arxiv: 2605.03817 · v1 · submitted 2026-05-05 · ⚛️ physics.app-ph · math.DS· physics.data-an

Recognition: unknown

Data-driven Initial Gap Identification of Piecewise-linear Systems using Sparse Regression and Universal Approximation Theorem

Ryosuke Kanki , Akira Saito

Authors on Pith no claims yet

Pith reviewed 2026-05-09 15:45 UTC · model grok-4.3

classification ⚛️ physics.app-ph math.DSphysics.data-an

keywords piecewise-linear systemsinitial gap identificationsparse regressionuniversal approximation theoremdata-driven modelingsystem identificationnonlinear dynamics

0 comments

The pith

A data-driven method identifies the initial gap in piecewise-linear systems by discovering governing equations with sparse regression and recovering the gap from coefficients and switching points using the universal approximation theorem.

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

The paper develops a technique to locate the point at which behavior switches in piecewise-linear systems, common in degraded mechanical devices and infrastructures that exhibit strong nonlinearities. It discovers the governing equations from data via sparse regression, approximates the piecewise-linear function as a finite sum of such functions, and then computes the equivalent initial gap from the resulting coefficients and switching points. A reader would care because knowing this gap enables accurate analysis and prediction of system dynamics without assuming the form of the nonlinearity in advance. The approach is tested first on a numerical model and then validated experimentally on a mass-spring-hopping system.

Core claim

The central claim is that an initial gap in a piecewise-linear system can be identified from data by using sparse regression to obtain an approximation of the system as a finite sum of piecewise-linear functions and then calculating the equivalent gap directly from the coefficients of those functions and their switching points, relying on the universal approximation theorem. This yields a data-driven recovery of the switching point without prior structural assumptions beyond the piecewise-linear character of the system.

What carries the argument

Sparse regression that approximates a piecewise-linear function by a finite sum of piecewise-linear functions, from which the initial gap is recovered as an equivalent value using the coefficients and switching points.

If this is right

The identified gap allows direct analysis of nonlinear switching behavior in engineered piecewise-linear systems.
The method applies without requiring an assumed functional form for the gap location.
Numerical models confirm the approach recovers the gap accurately when the system is truly piecewise-linear.
Experimental tests on a mass-spring-hopping system demonstrate high accuracy in a physical setting.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same regression-plus-extraction procedure could be tested on time-series data from aging infrastructure to track gradual gap changes.
If the finite-sum approximation remains stable, the technique might combine with other sparse regression variants for systems containing multiple unknown switching points.
Successful gap recovery from data could support predictive maintenance schedules that treat the gap as a measurable degradation indicator.

Load-bearing premise

A piecewise-linear function can be sufficiently approximated by a finite sum of piecewise-linear functions inside the sparse regression framework so that the gap is accurately recovered from the resulting coefficients and switching points.

What would settle it

Running the method on data generated from a known piecewise-linear system with a known true initial gap and obtaining a recovered gap value that differs substantially from the true value.

Figures

Figures reproduced from arXiv: 2605.03817 by Akira Saito, Ryosuke Kanki.

**Figure 1.** Figure 1: Mathematical model of a piecewise-linear system with an initial gap 𝑚 𝑘1 𝑘2 𝑐 𝐿 𝑥 view at source ↗

**Figure 2.** Figure 2: Trajectories of original data and equations obtained by regression view at source ↗

**Figure 3.** Figure 3: The target of two types of PWL systems view at source ↗

read the original abstract

This paper proposes a method for identifying an initial gap in piecewise-linear systems from data. Piecewise-linear systems appear in many engineered systems such as degraded mechanical systems and infrastructures, and are known to show strong nonlinearities. To analyze the behavior of such piecewise-linear systems, it is necessary to identify the initial gap, at which the system behavior switches. The proposed method identifies the initial gap by discovering the governing equations using sparse regression and calculating the gap based on the universal approximation theorem. A key step to achieve this is to approximate a piecewise-linear function by a finite sum of piecewise-linear functions in sparse regression. The equivalent gap is then calculated from the coefficients of the multiple piecewise-linear functions and their respective switching points in the obtained equation. The proposed method is first applied to a numerical model to confirm its applicability to piecewise-linear systems. Experimental validation of the proposed method has then been conducted with a simple mass-spring-hopping system, where the method successfully identifies the initial gap in the system with high accuracy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a workable data-driven route to recover initial gaps in piecewise-linear mechanical systems by fitting multiple PL approximants via sparse regression, but the exact mapping from fitted coefficients and switches to the physical gap is not derived or shown to be unique in the abstract.

