Damage Pattern Recognition Of Refractory Materials Based On K-means Clustering And BP Artificial Neural Network

Refractory materials are of great importance to the normal running of facilities under hightemperature, and the micro-damage pattern recognition of which has always been the emphasisof research. The damage process of three-point bending test of refractory materials is monitoredby using of acoustic emission technique (AET). Then, a database of experiment samples fordamage classification of refractories is built by acoustic emission (AE) parameters analysis andK-means clustering analysis. Finally, the type and degree of damage is recognized by BPArtificial Neural Network.The main work and conclusions of this dissertation are as follows:(1) Use differential high temperature stress and strain testing Machine（HMOR/STRAIN）and DISP acoustic emission testing system of PAC to collect AE signals of three-point bendingtest of refractories. Two types of damage are obtained by the theory of micro-damage: matrixdamage and interface damage.(2)The AE characteristics corresponding to each type of damage are determined bycombining AE parameters analysis and K-means clustering analysis of signals collected in theprocess of three-point bending test of refractories, which can build a database of experimentsamples for damage classification of refractories.(3)The pattern recognition system based on BP Artificial Neural Network (ANN) isestablished by using the AE parameters such as amplitude, counts, risetime, duration andcentroid frequency, which can recognize the damage pattern of refractories. The total recognitionrate of the damage types is up to95%, and the prediction error of the damage degree is relativelysmall and the maximum value is no more than0.2. It proved that this method has the value ofapplication and dissemination in the aspect of micro-damage type recognition and damagedegree prediction.