Reconhecimento automático de padrões de defeitos em motobombas utilizando análise de sinais de vibração

Abstract

Predictive maintenance plays an important role for the economy and safety of petroleum systems. Analysis of vibration signals obtained from machines involved in the petroleumextraction process allows subject matter experts to characterize and monitor the situation. However, because of the high cost and the lack of availability of those experts, the existence of automatic systems that support the analysis is desirable. This work presents an automatic procedure to recognize defect patterns in motorpump equipments. A set of techniques previously selected for each stage of the pattern recognition process is applied in the procedure. Signals processing techniques are used to obtain descriptive features from vibration signals. Two approaches are evaluated for the selection of relevant characteristics: using heuristics based on domain specialized knowledge (manual approach) and application of selection algorithms (automatic approach). Real examples are subjected to a supervised learning algorithm in order to compare the manual and the automatic selection approaches.