A novel model-based fault detection method for temperature sensor using fractal correlation dimension

The direct residual-based fault detection method compares the difference between measured and estimated data of a process variable. Its correct fault detection rate is low due to the noise in measured signals. A novel method using fractal correlation dimension (FCD) is developed, in which FCD deviation is adopted instead of direct residual. The method is validated by detecting fixed and drifting bias faults generated in supply air temperature sensor of air handling unit (AHU) system. The setting of three main parameters including embedding dimension, time delay parameter and length scale, is investigated to find out the influence on calculating FCD values. The results show that it is more efficient to detect relatively small bias fault under noise conditions, although it needs a period of time to collect measured data. As a promising and practical tool, a hybrid fault detection technique combining FCD with direct residual should be conducted in further investigation to identify the generated fault under inevitable noise conditions.