Research On Soft Sensor For The Characteristic Parameters About The Clinker From NSP Kiln

There are three characteristic parameters about the material from NSP kiln. They are the material flow from the kiln, the material temperature from the kiln and the particle size of the material from the kiln. Rotary kiln plays an important role in the process of the raw calcination. Grate cooler is the key link in clinker cooling and crushing. So, rotary kiln and grate cooler are both the important part in NSP cement production. Although the flow, temperature and the particle size of material from the kiln all play a vital role in the normal operation and real-time control about rotary kiln and grate cooler, but they could not get from direct measurement. That's why the paper uses the soft measurement to realize the characteristic parameters about the material on-line measurement indirectly.Combining actual technology of the uniseriate five grade cyclone pre-heater NSP cement production, we get the material flow from the kiln as the primary variable which calculated by the amount of clinker in clinker warehouse and the hoist current which can transport the clinker into the clinker warehouse that collected from the field of the cement production, choose the hoist current that load raw into kiln and the kiln motor current as instrumental variable, using RBF neural network to build soft-sensing model of the material flow from the kiln, and make compared with the soft-sensing model which built by single hidden layer BP neural network. Obtained, the material flow from the kiln soft sensor model based on RBF neural network is better than the material flow from the kiln soft sensor model based on BP network in the approximation ability, classification ability and learning speed. So, we choose soft-sensing model that based on RBF neural network to realize the material flow from the kiln on-line measurement indirectly.Combining actual technology of the diallel five grade cyclone pre-heater NSP cement production, we use high-temperature gun to measure the temperature of the clinker just from grate cooler. We get the material temperature from the kiln as the primary variable by heat balance equation under the severe convection condition when grate cooler works in stability situation. Select raw feed, outlet temperature of calciner, temperature of secondary air as instrumental variable of the soft sensor. Using double RBF neural network to build soft-sensing model of the material temperature from the kiln, and make compared with the soft-sensing model based on LS-SVM. The conclusion is that the material temperature from the kiln soft sensor model based on double RBF neural network and LS-SVM are all square in approximation ability and classification ability. Due to the network structure of the double RBF neural network is complex, therefore, its learning speed slightly slower than the soft-sensing model based on LS-SVM. But the LS-SVM model suitable for the study of small sample data, so, according to the actual need of production field, select the soft-sensing model based on double RBF neural network to realize the material temperature from the kiln on-line measurement indirectly.Under the premise of the material flow from the kiln stability, combining with workers'practical operation experience to determine the material particle size from the kiln as primary variable of this soft-sensing model. Select temperature of secondary air, tube pressure of grate cooler's one room as instrumental variable of the soft-sensing model. Use crusher current of the grate cooler's crusher as the calibration of this soft-sensing model. Using fuzzy mathematics to establish the particle size of the material from the kiln soft-sensing model, thus, realize the particle size of the material from the kiln on-line discrimination indirectly.According to the actual technology of NSP cement production, Using Visual C++ tool to develop software, the soft-sensing model based on RBF neural network, the soft-sensing model based on double RBF neural network and the soft-sensing model based on LS-SVM are applied to practical production, these obtain the good measurement effect.