Physics-Informed Framework for Impact Identification in Aerospace Composites

Natália Ribeiro Marinho; Richard Loendersloot; Jan Willem Wiegman; Frank Grooteman; Tiedo Tinga

cs.LGMar 30, 2026

Physics-Informed Framework for Impact Identification in Aerospace Composites

Authors: Natália Ribeiro Marinho, Richard Loendersloot, Jan Willem Wiegman, Frank Grooteman, Tiedo Tinga

Organizations: Engineering Technology Faculty, Department of Mechanics of Solids, Surfaces and Systems, Dynamics Based Maintenance (DBM) Group, University of Twente (UT), Enschede, NL · Department of Aerospace Vehicles Integrity and Life Cycle Support (AVIL), Royal Netherlands Aerospace Centre (NLR), Marknesse, NL

Abstract

This paper introduces a novel physics-informed impact identification (Phy-ID) framework. The proposed method integrates observational, inductive, and learning biases to combine physical knowledge with data-driven inference in a unified modelling strategy, achieving physically consistent and numerically stable impact identification. The physics-informed approach structures the input space using physics-based energy indicators, constrains admissible solutions via architectural design, and enforces governing relations via hybrid loss formulations. Together, these mechanisms limit non-physical solutions and stabilise inference under degraded measurement conditions. A disjoint inference formulation is used as a representative use case to demonstrate the framework capabilities, in which impact velocity and impactor mass are inferred through decoupled surrogate models, and impact energy is computed by enforcing kinetic energy consistency. Experimental evaluations show mean absolute percentage errors below 8% for inferred impact velocity and impactor mass and below 10% for impact energy. Additional analyses confirm stable performance under reduced data availability and increased measurement noise, as well as generalisation for out-of-distribution cases across pristine and damaged regimes when damaged responses are included in training. These results indicate that the systematic integration of physics-informed biases enables reliable, physically consistent, and data-efficient impact identification, highlighting the potential of the approach for practical monitoring systems.

Figures & tables

Figure 1: General framework of Physics-Informed Impact Identification (Phy-ID), showing how observational, inductive, and learning biases are combined in a structured integration step to achieve physics-informed impact energy estimation.

Figure 2: High-level overview of the sequential disjoint training process: the displacement network (top) generates a velocity estimate for the impactor mass network (bottom), which in turn produces a mass estimate fed back to the displacement model.

Parameter	Displacement network, $w_{\theta}(\bm{x},t)$	Mass network, $m_{\phi}(\bm{x})$
fully connected layer size, $n_{h}$	32, 64, 128, 256	16, 32, 64
number of hidden layers, $L_{h}$	2, 3, 4	2, 3
learning rate, $lr$	$10^{-2}$ , $10^{-3}$	$10^{-2}$ , $10^{-3}$
activation function	Sigmoid, Softplus, Tanh	Sigmoid, Softplus, Tanh

Table 1: Hyperparameter search space explored for the displacement and mass FCNNs.

Displacement network $w_{\theta}(\bm{x},t)$
Parameter	Value
architecture	FCNN, 3 hidden layers of size 256, activation $\tanh$
output	impact velocity, $\hat{v}_{0}$
learning rate	$1\times 10^{-2}$
loss function	$\mathcal{L}_{\text{disp}}=\lambda_{v0}\mathcal{L}_{v0}+\lambda_{\text{ic}}\mathcal{L}_{IC}+\lambda_{\text{ke}}\mathcal{L}_{KE}$
weighting factors	$\lambda_{v0}=1\times 10^{-4}$ , $\lambda_{\text{IC}}=1\times 10^{-6}$ and $\lambda_{\text{KE}}=1\times 10^{-4}$

Table 2: Implementation settings for the disjoint PINN architecture.

Panel of Figure 3 — Figure 3: Parallel coordinate plots showing the performance of the displacement and mass networks across hyperparameter combinations. ( $n_{h}$ : fully connected layer size; $L_{h}$ : number of fully connected layers; lr : learning rate; and $\sigma$ : activation functions.)

Panel of Figure 4 — Figure 4: Schematic impact test configuration; $\bullet$ piezoelectric sensors (PZT) and $\star$ Impact Locations (IE): (a) drop-tower assembly with impact height ( $h$ ) set by impact energy; (b) stiffened thermoplastic composite structure ( $1600\times 370$ mm) with varying thicknesses $t_{1}=8.26$ mm, $t_{2}=6.30$ mm, and $t_{3}=5.46$ mm; (c) experimental set-up, with stiffeners at the bottom-side of the panel. Adapted from ( Marinho et al., 2025b ; Marinho et al., 2024 ) .

