The kernel non-negative matrix factorization ( KNMF) method was introduced to the process moni-toring, and two new statistics K2 and SPE which responding to the fluctuation of raw data were designed to de-tect faults.According to the correlation between measured variables and nonlinear data, KNMF contribution plots were proposed to calculate contribution value and to draw contribution plots as the faulty variable re-quired.Simulation in Tennessee Eastman ( TE) model proves the detection performance of KNMF and making use of contribution plots can identify the faulty variable well.%将核非负矩阵分解方法引入到过程监控中,设计了K2和SPE统计量反映原始数据的能量波动情况,进而检测过程故障的发生。同时提出一种KNMF贡献图计算方法,根据变量和非线性数据的相关性,计算变量贡献值并绘制贡献图,用于故障辨识。在TE模型上的仿真结果验证了KNMF故障检测的良好性能,利用KNMF贡献图可以较好地辨识故障变量。
展开▼