Ajaya Kumar Pani

Development and comparison of neural network based soft sensors for online estimation of cement clinker quality.

The online estimation of process outputs mostly related to quality, as opposed to their belated measurement by means of hardware measuring devices and laboratory analysis, represents the most valuable feature of soft sensors. As of now there have been very few attempts for soft sensing of cement clinker quality which is mostly done by offline laboratory analysis resulting at times in low quality clinker. In the present work three different neural network based soft sensors have been developed for online estimation of cement clinker properties. Different input and output data for a rotary cement kiln were collected from a cement plant producing 10,000 tons of clinker per day. The raw data were pre-processed to remove the outliers and the resulting missing data were imputed. The processed data were then used to develop a back propagation neural network model, a radial basis network model and a regression network model to estimate the clinker quality online. A comparison of the estimation capabilities of the three models has been done by simulation of the developed models. It was observed that radial basis network model produced better estimation capabilities than the back propagation and regression network models.

A Survey of Data Treatment Techniques for Soft Sensor Design

Soft sensors have proved themselves as valuable alternatives to the traditional means for the acquisition of critical process variables, process monitoring and other tasks relating to process control. Most of the present day soft sensors for complex chemical processes are designed from actual industrial data because of the various difficulties associated with developing first principle models such as poor process understanding, impossible or difficult to determine model parameters and mathematical complexity of the models. This paper discusses characteristics of the process industry data which are critical for the design and development of data driven soft sensors. The focus of this paper is on the different shortcomings of the raw process data collected from the historical database and a review of different techniques available for processing of the raw data so as to make the data suitable for design of data driven soft sensors.

Online monitoring and control of particle size in the grinding process using least square support vector regression and resilient back propagation neural network

Particle size soft sensing in cement mills will be largely helpful in maintaining desired cement fineness or Blaine. Despite the growing use of vertical roller mills (VRM) for clinker grinding, very few research work is available on VRM modeling. This article reports the design of three types of feed forward neural network models and least square support vector regression (LS-SVR) model of a VRM for online monitoring of cement fineness based on mill data collected from a cement plant. In the data pre-processing step, a comparative study of the various outlier detection algorithms has been performed. Subsequently, for model development, the advantage of algorithm based data splitting over random selection is presented. The training data set obtained by use of Kennard-Stone maximal intra distance criterion (CADEX algorithm) was used for development of LS-SVR, back propagation neural network, radial basis function neural network and generalized regression neural network models. Simulation results show that resilient back propagation model performs better than RBF network, regression network and LS-SVR model. Model implementation has been done in SIMULINK platform showing the online detection of abnormal data and real time estimation of cement Blaine from the knowledge of the input variables. Finally, closed loop study shows how the model can be effectively utilized for maintaining cement fineness at desired value.

Data driven soft sensor of a cement mill using generalized regression neural network

Soft sensors play an important role in predicting the values of unmeasured process variables from knowledge of easily measured process variables. Most of the present day soft sensors for complex chemical processes are designed from actual industrial data because of the various difficulties associated with developing first principle models such as poor process understanding, impossible or difficult to determine model parameters and mathematical complexity of the models. No hardware sensors for online estimation of particle size is available for solid grinding units. In the present work a generalized regression neural network based model of a cement mill is developed based on the actual plant data for estimation of cement fineness. The collected raw industrial data is pre processed to get rid of the outliers and missing values followed by model development. Optimal selection of smoothing parameter for the regression model has been done by investigating the model performance at different spread values. Simulations results show satisfactory prediction capabilities of the developed model over that of linear regression model and quadratic response surface model.

Neural Network Soft Sensor Application in Cement Industry: Prediction of Clinker Quality Parameters

A soft sensor tries to estimate difficult to measure quality parameters from the knowledge of easy to measure online process variables. Empirical approach of soft sensor development has gained much popularity recently due to availability of huge quantity of actual process data stored in the industrial database. In this work a soft sensor based on back propagation neural network has been developed for rotary cement kiln. For this purpose, data for all variables associated with rotary cement kiln were collected over a period of one month from a cement industry having a capacity of 10000 tons of clinker production per day. Data preprocessing of the raw data has been performed to remove the anomalies present in the original data. The processed data was used to develop the neural network model of the kiln. Model simulation produced quite satisfactory prediction of free lime, C3S, C2S and C3A.

Pollutant monitoring in tail gas of sulfur recovery unit with statistical and soft computing models

Abstract In this article, data-driven models are developed for real time monitoring of sulfur dioxide and hydrogen sulfide in the tail gas stream of sulfur recovery unit (SRU). Statistical [partial least square (PLS), ridge regression (RR) and Gaussian process regression (GPR)] and soft computing models are constructed from plant data. The plant data were divided into training and validation sets using Kennard-Stone algorithm. All models are developed from the training data set. PLS model is designed using SIMPLS algorithm. Three different ridge parameter selection techniques are used to design the RR model. GPR model is designed using four hyper parameter selection methods. The soft computing models include fuzzy and neuro-fuzzy models. Prediction accuracy of all models is assessed by simulation with validation dataset. Simulation results show that the GPR model designed with marginal log likelihood maximization method has good prediction accuracy and outperforms the performance of all other models. The developed GPR model is also found to yield better prediction accuracy than some other models of the SRU proposed in the literature.

Inferential Sensing of Output Quality in Petroleum Refinery Using Principal Component Regression and Support Vector Regression

In this research, linear regression (ordinary least square and principal component) and non-linear regression (standard and least square support vector) models are developed for prediction of output quality from sulphur recovery unit. The hyper parameters associated with standard SVR and LS-SVR are determined analytically using the guidelines proposed in the literature. The relevant input-output data for process variables are taken from open source literature. The training set and validation set are statistically designed from the total data. The designed training data were used for design of the process model and the remaining validation data were used for model performance evaluation. Simulation results show superior performance of the standard SVR model over other models.