Banca de QUALIFICAÇÃO: DANIELA MAIA DA SILVA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : DANIELA MAIA DA SILVA
DATE: 22/07/2025
TIME: 14:00
LOCAL: Auditório do NPITI
TITLE:

Fault Detection in Industrial Pumps from Acoustic Signals Using Cyclostationary Analysis and Unsupervised Learning

KEY WORDS:

Fault detection, Industrial pumps, Spectral correlation density, Machine learning.

PAGES: 42
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

Early fault detection in industrial machinery is essential for predictive maintenance strategies, reduction of operational costs, and accident prevention. Among emerging methods, the analysis of acoustic signals stands out, as it enables the identification of changes in equipment operation without physical contact. However, the presence of noise and the spectral complexity of these signals hinder the application of conventional techniques. This work proposes a methodology for fault detection based on acoustic signals, combining cyclostationary analysis via Cyclic Spectral Density (SCD) with machine learning techniques. SCD is used to extract $\alpha$-profiles, which highlight cyclic modulations in the signal related to mechanical operation and, in some cases, to faults. Then, the Random Forest algorithm is applied to automatically select the most relevant attributes, which are used as input to an autoencoder trained to model the system’s normal behavior and detect anomalies based on reconstruction error. The methodology will be validated using acoustic data from industrial pumps, extracted from the MIMII dataset, which contains recordings under normal and faulty conditions, across different noise levels. As a complementary step, cepstral coefficients will be computed from the frequency spectra corresponding to the most relevant cyclic frequencies, representing a spectral compression strategy guided by the cyclostationary properties of the signal. The aim is to develop a robust, interpretable, and generalizable model applicable to diverse operating conditions of industrial pumps.

COMMITTEE MEMBERS:
Presidente - 1543191 - LUIZ FELIPE DE QUEIROZ SILVEIRA
Interno - 2579664 - ALLAN DE MEDEIROS MARTINS
Externo ao Programa - 1640260 - JOSE JOSEMAR DE OLIVEIRA JUNIOR - UFRN