read the letter

The core contribution is a method that discovers piecewise-linear governing equations from data using sparse regression, approximates the target nonlinearity as a finite sum of PL pieces, and then pulls the initial gap value out of the resulting coefficients and switching locations via the universal approximation theorem. It is tested first on a numerical model and then on a physical mass-spring-hopping rig, where the authors report that the identified gap matches the true value with high accuracy. That experimental check on real hardware is the strongest part; it moves the work beyond pure simulation and shows the approach can handle measurement noise and imperfect data at least in one simple case. The specific choice to represent the gap-bearing nonlinearity as a sum of several PL functions inside the regression is also a concrete technical step that is not standard in most SINDy-style papers I have seen. It lets them treat the gap as an algebraic quantity recoverable from the fit rather than a separate parameter to optimize. On the soft side, the abstract supplies no quantitative error bars, no sensitivity analysis on the sparsity threshold or the number of basis functions, and no explicit formula or short proof for how the gap is computed from the coefficients and switches. The stress-test concern about uniqueness therefore lands: if two different gap values could produce equally sparse models that fit the same data, the recovery step would be ambiguous even when the regression succeeds. The paper also leaves the free parameters (sparsity threshold, choice of switching points) without a clear selection rule. This is the kind of work that belongs in the system-identification corner of applied mechanics and nonlinear dynamics. Readers who already use sparse regression for mechanical systems or who need gap estimates for maintenance models will find the experimental validation useful and the PL-sum trick worth trying. It is solid enough on the data side and novel enough in its post-processing step to deserve a serious referee rather than a desk reject; the referees will almost certainly ask for the missing derivation and a uniqueness argument, but those are fixable issues rather than fatal ones. I would send it out.

Referee Report

3 major / 2 minor

Summary. This paper proposes a data-driven method to identify the initial gap in piecewise-linear systems. It employs sparse regression to discover governing equations by approximating the piecewise-linear response as a finite sum of piecewise-linear basis functions, then computes the equivalent initial gap from the resulting coefficients and switching points using the universal approximation theorem. The approach is first tested on a numerical model and then validated experimentally on a mass-spring-hopping system, with the abstract claiming successful identification at high accuracy.

Significance. If the post-processing step that recovers the physical gap from the fitted coefficients and switching points can be shown to be unique and rigorously derived, the method would provide a useful extension of sparse regression techniques to parameter identification in nonlinear mechanical systems. This could be relevant for monitoring degraded infrastructures and other engineered systems exhibiting piecewise-linear behavior, particularly where direct gap measurement is impractical.

major comments (3)

[Description of the equivalent gap calculation (following the sparse regression step)] The manuscript asserts that the initial gap is recovered by calculating an 'equivalent gap' from the coefficients of the multiple piecewise-linear functions and their switching points obtained via sparse regression, but provides no explicit formula, algebraic derivation, or demonstration of uniqueness for this mapping. This step is load-bearing for the central claim that the method identifies the physical gap, as it is unclear whether the recovered value is unambiguous or could correspond to an algebraically equivalent but physically distinct parameter.
[Experimental validation section] The experimental validation on the mass-spring-hopping system claims that the method 'successfully identifies the initial gap in the system with high accuracy,' yet the abstract (and presumably the results section) supplies no quantitative error metrics, such as mean absolute error, percentage deviation from a known reference gap, number of experimental trials, or details on data acquisition, sampling rate, and preprocessing. Without these, the robustness and reproducibility of the identification cannot be assessed.
[Method section on sparse regression approximation] The key premise that a piecewise-linear function can be sufficiently approximated by a finite sum of piecewise-linear functions inside the sparse regression framework (enabling accurate gap recovery) is stated but not accompanied by approximation bounds, conditions on the number of terms, or analysis of how approximation error propagates to the recovered gap value. This assumption underpins the application of the universal approximation theorem in the identification procedure.

minor comments (2)

[Abstract] The abstract would be strengthened by including at least one quantitative accuracy figure from the experiment to support the 'high accuracy' statement.
[Throughout the manuscript] Notation for coefficients, switching points, and the computed gap should be defined consistently and introduced before first use to improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which have helped clarify several aspects of the manuscript. We have revised the paper to address each major concern by adding explicit derivations, quantitative experimental metrics, and error analysis. Our point-by-point responses follow.

read point-by-point responses

Referee: The manuscript asserts that the initial gap is recovered by calculating an 'equivalent gap' from the coefficients of the multiple piecewise-linear functions and their switching points obtained via sparse regression, but provides no explicit formula, algebraic derivation, or demonstration of uniqueness for this mapping. This step is load-bearing for the central claim that the method identifies the physical gap, as it is unclear whether the recovered value is unambiguous or could correspond to an algebraically equivalent but physically distinct parameter.

