Fault Diagnosis And Quality Prediction In Batch Fermentation Based On Improved KPLS

With the continuous complexity and mechanization of modern industrial process,the effective process monitoring,fault diagnosis and quality prediction have been of great importance to guarantee the security of production course and improve the quality of products.There are more and more historical data can be obtained owning to the extensive development and application of DCS and intelligent instruments.Therefore,it has been research hotspot which extract the statistical rules of normal operation state and known fault data modeling by historical data of industrial process to realize the process monitoring and fault diagnosis of the current process.In this case,the process monitoring and fault diagnosis method based on multivariate statistical theory emerged at the right time.Aiming at the frequent start-stop,nonlinear,multistage and dynamics in the batch processes,taking the process of industrial Escherichia coli fermentation as the object,the method of fault diagnosis and quality prediction based on kernel partial least squares is improved and supplemented.An improved feature sampling method was proposed to solve the computational problem of KPLS.Then a fault diagnosis and quality monitoring method based on the standard vector kernel space contribution diagram is proposed which makes up the limitation of the traditional fault diagnosis methods.At last the kernel entropy partial least squares method was proposed for fault diagnosis and quality prediction with studying the traditional PLS and KECA method which solved the problems of instability and nonlinearity for batch processes.The primary contents of this paper can be summarized as follows:(1)Traditional PLS and KPLS method was studied carefully.Then,aiming at the large computational load and the blindness of traditional feature sampling,an improved feature sampling method was presented which classify the data firstly,then selected the initial base vector and get feature circularly for the different clusters.(2)Aiming at the traditional contribution plot of fault diagnosis couldn’t be used in kernel space,a standard vector kernel contribution diagram method of fault diagnosis method was proposed.The method reconstruct the monitoring samples directly,namely,this method aim at finding a normal and representative vector of modeling samples as standard vector,then replaced the corresponding variable of normal samples for the variable of fault once monitoring fault or skewing of quality prediction curve,and then monitored the replaced variable;Then compared the new statistical magnitude or new predicted value to the original statistical magnitude or normal prediction curve to find the fault source.(3)A kernel entropy partial least squares algorithm for process monitoring and quality prediction method was proposed combining with kernel partial least squares and kernel entropy analysis which took example by the improvement of KECA for KPCA.The data dimensionality reduction was conducted according to eigenvalues and eigenvectors of the Renyi entropy which improved the accuracy of monitoring and prediction model.The proposed method is applied to the process monitoring and fault diagnosis of batch fermentation process.The simulation results show the effectiveness of the proposed method.