Advanced Adaptive Control Approaches for Complex Batch or Semi-Batch Operations

Research Plan:

Batch/semi-batch processes are an important set of unit operations in industries such as biofuel synthesis, synthesis of active pharmaceutical ingredients (APIs), specialty chemicals, etc. Such applications, in general, have complex dynamics and are operated in conjunction with other unit operations such as distillation, separation, etc. The products of these applications need to meet stringent quality requirements. Given the complex dynamics of such systems, it becomes essential to develop advanced control schemes that can utilize its multivariable dynamics in order achieve the desired control objectives and optimal performance of the process. Developing an all-encompassing control algorithm in order to achieve these goals involves addressing the following main challenges-

Adaptive identification of the pdf of the disturbances for optimal estimation of states and parameters. While the states and uncertain parameters can be estimated using Bayesian estimation algorithms, these algorithms rely on the fact that the system model and pdfs of the unmeasured disturbances and measurement noise are perfectly known. However, this is not the case since there are uncertainties in the model parameters and the pdfs of the unmeasured disturbances and measurement noise are not known. Given the continuously time-varying nature of the batch/semi-batch processes I would like to work on developing online MLE-based approaches for estimation of the pdfs of the disturbances and noise. This would help maintain the optimal performance of the Bayesian state and parameter estimators, which will aid in improved state-feedback control and monitoring.

Stochastic iterative learning control and parameter learning for batch/semi-batch operations. The repeated nature of batch/semi-batch operations means that knowledge of the control moves and parameters used in the previous batches can be used to improve the controller performance and parameter estimates for the current batch. Given the stochastic nature of batch/semi-batch operations and uncertainties in the system model arising from the time-varying nature of the processes, I would like to work on developing iterative learning control and iterative learning parameter estimation approaches that account for the probabilistic nature of these stochastic uncertainties in computing the control moves and parameter estimates respectively. This would also enable the formulation of a probabilistic control objective, thereby allowing for greater control over the final product quality.

Research Experience:

My research experience has been primarily in the areas of nonlinear Bayesian state estimation, system identification and model predictive control. During my PhD, I developed a maximum likelihood estimates (MLE)-based approach for identification of the covariance matrices associated with the Gaussian process and measurement noise from input-output process data for state and parameter estimation using extended Kalman filter (EKF)¹. I have also developed an MLE-based approach for identification of grey-box models for state and paramter estimation, when the system under consideration is affected by time-correlated noise in the inputs or disturbances². We have validated these approaches on an experimental setup as well. As a post-doctoral fellow at University of Alberta, I have worked on data-based approaches for modeling and classification for welding processes. We used approaches such as principal component analysis (PCA) and partial least squares (PLS) for classification of gas-metal arch welding jobs³. This work has led to an industry-academia collaboration and is being taken forward by the concerned PIs. As a post-doctoral fellow at UC Berkeley, my work has focussed on control-relevant model identification for stochastic model predictive control with a focus on cold atmospheric plasma (CAP) system for biomedical applications⁴. As a result I have research experience in multiple fields of Process Systems Engineering, both in computational and experimental studies.

References:

1. V. A. Bavdekar, A. P. Deshpande and S. C. Patwardhan, "Identification of process and measurement noise covariance for state and parameter estimation using extended Kalman filter", Journal of Process Control 21 (4) (2011) 585-601.
2. V. A. Bavdekar and S. C. Patwardhan, "Development of grey box state estimators for systems subjected to time correlated unmeasured disturbances", Journal of Process Control 22 (9) (2012) 1543-1558.
3. R. Ranjan, A. Talati, M. Ho, H. Bharmal, V. A. Bavdekar, V. Prasad, P. F. Mendez, "Multivariate data analysis of gas-metal arc welding process", Proceedings of the 9th International Symposium on Advanced Control of Chemical Processes (AdCHEM 2015).
4. V. A. Bavdekar and A. Mesbah, "Model predictive control with integrated input design for nonlinear systems with probabilistic uncertainties", accepted for the 11th International Symposium on Dynamics and Control of Process Systems, 2016.

Select Publications:

1. V. A. Bavdekar and A. Mesbah; Stochastic nonlinear model predictive control with joint chance constraints, accepted for the 10th IFAC Symposium on Nonlinear Control Systems, 2016.
2. V. A. Bavdekar and A. Mesbah; Model predictive control with integrated input design for nonlinear systems with probabilistic uncertainties, accepted for the 11th International Symposium on Dynamics and Control of Process Systems, 2016.
3. V. A. Bavdekar and A. Mesbah; A polynomial chaos-based nonlinear Bayesian approach for estimating state and parameter probability density functions, accepted for the 2016 American Control Conference.
4. V. A. Bavdekar, N. Nandola and S. C. Patwardhan; Estimation of noise covariance matrices for state estimation of autonomous hybrid systems, under review, Computers and Chemical Engineering.
5. V. A. Bavdekar, J. Prakash, S. C. Patwardhan and S. L. Shah; A Moving window formulation for state estimation of systems with irregular and delayed measurements, Industrial and Engineering Chemistry Research 53(35) (2014) 13750-13763. 
6. V. A. Bavdekar and S. L. Shah; Computing point estimates from a non-Gaussian posterior distribution using a probabilistic k-means clustering approach, Journal of Process Control 24 (2) (2014) 487-497.
7. V. A. Bavdekar and S. C. Patwardhan; Development of grey box state estimators for systems subjected to time correlated unmeasured disturbances, Journal of Process Control 22 (9) (2012) 1543-1558.
8. V. A. Bavdekar, A. P. Deshpande and S. C. Patwardhan; Identification of process and measurement noise covariance for state and parameter estimation using extended Kalman filter, Journal of Process Control 21 (4) (2011) 585-601.

Research Interests:

My research interests are in the areas of nonlinear Bayesian state and parameter estimation, system identification, control-relevant model identification, stochastic MPC, data analysis uses methods such as PCA, PLS, etc. I'm interested in applications involving batch/semi-batch operations such as ethanol synthesis, algal biofuels, batch polymerizations, active pharmaceutical ingredient (API) synthesis, etc.

As a faculty member, I will use my expertise in the above areas of Process Systems Engineering to form a group that conducts leading research at the interface of systems' theory, applied mathematics and process engineering towards the development of advanced control algorithms for batch/semi-batch operations.

Teaching Interests:

At the undegraduate level, I would be interested in teaching the following courses (or their equivalents) and the laboratory/computational hours associated with them: Process Dynamics and Control, Computer-aided Process Design, Numerical Methods in Chemical Engineering and Process Optimization. I would also like to explore the option of introducing a course on Data Analysis and Statistics for Chemical Engineers, which will introduce students to basic experiment design and analyis techniques, with focus on the multivariate nature of chemical engineering data.

At the graduate level, I would be interested in teaching the following courses (or its equivalents): Digital Control and Advanced Numerical Methods for Chemical Engineers. I am interested in offering a course on Bayesian state estimation, given my research experience in this area and a course on System Identification, which can be a nice complement to the course on Digital Control.