Pith. sign in

REVIEW 4 major objections 4 minor 56 references

Data-Driven Dimensional Synthesis of Diverse Planar Four-bar Function Generation Mechanisms via Direct Parameterization

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Four-bar function-generator synthesis can be recast as sequence-to-one regression: feed ordered input-output angle pairs to an LSTM and read off the four link lengths, with a type-specifying layer guaranteeing the requested mechanism type.

desk verdict A sensible data-driven synthesis pipeline whose own Table 2 undercuts its 'accurate for all types' claim; fix the rocker-crank experts and the eval and it's a useful tool. read the letter →

arxiv 2507.08269 v1 pith:72EPX5LL submitted 2025-07-11 cs.LG

Woon Ryong Kim , Jaeheun Jung , Jeong Un Ha , Donghun Lee , Jae Kyung Shim This is my paper · ORCID
classification cs.LG
keywords dimensionalsynthesisfour-barlinkagefunctiongenerationmechanismdefect-freesimultaneousdeeplearningLSTMmixtureofexperts
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that dimensional synthesis of four-bar function generators—normally an inverse problem solved by loop-closure equations or optimization—can be replaced by a supervised learning pipeline. It generates a large synthetic dataset of linkages whose type is fixed in advance, trains sixteen LSTM experts, one per four-bar type and geometric inversion, and adds a type-specifying layer that forces every predicted set of link lengths to belong to the requested type. Given target precision points, the system outputs dimensions whose simulated motion passes through those points, and the paper demonstrates this for five and twenty absolute precision points as well as seven and twenty relative precision points, with maximum output-angle errors below a tenth of a degree in the single-type examples. If this works, a designer can explore many candidate mechanisms instantly without solving equations, and the usual circuit, branch, and order defects never appear because the training data are generated from actual moving mechanisms.

What carries the argument

The machinery is a direct parameterization of link lengths by an orthogonal type matrix, $M$, and the four-vector $\vec{T} = M \vec{r}$, with $\vec{r} = \frac{1}{4} M^T \vec{T}$. The signs of $T_1, T_2, T_3$ classify the linkage type, so a type-specifying layer computes $T_j = \operatorname{Softplus}(h_j) \, \operatorname{sgn}(T_j)$ from the LSTM output and guarantees the predicted type. An LSTM encodes the ordered precision points, handling variable sequence length without fixing the point count, and a Mixture of Experts with 16 type-specific LSTM experts supports both single-type and multi-type synthesis. The simulation metric $S_{\mathrm{simul}} = 1 - \frac{1}{n}\sum_{i=1}^{n} \cos(\bar{\theta}^{\mathrm{out}}_i - f_{\vec{r}_{\mathrm{pred}}}(\bar{\theta}^{\mathrm{in}}_i))$ ranks predictions by how closely the simulated mechanism reproduces the desired angular relationship, without requiring the ground-truth dimensions.

What would settle it

Take precision points drawn from a distribution deliberately different from the dataset generator—for example, target motions whose link ratios exceed the paper's unstated bound $m$, or relative points whose random initial configurations in Section 7 are replaced by configurations outside the sampled range—and evaluate the simulation metric $S_{\mathrm{simul}}$. If the metric degrades well above the reported 0.0001–0.02 levels, or if the simulated mechanism misses some precision points, the method is interpolating its training manifold rather than solving synthesis generally.

Watch

Extended reading notes

Core claim

The central claim is that dimensional synthesis of four-bar function generators is learnable as a direct mapping from ordered precision points to link lengths. The paper constructs the dataset by sampling the type parameters $T_j$ rather than raw link lengths, computing motions from the loop-closure equations, and treating each of the eight four-bar types in each of its two geometric inversions as a separate class, giving 16 experts. A type-specifying layer applies $T_j = \operatorname{Softplus}(h_j)\,\operatorname{sgn}(T_j)$ and recovers link lengths via $\vec{r} = \frac{1}{4} M^T \vec{T}$, so the predicted linkage is guaranteed to have the requested type. Predictions are ranked by a simulation metric $S_{\mathrm{simul}}$ that compares desired and simulated output angles without needing ground-truth dimensions. The paper reports that all 16 experts achieve low simulation metrics, and the worked examples pass through all specified precision points with maximum absolute errors around 0.03 degrees for five-point synthesis and 0.02 degrees for twenty-point synthesis.

