Abstract
This paper presents the framework of a physics-informed neural network (PINN) with a boundary condition-embedded approximation function (BCAF) for solving common problems encountered in flexible mechatronics and soft robotics; both forward and inverse problems are considered. Unlike conventional PINNs that minimize a lumped loss function including the errors contributed by the initial or boundary conditions (ICs or BCs), the BCAF-PINN completely satisfies the ICs and/or BCs while minimizing a loss function for parameter identification and boundary force estimation, overcoming a common erroneous-convergence problem due to unbalanced gradients in training a PINN. The formulation and implementation of a BCAF-PINN are illustrated with three practical applications, including a nonlinear system where solutions are available for numerical verification and for comparing with conventional PINNs, and a biomechanical system where a BCAF-PINN uses multiple cycles of natural foot flexion to identify its dynamic parameters experimentally. While overcoming several problems associated with traditional studies based on perturbation models with certain level of muscle contraction, the damping ratio identified by the BCAF-PINN indicates that the ankle joint is an overdamped system during flexion, consistent with that observed in published experiments.
Similar content being viewed by others
Abbreviations
- ANN:
-
Artificial neural network
- BC:
-
Boundary condition
- BCAF:
-
BC-embedded approximation function
- BCLF:
-
BC loss function
- BVP:
-
Boundary value problem
- DOF:
-
Degree of freedom
- DF:
-
Dorsiflexion
- FM:
-
First metarsal heads
- FSM:
-
Full state measurement
- HF:
-
Head of fibula
- IBK:
-
Inertia I, viscous B and elastic K model
- IC:
-
Initial condition
- IVP:
-
Initial value problem
- LM:
-
Lateral malleolus
- MM:
-
Medial malleolus
- MSO:
-
Multiple single output
- ODE:
-
Ordinary differential equation
- PDE:
-
Partial differential equation
- PF:
-
Plantarflexion
- PINN:
-
Physics informed neural network
- PSM:
-
Partial state measurement
- RC:
-
Ridge of the calcaneus
- SMO:
-
Single multiple output
- SS:
-
State space
- VM:
-
Fifth metarsal heads
- A :
-
ICs/BCs satisfied function entry
- E, G :
-
Elastic and shear modulus
- F :
-
ICs/BCs no contribution function entry
- I :
-
Area moment of inertia
- J :
-
Average angular moment of inertia of foot
- L :
-
Length
- N :
-
Network output entry
- M :
-
Number of measurements
- A :
-
ICs/BCs satisfied function vector
- F :
-
ICs/BCs no contribution function vector
- N :
-
Network output vector
- P C :
-
Position vector of beam end point
- a :
-
Initial/boundary conditions
- b, h :
-
Beam width, thickness
- s, t :
-
Path-length,, time
- x, y :
-
Independent, dependent variable
- y a :
-
BCAF dependent variable entry
- \(\hat{y}_{a}\) :
-
BCLF dependent variable entry
- u :
-
Applied force
- u 1, u 2 :
-
Applied force in y1, y2 directions
- u M :
-
Ankle joint net torque
- e 0 :
-
Initial error
- e b :
-
Boundary error
- e i :
-
Measurement error
- p :
-
Network weights and biases
- r p :
-
Residual
- x, y :
-
Independent, dependent variable vectors
- y a :
-
BCAF dependent variable vector
- \({\hat{\mathbf{y}}}_{a}\) :
-
BCLF dependent variable vector
- Γ:
-
Structural rigidity
- \({{\mathcal{G}}}\) :
-
Differential operator
- \({{\mathcal{L}}}_{\text{F}}\) :
-
BCAF forward loss function
- \({{\mathcal{L}}}_{\text{I}}\) :
-
BCAF inverse loss function
- \({\hat{\mathcal{L}}}_{\text{F}}\) :
-
BCLF forward loss function
- \({\hat{\mathcal{L}}}_{\text{I}}\) :
-
BCLF inverse loss function
- α, α :
-
Coefficient entry/vector
- λ :
-
Magnitude tuning factor
- ρ,ν :
-
Density and Poisson ratio
- ψ,κ :
-
Angle of rotation and curvature
- η :
-
Initial angle of rotation
- θ :
-
Foot flexion angle
- ζ,ω n :
-
Damping ratio and natural frequency
References
Chen, Y., Lu, L., Karniadakis, G.E., Dal Negro, L.: Physics-informed neural networks for inverse problems in nano-optics and metamaterials. Opt Express 28(8), 11618–11633 (2020)
Gunes Baydin, A., Pearlmutter, B.A., Andreyevich Radul, A., and Siskind, J.M.: Automatic differentiation in machine learning: a survey. arXiv e-prints, p. arXiv:1502.05767, 2015.
