Design of patterns in tubular robots using DNN-metaheuristics optimization
Graphical abstract
Introduction
Concentric-tube robots represent a type of continuum tubular robot comprising concentrically assembled precurved elastic tubes [1,2]. They are particularly well-suited for minimally invasive surgery because they can assume curves of complex shapes in three-dimensional space and possess sufficient stiffness to steer tissue and manipulate tools inside body cavities [3,4]. The most distinctive characteristic of this type of robot is that its distal end is steered and advanced through the rotation and translation of the precurved tubes relative to each other. Actuators are attached only to the proximal ends of the tubes, affording mechanical simplicity and small-scale design.
However, under certain conditions, torsion of the precurved tubes causes structural instability. Torsional deformation caused by the relative rotation of the precurved tubes is unavoidable and causes a lag in the transmission of the rotation angle. In particular, the tubes may abruptly move from one minimum-energy point to another during operations, which is often called the “snapping problem” [4,5]. In prior research [4,5], the condition for stability has been derived. It can be summarized aswhere L is the overlapping length, and κ is the curvature of the tube. In Eq. (1), EI and GJ represent bending and torsional stiffness, respectively. This condition shows that the key factor for eliminating instability is the low ratio of bending stiffness to torsional stiffness (EI/GJ).
In this study, we aim to optimize the anisotropic pattern design of continuum tubular robots to solve the instability problem. An optimization process was devised to achieve this goal using a deep neural network (DNN)-metaheuristic optimization framework. The main contribution of this study is the development of an optimization methodology for reducing the EI/GJ of tube robots while satisfying the minimum compliance constraint for securing the desired tube performance. A DNN model was trained as a finiteelement (FE) surrogate from the data generated by the FE analysis (FEA) of a topology-optimized tube pattern. Metaheuristic-based optimization of the tube design variables with this DNN-based surrogate model is used to generate an optimized design.
Many studies have been conducted to solve the snapping problem [5], [6], [7], [8], [9]. An intuitive and trivial solution limits the curvature but restricts the workspace. Recently, intensive research has been conducted on the patterning methods. Researchers have refined the material properties to solve instability problems [10], [11], [12]. Laser machining of through-hole patterns on precurved tubes is an efficient technique to change the structural bending and torsional stiffness of the tube. Several patterning methods for continuum tubular robots have been studied to provide an anisotropic mechanical property [10,[13], [14], [15], [16]].
Fig. 1 summarizes previous designs for through-hole patterning on tubes used to assign anisotropic mechanical properties. Different pattern shapes and distributions were employed for the applications in each study. Dog bone shapes [12], multilayer helical tubes [11], and long horizontal slit patterns [10,17] have been proposed to mitigate the snapping issue. A horizontal split pattern [15] was used to maximize the variable stiffness functionality. However, in these studies, the general pattern shape was determined based on engineering intuition.
Determining the general pattern shape is the starting point of the parametric design optimization process. The basic shape significantly affects the final performance of the patterned tube; therefore, selecting the general shape through an optimization process, rather than an intuitive selection, may be beneficial. Topology optimization has been proposed in various fields (aerospace, mechanical, biomechanical, and civil engineering) [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30]. Luo et al. [14] first attempted to employ topology optimization to identify a pattern shape that reduces EI/GJ. Although the performed topology optimization reached non-converged results due to the mesh dependence or local minimum problem [18], a diamond shape was determined, and a low EI/GJ value (0.14) was obtained. Park et al. [16] also determined rectangular columns as the general shape of the pattern, by utilizing topology optimization to maximize the variable stiffness functionality, where three different setups for the optimization were considered to confirm the convergence of the results.
Once the general shape of the pattern has been obtained and the design variables have been determined accordingly, a follow-up parametric study is necessary to determine the optimal values of the variables. However, because of the large number of possible combinations, this parametric study has often been performed through limited sensitivity analysis with the assumption of independence of variables [[10], [11], [12],14,15,17,31,32].
Recently, research has been conducted on the full search of the design space to optimize the pattern shape with an automated FE analysis [13,16]. Researchers have attempted to search a design space fully for auxetic [13] and rectangular [16] patterns. In [16], the design for the given stiffness constraints was optimized manually using the results of a fully searched design space.
However, as the shape of the pattern becomes more complex, the numbers of design variables and possible combinations increase exponentially. The design space to be searched is excessively wide in the case of the optimization problem for multivariable functions. Therefore, optimizing the design using the existing fullsearch method is time consuming. In addition, manual optimization based on the results of the fully searched space still requires appropriate engineering decisions.