Load-bearing premise

The load-bearing premise is that the synthetic dataset used for training is representative of every set of precision points a designer might ask for; if a target set lies far from the sampled link lengths and angular ranges, the network can silently return a linkage whose motion misses the requested points.

Editorial extensions

If this is right

  • A designer can specify five or twenty precision points and receive dimensions for a requested four-bar type immediately, with the worked examples showing maximum output-angle errors around 0.03 degrees and 0.02 degrees respectively.
  • The unified system can also return several distinct linkage types for the same precision points, so a user can compare alternative mechanisms before committing to one.
  • Relative precision points, whose exact-synthesis limit is seven, are handled by expanding each sequence into 100 random absolute variants and ranking the resulting 1,600 predictions; the paper reports this takes 0.27 seconds per task.
  • Because the dataset treats the 16 type-inversion configurations as separate classes and uses inertia-based motion at dead centers, the generated linkages carry no circuit, branch, or order defects.
  • The same framework can seed traditional optimization: its instant proposals are intended as starting points for refinement, not just final answers.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A consequence the paper leaves implicit is that the 20-point cap is an implementation choice, not an architectural limit: an LSTM accepts arbitrary sequence length, so the same pipeline should extend to more precision points provided the dataset does.
  • The type-specifying layer is a transferable idea: any inverse problem whose solution classes are encoded by sign constraints can reuse this trick, so the architecture should carry over to six-bar linkages, path generation, or motion generation once a dataset generator exists.
  • The relative-point expansion into 100 random absolute variants is a Monte Carlo procedure; a natural extension would test how the number of variants affects ranking quality, since a deterministic cover of the initial-orientation circle might reach the same accuracy faster.
  • A distributional stress test would clarify the method's scope: target precision points far from the training generator, for example link ratios beyond the unstated bound $m$, would show whether the regression generalizes or interpolates.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a supervised learning pipeline for dimensional synthesis of planar four-bar function generators. It constructs synthetic datasets for 16 type-inversion configurations, trains one LSTM expert per configuration, applies a type-specifying layer to enforce linkage type, and uses a Mixture-of-Experts ranking procedure based on a simulation metric. Experiments report single-type and multi-type synthesis with 5, 7, and 20 precision points in both absolute and relative forms, claiming accurate, defect-free linkages across all types.

Significance. If the accuracy claims held for all sixteen configurations, the framework would be a useful complement to exact and optimization-based synthesis: it avoids solving nonlinear systems, handles variable precision-point counts, and offers multiple mechanism alternatives without retraining. The design of type-specified synthetic data, the LSTM-based sequence-to-one mapping, and the simulation metric independent of ground-truth dimensions are sensible and clearly described. The examples in Tables 3–6 show sub-degree errors for several types, which is encouraging evidence for those configurations. However, the aggregate results show that rocker-crank mechanisms are synthesized with roughly 11 degrees of RMS output-angle error, and one relative-point example produces a negative link length, so the central claims of accurate synthesis for all types and defect-free outputs are not supported as stated.

