Updated Finite Element Model of Axial Piston Pump for Machine Learning-Based Failure Detection

Axial piston pumps (APP) are energy efficient while operating, but they are prone to catastrophic failures by virtue of their construction and tight tolerances. Reliability can be improved either by (i) increasing the robustness of each component used in the APPs or (ii) predicting their health status by using different techniques, so that failures can be prevented. To predict the characteristics of APPs, the underlying model and physics of the pump operation need to be studied before deploying any of the condition monitoring algorithms. Although studies indicate that deploying a machine learning (ML) algorithm can be helpful, it requires a considerable amount of failure data. The typical nature of failure in APPs is quite catastrophic, and it develops very fast and, consequently, the time to react is very limited. The goal of this work is to achieve a validated FE model of the APP that can subsequently be used for creating simulated failure data. This, in turn, can be used as training data for developing an ML-based failure detection system, where specific features can be tailored to the respective simulated failures. In this paper, model updating techniques are used to validate the FE model against experimental data, which have been obtained from modal testing of an APP and its parts. A simplified APP consisting of a swashplate, housing, end cover, and port flange are considered for the initial experimental testing. Here, the APP is subjected to impact testing, where the structural response is measured using accelerometers. The accelerometer signals are then analyzed using MATLAB to derive the first few modal frequencies.