A Novel 2-D Current Signal-Based Residual Learning With Optimized Softmax to Identify Faults in Ball Screw Actuators

Abstract

Ball screw electro-mechanical actuators are commonly found in high precision motion control applications including aerospace systems as well as automated setups for industries. These actuators perform flight / application critical job and ball screw drives are responsible to provide precise linear motion while carrying thrust loading. A failure in ball screw drive may disturb positioning accuracy of overall system. At present, few techniques are available to monitor electro-mechanical actuators for aerospace and industrial systems. This paper provides a deep learning based intelligent technique to monitor condition of ball screw actuators. The proposed scheme utilizes modified residual learning scheme to extract features from two-dimensional transformed motor current signals. The current signal data was collected under different load domains in terms of magnitude and direction reversal. A 2D-Remanant-CNN (2D-Rem-CNN) model was developed for features extraction with proposed optimized softmax for classification of mechanical faults. The proposed technique was validated against different ball screw fault cases. The testing results prove the superiority of 2D-Rem-CNN model against different state of the art techniques. The proposed framework was also tested for system's stability under different load domains.