Invention Grant
- Patent Title: System and methods for machine anomaly detection based on sound spectrogram images and neural networks
-
Application No.: US17173946Application Date: 2021-02-11
-
Publication No.: US11475910B2Publication Date: 2022-10-18
- Inventor: Martin Byung-Guk Jun , Hanjun Kim
- Applicant: Purdue Research Foundation
- Applicant Address: US IN West Lafayette
- Assignee: Purdue Research Foundation
- Current Assignee: Purdue Research Foundation
- Current Assignee Address: US IN West Lafayette
- Agency: Purdue Research Foundation
- Main IPC: H04R29/00
- IPC: H04R29/00 ; G10L25/51 ; G10L25/18 ; G06N3/08 ; G10L25/30 ; H04R1/08

Abstract:
A system and methods for machine anomaly and behavior classification is provided. An audio capture device may attach to a mechanical apparatus comprising a first component and a second component. The first component and second component may separately generate audible noise. The capture device may include a body, a diagraph disposed in the body, and a microphone. The diagraph may vibrate in response to sound generated by the mechanical device. The microphone may generate a signal in response audio caused by the vibrating diagram. The system may receive the signal generated by the microphone of the audio capture device. The system may determine, based on a machine learning model and the signal, an anomalous event associated with the first component, a second component, or a combination thereof. Alternatively, or in addition, the system may classify operation of the machine based on second machine learning model.
Public/Granted literature
- US20210256991A1 SYSTEM AND METHODS FOR MACHINE ANOMALY DETECTION BASED ON SOUND SPECTROGRAM IMAGES AND NEURAL NETWORKS Public/Granted day:2021-08-19
Information query