major comments (4)
  1. [Table 2, Eq. (10)] The claim of accurate synthesis for all 16 configurations is contradicted by the paper's own aggregate metric. In Eq. (10), for small angular errors S_simul is approximately (1/(2n))Σ δ_i^2, so the reported value S_simul = 0.0186 for Rocker-Crank+ corresponds to an RMS output-angle error of about 11 degrees, roughly two orders of magnitude worse than the 0.0001 values reported for Crank-Rocker, Double-Crank, and Double-Rocker. Table 11 shows the same problem for relative precision points (0.0213 and 0.0199 for the two rocker-crank configurations). No rocker-crank synthesis example appears in Sections 5–7, and all multi-type 'top three' lists exclude rocker-crank, which is consistent with a poorly performing type being omitted. The abstract and Section 8 state 'accurate, defect-free linkages across various configurations' and 'high accuracy for both absolute and relative precision point specifications'; these statements are not supported for rocker-crank mechanisms. Table 2 also gives no sample size or variance, so it is not possible to assess whether the point estimates are stable.
  2. [Table 13, Section 7.2] The relative-point synthesis examples include physically invalid predicted dimensions. Table 13 lists Mechanism 1 (Double Rocker−) with r3 = -0.00903, which violates the validity condition ri > 0 stated in Section 4.1.1. The type-specifying layer in Eq. (7) constrains only the signs of T1,...,T3 and the non-negativity of Tj; it does not guarantee that the vector r returned by Eq. (6) has positive entries. Thus the pipeline can output a negative link length, which does not correspond to a four-bar linkage. This undermines the 'defect-free' claim for relative precision points and indicates that a validity projection or additional constraint is missing from the model output.
  3. [Section 4.1.1] The sampling interval for Tj is written as [0, sgn(Tj) × m], which is not a well-defined interval when sgn(Tj) is negative. If the intent is to sample |Tj| and then apply the sign, the text should say so explicitly. In addition, m is never given a numerical value, and the distribution of n (the number of precision points per sample) is not specified. Because the synthetic dataset is the only source of supervision and the paper emphasizes the diversity of the data, the missing parameter values and the ambiguous sign convention prevent replication and make it difficult to judge how representative the training distribution is.
  4. [Sections 5–7] Generalization beyond the training distribution is not demonstrated. All evaluation tasks appear to be generated by the same synthetic procedure as the training set, and the relative-point expansion in Section 7 randomly assigns initial configurations rather than testing on an independent family of design specifications. The paper claims applicability 'regardless of the number of precision points' (Section 1), but training and evaluation are capped at n = 20 (Section 4.2). No held-out distributional analysis or ablation is provided, so silent degradation on precision-point sets far from the synthetic manifold remains a plausible risk rather than a tested property.
minor comments (4)
  1. [Tables 12–13] Several entries labeled 'Absolute error' are negative (e.g., -0.86082 in Table 13), which contradicts the definition of absolute error in Eq. (11). These values appear to be signed errors and should be relabeled or recomputed.
  2. [Figures 9–12] The figure captions and axis labels render triple-rocker type names using symbols such as '::' and ':0' instead of ππ and π0; please use consistent mathematical notation.
  3. [Title page] The affiliation contains a typo: 'Mathemetics' should be 'Mathematics'.
  4. [Section 4.2.2] The text says S_simul is 'subtracted from 1' to align with minimization, but Eq. (10) defines S_simul itself as 1 minus the average cosine. Please rephrase to avoid the impression that the final metric is computed by an additional subtraction.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: predictions are produced by supervised LSTM experts trained on synthetic forward-kinematics data, with no target-time parameter fitting.

full rationale

The paper's derivation chain is self-contained as a supervised learning pipeline. Linkage dimensions are sampled in the T-coordinate system, validated by the quadrilateral inequality, and converted to precision points using the standard loop-closure equations (Eqs. 2-4). Each expert LSTM maps a sequence of precision points to a four-dimensional r vector; the type-specifying layer (Eq. 7) simply enforces the sign pattern of the target type, and Eq. 6 is the inverse of the T-to-r transform, so this is a hard architectural constraint rather than a fitted input. The simulation metric (Eq. 10) computes the output angles of the predicted mechanism via the same forward kinematics and compares them to the specified precision points; since the model is trained with MSE on r and never optimizes Ssimul against the target at inference, this is an external evaluation. In multi-type synthesis, ranking the 16 fixed experts by Ssimul is model selection, not fitting. The relative-precision-point protocol expands each problem into 100 random absolute variants and ranks the resulting predictions; this is a search over the unspecified initial configuration, disclosed in Section 7, not a hidden reuse of the target in the regressor. The only self-citation, [55], supplies the standard loop-closure equation; that equation is textbook kinematics and is not load-bearing in any circular sense. Concerns about generalization outside the synthetic training distribution, the unsupported inference from defect-free training data to defect-free predictions, and the unvalidated claim in Table 2 are correctness/evidence issues, not circularity.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical entities. Its free parameters are dataset-scale and training choices (m, n_max, 100 variants, hyperparameters), while the axioms are standard kinematics plus dataset-construction assumptions. The most consequential assumption is that the hand-built sampling distribution yields a dataset representative enough for generalization to arbitrary precision-point inputs.

