Development of internal sound sensor using stethoscope and its applications for machine monitoring

With the development of Internet of Things (IoT), predictive maintenance (PdM) for smart manufacturing receives attentions recently. For the monitoring of machines with rotary components, sound and vibration emitted from the machines have been utilized as the meaningful information. However, sound sensors are susceptible to external noise and the costs for signal conditioning should be considered. In this paper, a stethoscope is utilized as an internal sound sensor which are capable of noise reduction and affordable as IoT solutions. After the stethoscope was assembled with a microphone, a frequency range for measurement was identified as 0~255Hz using the swept-sine technique. Two stethoscope sensors were attached to an industrial robot, and the capability of noise reduction is verified by similarity of Short-Time Fourier Transform (STFT) spectrograms from sounds with and without existence of factory noise. Next, impact tests were performed to detect collision when the robot was stationary, and collisions by human hand during robot operation were detected from STFT spectrogram analysis. Finally, the Convolutional Neural Network (CNN) model from the spectrograms was established to estimate a rotating joint axis during operation, showing 91% and 90.83% accuracy for each stethoscope, respectively.