Research On A Class Of Non-linear Process Quality Related Monitoring Methods

The quality monitoring and fault diagnosis play an important role in the complex industrial production processes.However,the detection data of quality variable is often difficult to obtain in real time or the cost is relatively high,most of the fault diagnosis methods in the past are based on monitoring the abnormal changes of process variables.This kind of monitoring method cannot judge whether the changes of process variables will affect the quality of the output product.In practice,it is necessary to focus on the faults that affect product quality while ignoring quality-independent faults.Therefore,a reliable,real-time monitoring method that can reveal the correlation between process variables and quality variables is particularly important.To solve this problem,this paper analyzes the advantages and disadvantages of Partial Least Squares(PLS)and Local Linear Embedding(LLE).Then according to the idea of complementary advantages,this paper proposes two kinds of fusion schemes,including partial fusion and full integration.1.The global structural information of PLS and the local geometric structure information of LLE are directly combined by introducing the balance parameter,and the LLE enhanced type partial least squares(LLEEPLS)algorithm is proposed.The simulation results show that the algorithm has good local geometric structure retention ability and can establish the correlation between process measurement and quality measurement well.2.Based on 1,this paper further analyzes the deficiencies of the LLEEPLS method in the fusion of PLS and LLE.This paper proposes a modified Locally Linear Embedding Enhanced Partial Least Squares Statistical algorithm(MLLEEPLS).The algorithm can completely integrate the local geometric features of LLE into the PLS algorithm.The simulation results show that the algorithm not only has good nonlinear dimensionality reduction ability,but also has good predictability.