Guo, J., Lee, K., Zhu, D., Yi, X., Wang, Y.: Large-deformation analysis and experimental validation of a flexure-based mobile sensor node. IEEE/ASME Trans. Mechatron. 17(4), 606–616 (2012)
Hornik, K., Stinchcombe, M., White, H.: Multilayer feedforward networks are universal approximators. Neural Netw. 2(5), 359–366 (1989)
Jiang, J., Li, W., Lee, K.-M.: a novel pantographic exoskeleton based collocated joint design with application for early stroke rehabilitation. IEEE/ASME Trans. Mechatron. 25(4), 1922–1932 (2020)
Jiang, J., Li, W., Lee, K., Ji, J.: Physics-based Ankle Kinematics for Estimating Internal Parameters. In IEEE/ASME Int. Conf. advanced intelligent mechatronics. Hong Kong, China, pp. 471-476 (2019)
Kearney, R.E., Hunter, I.W.: Dynamics of human ankle stiffness: variation with displacement amplitude. J. Biomech. 15(10), 753–756 (1982)
Lagaris, I.E., Likas, A., Fotiadis, D.I.: Artificial neural networks for solving ordinary and partial differential equations. IEEE Trans. Neural Netw. 9(5), 987–1000 (1998)
Lan, C.-C., Lee, K.-M.: generalized shooting method for analyzing compliant mechanisms with curved members. J. Mech. Design 128(4), 765–775 (2006)
Lee, H., Krebs, H.I., Hogan, N.: Multivariable dynamic ankle mechanical impedance with active muscles. IEEE Trans. Neural Syst. Rehabil. Eng. 22(5), 971–981 (2014)
Mirbagheri, M.M., Barbeau, H., Kearney, R.E.: Intrinsic and reflex contributions to human ankle stiffness: variation with activation level and position. Exp. Brain Res. 135(4), 423–436 (2000)
Misgeld, B.J., Zhang, T., Luken, M.J., Leonhardt, S.: Model-based estimation of ankle joint stiffness. Sensors 17(4), 713 (2017)
Petri, E., Hao, G., Kavanagh, R.C.: Design and hybrid control of a two-axis flexure-based positioning system. Int. J. Intell. Robot. Appl. (2021). https://doi.org/10.1007/s41315-021-00162-7
Pun, G.P.P., Batra, R., Ramprasad, R., Mishin, Y.: Physically informed artificial neural networks for atomistic modeling of materials. Nat. Commun. 10(1), 2339 (2019)
Raissi, M., Perdikaris, P., Karniadakis, G.E.: Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J. Comput. Phys. 378, 686–707 (2019)
Rastgaar, M.A., Ho, P., Lee, H., Krebs, H.I., Hogan, N.: Stochastic estimation of multi-variable human ankle mechanical impedance. ASME Dyn. Syst. Control Conf. 2, 45–47 (2009). (Hollywood, California, USA)
Thomas, T.L., Kalpathy Venkiteswaran, V., Ananthasuresh, G.K., Misra, S.: Surgical applications of compliant mechanisms: a review. J. Mech. Robot. (2021). https://doi.org/10.1115/1.4049491
Wang, S., Teng, Y., Perdikaris, P.: Understanding and mitigating gradient pathologies in physics-informed neural networks. arXiv e-prints, arXiv:2001.04536, 2020
Wang, J.-Y., Lan, C.-C.: A constant-force compliant gripper for handling objects of various sizes. J. Mech. Design 136, 071008 (2014)
Weiss, P.L., Kearney, R.E., Hunter, I.W.: Position dependence of ankle joint dynamics—I. Passive mechanics. J. Biomech. 19(9), 727–735 (1986)
Wessels, H., Weißenfels, C., Wriggers, P.: The neural particle method—an updated Lagrangian physics informed neural network for computational fluid dynamics. Comput. Methods Appl. Mech. Eng. 368, 113127 (2020)
Yazdani, A., Lu, L., Raissi, M., Karniadakis, G.E.: Systems biology informed deep learning for inferring parameters and hidden dynamics. PLoS Comput. Biol. 16(11), e1007575 (2020)
Acknowledgements
This work was supported in part by the U.S. National Science Foundation under Grant CMMI-1662700.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Li, W., Lee, KM. Physics informed neural network for parameter identification and boundary force estimation of compliant and biomechanical systems. Int J Intell Robot Appl 5, 313–325 (2021). https://doi.org/10.1007/s41315-021-00196-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s41315-021-00196-x