free parameters (4)
  • m = not specified
    Upper bound for sampling Tj in [0, sgn(Tj) * m] in Section 4.1.1; the value is never stated and controls the scale and diversity of the synthetic dataset.
  • n_max = 20
    Practical cap on precision-point sequence length despite the paper's claim of unrestricted counts; training uses n=20 as the upper bound.
  • Number of random initial configurations for relative points = 100
    Section 7 expands each relative-precision-point task into 100 random absolute variants; this number is chosen by hand and affects the search quality.
  • Model hyperparameters = 7 LSTM layers, dropout 0.3, lr 1e-4, weight decay 2e-3, 2000 epochs, 1024 samples per epoch
    Architecture and training choices in Section 4.2.4 are hand-selected and may not be optimal; they influence the reported performance.
assumptions (6)
  • standard math M is orthogonal with M^T M = 4I, so r = (1/4) M^T T
    Used in Eq. 6 to invert the type parameterization; standard linear algebra.
  • standard math The loop-closure equation (Eq. 2) with the half-angle substitution (Eq. 3) fully determines four-bar output angles
    Dataset generation in Section 4.1 relies on this standard kinematic analysis.
  • domain assumption All links are rigid bodies with zero-clearance joints
    Stated in Section 4.1 as the only physical assumption; necessary for exact kinematic relations.
  • domain assumption Inertia-based motion after dead center positions for rocker-input linkages
    Section 4.1.3: when the radical in Eq. 3 vanishes, the output link is assumed to switch geometric inversion, following treadle-sewing-machine behavior; the opposite motion is discarded.
  • ad hoc to paper Sampling Tj from signed intervals yields a representative, diverse dataset
    The sampling distribution is not stated to be uniform, and m is unstated; the coverage and generalization arguments depend on this assumption.
  • domain assumption Treating precision points as ordered sequences suffices to avoid order defects
    Section 4.1.4 claims order-defect avoidance by the sequential LSTM structure; this is asserted rather than proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-Driven Dimensional Synthesis of Diverse Planar Four-bar Function Generation Mechanisms via Direct Parameterization." pith.science (2026). https://pith.science/paper/72EPX5LL

@misc{pith2026250708269,
  author       = {Pith},
  title        = {Pith review of: Data-Driven Dimensional Synthesis of Diverse Planar Four-bar Function Generation Mechanisms via Direct Parameterization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/72EPX5LL}},
  note         = {Machine review of arXiv:2507.08269}
}
read the original abstract

Dimensional synthesis of planar four-bar mechanisms is a challenging inverse problem in kinematics, requiring the determination of mechanism dimensions from desired motion specifications. We propose a data-driven framework that bypasses traditional equation-solving and optimization by leveraging supervised learning. Our method combines a synthetic dataset, an LSTM-based neural network for handling sequential precision points, and a Mixture of Experts (MoE) architecture tailored to different linkage types. Each expert model is trained on type-specific data and guided by a type-specifying layer, enabling both single-type and multi-type synthesis. A novel simulation metric evaluates prediction quality by comparing desired and generated motions. Experiments show our approach produces accurate, defect-free linkages across various configurations. This enables intuitive and efficient mechanism design, even for non-expert users, and opens new possibilities for scalable and flexible synthesis in kinematic design.

Figures

Figures reproduced from arXiv: 2507.08269 by the authors.

Figure 1
Figure 1. Four-bar function generation mechanism allowing seamless conversion between the two when necessary. For completeness, results based on relative precision points are discussed in Section 7. In the exact synthesis of four-bar function generators, the maximum number of relative precision points that can be specified is seven, while the maximum number of absolute precision points is five [3, 4]. Thus, synthesis methods … view at source ↗
Figure 2
Figure 2. Geometric inversions of (a) a rocker-driven four-bar, (b) a crank-driven four-bar [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Design parameters of a four-bar linkage at the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Dataset generation process 4.1.1 Type-Specified Dimension Generation Recall that the type of a four-bar linkage can be determined by the signs of T1, T2, and T3, which are given by (1), [T1, T2, T3] T = M0⃗r. By introducing the additional parameter T4 = r1 + r2 + r3 + …
Figure 5
Figure 5. Figure 5: Dead center positions (DCPs) of four-bar linkage [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Two possible motions of a rocker-input four-bar after passing a dead center position (DCP) [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Model training process Cosine similarity is a widely used metric in deep learning that measures the angular similarity between two vectors. It is invariant to scale, making it suitable for assessing how closely the predicted ⃗rpred aligns with the ground truth ⃗r in fu…
Figure 8
Figure 8. Figure 8: Mixture of Experts architecture a single linkage type—and broad multi-type exploration across all types, thereby increasing both reliability and diversity in the resulting linkage configurations. 4.2.4 Implementation Details A unidirectional LSTM architecture with seve…
Figure 9
Figure 9. Figure 9: Displacement plots for five-precision-point synthesis: (a) Crank-Rocker [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Displacement plots for twenty-precision-point synthesis: (a) Double Crank [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Displacement plots for five-precision-point synthesis: (a) Case 1 (b) Case 2 [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: Displacement plots for twenty-precision-point synthesis: (a) Case 1 (b) Case 2 [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 55 canonical work pages

