In this paper, a new data-driven sensor fault detection and isolation (FDI) technique for interval-valued data is developed. The developed approach merges the benefits of generalized likelihood ratio (GLR) with interval-valued data and principal component analysis (PCA). This paper has three main contributions. The first contribution is to develop a criterion based on the variance of interval-valued reconstruction error to select the number of principal components to be kept in the PCA model. Secondly, interval-valued residuals are generated, and a new fault detection chart-based GLR is developed. Lastly, an enhanced interval reconstruction approach for fault isolation is developed. The proposed strategy is applied for distillation column process monitoring and air quality monitoring network.
KEYWORDSdata-driven process monitoring, fault detection and isolation, generalized likelihood ratio, interval-valued data, principal component analysis, reconstruction
INTRODUCTIONModern chemical plants are becoming increasingly complex, making the need for enhanced process safety, consistent high product quality, and operational efficiency very crucial. These challenges can be addressed by using effective real-time process monitoring methods for abnormal process conditions.Various data-driven process monitoring approaches, such as principal component analysis (PCA), have been used in various industrial applications. 1-3 The application of multivariate approaches for abnormal events detection has been developed mainly in the area of process monitoring. [4][5][6][7][8][9][10][11] The process monitoring scheme involves building a PCA model of the system under the normal operating condition (NOC). The identification of the PCA model relies on estimating the process structure by using an eigendecomposition problem. 3 Several extensions of the PCA model with different applications were proposed in literature to deal with dynamic and nonlinear behavior of processes. 3,9,12 Abnormal situations that occur due to sensor drifts induce changes in sensor measurements. PCA is used to model normal process behavior, and faults are then detected by checking the observed behavior against this model. 6,12 Hotelling's T 2 and squared prediction error SPE statistic are two statistics commonly used in data-driven process monitoring methods. 3,10,13 Regarding fault detection, several univariate and multivariate techniques have been developed in literature for process monitoring. [14][15][16][17] The multivariate detection techniques include latent variable-based techniques, which are the well-known empirical data-based techniques including partial least squares (PLS) and PCA. [18][19][20][21] The univariate Journal of Chemometrics. 2020;34:e3222.wileyonlinelibrary.com/journal/cem