(7ik) Data Analytics for Complex Systems

In my research, I create, modify, and adapt machine learning algorithms for applications in chemical and biological engineering. Machine learning is an exciting field due to improved sensor systems yielding new types of datasets, decreasing data storage costs, and increasing computational power. This combination of resources gives engineers unprecedented access to data to answer questions about complex systems.

During my PhD work, I have participated in several collaborations. A key aspect of my work is the translation of modeling goals to algorithms. For instance, in a collaboration with a biomanufacturing company, I designed a methodology to learn a sparse model such that the most important control variables could be identified [1]. The resulting approach was able to achieve 95% prediction accuracy for the desired tolerance on an unseen test dataset. In an academic collaboration, I developed an algorithm that could use high-dimensional bio-assay data with missing entries by simultaneously performing imputation and model building [2]. The resulting models were at least as accurate and sparser than competing methods. More recently, I have collaborated with industrial and academic partners on energy applications. I collaborated with an oil and gas company to predict anomalous well conditions using a semi-supervised likelihood ratio test [3] and have started work on data-driven capacity monitoring for lithium ion batteries. These projects use a combination of statistics, optimization, numerical methods, and chemical engineering to improve decision making and increase system understanding.

Data analytics can help solve many open problems and chemical and biological engineers have an important role to play in designing techniques that will work for these applications. My future research will use data-driven techniques to improve process monitoring [4], better characterize parametric uncertainty, and build predictive models. I will focus on solving chemical and biological engineering problems by developing and employing techniques that handle issues specific to chemical processes such as missing data [2,5], unstructured data, and heterogeneous measurements as well as biological data issues such as high dimensionality and small sample size [6]. I will establish collaborations with manufacturing companies in the areas of time series analysis, unsupervised datasets, feature engineering, and manifold learning. I will also work with other faculty members to help develop methods to use data from new sensor systems and high throughput analysis.

Teaching Interests:

I am particularly interested in the important role software and computational tools play in engineering and would hope to integrate coding into any class I teach. I would prefer teaching applied mathematics, process design, unit operations, control, and transport phenomenon.

References:

[1] K. Severson, J.G. Van Antwerp, V. Natarajan, C. Antoniou, J. Thömmes, and R.D. Braatz. Elastic net with Monte Carlo sampling for data-based modeling in biopharmaceutical manufacturing facilities. Computers & Chemical Engineering, 80:30-36, 2015.

[2] K.A. Severson, B. Monian, J.C. Love, and R.D. Braatz. A method for learning a sparse classifier in the presence of missing data for high-dimensional biological datasets. Bioinformatics, in press. doi: 10.1093/bioinformatics/btx224

[3] K.A. Severson, P. Chaitwatanodom, M.C. Molaro, R.S. Bailey, and R.D. Braatz. Anomaly detection and diagnosis using semi-supervised models for industrial time-series data, in preparation.

[4] K. Severson, P. Chaiwatanodom, and R. D. Braatz. Perspectives on process monitoring of industrial systems. Annual Reviews in Control, 42:190-200, 2016.

[5] K.A. Severson, M.C. Molaro, and R.D. Braatz. Methods for applying principal component analysis to process datasets with missing values. Special Issue on Process Data Analytics, Processes, accepted.

[6] K.A. Severson, J.G. VanAntwerp, V. Natarajan, C. Antoniou, J. Thömmes, and R.D. Braatz. A systematic approach to process data analytics in pharmaceutical manufacturing: The data analytics triangle and its application to the manufacturing of a monoclonal antibody. In Multivariate Analysis in the Pharmaceutical Industry, edited by A. P. Ferreira, J. C. Menezes, and M. Tobyn, Elsevier, in press.