  1. [1]

    An analytical approach to the design of four-link mechanisms,

    F. Freudenstein, “An analytical approach to the design of four-link mechanisms,” Transactions of the American Society of Mechanical Engineers , vol. 76, no. 3, pp. 483–489, 1954

    work page 1954

  2. [2]

    G. N. Sandor, A general complex-number method for plane kinematic synthesis with applications . Columbia University, 1959

    work page 1959

  3. [3]

    A. G. Erdman and G. N. Sandor, Mechanism design analysis and synthesis (Vol. 1). Prentice-Hall, Inc., 1997

    work page 1997

  4. [4]

    Norton, Design of Machinery: An Introduction to the Synthesis and Analysis of Mechanisms and Machines

    R. Norton, Design of Machinery: An Introduction to the Synthesis and Analysis of Mechanisms and Machines . McGraw-Hill series in mechanical engineering, McGraw-Hill Higher Education, 2003

    work page 2003

  5. [5]

    Three and four precision point kinematic synthesis of planar linkages,

    A. G. Erdman, “Three and four precision point kinematic synthesis of planar linkages,”Mechanism and Machine Theory , vol. 16, no. 3, pp. 227–245, 1981

    work page 1981

  6. [6]

    Synthesis of path-generating mechanisms by means of a programmed digital computer,

    F. Freudenstein and G. N. Sandor, “Synthesis of path-generating mechanisms by means of a programmed digital computer,” Journal of Engineering for Industry , vol. 81, no. 2, pp. 159–167, 1959

    work page 1959

  7. [7]

    Synthesis of Plane Linkages With Use of the Displacement Matrix,

    C. H. Suh and C. W. Radcliffe, “Synthesis of Plane Linkages With Use of the Displacement Matrix,” Journal of Engineering for Industry , vol. 89, pp. 206–214, 05 1967

    work page 1967

  8. [8]

    Solving a planar four-bar linkages design problem,

    X. Huang, G. He, Q. Liao, S. Wei, and X. Tan, “Solving a planar four-bar linkages design problem,” in 2009 International Conference on Information and Automation , pp. 1586–1590, IEEE, 2009

    work page 2009

Show all 56 references
  1. [9]

    Complete Solution of the Nine-Point Path Synthesis Problem for Four-Bar Linkages,

    C. W. Wampler, A. P. Morgan, and A. J. Sommese, “Complete Solution of the Nine-Point Path Synthesis Problem for Four-Bar Linkages,” Journal of Mechanical Design , vol. 114, pp. 153–159, 03 1992

    1992

  2. [10]

    Numerical synthesis of six-bar linkages for mechanical computation,

    M. M. Plecnik and J. Michael McCarthy, “Numerical synthesis of six-bar linkages for mechanical computation,” Journal of Mechanisms and Robotics , vol. 6, no. 3, p. 031012, 2014

    2014

  3. [11]

    Computational design of stephenson ii six-bar function generators for 11 accuracy points,

    M. M. Plecnik and J. Michael McCarthy, “Computational design of stephenson ii six-bar function generators for 11 accuracy points,” Journal of Mechanisms and Robotics , vol. 8, no. 1, p. 011017, 2016

    2016

  4. [12]

    Kinematic synthesis of stephenson iii six-bar function gen- erators,

    M. M. Plecnik and J. M. McCarthy, “Kinematic synthesis of stephenson iii six-bar function gen- erators,” Mechanism and Machine Theory , vol. 97, pp. 112–126, 2016

    2016

  5. [13]

    A function generation synthesis methodology for all defect-free slider-crank solutions for four precision points,

    A. Almandeel, A. P. Murray, D. H. Myszka, and H. E. Stumph III, “A function generation synthesis methodology for all defect-free slider-crank solutions for four precision points,” Journal of Mechanisms and Robotics , vol. 7, no. 3, p. 031020, 2015. 22

    2015

  6. [14]

