Soft Sensor Methods For Clinker Calcination Process

Cement industry belongs to one of the cornerstone industries for national economy.An important way to improve product quality and to reduce energy consumption is the improvement of informatization and automatization level of the cement clinker calcination process in rotary kiln.However,cement clinker calcination is a nonlinear dynamic process with large time delay and strong coupling.Such a closed and harsh production environment(high temperature and dust)makes it difficult or even impossible to measure directly the process information,which challenges the control theories and technologies for automatic control of cement production.In order to solve this problem,soft sensor methods are proposed in the present work to measure important process information(i.e.free lime content in clinker,heat loss through kiln shell and material depth)of cement clinker calcination process,based on mechanism knowledge and process data.The present work was funded by National Natural Science Foundation project,and the proposed soft sensor methods can provide new quantitative process information for high-performance operation and energy-saving control of cement clinker calcination process.The main contributions of this work are:(1)Free lime(f-CaO)content in clinker reflects the clinker quality as well as the energy consumption cost.A time-series analysis method for model inputs and output is proposed in the present work to establish an accurate f-CaO content soft sensor model,in which important process features(such as process continuousness and time-delay)ingored in most present methods are considered.Firstly,a time-matching strategy for model inputs is developed based on the transport mechanism of the material in each process equipment.Then,the time-series of the input variables are weighted near the matched time point to obtain input-output samples with abundant time information.The proposed time-series analysis method is helpful for improving the accuracy of the soft sensor model for f-CaO content measurement.(2)Current single global models may fail to deliver accurate predictions due to their limited generalization ability.Therefore,a soft sensor method for f-CaO content is established in this paper,based on time series analysis and ensemble learning.Firstly,six different individual machine learning algorithms are selected to build each sub-model.The inputs of each sub-model are pre-processed by time series analysis method proposed in this paper.Secondly,mutual information method is adopted to select sub-models with better performance,which are then combined to realize the measurement of f-CaO content in clinker.The proposed soft sensor model is trained and tested by process data sampled from a cement production line.Results show that the proposed model can predict qualitatively and quantitatively well the trend of f-CaO content with time.Comparison with single global models and conventional ensemble learning(without ensemble pruning)based model,the model proposed in this paper has higher measuring accuracy and can meet the requirement for online measurement.It has theoretical and practical values for improving the automation level of clinker calcination process.(3)Due to high temperature characteristics of the clinker calcination process,heat lost through kiln shell is inevitable and it fluctuates with the change of operation variables.Thus,measurement of heat loss through kiln shell and analysis of its influencing factors are the prerequisite and foundation for energy saving control and strategy of cment clinker calcination process.In the estimation of heat loss through kiln shell,the spatial and temporal heterogeneousness of kiln shell temperature is mostly ignored.In order to improve the measuring accuracy,a soft sensor method for heat loss is propsed in the present work,based kiln shell temperatures from infrared thermography.By analyzing convection and radiation heat transfer mechanism through the kiln shell,equations for heat loss through kiln shell are established;basd on temperature data obtained by the real-time infrared images,the heat loss through kiln shell is then determined.Meausring tests on an industrial cement kiln show that heat loss through kiln shell in the calcination zone of the investigated kiln is about 80.68 kw/m,and the heat losses in the calcination zone by radiation and convection are comparable.Less than 12% of the energy input into the rotary kiln is lost through the whole kiln shell.Finally,in order to find out the main operating variables that greatly affect the heat loss of the investigated rotary kiln,an analysis method combining Random Forest and Pearson correlation coefficient is adopted.The results show that five main operating variables(the opening degree of 2# fan at the grate cooler,feeding rate of raw material and coal feeding rate for precalciner and so on)have greatest influence on the amount heat loss through kiln shell.The proposed model and analysis method can provide quantitative information for energy saving control and decision-making of the clinker calcination process.(4)Material depth is a key factor affecting the heat-and mass transfer processes as well as chemical reactions in rotary kiln.It is also one of the decisive factors affecting high performance operation control of clinker calcination process.However,the hostile environment(high temperature and enclosed)of the cement clinker calcination process makes it difficult to measure the material depth.In the present work,a soft sensor method is proposed for measuring the material depth in a directly-heated pilot rotary kiln,based on temperature fields in the kiln cross section.By analyzing the motion behavior of the material bed and the periodic pattern of the temperatures profile,the material region and the gas region are qualitatively identified.The statistical analysis method is then used to estimate the average temperature of the active layer of the material bed,with which the central angle covered by the material bed at a given raditional location is determined and thus the material depth can be measured using geometrical relationship.The proposed method is tested on a directly heated pilot rotary kiln and compared with manual measurement.It is found that the proposed soft sensor model for material depth has good accuracy(maximal error 7%)as well as stability for all the fourteen testing cases.It provides a new thought for the measurement of the material depth in industrial cement rotary kiln.