Advanced digital signal processing technique for asset health monitoring

Abstract

Asset health monitoring application is critical to ensure structures are operating in a healthy state and damages can be detected earlier for efficient maintenance scheduling to save cost. RAC Foundations has reported that there were more than three thousand bridges identified as substandard bridges in the United Kingdom in 2019. This is due to destructive damage such as deterioration, corrosion, overloading and lack of maintenance etc. These factors have accelerated the deterioration of bridges, reduced performance in normal operation and also potential hazard of bridges at risk of collapse. A figure of an estimated £6.7 billion bill is required to bring substandard bridges back to good condition. In order to improve the bridge health monitoring process and save cost with well-planned inspection schedule, the development of efficient damage detection algorithms is needed. This study has investigated the current state-of-art in bridge health monitoring applications in terms of types of signals, sensor development and signal processing techniques to extract damage sensitive features. Based on the research outcome in terms of accuracy and efficiency of applications, the author has proposed an advanced digital signal processing technical to develop a vibration-based fault analysis algorithm for early damage detection using optimal filtering. A Spectral Kurtosis (SK) based optimal filter is designed to extract frequencies that are generated by damages. The proposed technique is validated by two applications to detect small defects such as bolt looseness on bolted joint structures as well as applying the technique on bearing fault detection. The two applications has proved that the proposed technique can detect damages in both stationary infrastructure such as bridges and operating rotary machine to show its versatility. The results provide confidence that Spectral Kurotsis is capable of detecting non-linear and non-stationary components buried in noisy signal. The outcome has contributed a method to detect small defects that are hard to find and therefore early repair and better maintenance schedule can be achieved to save the cost and ensure structures are operating in a healthy status. Apart from that, the developed algorithm can also be used to extract features for an automated damage detection application; for instance, feeding the damage sensitive features into machine learning models such as support vector machine and random forest to classify damages. This combined method will reduce the chance of false alarm.