BIG Data Analytics for Soft-Sensing and Data Reconciliation

For effective control, monitoring and real-time optimization of any industrial processes, real-time information of the quality variable which is being controlled or monitored is of vital importance. Measurement of these quality variable are available at irregular intervals with long delays (12 hrs/24 or 48 hrs). Further, analyzers for measuring this quality variables are costly and need constant maintenance. Soft-Sensors or Inferential models, which can predict the quality variable information have gained increasing popularity in the process industry, which increases the economic performance of the processes, can control and monitor the system effectively. In developing inferential sensors, quality of the process data, i.e., missing data due to sensor failure, outliers in process data, noisy sensor data, process/sensor drift, unknown disturbances, nonlinearity, high dimensionality of the process, multi-modal process behavior, play a very important role in designing the soft sensors effectively. Our group focusses on theoretical development of robust soft sensors using first-principles model based, completely data driven and hybrid estimation algorithms (model based + machine learning techniques) using BIG data techniques which can account for the above-mentioned abnormalities in the process data. Here, interests of application of the developed algorithms is on Oil & Gas, Mining, Agricultural and Energy storage systems.

Probabilistic Graphical Model based Soft-Sensors

1. Khosbayar, A., Valluru, J., Huang, B., Multi-rate Gaussian Bayesian network soft sensor development with noisy input and missing data, Journal of Process Control 105 (2021) 48-61.

2. Khosbayar, A., Valluru, J., Huang, B., Adaptive Inference for Bayesian Networks based Soft-Sensor in the presence of Process and Sensor Drift. Special issue in Canadian Journal of Chemical Engineering, Vol. 100 (9) page no. 2119-2134 (2022).

Simultaneous leak detection and bias removal using Robust Bayesian State and Parameter Estimators

1. Valluru, J., Lakhmani, P., Patwardhan, S.C., Biegler, L.T., Development of Moving Window State and Parameter Estimators under Maximum Likelihood and Bayesian Frameworks. Journal of Process Control, 60 (2017) 48-67. (Part of Special issue on selected papers from DYCOPS-2016).

2. Valluru, J., Patwardhan, S.C., Biegler, L.T., Development of Robust Extended Kalman Filter and Moving Window Estimator for Simultaneous State and Parameter/Disturbance Estimation. Journal of Process Control, 69 (2018) 158-178.

3. Valluru, J., Purohit, J.L., Patwardhan, S.C., Biegler, L.T., Development of a Moving Window Maximum Likelihood Parameter Estimator and its application on Ideal Reactive Distillation System. Proc. of 11^th IFAC Symposium on Dynamics and Control of Process Systems, including Bio-systems (DYCOPS-2016),Trondheim, Norway, 6-8 June, 2016. (shortlisted for a special issue of Journal of Process Control based on DYCOPS-2016 submissions).

4. Lakhmani, P., Valluru, J., Chandrasekharan, G., Patwardhan, S.C., Application of a Moving Window Parameter Estimator for Leak Identification in the Quadruple Tank System. Proc. of 4^th IFAC Conference on Advances in Control and Optimization of Dynamical Systems (ACODS-2016), Tiruchirapalli, India, 1-5 February, 2016.

Probabilistic Graphical Model based Process Data Reconciliation

1. Sundaramoorthy, A.S., Valluru, J., Huang, B., Bayesian networks based data reconciliation with state uncertainties and recycle streams, Chemical Engineering Science, 246 (2021) 116996-117010. (Part of Special Issue on Digitalisation in Chemical Engineering Science).