CNN and ANN Based Tool Condition Monitoring in Gear Machining Using Audio and Vibration Signals Via Cost Effective Sensors

Effective tool condition monitoring is crucial in modern automated gear machining environments to detect the state of cutting tools and initiate appropriate actions. This paper proposes a way to detect the state of gear milling cutters using audible sound signals and vibration signals through cost-effective sensors. Various gear milling experiments were conducted under a series of cutting conditions to gather signals of audio and vibration. Despite raw audio and vibration signals, as well as their Fast Fourier Transform (FFT), not visually revealing direct fault-related aspects, Mel-Frequency Cepstral Coefficients (MFCC) features of audio signals exhibit noticeable differences between healthy and faulty tools. The assessment of the suggested monitoring system's performance indicates fairly precise forecasts in identifying instances of tool wear during gear milling. The Convolutional Neural Network (CNN) and Artificial Neural Network (ANN) models, trained with a combination of MFCC from the tool audio signal and FFT from the vibration signal, demonstrate higher accuracy when compared to models trained exclusively on either MFCC from the audio signal or FFT from the vibration signal. The feasibility of using the Raspberry Pi to diagnose cutting tools online based on audio and vibration signals via cost-effective sensors was assessed, requiring at most 180 ms for processing.