Digital Twins for Mechanical Condition Assessment of Reusable Surgical Instruments
L.B. van Genderen (TU Delft - Mechanical Engineering)
T. Horeman – Mentor (TU Delft - Mechanical Engineering)
R. Horeman – Mentor (TU Delft - Mechanical Engineering)
J.M. Prendergast – Graduation committee member (TU Delft - Mechanical Engineering)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Assessing the mechanical condition of reusable robotic surgical instruments is important for determining whether they remain suitable for continued clinical use. This study evaluates a Digital Twin framework for predicting healthy behaviour and detecting mechanical deviations in the Shaft-Actuated Tip Articulating (SATA) system. Data-driven Motor DTs predicted encoder positions and currents, a physics-based Gearbox DT estimated internal shaft rotations and translation, and physics-based and hybrid Instrument DTs predicted instrument tip motion. Motor predictions were compared with recorded motor signals, while instrument motion predictions were compared with video-based measurements of tip bending and jaw opening.
A healthy-condition benchmark using leave-one-run-out cross-validation of the selected models gave an encoder mean absolute error (MAE) of 1.41 ± 0.77 pulses and a current MAE of 4.57 ± 1.35 mA for the full-system configuration with the gripper. The hybrid bending model reduced MAE from 5.59 ± 0.73° to 1.63 ± 0.48°, a reduction of 70.8%.
Three experiments examined increased gearbox resistance, increased instrument-shaft resistance and scissor-blade abrasion. Linear discriminant analysis (LDA) with cross-validation was used to assess whether current residual parameters could distinguish healthy and degraded conditions. For gearbox resistance, M0 current residual RMS alone correctly classified all held-out observations at all three window lengths (five, ten and thirty M0 revolutions) in leave-one-run-out cross-validation across three healthy and two degraded recordings.
LDA using M1 absolute current residual peak correctly classified 100% of the 216 bending cycles from four recordings. For scissors, selected parameter pairs correctly classified 100% of the 28 observations from cuts 2 and 3 as healthy or degraded. Three-class classification correctly labelled 78.6% of observations (22/28) when all intermediate abrasion levels were included.
These exploratory results demonstrate the potential of using predictions of healthy behaviour for detecting mechanical changes under controlled conditions. The framework provides a basis for developing standardised mechanical condition tests to support inspection and maintenance decisions for reusable surgical instruments.
Files
File under embargo until 27-11-2028