    Synthesis of function-generation mechanisms using variable length driving links,

    R.-C. Soong and S.-B. Chang, “Synthesis of function-generation mechanisms using variable length driving links,” Mechanism and Machine Theory , vol. 46, no. 11, pp. 1696–1706, 2011

    2011

  7. [15]

    Fourth order synthesis of spherical 4-bar function generators,

    J.-S. Chen and C. Chiang, “Fourth order synthesis of spherical 4-bar function generators,” Mech- anism and machine theory , vol. 18, no. 6, pp. 451–456, 1983

    1983

  8. [16]

    Synthesis of bistable compliant four-bar mechanisms using polynomial homotopy,

    H.-J. SU and J. M. MCCARTHY, “Synthesis of bistable compliant four-bar mechanisms using polynomial homotopy,” Journal of mechanical design (1990) , vol. 129, no. 10, pp. 1094–1098, 2007

    1990

  9. [17]

    The synthesis of planar four-bar linkage for mixed motion and function generation,

    B. Wang, X. Du, J. Ding, Y. Dong, C. Wang, and X. Liu, “The synthesis of planar four-bar linkage for mixed motion and function generation,” Sensors, vol. 21, no. 10, p. 3504, 2021

    2021

  10. [18]

    Five position synthesis of a slider-crank function generator,

    M. M. Plecnik and J. M. McCarthy, “Five position synthesis of a slider-crank function generator,” in International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, vol. 54839, pp. 317–324, 2011

    2011

  11. [19]

    Synthesis of Six-Link Plane Mechanisms by Numerical Analysis,

    C. W. McLarnan, “Synthesis of Six-Link Plane Mechanisms by Numerical Analysis,” Journal of Engineering for Industry , vol. 85, pp. 5–10, 02 1963

    1963

  12. [20]

    Approximate synthesis of four-bar linkages,

    F. Freudenstein, “Approximate synthesis of four-bar linkages,” Transactions of the American Society of Mechanical Engineers , vol. 77, no. 6, pp. 853–859, 1955

    1955

  13. [21]

    Unconstrained nonlinear least-square optimization of planar linkages for rigid-body guidance,

    R. Akhras and J. Angeles, “Unconstrained nonlinear least-square optimization of planar linkages for rigid-body guidance,” Mechanism and Machine Theory , vol. 25, no. 1, pp. 97–118, 1990

    1990

  14. [22]

    Optimal design of mechanisms with the use of matrices and least squares,

    C. Suh and A. Mecklenburg, “Optimal design of mechanisms with the use of matrices and least squares,” Mechanism and Machine Theory , vol. 8, no. 4, pp. 479–495, 1973

    1973

  15. [23]

    Optimal synthesis of function generator of four-bar linkages based on distribution of precision points,

    M. Shariati and M. Norouzi, “Optimal synthesis of function generator of four-bar linkages based on distribution of precision points,” Meccanica, vol. 46, pp. 1007–1021, 2011

    2011

  16. [24]

    Optimal Synthesis of Function Generators Without the Branch Defect,

    S. O. Tinubu and K. C. Gupta, “Optimal Synthesis of Function Generators Without the Branch Defect,” Journal of Mechanisms, Transmissions, and Automation in Design, vol. 106, pp. 348–354, 09 1984

    1984

  17. [25]

    On mechanism design optimization for motion gener- ation,

    P. J. Martin, K. Russell, and R. S. Sodhi, “On mechanism design optimization for motion gener- ation,” Mechanism and machine theory , vol. 42, no. 10, pp. 1251–1263, 2007

    2007

  18. [26]

    A novel analytical method for function generation synthesis of planar four-bar linkages,

    X. Li, S. Wei, Q. Liao, and Y. Zhang, “A novel analytical method for function generation synthesis of planar four-bar linkages,” Mechanism and Machine Theory , vol. 101, pp. 222–235, 2016

    2016

  19. [27]

    Concurrent type and dimensional continuous approximate function generator synthesis for all planar four-bar mech- anisms,

    Z. Copeland, M. Rotzoll, M. J. D. Hayes, S. Nokleby, and P. Cardou, “Concurrent type and dimensional continuous approximate function generator synthesis for all planar four-bar mech- anisms,” in 11th CCToMM Symposium on Mechanisms, Machines, and Mechatronics, Ontario Tech Univ...

