ARTIFICIAL INTELLIGENCE ENABLED MACHINERY FAULT DETECTION AND DIAGNOSIS USING VIBRO-ACOUSTIC SIGNALS
Author ORCID Identifier
https://orcid.org/0000-0002-7404-4335
Date Available
8-8-2025
Year of Publication
2024
Document Type
Doctoral Dissertation
Degree Name
Doctor of Philosophy (PhD)
College
Engineering
Department/School/Program
Mechanical Engineering
Advisor
Dr. David W. Herrin
Co-Director of Graduate Studies
Dr. Peng Wang
Abstract
In various industries, the early detection of faults in rotating machinery is crucial to prevent system failures and ensure customer satisfaction. Typically, vibration measurement and diagnosis are employed for fault detection, but this process faces challenges in automation due to the complexity of installing and maintaining accelerometers, particularly in end-of-line quality control or pre-installed machinery health assessments. Acoustic signals, as a form of mechanical wave, offer an alternative for monitoring machinery while in operation. Unlike accelerometers, acoustic transducers are non-contact and easy to set up, enabling real-time data collection without interrupting equipment operation. However, utilizing acoustic signals in manufacturing poses challenges, including the scalar nature of traditional sound pressure measurements, susceptibility to background noise, and the need for extensive manual intervention in data analysis.
The advent of artificial intelligence (AI) has revolutionized various industries, offering advanced capabilities in fields like computer vision and natural language processing. Machine learning (ML) and deep learning (DL) have particularly facilitated the development of innovative algorithms for enhancing design, manufacturing, and quality control processes. AI-driven diagnostic models now exhibit remarkable accuracy, often exceeding human performance.
This dissertation proposes three AI-based fault detection methodologies leveraging vector-based acoustic sensing equipment to address manufacturing challenges. Firstly, supervised ML methods are employed for fault classification, using vibro acoustic signal features extracted by experts to train models to detect anomalies in machinery samples. Secondly, a DL approach using one-dimensional convolutional neural networks is proposed to process raw acoustic time sensing data, eliminating manual feature extraction and reducing preprocessing latency. This DL method proved effective in detecting mechanical faults in electric motors, offering a viable alternative to accelerometer-based approaches.
However, the success of DL methods is highly dependent on the quantity and availability of data, which can be limited when transitioning to new acoustic sensor technology. Transfer learning is introduced as a solution, allowing DL models trained on accelerometer signals to be adapted for acoustic signals. This domain adaptation methodology facilitates the use of historical accelerometer-based data when training DL models for acoustic signal analysis.
Overall, these methods aim to streamline part quality inspection in manufacturing testing frameworks and lay the groundwork for analyzing multi-domain sensing data for future machinery fault detection, potentially incorporating additional data types such as laser displacement, temperature, pressure, and imagery.
Digital Object Identifier (DOI)
https://doi.org/10.13023/etd.2024.334
Funding Information
This work was supported by Vibro-Acoustics Consortium and National Science Foundation Grant No. 2015889 from 2020-2024.
Recommended Citation
Ippili, Srinivasa Rao, "ARTIFICIAL INTELLIGENCE ENABLED MACHINERY FAULT DETECTION AND DIAGNOSIS USING VIBRO-ACOUSTIC SIGNALS" (2024). Theses and Dissertations--Mechanical Engineering. 223.
https://uknowledge.uky.edu/me_etds/223
Included in
Acoustics, Dynamics, and Controls Commons, Artificial Intelligence and Robotics Commons, Manufacturing Commons, Signal Processing Commons
