A framework for data-driven fault detection and identification with multiscale kernel fisher discriminant analysis in chemical process systems / Norazwan Md Nor

• Main Author: Norazwan , Md Nor
• Format: Thesis
• Published: 2018
• Subjects: TP Chemical technology
• Online Access: http://studentsrepo.um.edu.my/9357/1/Norazwan_Md_Nor.pdf; http://studentsrepo.um.edu.my/9357/6/norazwan.pdf; http://studentsrepo.um.edu.my/9357/

Fault detection and identification (FDI) framework plays an important role to ensure consistent and reliable operation of chemical process systems. The FDI framework has two main tasks, namely to detect the presence of a fault and to classify the location and type of the fault. In most cases, it is impractical to develop precise model from first principles as it requires the involvement of process complex physics and the interactions among the different components creating the process. Therefore, data-driven FDI methods, which can make use of process data to capture their trends and dynamics, provide an attractive alternative for the quick development and deployment of FDI solutions. One of the main objectives of this thesis is to develop a hybrid framework for data-driven FDI in chemical process systems. This framework integrated a novel multiscale dimensional reduction method for pre-processing step and an improved datadriven FDI framework. This thesis focuses on proposing a dimensionality reduction method based on multi-scale kernel Fisher discriminant analysis (multi-scale KFDA), in which discrete wavelet transform (DWT) was combined with kernel Fisher discriminant analysis (KFDA) method. Initially, DWT was applied to extract the dynamics of the process at different scales. The wavelet coefficients obtained during the analysis were reconstructed using the inverse discrete wavelet transform (IDWT) method and then, they were fed into the KFDA to produce discriminant vectors. Finally, the discriminant vectors were used as inputs for the classification task in fault identification step. Apart from that, complete fault identification procedures based on adaptive neuro-fuzzy inference system (ANFIS), support vector machine (SVM), Gaussian mixture model (GMM), and k-nearest neighbor (kNN) were developed to investigate the parameters that could be optimised for better fault identification. Furthermore, this thesis extended the proposed multi-scale KFDA-based fault identification methods to a hybrid datadriven FDI framework. In the hybrid FDI framework, all classification methods used previously were combined into a single classification framework. Hence, a complete data-driven hybridisation FDI framework for chemical process systems was proposed and analysed. The proposed FDI frameworks were applied in three different chemical processes: the simulation of Tennessee Eastman process, the fed-batch penicillin fermentation process, and a real industrial data set of semiconductor etch process. Notably, the fault detection and classification results demonstrated the effectiveness of the proposed methods.