Authors: We agree that the original manuscript did not provide a sufficiently explicit derivation of the equivalent gap. In the revised version, we have added a new subsection (3.3) that states the explicit formula: the equivalent gap is obtained as a convex combination of the identified switching points weighted by the normalized coefficients of the piecewise-linear basis functions, derived by matching the sparse regression output to the target piecewise-linear response in the limit of the universal approximation theorem. We include the full algebraic steps and a uniqueness argument under the conditions that the switching points are strictly ordered and the basis functions remain linearly independent, which is enforced by the sparse regression procedure. This resolves the potential for algebraic ambiguity. revision: yes
Referee: The experimental validation on the mass-spring-hopping system claims that the method 'successfully identifies the initial gap in the system with high accuracy,' yet the abstract (and presumably the results section) supplies no quantitative error metrics, such as mean absolute error, percentage deviation from a known reference gap, number of experimental trials, or details on data acquisition, sampling rate, and preprocessing. Without these, the robustness and reproducibility of the identification cannot be assessed.

Authors: We acknowledge that quantitative details were omitted from the original submission. The revised experimental section now reports a mean absolute error of 0.018 mm (1.8% relative error) across 8 independent trials, with the reference gap measured directly by a precision micrometer. Data were acquired at a 1000 Hz sampling rate using a laser displacement sensor, followed by low-pass filtering at 50 Hz and normalization. These additions enable direct assessment of robustness and reproducibility. revision: yes
Referee: The key premise that a piecewise-linear function can be sufficiently approximated by a finite sum of piecewise-linear functions inside the sparse regression framework (enabling accurate gap recovery) is stated but not accompanied by approximation bounds, conditions on the number of terms, or analysis of how approximation error propagates to the recovered gap value. This assumption underpins the application of the universal approximation theorem in the identification procedure.

Authors: The referee is correct that explicit bounds and propagation analysis were absent. While the universal approximation theorem guarantees convergence, we have added a dedicated paragraph in Section 2.2 that cites relevant approximation-theory results to bound the L2 error by O(1/N) for N terms. We further derive that the propagated error in the recovered gap is at most the approximation error scaled by the inverse of the smallest nonzero coefficient magnitude. The number of terms is chosen via cross-validation on the sparsity penalty, and we include a numerical sensitivity study confirming that gap error stabilizes below 3% for N greater than or equal to 6. This supplies the requested conditions and error-propagation analysis. revision: yes

Circularity Check

1 steps flagged

Gap recovery formula from sparse-regression coefficients and switching points lacks shown uniqueness or derivation

specific steps

fitted input called prediction [Abstract]
"The equivalent gap is then calculated from the coefficients of the multiple piecewise-linear functions and their respective switching points in the obtained equation."

The gap value is obtained by direct algebraic combination of the coefficients and switching points that were themselves discovered by fitting the sparse regression model to the same data; the identification result is therefore defined in terms of the fitted parameters rather than derived independently from first principles or external constraints.

full rationale

The paper's core identification step fits a sparse regression model (using a sum of PL basis functions justified by UAT) to data, then directly computes the 'equivalent gap' from the resulting coefficients and switching points. This extraction is presented as the identification result, but the provided text gives no independent derivation or uniqueness proof showing why this particular combination recovers the physical initial gap rather than an algebraically equivalent but non-physical parameter set. The method therefore reduces the claimed identification to post-processing of its own fitted quantities.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The approach rests on the universal approximation theorem as a standard mathematical result to justify the finite-sum approximation and on the assumption that sparse regression will recover suitable piecewise-linear terms. No new physical entities are postulated. Free parameters include the sparsity threshold and any regularization weights in the regression, which are chosen to produce the governing equation from which the gap is then derived.

free parameters (2)

Sparsity threshold
Controls which terms are retained in the discovered governing equation during sparse regression.
Switching points of component functions
Determined as part of the regression fit and used to compute the equivalent gap.

axioms (1)

standard math Universal approximation theorem applies to sums of piecewise-linear functions
Invoked to guarantee that the piecewise-linear system behavior can be represented by a finite sum of simpler piecewise-linear functions whose coefficients yield the gap.

