Reliability control of multichannel telemetry of a robotic manipulator based on an inter-channel consistency graph

Abstract

The paper addresses data quality control in the multichannel telemetry of a robotic manipulator. The aim of the study is to develop and experimentally evaluate a method that detects and localises a corrupted channel without training a neural network and without a physical test bench. The method combines indicators of completeness, temporal correctness and stability of each stream with kinematic and dynamic residuals between the encoders, the drive currents, the inertial channel and the external estimate of the end-effector position. Threshold values are computed over a clean calibration interval using the median and the median absolute deviation; localisation is performed on a graph of inter-channel relations. The reproducible software experiment covers 864 scenarios, six types of distortion, four channels, three intensity levels and twelve repetitions. The proposed method reached a mean F1 of 0,862 with a 95% bootstrap interval [0,845; 0,879], whereas fixed rules and adaptive single-channel control yielded 0,622 and 0,791 respectively. The mean localisation accuracy was 0,843 and the false positive rate was 0,0059. The largest gain was obtained for constant offsets; for outliers and freezing, single-channel methods proved more competitive. The result is an interpretable module for input telemetry control suitable for embedding before data storage and analytical processing.