Research On Fault Diagnosis Of Process Industry Based On PCA

Fault diagnosis of the process industry plays an important role in ensuring the safety of industria l production and improving product quality. This paper studies on data information extraction, mult i- mode and complex distributed process data of fault diagnosis for process industry. The ma in contents are as follows:(1) For traditional PCA method only considers data's global structure in the process of data dimensionalit y reduction, this paper applies a method called local and global structure preserving pro jections(LGSPP), which can project the data to a low-dimensional feature space that has similar locality neighborhood structure and global structure with original space. After projection, statistics and Bayesian classifier are used to detect and identify fault. Considering the dynamic problem in data, augmented matrix containing the first h observations is constructed. Simulation on TE process identifies its effectiveness.(2) To solve the problem of mult i- mode in industrial process data, this paper makes some improvements from the perspective of data preprocessing. Local neighborhood standardization(LN S) is introduced to data preprocessing, then the paper proposes LNS- LGSPP. This method applies LNS to preprocess data, so it can not only solve the influence of different magnitudes but also remove the mult i-distribution feature. Then the processed data is applied to LGSPP to detect fault of mult i- modal process. The valid ity of the proposed method is illustrated through numerical and TE process.