Industrial application of hydrophone for condition monitoring of water pump

The hydrophone, also known as an underwater microphone, is placed in industrial tanks and converts sounds under water (or other liquids) into an electrical analogue signal. The bandwidth is typically the audible range, but they can also operate in the ultrasonic range, up to 1 MHz. The analogue signal from the hydrophone is digitized and the time-domain wave file is processed using Fast Fourier Transform to produce a power spectrogram, which is one of the key inputs for further data and signal processing. Artificial Intelligence Deep-Learning techniques help in the machine processing of large amounts of data and automatic classification. The Python package Librosa utilized for the features extraction of audio files. The waterfall spectrogram and RMS are basic tools in this analysis. The csv, numpy and pandas python packages are used to convert .wav files into .csv files on which data processing can be performed. The sklearn python package is used for preprocessing, and the keras and tensorflow python packages are used for model training and evaluation, and for building the neural networks. The trained model can be used to estimate which previous classified type a new sound sample may belong to. The precision of the model improved by raising the number of samples, but only to a certain limit, which must be determined appropriately.