Design optimization algorithms have been frequently used in various fields of science and engineering to solve optimization problems [33], [34], [35], [36], [37], [38], [39], [40], [41] converted from real-world application problems [42], [43], [44], [45]. Gradient-based optimization methods, such as solid isotropic material with penalization method or ground structure approach [20], are highly efficient in searching for optimal solutions based on gradient information. However, finding a global optimal solution through gradient-based optimization when solving constrained nonlinear problems is challenging. The optimization results are highly sensitive to the initial values and are prone to becoming trapped in the local optima. In contrast, metaheuristic-based optimization methods, such as genetic algorithms (GAs) [46], particle swarm optimization (PSO) [47], ant colony optimization (ACO) [48], wild horse optimizer [49], growth optimizer [50], and teaching–learning-based optimization (TLBO) [51], may effectively find the global optimum in the entire design space. However, these methods require thousands or even millions of objective evaluations. To reduce the computational cost of numerous evaluations during optimization iterations, a surrogate model is required instead of an FE method (FEM)-based evaluation. Objective evaluations performed using surrogate models, such as polynomial response surfaces, support vector machines, and artificial neural networks (NNs), can provide a global optimal solution within a reasonable amount of time [52], [53], [54], [55], [56], [57], [58].
Section snippets
Proposed optimization procedure
The overall procedure of the framework is outlined as follows.
Step 1. Determination of the general shape of the pattern by topology optimization, and parametrization
Step 2. Data generation using FEA
Step 3. Construction of the DNN model as an FE surrogate
Step 4. Metaheuristic-based optimization of design variables with the DNN-based surrogate model
Step 5. Experimental validation
First, the general pattern shape was determined through topology optimization using the ground structure approach [20].
General pattern shape
To determine the general shape of the pattern to be designed, the ground structure method [20] was employed. This method determines the cross-sectional areas of truss elements to generate an optimized topology [59], [60], [61], [62], [63], [64].
Because the flexural stiffness (EI) decreases as the volume of the tube decreases, the optimization problem was framed as the maximization of torsional stiffness (GJ) while reducing the volume of the cylinder. Fig. 2(a) shows the cylinder of interest
FE surrogate model
Numerous evaluations of the performance of design variables during optimization iterations are essential for population-based metaheuristic optimization. A well-trained network such as an FE surrogate model may facilitate rapid and accurate evaluation. Once trained, a network model can perform the evaluation in milliseconds, whereas FEA requires several tens of minutes to complete. Hence, in this study, a DNN model was trained to predict the EI and GJ of the designs.
The basic architecture of
Optimization problem
An optimized design variable set (nr,lt,θh) that minimizes the ratio of torsional stiffness to bending stiffness while satisfying the given stiffness constraints and geometric restrictions was determined. The optimization problem can be expressed as
Subject to:
- (1)
B* ≥ β
- (2)
2 ≤ nr ≤ 14; where nr is an integer
- (3)
10° ≤ θh ≤ 70°
- (4)
0.2 ≤ lt ≤ 1
- (5)
t ≥ ta
The value of the objective function of a design set (nr, θh, lt) is calculated using the NN outputs B* and T*. The bending
Laser machining for patterning
To validate the NN and FEA results experimentally, nitinol tubes were laser patterned, and bending and torsion tests were performed. The tubes had diamond patterns in the proposed mechanism, and the designs could be scaled down to tens of micrometers. Therefore, an accurate manufacturing process was necessary to fabricate them.
The through-hole patterns on the nitinol tube were created by laser cutting (RT1000 Laser Tube Cutting Machine, Preco Inc., KS, USA), with a feed rate of 127 mm/min, duty
Discussion and conclusions
In this study, a strategy for optimizing patterned tubes to decrease the ratio of bending to torsional stiffness while maintaining sufficient compliance was explored. The patterns found with the proposed DNN-metaheuristic optimization procedure have lower EI/GJ values than those obtained in previous studies under the same B* constraint conditions (Table 3). Using the DNN-based surrogate model and a metaheuristic approach, design optimization can be performed in an extended search space in a
CRediT authorship contribution statement
Soyeon Park: Conceptualization, Methodology, Software, Writing – original draft. Jongwoo Kim: Investigation, Validation, Formal analysis. Jieun Park: Writing – review & editing, Formal analysis, Visualization. Jessica Burgner-Kahrs: Writing – review & editing. Gunwoo Noh: Conceptualization, Supervision, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work was supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIP) (No. 2022-0-00824-001) and Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No.RS-2022-00155911) and the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 21R1C1C1011494 and No. 2022R1F1A1074359).
