Design and Implementation of Inline Current Sensing for Artificial Neural Network Based Torque Estimation in an Industrial Robot System

Abstract

Industrial robotic manipulators are typically operated by integrated servo drive systems in which motor currents are continuously monitored for control and diagnostic purposes. However, detailed axis-level analysis remains challenging due to limited access to high-fidelity electrical data, despite its value for energy optimization and trajectory planning. This paper presents the design and implementation of an inline, isolated, multiphase current-sensing system for an industrial robot, positioned electrically in series with the motor phase conductors. The proposed architecture introduces an independent measurement layer at the motor interface, enabling direct observation of motor phase currents while remaining non-intrusive to the existing control system. The sensing is experimentally validated on a SCARA robot, demonstrating stable waveform acquisition and phase consistency under dynamic motion. The measured currents are further analyzed to reconstruct torque-related components. A data-driven approach is employed, in which a function-fitting artificial neural network is used to estimate the electromagnetic torque from the measured current signals, achieving a root mean square error of 0.0175 Nm with respect to controller telemetry. The results demonstrate that the proposed sensing architecture provides a reliable and transparent interface for current measurement in industrial robots, forming a foundation for future extensions toward voltage sensing, power computation, and energy-aware control.