Abstract

The paper presents a generalization of multi-dimensional linear regression to facilitate multi-sensor fault detection and signal reconstruction through the use of analytical optimization. The proposed methodology is founded upon the solution of an optimal signal reconstruction problem. The technique is applied to the real time monitoring of exhaust gas temperature sensors and burner-tip temperature sensors, of a 14 MW industrial gas turbine. Key benefits of the proposed technique are that it facilitates (i) real-time detection of sensor faults and the number of sensors that are at fault in a multi-sensor system; (ii) reconstruction of measurements that would normally be expected from the sensor at fault - thereby facilitating improved unit availability; (iii) determining the minimum number of non-faulty sensors that are required to be available to continue unit operation without unduly compromising performance. The use of an analytical formulation to determine (i-iii) means that the resulting technique incurs low computational overhead and is readily applied to real-time monitoring and subsequent remedial action. Experimental results demonstrate the efficacy of the developed procedures to facilitate continued unit operation in the event of sensor faults. Whilst the application to industrial gas turbine machinery is used to focus the study, it should be noted that the proposed techniques are much more widely applicable to numerous industrial and commercial systems. © 2013 Elsevier Ltd.All rights reserved.