Diagnóstico via análise sonora da condição de funcionamento e nível de severidade em um redutor coroa sem-fim utilizando redes neurais artificiais

Abstract

This work developed an intelligent diagnostic system for identifying the operational conditions of a worm gear reducer. An experimental setup was built to reproduce different operating conditions and severity levels, starting from the reference state (good condition). Among the tested conditions are angular misalignment, overload, and inefficient lubricat ion. The setup incorporated a data acquisition system for vibration analysis, primarily using an Arduino and an MPU 6050 accelerometer, as well as sound analysis via a smartphone. The vibration data were used to validate the system's dynamics and perform a comparative analysis with the sound signals through Fast Fourier Transform (FFT). For the sound signals, analyses were performed in the time domain using statistical techniques such as Kurtosis, Crest Factor, and RMS; in the frequency domain using the Fas t Fourier Transform (FFT); and in the time frequency domain using the Wavelet Transform. An intelligent diagnostic system was developed, consisting of two neural networks: one responsible for classifying the operational condition (reference state, angular misalignment, overload, and inefficient lubrication) and another for identifying severity levels (mild, medium, and severe). The accuracy achieved for operational condition classification was 98.889%, and for severity classification, it was 98.045%. In cro ss validation, the average accuracy obtained was 97.738% for the operational condition neural network and 97.403% for the severity level neural network. Based on the results, it was found that sound signals, in conjunction with the neural network, were eff ective in determining the operational conditions.