Structural analysis based sensor measurement fault diagnosis in cement industries

• Published in: Control engineering practice Vol. 64; pp. 148 - 159
• Main Authors: Gomathi, V., Srinivasan, Seshadhri, Ramkumar, K., Muralidharan, Guruprasath
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-07-2017
• Subjects: Cement kiln; Fault diagnosis; Measurement faults; SaTool; Structural analysis; Structural model; Fault diagnosis; Measurement faults; Structural model; Cement kiln; SaTool; Structural analysis
• ISSN: 0967-0661; 1873-6939
• DOI: 10.1016/j.conengprac.2017.02.012

This investigation presents a fault diagnosis methodology for detecting sensor faults in cement industries pyro processing section. It works in three steps: (a) modelling, (b) analysis, and (c) validation. In the modelling, the actual data from the cement pyro processing is used to do a correlation analysis between output and input variables. The structural model is obtained from the correlation tests. During the analysis phase the Structural analysis Tool (SaTool) is used to detect the detectability and isolability of the faults. The results of the structural analysis are validated in a cement industry using residual analysis performed using structural sensor model and real-time measurements. The main advantages of this fault diagnosis technique are: (a) it requires only correlation analysis to obtain the structural model without a detailed physical model as in other methods, (b) conclusions regarding detectability and isolability can be easily drawn during the analysis stage itself, and (c) the method is simple compared to the model-based, and data-history based methods. The effectiveness of the proposed method is illustrated using data from cement pyro processing plant and its performance is compared with model based approaches for four different types of sensor faults: (1) bias, (2) drift, (3) stuck, and (4) measurement failures. Our results demonstrate that the structural method is able to detect the sensor faults even in the presence of noisy information, and its performance is comparable with that of model based approaches without employing a physical model.