On-line monitoring and diagnosis of membrane fouling using ultrasonic techniques

•Easy collection of the on-line fouling status using non-invasive UTDR.•l Wavelet transform to extract UTDR signals to well represent the features.•l Decision tree based on UTDR features for early detection of the system anomaly.•l Successful application of the proposed method to a real membrane filtration system.

In membrane filtration, fouling and its performance can be affected by different operation conditions, so there is a need for diagnosis of the filtration system. It is difficult to identify the status of the filtration simply from sensor reading; in addition, some observation data cannot be obtained online but from the lab test. In this research, the ultrasonic detection of the local region of the membrane is proposed. The wavelet packet transform is used for feature extraction of the reflection signal. Principal component analysis with Gaussian smoothing is then used to classify the data and C4.5 is used to construct the diagnosis system for filtration processes. The proposed method applied to a degummed process of vegetable oil shows a good performance.