    2021

  20. [28]

    Optimum design method for four-bar function generator using ais and genetic algorithms,

    M. S. Javash, M. M. Ettefagh, and Y. E. Hamidi, “Optimum design method for four-bar function generator using ais and genetic algorithms,” in 2013 IEEE INISTA , pp. 1–6, IEEE, 2013

    2013

  21. [29]

    Synthesis of 4-bar function-generators using geometric programming,

    A. Rao, “Synthesis of 4-bar function-generators using geometric programming,” Mechanism and Machine theory, vol. 14, no. 2, pp. 141–149, 1979

    1979

  22. [30]

    Box Approximations of Planar Linkage Configuration Spaces,

    J. M. Porta, L. Ros, T. Creemers, and F. Thomas, “Box Approximations of Planar Linkage Configuration Spaces,” Journal of Mechanical Design , vol. 129, pp. 397–405, 04 2006

    2006

  23. [31]

    A Fast Learning Algorithm for Deep Belief Nets,

    G. E. Hinton, S. Osindero, and Y.-W. Teh, “A Fast Learning Algorithm for Deep Belief Nets,” Neural Computation, vol. 18, pp. 1527–1554, 07 2006

    2006

  24. [32]

    Backpropagation applied to handwritten zip code recognition,

    Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel, “Backpropagation applied to handwritten zip code recognition,” Neural computation, vol. 1, no. 4, pp. 541–551, 1989. 23

    1989

  25. [33]

    Learning representations by back- propagating errors,

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning representations by back- propagating errors,” nature, vol. 323, no. 6088, pp. 533–536, 1986

    1986

  26. [34]

    Learning long-term dependencies with gradient descent is difficult,

    Y. Bengio, P. Simard, and P. Frasconi, “Learning long-term dependencies with gradient descent is difficult,” IEEE transactions on neural networks , vol. 5, no. 2, pp. 157–166, 1994

    1994

  27. [35]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Comput. , vol. 9, p. 1735–1780, nov 1997

    1997

  28. [36]

    A comprehensive overview and com- parative analysis on deep learning models: Cnn, rnn, lstm, gru,

    F. M. Shiri, T. Perumal, N. Mustapha, and R. Mohamed, “A comprehensive overview and com- parative analysis on deep learning models: Cnn, rnn, lstm, gru,” arXiv preprint arXiv:2305.17473, 2023

    2023 arXiv

  29. [37]

    Synthetic data approach for classification and regression,

    Y. Yue, Y. Li, K. Yi, and Z. Wu, “Synthetic data approach for classification and regression,” in 2018 IEEE 29th International Conference on Application-specific Systems, Architectures and Processors (ASAP), pp. 1–8, IEEE, 2018

    2018

  30. [38]

    Investigating the impact of synthetic data distribution on the performance of regression models to overcome small dataset problems,

    T. Sutojo, A. Syukur, S. Rustad, G. F. Shidik, H. A. Santoso, P. Purwanto, and M. Muljono, “Investigating the impact of synthetic data distribution on the performance of regression models to overcome small dataset problems,” in 2020 International Seminar on Application for Tec...

    2020

  31. [39]

    Using synthetic data to evaluate multiple regression and principal component analyses for statistical modeling of daily building energy consumption,

    T. Reddy and D. Claridge, “Using synthetic data to evaluate multiple regression and principal component analyses for statistical modeling of daily building energy consumption,” Energy and buildings, vol. 21, no. 1, pp. 35–44, 1994

    1994

  32. [40]

    Dimensional synthesis of planar mechanisms using neural networks: application to path generator linkages,

    A. Vasiliu and B. Yannou, “Dimensional synthesis of planar mechanisms using neural networks: application to path generator linkages,” Mechanism and Machine Theory , vol. 36, no. 2, pp. 299– 310, 2001

    2001

  33. [41]

    A machine learning approach to kinematic synthesis of defect-free planar four-bar linkages,

    S. Deshpande and A. Purwar, “A machine learning approach to kinematic synthesis of defect-free planar four-bar linkages,” Journal of Computing and Information Science in Engineering , vol. 19, no. 2, p. 021004, 2019

    2019

  34. [42]

    Synthesis of mechanical linkages using artificial neural networks and optimization,

    J. C. Hoskins and G. A. Kramer, “Synthesis of mechanical linkages using artificial neural networks and optimization,” in IEEE International Conference on Neural Networks, pp. 822–J, IEEE, 1993

    1993

  35. [43]

