Título: Armature fault diagnosis of universal motors using time-series data in neural networks
Autores: Guilherme da Silva, Bernard da Silva, Douglas Wildgrube Bertol
Resumo: A common issue of DC and universal motors is the quality of the armature circuit and the commutation process. Some faults that could happen during the final assembly process, such as a short circuit between two segments or a broken segment in the armatures commutator, may not be detected by a visual inspection or the regular no-load end-of-line test. A more detailed analysis, using the feedback of various auxiliary sensors, usually requires implementing systems with more complex and expensive hardware. It also may take longer to get a precise analysis, which may not be suitable for the production process. As a method to do a fast and accurate detection and classification of faults in the armatures circuit of universal motors, this paper suggests the analysis of the armature voltage and current waveform by an artificial neural network (ANN). The current and voltage signals acquired during the no-load test formed a time-series waveform that was applied to an ANN trained to classify the signal in three possible outcomes: healthy armature, armature with shortcircuited segments, and armature with a broken part. The results achieve effectiveness above 98% in the task of detection and classification of faults. The solution was tested in the production line of universal motors and proved to be fast, reliable, and efficient for the usual no-load test, proving to be suitable in an industrial context
Palavras-chave: Artificial neural networks, fault diagnosis, universal motors
Páginas: 8
Código DOI: 10.21528/CBIC2023-020
Artigo em pdf: CBIC_2023_paper020.pdf
Arquivo BibTeX: CBIC_2023_020.bib