerging area of great interest among data scientists and practicing engineers to extract meaningful features that represent data and their underlying processes. Unlike neural network learning-based approaches that typically extract features without clear physical meanings, most statistical feature extractors have physical interpretations. This presentation will give a historical overview of big data analytics along with illustrative examples related to some popular feature extraction methods.

报告二：Advancing Causal Analysis for Fault Detection and Root Cause Analysis in Process Systems Engineering

报告人：Prof. Biao Huang, University of Alberta

报告时间：2024年5月24日14:00

报告地点：物联网工程学院B222室

简介：Causality analysis, a well-established data-driven technique for root cause identification, has garnered extensive attention across multiple disciplines. Utilizing causal analysis tools, engineers can construct causal maps crucial for fault prediction and diagnostic applications. However, relying solely on conventional data analytics for reconstructing causal maps raises challenges associated with data quality. High-quality data are imperative to ensure the reliability of results. In causality analysis, issues stemming from data quality manifest as spurious causations and the failure to identify the existence of causations. While causal maps can be constructed based on expert knowledge and process flow diagrams, this approach may prove inadequate for complex and tightly integrated processes. The emerging field of physics-informed modelling offers a promising avenue, having been successfully applied in various domains. However, combining physics information with observed data for reconstructing causal maps remains a relatively unexplored challenge. Motivated by these considerations, we introduce a novel framework to reconstruct causal maps for linear time-invariant dynamical systems. This innovative approach integrates observed data with physics information, enhancing the reliability of identifying the sources of process faults.