Several interesting problems arise in the processing of plant measurements. These range from the efficient storage and retrieval of data using data compression techniques to determining the health of process units, sensors and actuators using fault diagnosis techniques. The problems we are working on are concerned with processing of plant measurements for obtaining useful information (also known as data mining). In particular, we have made significant contribution to the subject of data reconciliation which deals with the problem of obtaining accurate and consistent estimates of process variables to be used in simulation, control, optimization or plant maintenance. We have also developed techniques for detecting gross errors in data caused by biases in sensors or material and energy leaks from process units. A general purpose commercial software package called RAGE incorporating these ideas have been developed in coordination with the R&D Centre of Engineers India Ltd., New Delhi. This package is used widely by EIL for processing data from mineral and petrochemical processes and has also been implemented in a few refineries in India. This subject has been comprehensively described in our book (co-authored with Dr. Cornelius Jordache of Simulation Sciences, Houston and published by Gulf Hemisphere, USA).

Process data can be used to judge the health of a plant in much the same way as we determine the health of a person by monitoring several parameters. In recent years, the problem of fault diagnosis has been the focus of several research groups all over the world. We have extended the techniques developed for gross error detection to fault diagnosis (FDI) in dynamic systems. In collaboration with Prof. Sachin Patwardhan of IIT Bombay, we are one of the first groups to integrate the FDI techniques with control systems to recently develop a fault-tolerant control strategy. We are currently investigating several related issues such as model identification, fault tolerant control using identified models, fault tolerant control of complex high dimensional processes and nonlinear processes. A project funded by the Department of Science and Technology is ongoing for testing the fault tolerant control strategy on a laboratory reactor set-up.

A specific fault diagnosis problem we are currently investigating in collaboration with Prof. Murty of Civil Engineering, IITM, is that of leak detection in gas transportation pipeline networks. Recently, the Gas Authority of India have funded a project for setting up a laboratory facility for developing and testing several leak detection techniques and to implement it online in one of the gas pipelines they own and operate in Andhra Pradesh.

The area of Process Design has seen significant developments in the last two decades. It is now recognized that optimizing the structure of the process can lead to substantially reduce energy or water consumption and waste generation. Our research deals with the use of graph theoretic approaches for generating optimal process structures. In particular we are dealing with the problem of sensor network design for optimal choice and location of sensors for better process reliability and controllability. Other problems where we have made effective use of graph theory are in the design of heat exchanger networks and in efficient design and solution of water distribution networks.

Research in Artificial Intelligence has led to several useful tools of which neural networks find wide use in engineering applications. Our focus is to use neural networks for developing black-box models of chemical processes and to use them in nonlinear model based control. A commercial software for this purpose has been recently developed in collaboration with Envision Systems India Ltd. Chennai, which will be incorporated in a training simulator for training plant operators on the use of this technology. We are currently studying different issues for obtaining better control for a wide range of nonlinear processes.

RAGE is a general-purpose, user-friendly software for data reconciliation and gross error detection. The software incorporates state-of-the-art techniques and is applicable in chemical and mineral processing industries. It was field tested on crude-preheat train data at Madras Refineries Limited. Currently, it is widely used in the R&D centre of EIL for many applications. A copy of the software has been sold to Bharat Petroleum Refineries Ltd. and Hindustan Petroleum Corporation Ltd.

This software is useful for short term scheduling of operations in batch process industries. A sparse mixed integer linear programming approach has been used to solve the scheduling problem. The software can handle any complex processing sequence and satisfies constraints on material, equipment utilization and manpower, utility consumption. It is currently being tested on data from a pharmaceutical plant.

This package is useful in developing neural network models given input-output data for any process. Currently it automatically builds multilayer perceptron models. State of the art neural network training strategies are used to train the network. Dynamic network building and pruning strategies are incorporated along with dynamic training techniques.The software provides the necessary interface for directly using the neural network model in model predictive control strategies. Furthermore, using the basic building blocks (objects) incorporated in the software any network architecture can be easily implemented. This package will be incorporated in a process control training simulator which is currently marketed by Envision Systems India Pvt. Ltd.