Smart Data Analytics for Manufacturing Processes
Name
mohr-fmohr-phdcep-cheme-2024-thesis.pdf
Description
Thesis PDF
Size
5.16 MB
Format
Adobe PDF
Checksum (MD5)
9d7000afb72a34ca9f31bb8f59e86443
Author(s)
Mohr, Fabian
Advisor(s)
Braatz, Richard D.
Date Issued
May 2024
Publisher
Massachusetts Institute of Technology
Abstract
The increasing multitude of powerful data analytic tools in recent years has motivated the development of smart data analytics approaches, i.e. a decision tree based super algorithm that automatically selects the most suitable method based on a systematic interrogation of the dataset. This work introduces two different smart data analytics approaches for the objectives of supervised classification and fault detection, as they arise for example in the context of process monitoring schemes for chemical manufacturing processes. For both objectives, a visual representation of the method selection process is presented in form of a data analytics triangle. The necessary interrogation framework for the model selection process is introduced and the overall approaches for both objectives are demonstrated in case studies showing great performance over a variety of different supervised classification and fault detection problems. Additionally, predictive modeling techniques are applied to industrial endto-end biomanufacturing datasets for two monoclonal antibody products to predict critical quality attributes. Different approaches are introduced to consider both secondorder and third-order tensorial data combined as possible inputs to the predictive modeling. It is shown that the utilization of the proposed methods is capable of significantly improving the prediction performance if the dataset is analyzed correctly beforehand. Moreover, the ability to predict and classify the cycle life of LiNiMnCo (NMC) battery half-cells based on acoustic measurements capturing degradation events such as grain fracture and gas formation in combination with prior introduced data analytical methods is demonstrated. Lastly, the application of a Kalman Filter model to predict moats around companies in the stock market is explored.
MIT Department
Massachusetts Institute of Technology. Department of Chemical Engineering
Sloan School of Management
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
