Constantly increasing decarbonization requirements lead to a distribution of electric technologies in Non-road Mobile Machinery (NRMM). Traditional Internal Combustion Engine (ICE)-driven systems are replaced by Electric Motor (EM)-based solutions, not only in powertrains but also in working hydraulics. An example of such a replacement is Direct Driven Hydraulics (DDH). DDH systems have already demonstrated great dynamics, reliability, robustness, and fast response in early studies. In addition, the electric motor also enables the implementation of condition monitoring via Artificial Intelligence (AI) techniques such as Machine Learning (ML). This approach avoids the need for hydraulic sensors completely. This study raises the questions: how accurately can a hydraulic failure be determined using only the signals obtained from the electric part of the DDH and how can the architecture of an electrical drive affect the final result?

A DDH system model with the detailed check-valve model (under healthy and faulty conditions) was implemented in MATLAB Simulink. The electric part of the model consists of a Permanent Magnet Synchronous Machine (PMSM) model and various structures of the electric drive. As it turned out, the electric drive and especially the inverter have a direct impact on the fault detection of the hydraulic parts of the system. Thus, this paper considers the influence of different types of inverters (i.e., two-level, three-level without optimized switching, and three-level with optimized switching) on check valve fault detection via electric signals. The complete model is used for the generation of data, which is fed to the AI-based classifier for training, validation, and testing. This process is repeated for the system with different types of inverters. The simulation experiment demonstrated that a three-level inverter-fed system illustrates the highest accuracy (86.9%) of fault detection. The accuracy difference between two-level and three-level systems varies around 5–15%. The overall accuracy of fault detection is between 72–87%.

This content is only available via PDF.
You do not currently have access to this content.