    Dimensional synthesis of mechanical linkages using artificial neural networks and fourier descriptors,

    N. Khan, I. Ullah, and M. Al-Grafi, “Dimensional synthesis of mechanical linkages using artificial neural networks and fourier descriptors,” Mechanical Sciences, vol. 6, no. 1, pp. 29–34, 2015

    2015

  36. [44]

    Optimization of transmission angle for slider-crank mechanism with joint clearances,

    S. Erkaya and I. Uzmay, “Optimization of transmission angle for slider-crank mechanism with joint clearances,” Structural and Multidisciplinary Optimization , vol. 37, pp. 493–508, 2009

    2009

  37. [45]

    Path synthesis of crank-rocker mechanism using fourier descriptors based neural network,

    X. Mo, W. Ge, D. Zhao, and Y. Zhang, “Path synthesis of crank-rocker mechanism using fourier descriptors based neural network,” in Recent Advances in Mechanisms, Transmissions and Appli- cations: Proceedings of the Fifth MeTrApp Conference 2019 5 , pp. 32–41, Springer, 2020

    2019

  38. [46]

    Big data approach for the simultaneous determination of the topology and end-effector location of a planar linkage mechanism,

    N. H. Yim, J. Lee, J. Kim, and Y. Y. Kim, “Big data approach for the simultaneous determination of the topology and end-effector location of a planar linkage mechanism,”Mechanism and Machine Theory, vol. 163, p. 104375, 2021

    2021

  39. [47]

    Big data approach for synthesizing a spatial linkage mech- anism,

    N. H. Yim, J. Ryu, and Y. Y. Kim, “Big data approach for synthesizing a spatial linkage mech- anism,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) , pp. 7433– 7439, IEEE, 2023

    2023

  40. [48]

    Synthesis of a six-bar mechanism for generating knee and ankle motion trajectories using deep generative neural network,

    A. Kapsalyamov, S. Hussain, N. A. Brown, R. Goecke, M. Hayat, and P. K. Jamwal, “Synthesis of a six-bar mechanism for generating knee and ankle motion trajectories using deep generative neural network,” Engineering Applications of Artificial Intelligence , vol. 117, p. 105500, 2023

    2023

  41. [49]

    Appropriate analysis of the four-bar linkage,

    J. K. Pickard, J. A. Carretero, and J.-P. Merlet, “Appropriate analysis of the four-bar linkage,” Mechanism and Machine Theory , vol. 139, pp. 237–250, 2019. 24

    2019

  42. [50]

    J. M. McCarthy and G. S. Soh, Geometric design of linkages , vol. 11. Springer Science & Business Media, 2010

    2010

  43. [51]

    Range of the six fundamental position angles of a planar four-bar mechanism,

    C. R. Barker and Y.-R. Jeng, “Range of the six fundamental position angles of a planar four-bar mechanism,” Mechanism and machine theory , vol. 20, no. 4, pp. 329–344, 1985

    1985

  44. [52]

    A. K. Mallik, Kinematic analysis and synthesis of mechanisms . Crc Press, 2021

    2021

  45. [53]

    Defects in link mechanisms and solution rectification,

    S. S. Balli and S. Chand, “Defects in link mechanisms and solution rectification,” Mechanism and Machine Theory, vol. 37, no. 9, pp. 851–876, 2002

    2002

  46. [54]

    Circuits and Branches of Single-Degree-of-Freedom Planar Link- ages,

    T. R. Chase and J. A. Mirth, “Circuits and Branches of Single-Degree-of-Freedom Planar Link- ages,” Journal of Mechanical Design , vol. 115, pp. 223–230, 06 1993

    1993

  47. [55]

    Kinematic conceptual design of in-line four-cylinder variable compression ratio engine mechanisms considering vertical second harmonic acceleration,

    S. W. Kwak, J. K. Shim, and Y. K. Mo, “Kinematic conceptual design of in-line four-cylinder variable compression ratio engine mechanisms considering vertical second harmonic acceleration,” Applied Sciences, vol. 10, no. 11, p. 3765, 2020

    2020

  48. [56]

    Deep sparse rectifier neural networks,

    X. Glorot, A. Bordes, and Y. Bengio, “Deep sparse rectifier neural networks,” in Proceedings of the fourteenth international conference on artificial intelligence and statistics , pp. 315–323, JMLR Workshop and Conference Proceedings, 2011. 25

    2011

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.