Fault Diagnosis Of Batch Process Based On Improved Neighborhood Preserving Embedding Algorithms

With intelligent, large-scale and integrated development of industrial process,production process is more and more complex. Complicated systems may produce some faults because of all kinds of external environment interference and their own equipment aging. To improve production efficiency, control system must be maintained steady state and production process needs to real-time monitor. Batch production mode is applied in more and more widely actual production processes because of its own merits. The products of batch processes are batch outputs. Once a fault of a production process variable at a certain moment can’t be timely solved, the whole batch products can not meet the requirements and lead to huge economic losses.So process monitoring and fault diagnosis of batch process are more important. It is very hard to obtain accurate process model for complex batch process. Multivariate statistical analysis methods based on data driven don’t need to get precise process model and just use history and online process data to realize process monitoring and fault diagnosis. With development of sensor technology and widely using in industrial processes, historical and real-time process data becomes relatively easy acquisition.How to fully extract useful information from massive amounts process data is a key for process monitoring and fault diagnosis. Comparing with principal component analysis(PCA) algorithm, which focuses on global feature information of process,neighborhood preserving embedding(NPE) algorithm can dig out more detail local feature information of batch process. This thesis focuses on actual problems of monitoring and fault diagnosis of batch process and improves NPE algorithm. The main works of this thesis are as follows:1. Aiming at strong nonlinearity of batch process, traditional methods only decompose the covariance matrix of data and ignore higher order statistics information. The effect of diagnosis is bad because effective information cannot be fully extracted in nonlinear batch process. The improved multi-way kernel neighborhood preserving embedding(SPA-MKNPE) algorithm based on statistics pattern analysis is proposed. First, statistics pattern analysis method is introduced to make sample data map to statistics sample space, so information of higher order statistics of nonlinear data can be fully extracted. Then statistic samples are mapped to high-dimensional kernel space through kernel function in statistic space to solve nonlinear problem of data. Finally the local structure of data is fully extracted byneighborhood preserving embedding algorithm in high dimensional kernel space. The contribution plot method is used to diagnose fault variables after detecting faults. The penicillin fermentation process simulation verified the effectiveness of the proposed algorithm.2. Batch production process data usually contains mixture ingredients of the Gaussian and non-Gaussian distribution. Traditional monitoring methods require data to meet the requirements of Gaussian distribution and do not take into account the global and local features for feature extraction. So a multi-way global neighborhood preserving embedding-local independent component analysis(MGNPE-LICA)algorithm is proposed. Firstly, raw data is divided into Gaussian and non-Gaussian space by D-test. For the Gaussian space, MGNPE algorithm is used to fully extract local structural features and global structural features of data. For the non-Gaussian space, MLICA algorithm is used to solve non-Gaussian problems and at the same time to reserve global and local information of data. Then the monitoring index of two spaces synthesizes a joint monitoring indicator to monitor the process. The contribution plot method is used to diagnose fault variables after detecting faults. The simulation results of penicillin fermentation process verified the effectiveness of the proposed algorithm.3. The traditional methods unfold three-dimensional data to two-dimensional data.The process inevitably leads to destruct internal structure of the data, and usually only considering the global information data or local information data. Useful process information can not be fully extracted, which leads to poor diagnosis. Aiming to above problems, a tensor global-local neighborhood preserving embedding(TGNPE)algorithm is proposed. First tensor factorization is used to deal with three-dimensional data directly which effectively save the internal structure of the data. Then the neighborhood preserving embedding algorithm is used to extract the local structure of the data information, at the same time considering the global information of the data,data information is fully extracted under keeping internal data structure. The contribution plot method is used to diagnose fault variables after detecting faults. The simulation results of penicillin fermentation process verified the effectiveness of the proposed algorithm.4. Aiming at dynamic characteristics of batch process in time and space, a tensor dynamic neighborhood preserving embedded(TDNPE) algorithm is proposed. First batch process data is deemed to a second order tensor. The tensor factorization method is used to model that can avoid destroying internal structures of data. Thendynamic neighborhood preserving embedded algorithm is used to extract process feature information by considering local features of space and time in tensor space and that can effectively deal with process dynamic characteristic. The contribution plot method is used to diagnose fault variables when faults are detected. The simulation results of penicillin fermentation process verified the effectiveness of the proposed algorithm.