Fault diagnosis for sensors in air handling unit based on neural network pre-processed by wavelet and fractal

This paper presents a new fault diagnosis method for sensors in an air-handling unit based on neural network pre-processed by wavelet and fractal (NNPWF). Three-level wavelet analysis is applied to decompose the measurement data, and then fractal dimensions of each frequency band are extracted and used to depict the failure characteristics of the sensors. With these procedures, a signal is extracted into an eigenvector which consists of several fractal dimensions. Following, the eigenvector is introduced into a neural network developed and trained to diagnose the sensor faults. When new measurement data are obtained, similar way is applied to get the eigenvector and the prediction. By comparing the prediction with the objective vectors, the sensor faults can be diagnosed. The fault diagnosis method has been validated and the results show that the proposed method can diagnose different kinds of fault conditions efficiently. Moreover, comparing to the previous work, there is an increase in diagnosis efficiency as large as 15% for the same type of fault.

► A new FDD method so-called NNPWF is proposed.
► Integrating two fractal dimensions greatly improves the diagnosing performance.
► Requiring as less as 10 samples for training.
► Increasing of diagnosis efficiency by as large as 14% compared to the previous work.
► The new method is validated efficiently to diagnose sensor faults.