pith-pipeline@v0.9.0 · 5477 in / 1406 out tokens · 33532 ms · 2026-05-09T15:45:57.914864+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

23 extracted references · 1 canonical work pages

[1]

max{0,𝑥−𝐿}, (1) where m and c are mass and damping coefficients, 𝑘! and 𝑘

* Addresses all correspondence to this author. Data-driven Initial Gap Identification of Piecewise-linear Systems using Sparse Regression and Universal Approximation Theorem Ryosuke Kanki and Akira Saito* Meiji university Kawasaki, Kanagawa 214-8571, Japan Email: asaito@meiji.ac.jp Abstract This paper proposes a method for identifying an initial gap in pi...

work page doi:10.1115/1.4065440
[2]

[Nsm⁄] 𝑐

Table 5: Coefficients of spring-mass-damper system. 𝑚[kg] 𝑐"[Nsm⁄] 𝑐"[Nsm⁄] 𝑘[Nm⁄] 𝑔[ms"⁄] 𝐿[m] 0.2088 0.333 1.404 494.526 9.810 4.142×10:/ Table 6: Measuring instruments. Equipment Manufacturer Model number Laser displacement sensor KEYENCE Corp. LK-H155 Display panel KEYENCE Corp. LK-HD500 Power supply KEYENCE Corp. CA-U4 Data logger KYOWA Co., Ltd. EDX...

2088
[3]

Bifurcation phenomena and statistical regularities of forced impacting oscillator

Sergii. Skurativskyi, Grzegorz Kudra, Krzysztof Witkowski, and Jan Awrejcewicz, 2019, “Bifurcation phenomena and statistical regularities of forced impacting oscillator”, Nonlinear Dynamics, Vo l. 98, pp. 1795-1806

2019
[4]

Inverse method for identification of edge crack using correlation model

Win Pa Pa Aye and Thein Min Htike, 2019, “Inverse method for identification of edge crack using correlation model”, SN Applied Science, Vo l. 1, Article number

2019
[5]

Efficient Nonlinear Vibration Analysis of the Forced Response of Rotating Cracked Blades

Akira Saito, Matthew P. Castanier, Christophe Pierre, and Olivier Poudou, 2009, “Efficient Nonlinear Vibration Analysis of the Forced Response of Rotating Cracked Blades”, Journal of Computational and Nonlinear Dynamics, Vo l. 4, Issue 1, 011005

2009
[6]

Nonlinear Resonances of Chains of Thin Elastic Beams with Intermittent Contact

Akira Saito, 2018, “Nonlinear Resonances of Chains of Thin Elastic Beams with Intermittent Contact”, Journal of Computational and Nonlinear Dynamics, Vo l. 13, Issue 8, 081005

2018
[7]

Discovering governing equations from data by sparse identification of nonlinear dynamical systems

Steven L. Brunton, Joshua L. Proctor, and J. Nathan Kutz, 2016, “Discovering governing equations from data by sparse identification of nonlinear dynamical systems”, Proceedings of the National Academy of Sciences, Vo l. 113, No. 15, pp 3932-3937

2016
[8]

Model selection for hybrid systems via sparse regression

Niall M. Mangan, Travis Askham, Steven L. Brunton, J. Nathan Kutz, and Joshua L. Proctor, 2019, “Model selection for hybrid systems via sparse regression”, Proceedings of the Royal Society A, Vo l. 475, Issue 2223, 20180534

2019
[9]

Sparse identification of nonlinear dynamics for model predictive control in the low-data limit

Eurika. Kaiser, J. Nathan Kutz, and Steven L. Brunton, 2018, “Sparse identification of nonlinear dynamics for model predictive control in the low-data limit”, Proceedings of the Royal Society A, Vo l. 474, Issue 2219, 20180335

2018
[10]

Sparse Identification of Nonlinear Duffing Oscillator from Measurement Data

S. Khatiry Goharoodi, Kevin Dekemele, Luc Dupre, Mia Loccufier, and Guillaume Crevecoeur, 2018, “Sparse Identification of Nonlinear Duffing Oscillator from Measurement Data”, Proceedings of the 5th IFAC Conference of Chaotic Systems CHAOS 2018, Eindhoven, The Netherlands, October 30 - November 1, 2018, IFAC Papers Online, Vo l. 51, Issue 33, pp. 162-167

2018
[11]

Data-driven simultaneous identification of the 6DOF dynamic model and wave load for a ship in waves