References (72)
- et al.
Harnessing distinct deformation modes of auxetic patterns for stiffness design of tubular structures
Mater Des
(2021) - et al.
Designing a mandibular advancement device with topology optimization for a partially edentulous patient
J Prosthet Dent
(2020) - et al.
Topology optimization of thin-walled tubes filled with lattice structures
Int J Mech Sci
(2022) - et al.
Residual stress constrained self-support topology optimization for metal additive manufacturing
Comput Methods Appl Mech Eng
(2022) - et al.
Topology optimization of imperfect frame structures with improved manufacturability
Int J Mech Sci
(2022) - et al.
Numerical study and topology optimization of vibration isolation support structures
Int J Mech Sci
(2022) - et al.
Structural topology optimization with an adaptive design domain
Comput Methods Appl Mech Eng
(2022) - et al.
The origami inspired optimization design to improve the crashworthiness of a multi-cell thin-walled structure for high speed train
Int J Mech Sci
(2019) - et al.
Energy absorption of thin walled tube filled with gradient auxetic structures-theory and simulation
Int J Mech Sci
(2021) - et al.
Parameter optimization of bidirectional re-entrant auxetic honeycomb metamaterial based on genetic algorithm
Compos Struct
(2021)
Finite element-based optimisation of an elastomeric honeycomb for impact mitigation in helmet liners
Int J Mech Sci
Improved particle swarm optimization algorithms for aerodynamic shape optimization of high-speed train
Adv Eng Softw
Thermal design optimization of electronic circuit board layout with transient heating chips by using Bayesian optimization and thermal network model
Int J Heat Mass Transf
Crashworthiness optimization of crash box with 3D-printed lattice structures
Int J Mech Sci
Structure design and mechanical properties of a novel anti-collision system with negative Poisson's ratio core
Int J Mech Sci
TPMS-based interpenetrating lattice structures: design, mechanical properties and multiscale optimization
Int J Mech Sci
Design, mechanical properties and optimization of lattice structures with hollow prismatic struts
Int J Mech Sci
Dynamic axial crushing of origami crash boxes
Int J Mech Sci
Crashworthiness design for trapezoid origami crash boxes
Thin Wall Struct
Energy absorption of thin-walled square tubes designed by kirigami approach
Int J Mech Sci
Growth Optimizer: a powerful metaheuristic algorithm for solving continuous and discrete global optimization problems
Knowl Based Syst
Design of space trusses using modified teaching–learning based optimization
Eng Struct
Recent advances in surrogate-based optimization
Prog Aerosp Sci
Machine learning-based inverse design of auxetic metamaterial with zero Poisson's ratio
Mater Today Commun
Multi-objective optimization of a biomass gasification to generate electricity and desalinated water using Grey Wolf Optimizer and artificial neural network
Chemosphere
Micro-EDM optimization through particle swarm algorithm and artificial neural network
Precis Eng
Buckling performance of fuselage panels reinforced with Voronoi-type stiffeners
Int J Mech Sci
Teaching–learning-based optimization: a novel method for constrained mechanical design optimization problems
Comput Aided Des
Progressive Latin Hypercube Sampling: an efficient approach for robust sampling-based analysis of environmental models
Environ Model Softw
Auxetic meta-disk for independent control of flexural and torsional waves
Int J Mech Sci
Design and control of concentric-tube robots
IEEE Trans Robot
A dynamic model for concentric tube robots
IEEE Trans Robot
Minimally invasive surgery: national trends in adoption and future directions for hospital strategy
Surg Endosc
Continuum robots for medical applications: a survey
IEEE Trans Robot
Closed-form differential kinematics for concentric-tube continuum robots with application to visual servoing
Experimental robotics
Designing snap-free concentric tube robots: a local bifurcation approach
Cited by (5)
Auxetic pattern design for concentric-tube robots using an active DNN-metaheuristics optimization
2024, Thin-Walled StructuresA machine learning method of accelerating multiscale analysis for spatially varying microstructures
2024, International Journal of Mechanical SciencesIn silico optimization of actuation performance in dielectric elastomer composites via integrated finite element modeling and deep learning
2024, Theoretical and Applied Mechanics LettersSoft actuators in surgical robotics: a state-of-the-art review
2024, Intelligent Service Robotics
- 1
These authors contributed equally to this work.