Detection of Unit of Measure Inconsistency by means of a Machine Learning Model

The evaluation of sensor reliability is fundamental for all types of energy systems. Regardless of the environment in which a sensor is installed, only a reliable measurement can lead to optimal decisions about system operation and health state. In fact, a faulty sensor may provide misleading information for decision making, at the expense of business interruption and maintenance-related costs. In the literature, several checks are performed in order to detect sensor faults, such as out of range, stuck signal, dithering, standard deviation, trend coherence, spike and bias. However, process anomalies have been rarely investigated, though they often introduce errors whereby the unit of measure of a sensor is wrongly assumed. In this paper such a situation is named Unit of Measure Inconsistency (UMI). To this aim, the authors exploit a Machine Learning model, namely a Support Vector Machine, by considering four different approaches for UMI detection and analyzing two field datasets acquired from Siemens gas turbines. Among all tested approaches, the Radial Basis Function with One-vs-One decomposition proved to be the most effective approach, since it demonstrated its effectiveness and robustness in the majority of the analyses. Thanks to the selected approach, the actual information regarding sensor unit of measure can be provided and further sensor and engine diagnoses can be safely performed.