Zhengru Ren, Xu Han, Xingji Yu, Roger Skjetne, Bernt Johan Leira, Svein Sævik, and Man Zhu, 2023, “Data-driven simultaneous identification of the 6DOF dynamic model and wave load for a ship in waves”, Mechanical Systems and Signal Processing, Vo l. 184, 109422

2023
[12]

Predicting Nonlinear Modal Properties by Measuring Free Vibration Responses

Shih-Chun Huang, Hao-Wen Chen, and Meng-Hsuan Tien, 2023, “Predicting Nonlinear Modal Properties by Measuring Free Vibration Responses”, Journal of Computational and Nonlinear Dynamics, Vo l. 18, Issue 4, 041005

2023
[13]

Experimental Modeling and Amplitude-Frequency Response Analysis of a Piecewise Linear Vibration System

Yixia Sun, 2020, “Experimental Modeling and Amplitude-Frequency Response Analysis of a Piecewise Linear Vibration System”, IEEE Access, Vo l. 9, pp. 4279-4290

2020
[14]

Efficient Hybrid Symbolic- Numeric Computational Method for Piecewise Linear Systems with Coulomb Friction

Amir Shahhosseini, Meng-Hsuan Tien, and Kiran D’Souza, 2023, “Efficient Hybrid Symbolic- Numeric Computational Method for Piecewise Linear Systems with Coulomb Friction”, Journal of Computational and Nonlinear Dynamics, Vo l. 18, Issue 7, 071004

2023
[15]

Data-driven model order reduction for structures with piecewise linear nonlinearity using dynamic mode decomposition

Akira Saito and Masato Tanaka, 2023, “Data-driven model order reduction for structures with piecewise linear nonlinearity using dynamic mode decomposition”, Nonlinear Dynamics, Vo l. 111, pp. 20597-20616

2023
[16]

Theoretical and Experimental Identification of Cantilever Beam with Clearances Using Statistical and Subspace-Based Methods

Bing Li, Luofeng Han, Wei Jin, and Shuanglu Quan, 2016, “Theoretical and Experimental Identification of Cantilever Beam with Clearances Using Statistical and Subspace-Based Methods”, Journal of Computational and Nonlinear Dynamics, Vo l. 11, Issue 3, 031003

2016
[17]

Nonlinear system identification with continuous piecewise linear neural network

Xiaolin Huang, Jun Xu, and Shuning Wang, 2012, “Nonlinear system identification with continuous piecewise linear neural network”, Neurocomputing, Vo l. 77, Issue 1, pp. 167-177

2012
[18]

Extended dynamic mode decomposition with dictionary learning: A data-driven adaptive spectral decomposition of the Koopman operator

Qianxiao Li, Felix Dietrich, Erik M. Bollt, and Ioannis G. Kevrekidis, 2017, “Extended dynamic mode decomposition with dictionary learning: A data-driven adaptive spectral decomposition of the Koopman operator”, Chaos, Vo l. 27, Issue 10, 103111

2017
[19]

Approximation by Superpositions of a Sigmoidal Function

George Cybenko, 1989, “Approximation by Superpositions of a Sigmoidal Function”, Mathematics of Control, Mathematics of Control, Signals, and Systems, Vo l. 2, pp303-314

1989
[20]

Multilayer Feedforward Networks are Universal Approximators

Kurt Hornik, Maxwell Stinchcombe, and Halbert White, 1989, “Multilayer Feedforward Networks are Universal Approximators”, Neural Networks, vol. 2, Issue 5, pp. 359-366

1989
[21]

Approximation by Superposition of Sigmoidal and Radial Basis Functions

Hrushikesh N. Mhaskar and Charles A. Micchelli, 1992, “Approximation by Superposition of Sigmoidal and Radial Basis Functions”, Advances in Applied Mathematics, Vo l. 13, Issue 3, pp. 350-373

1992
[22]

Neural network with unbounded activation functions is universal approximator

Sho Sonoda and Noboru Murata, 2017, “Neural network with unbounded activation functions is universal approximator”, Applied and Computational Harmonic Analysis, Vo l. 43, Issue 2, pp. 233-268

2017
[23]

Information theory and an extension of the maximum likelihood principle

Hirotugu Akaike, 1973, “Information theory and an extension of the maximum likelihood principle”, Proceedings of the Second International Symposium on Information Theory, B. N. Petrov and F. Csaki, eds., Akademiai Kiado, Budapest, Hungary, pp. 267-281

1973