All systems, when functioning according to given design criteria and given conditions, generate a specific sound with given harmonics. As the system or its associated components age with time, or when the system is having to operate in different non-favourable conditions, its reliability degrades and a different sound is generated. If this sound is decoded early, it can be used as a continuous risk assessment tool for detecting early failures and providing appropriate warning signals for action. The key idea here is that each failure will have a unique digital sound signature, which if detected, could readily direct targeted and timely maintenance strategies and thus contribute to the avoidance of a variety of failures including catastrophic events.
This research focuses on the use of Fast Fourier Transform and associated technologies to develop a persistent monitoring system which continuously analyses sound captured from a system and compares it to a known sound reference dataset in order to provide early risk warnings. Such system can be used in several applications and in many engineering disciplines as it provides a non-intrusive real-time reliability and safety engineering monitoring tool. This research builds on work carried out using sound fingerprinting methods for monitoring daily activity of elderly people living alone at home.
The required methodology will be achieved using various staging methods, including pre-processing, framing, windowing, time/frequency domain feature extraction, and post-processing. Time/frequency domain feature extraction tools include Fourier Transforms (FTs), Modified Discrete Cosine Transform (MDCT), Principal Component Analysis (PCA), Mel-Frequency Cepstrum Coefficients (MFCCs), Constant Q Transform (CQT), Local Energy centroid (LEC), and Wavelet transform. In this research, Fast Fourier Transform (FFT) is applied on 0.1 seconds intervals of the recorded sound with minimisation of the spectral leakage using the Hamming window. The frequency peaks are detected from the spectrogram matrices to get the most appropriate FPM between the reference and sample data.
The required methodology will use the following 5 phases:
- Set-up phase that defines the targeted failures
- Configuration phase that optimises the implementation of the required sensors
- Learning phase whereby sounds data of the targeted failures are collected and stored in a fingerprint reference data set
- Listening phase whereby real-time data is collected and compared against the reference data set to provide information about the given failure, when, and for how long
- Alert phase whereby a reliability or safety risk score is generated in real-time.
The methodology will be enhanced with Artificial Neural Networks (ANNs), Deep Learning (DL) and probabilistic risk assessment algorithms to classify and predict different types of failure.