Monitoramento de bombas submersas, do tipo BCSS, através da técnica MCSA, para fins de manutenção preditiva

Abstract

Among the artificial lift methods used in subsea completions in Brazil, particularly in the production of heavy and viscous oils, Electrical Submersible Pumping (ESP) stands out by providing a significant increase in production flow, which can reach up to 100% compared to a well without this system. However, the failure of a Subsea Electrical Submersible Pump (ESP) entails high operational costs, including complex interventions, expensive maintenance, and considerable financial losses associated with lost profits. In this context, the pursuit of strategies to enhance the reliability of these systems becomes essential. Aiming to improve the predictive monitoring of the skid-ESP/ESP system, this work employs Electrical Signature Analysis techniques—an essential approach given the extremely limited physical access to the machine in the subsea environment. This study proposes the identification of fault patterns in the electrical signature of the motors responsible for driving the skid-ESP, as well as the evaluation of the potential of signal separation algorithms in scenarios where two motors are powered in parallel by a single cable. To handle non-stationary signals, advanced digital signal processing techniques were applied, including statistical methods. As a central part of the monitoring strategy, a Bayesian prognostic algorithm was developed. Tests were carried out using acceleration signals from a three-phase induction motor that experienced a bearing failure during monitoring. The results showed that the model was able to estimate the system’s remaining useful life (RUL) with increasing accuracy. Initially, uncertainty in the predictions led to errors of up to 72 hours, which were progressively reduced to zero as the moment of failure approached, reaching its highest precision at a critical point. Although small deviations were observed in the final estimates, the practical impact of these variations proved to be negligible, reinforcing the reliability of the model. Additionally, the Bayesian algorithm was effective in identifying the evolution of a spectral component in the acceleration signal associated with motor degradation, enabling the correlation between spectral variations and the progressive development of system faults. The optimized signal concatenation technique contributed to improving spectral resolution, minimizing issues arising from the non-stationarity of the monitored signals. As a continuation of this study, further prognostic analyses will be conducted using electrical signals from the skid-ESP and the ESP, with the objective of evaluating the model’s ability to anticipate failures compared to other online monitoring techniques used in the operation of submersible pumps.