Material Type  

 학위논문

 

0016935200

Date and Time of Latest Transaction  

20240214101904

ISBN  

9798380712613

DDC  

330

Author  

Zhou, Chengyu.

Title/Author  

High Dimensional Data Analysis for System Condition Monitoring - [electronic resource]

Publish Info  

[S.l.] : North Carolina State University., 2023

Publish Info  

Ann Arbor : ProQuest Dissertations & Theses, 2023

Material Info  

1 online resource(131 p.)

General Note  

Source: Dissertations Abstracts International, Volume: 85-05, Section: B.

General Note  

Advisor: Fang, Xiaolei.

학위논문주기  

Thesis (Ph.D.)--North Carolina State University, 2023.

Restrictions on Access Note  

This item must not be sold to any third party vendors.

Abstracts/Etc  

요약This thesis presents new methodologies focusing on the challenges of high dimensional data analysis for condition monitoring in modern manufacture, service sector, and energy area. Chapter 1 introduces the research background, motivation, and challenges, and briefly discusses the research topics in this dissertation.Chapter 2 proposes a novel convex two-dimensional variable selection method that can inspire both group-wise and element-wise sparsity. It is based on a generalized matrix regression that regresses the quality index from the exponential family against a predictor matrix, in which the rows represent the process variables and columns represent stages. Two types of norms are introduced for regularization. The rows and columns of the regression coefficient matrix are simultaneously penalized using an ℓ2 norm to inspire group-wise sparsity. In the meantime, all the elements of the coefficient matrix are penalized using an ℓ1norm to inspire element-wise sparsity.Chapter 3 develops a supervised dimension reduction-based prognostic model that uses an asset's incomplete degradation images to predict its TTF. The proposed supervised dimension reduction method works as follows: First, it constructs an optimization criterion that comprises a feature extraction term and a regression term. The first term extracts low-dimensional features from complete/incomplete degradation image streams of training assets, and the second term builds the connection between these assets' TTFs and the extracted features using LLS regression. Solving the optimization criterion yields a set of tensor basis matrices for dimension reduction. The TTFs of the training assets are then regressed against their tensor features using LLS regression, and the parameters are estimated using maximum likelihood estimation. Next, a block updating algorithm is proposed to solve the optimization criterion and closed-form solutions are derived under normal or lognormal distribution no matter the degradation image streams are complete or incomplete.Chapter 4 proposes a Federated Multilinear Principal Component Analysis (FMPCA) framework to achieve tensor dimension reduction for decentralized data. It is a federated learning-based methodology which is a collaboratively decentralized privacy-preserving technology aiming to overcome data silo. The motivation is achieved by proposing three algorithms for the data leakage steps in Multilinear Principal Component Analysis (MPCA)- those are - secure centralization of input tensor samples, secure derivation of projection matrices in initialization, and secure derivation of projection matrices in local optimization. Chapter 5 concludes the dissertation.

Subject Added Entry-Topical Term  

Sparsity.

Subject Added Entry-Topical Term  

Missing data.

Subject Added Entry-Topical Term  

Feature selection.

Subject Added Entry-Topical Term  

Privacy.

Subject Added Entry-Topical Term  

Sensors.

Subject Added Entry-Topical Term  

Industrial engineering.

Added Entry-Corporate Name  

North Carolina State University.

Host Item Entry  

Dissertations Abstracts International. 85-05B.

Host Item Entry  

Dissertation Abstract International

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

소장사항  

202402 2024

Control Number  

joongbu:643890