Publication

Abstract : Underwater electroacoustic sensors used for marine application and underwater pipeline inspection degrade over time due to water ingress through the water-proof polymer encapsulation. The degradation of the sensor can be assessed by measuring the insulation resistance due to the leakage resistance added by polymer over a long time. An experimental study is conducted on a sensor dipped in a sea water bath, measuring insulation resistance in regular intervals of time. The data is analysed using a deep learning algorithm for predicting the end-of-life of the sensor. A type of Recurrent Neural Network (RNN) called Long Short-Term Memory (LSTM) is employed to study the degradation pattern of the sensor, as LSTM-RNN can efficiently learn the long-term dependence of degradation data. The actual end of life of the sensor measured experimentally is compared with that obtained using LSTM-RNN for verification of the model. Main advantage of this study is, this methodology does not require disassembly of sensor from the system to make decisions on maintenance or replacement.