水泥回转窑故障诊断方法研究

Research on Fault Diagnosis Method of Cement Rotary Kiln

受传感器和复杂外界环境等因素影响,回转窑数据检测难度大,存在故障信息获取不全和不同时刻故障表征存在差异的问题,且单一算法诊断能力有限。针对上述问题,构建了一种免疫进化网络理论分类器(IENC)和反向传播(BP)神经网络决策融合的故障诊断模型。该模型对回转窑正常工况及5类故障情况下的数据进行分析,分别用两种算法对故障类型进行初步判定,采用D- S证据理论获得证据体的基本信任分配。如果判定结果一致,则直接得出结论。否则,用Dempster组合规则进行决策融合,得到更合理的诊断结果。将某水泥厂的监测数据用于仿真。试验结果表明,该方法的综合分类正确率达到97.04%,比IENC算法提高了10%,比BP神经网络算法提高了1.48%,可有效改善单一算法“一票否决”问题,提高诊断结果可信度。

Due to the influence of sensors and complex external environment, it is difficult to detect the data of rotary kiln. There are problems of incomplete acquisition of fault information and differences in fault characterization at different times, and moreover, single algorithm has limited diagnostic ability. In order to solve these problems, a fault diagnosis model based on immune evolutionary network classifier (IENC) and back propagation(BP) neural network decision fusion was constructed, which analyzing the data under normal conditions and five fault conditions, using two algorithms to preliminarily determine the fault types, and adopting D- S evidence theory to obtain the basic probability assignment of the evidence body. If the determination results were consistent, the conclusion would be reached directly. Otherwise, decision fusion would be conducted according to Dempster combination rules to obtain a more reasonable diagnosis result. Simulation experiments were conducted with the monitoring data of a cement plant, and the results showed that the comprehensive classification accuracy of the method reached 97.04%, which was 10% higher than the IENC algorithm and 1.48 higher than the BP neural network algorithm, effectively solved the problem of “one- vote veto” of the single algorithm and improved the reliability of diagnosis results.