Fault diagnosis of electrohydraulic actuator based on multiple source signals: An experimental investigation

As a comparatively complicated and compact system with fast response, accurate control precision and high load-bearing capacity, electrohydraulic actuator (EHA) is generally composed of electronic control, hydraulic power, and mechanical drive systems, and has been widely used in aircrafts, mining machines, and transportation vehicles. Although a lot of redundancy designs are used in EHA to improve its operational reliability, failures are still inevitable due to long-term operation and harsh working environments. This paper conducts an experimental investigation on EHA fault diagnosis based on numerical simulation tests and real experimental tests. Multiple source domain signals are sampled from three types of sensors with multiple channels in the EHA's test platforms under variable control commands, thereby showing high redundancy of information. Another challenge is that the fault data sampled from an experimental test platform are more complex than that of the simulated data obtained from the AMESim simulation test platform. These characteristics may cause a huge challenge for traditional fault diagnosis methods. Recent development on deep learning has accelerated many classification tasks because of its end-to-end adaptive learning ability, while the application of deep learning in fault diagnosis of EHAs remains relatively rare. Therefore, a deep convolutional neural network (CNN) is proposed for EHA fault diagnosis, and comparison with several popular data-driven methods are conducted using two datasets sampled from the AMESim simulation test platform and experimental test platform. Among these classifiers, the proposed convolutional neural network is more robust, especially when handling complicated real experimental test data.