Page 76 - IJAMD-1-1

International Journal of AI
for Material and Design Machine learning for gripper state prediction

A B

Figure 7. The kinematics of joint stiffness-tuneable gripper when (A) joint 1 and (B) joint 2 are at the softest state (70°C), respectively.

A B C

D

Figure 8. (A) The mean absolute error of LR, DTR, and KNN from the cross-validation. (B) Spearman correlation of the variables in the dataset. (C) The
mean absolute error of DTR with varying max depth from 3 to 9. (D) The obtained angle from the simulation compared to the input angle in the test set
for LR, DTR, and KNN.
Abbreviations: DTR: Decision tree regression; KNN: K-nearest neighbors regression; LR: Linear regression.

2 is at its softest state (70°C). A comparable trend is noted a sufficient data point for model training, we replaced these
here, where a larger difference in joint temperature leads to data points with an alternative, where the pull distance
increased bending at the softer joint upon cable actuation. is 25 mm. This replacement of data points explains the
The larger angular displacement at the softer joint can be abnormal data point with a pull distance of 25 mm within
explained by the decreased stiffness of the material at higher the dataset, as shown in Figures 6 and 7B. In addition, it is
temperatures, allowing for greater deformation under the crucial to approach the simulation results obtained in this
same load. Concurrently, as the pull distance of the cable work with caution. The joint sections are assumed to have
increases, the ratio between the joint angles gradually a uniform temperature distribution throughout each joint
converges toward 1. This convergence is attributed to the section, which may not be the case in practical applications
geometric constraints of the gripper design, which limit due to heat loss through convection and conduction to the
the extent of differential bending as the joints reach their surrounding environment and structures. Nevertheless,
maximum allowable displacement. It should be pointed this work provides insight into the kinematics of the
out that the simulation solution for certain data points adjustable joint stiffness gripper and lays the groundwork
with a pull distance of 30 mm failed to converge. To ensure for further research, which could incorporate more

Volume 1 Issue 1 (2024) 70 https://doi.org/10.36922/ijamd.2328