A novel quality-related statistical process monitoring method based on global and local partial least-squares projection (QGLPLS) is proposed in this paper. The main idea of the QGLPLS method is to integrate the advantages of locality-preserving projections (LPP) and partial least squares (PLS) and extract meaningful low-dimensional representations of high-dimensional process and quality data. QGLPLS can exploit the underlying geometrical structure that contains both global and local information pertaining to the sampled data, including the process variable and quality variable measurements. It is well-known that the PLS method can find only the global structure information and ignores the local features of data sets and that the LPP method can preserve local features of data sets well without considering the product quality variables. The capacity for the preservation of global and local projections of the proposed method is compared to that of the PLS and LPP methods; the comparison results demonstrate that the QGLPLS method can effectively capture meaningful information hidden in the process and quality data. Next, a unified optimization framework, i.e., global covariance maximum and local graph minimum in the process measurement and quality data space, is constructed, and QGLPLS-based T2 and squared prediction error statistic control charts are developed for online process monitoring. Finally, two typical chemical processes, the Tennessee Eastman process and the penicillin fermentation process, are used to test the validity and effectiveness of the QGLPLS-based monitoring method. The experimental results show that the obtained process monitoring performances are better than those when using traditional monitoring methods, such as PLS, principal component analysis, LPP, and global–local structure analysis.

Process variables can be classified into three stages: normal operation, incipient fault, and significant fault stage. A two-step incipient fault detection strategy was proposed for monitoring the complex industrial process. The first step aims at the significant fault detection using the traditional multivariate statistical process monitoring methods. Then a method combining the wavelet analysis with the residual evaluation was carried out for monitoring the incipient fault. Wavelet analysis aims at extracting the incipient fault features from process noise. The residual generation is optimization based on the robustness and sensitivity index, which can be realized directly using the test data. An improved kernel density estimation based on signal to noise ratio is proposed to adaptively determine the detection threshold. The proposed incipient fault detection scheme is tested on a numerical example and the Tennessee Eastman process. Compared to other traditional fault detection methods, good monitoring performances, such as higher fault detection rate and lower false alarm rate, are obtained.

Inaccurate substage division problems often emerge when multiway principal component analysis is applied in fault monitoring of multistage batch processes. A new two-step stage division method based on support vector data description (SVDD) is proposed in order to avoid the hard-division and misclassification problems. The loading matrices of the MPCA model are modified using the idea of combining the mechanism knowledge with field data in the rough division step. The model differences are increased by introducing the sampling time to loading matrices, which can avoid division mistakes caused by the fault data. Detailed stage separation is realized here based on the SVDD hypersphere distance to divide the process strictly into steady or transition stages. Then a soft-transition sub-PCA model is given based on the hypersphere distance. The method is applied to monitoring a penicillin fermentation process online. Simulation results show that the proposed method can describe transition stage information in more detail. It can detect the fault earlier and avoid the false alarm compared with traditional sub-PCA monitoring.

•Residual evaluation and contribution plot is unified into a novel fault isolation scheme.•Smearing effect can be eliminated based on a new contribution index.•Fault evolution can be acquired according to current and previous residuals.•The primary faulty variable can be located when several different faults occur simultaneously.A new fault isolation strategy for industrial processes is presented based on residual evaluation and contribution plot analysis. Based on the space projection, the residual evaluation and contribution plot are unified into a framework. First, parity space and subspace identification methods are used to generate residuals for fault detection. Then, the optimal residuals are utilized to obtain a process fault isolation scheme. A new contribution index is calculated according to the average value of the current and previous residuals. The smearing effect can be eliminated, and the fault evolution can be acquired based on this index. This would be helpful for engineers to find the fault roots and then eliminate them. A numerical model and the Tennessee Eastman process are considered to assess the isolation performance of the proposed approach. The results demonstrate that the smearing effect is improved and the primary faulty variable can be located accurately when several different faults occur simultaneously. A superior isolation performance is obtained compared with the PCA-based isolation method. The reasonability of the isolation results is analyzed using fault propagation.