An online method for condition monitoring and prognostics of hydraulic systems

Engineering systems, such as aircrafts, hydraulic, electronic and electrical systems are becoming more complex and are subjected to failure modes that impact adversely their reliability, availability, safety and maintainability. In particular, hydraulic systems are challenging from the condition monitoring point of view due to the non-linear equations that describe behavior of the fluid. For this reason, numerous efforts have been made to improve the reliability of hydraulic system, leading to the development of complex algorithms for diagnostics and prognostics. In the present research, both diagnostic and prognostic algorithms have been developed, considering the case of an hydraulic crane available at the Purdue's Maha Fluid Power Research Center. Based on the architecture, three components have been analyzed as possible faults in the system: the fixed-displacement pump, the meter-in valve and the cylinder. Among all the modern approaches, a data-driven, neural network based method has been exploited based on a simulation model of the machine through which the behavior of the system is predicted. Moreover, a realistic simulation has been designed, in order to be as close as possible to the real system set-up. Then, a validation of this approach has been performed on the target machine. In summary, the diagnostic algorithm is capable to understand the intensity of the fault and to discern which is the component that is failing also during simultaneous failures; the prognostic algorithm can properly estimate the Remaining Useful Life (RUL) based on the Weibull distribution. Notably, both algorithms use a limited set of sensors taking also advantage of the implemented controller.