A new approach to load measurement for industrial scale ball mill

Precise load measurement is important and difficult for an industrial scale ball mill in thermal power plant. This paper presents a new approach which combines a novel feature extraction method and a hierarchical model to achieve this purpose. By analyzing the sensitivity distributions of working condition transitions and removing the ineffective information, the characteristic power spectrums (CPS) of noise and vibration signals are obtained. Then, three kinds of features: the feature energy, the centroid frequency and the frequency domain variance are extracted from the CPS. Based on adaptive network-based fuzzy inference system (ANFIS), a hierarchical model which consists of three ANFIS model is designed. The rules and initial parameters of each ANFIS model are obtained by subtractive clustering method. Experimental results show that the potential advantages of the proposed method are low model complexity, high estimated accuracy and good generalization ability. Moreover, as multi-sensors´ information are adopted, the measurement results may be more reliable.