ID	Coordinates [mm]
	x	y
PZT1	255	428
PZT2	252	818
PZT3	250	945
PZT4	187	945
PZT5	125	945

Table 3: Impact locations and sensor network coordinates.

Number of unique targets	73
Minimum energy [J]	3.74
Maximum energy [J]	80.95
Mean energy [J]	35.92
Median energy [J]	29.83
Standard deviation [J]	24.04

Table 4: Impact energy statistics.

Panel of Figure 5 — Figure 5: Phased array ultrasonic scans for different impact locations acquired after the impact events that precede and produce damage, highlighting the impact energy associated with damage onset. Each scan shows backwall attenuation, reflection view and cross-sectional view.

Case a,b	Description	Impact data		Training scenario	Samples
Case a,b		Data avail. [%]	Noise [%]	Energy range c [J]	Samples
P1	Predictive analytics	100	1.0	4–80	146
R1	Data availability sweep	25–100	0.0	4–80	variable
R2	Noise sweep	100	0–5	4–80	73
R3	Combined availability and noise	25–100	0–5	4–80	variable
G1	Out-of-distribution training ranges	100	1.0	variable	146

Table 5: Summary of data processing cases for predictive analytics, robustness analysis and generalisation assessment.

Figure 6: Data-processing workflow highlighting the associated analysis cases: P1 (predictive analytics), R1 (data availability sweep), R2 (noise sweep), R3 (combined availability and noise), and G1 (out-of-distribution training ranges).

Figure 7: Comparison between the original and noisy signals recorded by PZT5 due to an impact event of 10 J at impact location IE1. PZT: piezoelectric

Domain	Group	ID	$\mathbf{w^{m}}$	$\mathbf{r^{m}}$	$\mathbf{s^{m}}$	evaluation
Time	T1	RMS	0.58	0.97	0.57	$\ast\ast$ / $\bullet\bullet$
	T2	TE	0.59	0.95	0.56	$\ast\ast$ / $\bullet\bullet$
	T3	PA	0.61	0.99	0.60	$\ast\ast$ / $\bullet\bullet$
	T4	EPR	0.61	0.95	0.58	$\ast\ast$ / $\bullet\bullet$
	T5	RA	0.34	0.80	0.28	$\ast\ast$ / $\diamond$
Frequency	F1	PCR	0.34	0.81	0.28	$\ast\ast$ / $\bullet\bullet$

Table 6: Feature ranking and selection for defining the physics-based energy indicators.

Panel of Figure 8 — Figure 8: Predictive analytics results for Case P1. MAPE: Mean Absolute Percentage Error

Figure 9: Impact energy inferred after removing architectural priors, surrogate components, and hybrid loss terms.

Figure 10: Impact energy robustness analysis. Measured error across energy bins for (left) varying data availability at zero noise (Case R1), (centre) increasing noise levels at full data availability (Case R2), and (right) combined reductions in data availability and increases in noise (Case R3). MAPE: Mean Absolute Percentage Error

Panel of Figure 11 — Figure 11: Generalisation behaviour for varying training intervals. Predictions are evaluated on Out-Of-Distribution (OOD) samples outside each training range, illustrating extrapolation in impact energy estimation.

Figure 12: Generalisation behaviour without physics-motivated architectural priors, surrogate components, and hybrid loss terms. Predictions on out-of-distribution samples illustrate extrapolation in impact energy estimation.

Criterion	Data-driven	Physics-based	Probabilistic	Physics-informed (Phy-ID)
Accuracy	Low–Medium ( Malekloo et al., 2021 ; Azimi et al., 2020 ; Jia and Li, 2023 )	Low–Medium ( Kralovec and Schagerl, 2020 ; Fan et al., 2021 )	Medium ( Blakseth et al., 2022 ; Arias Chao et al., 2022 )	High
Robustness	Low ( Azimi et al., 2020 ; Nicassio et al., 2025 )	Low–Medium ( Fan et al., 2021 ; Hassani et al., 2021 )	Medium ( Blakseth et al., 2022 ; Arias Chao et al., 2022 )	High
Generalisation	Low ( Azimi et al., 2020 ; Nicassio et al., 2025 )	Low–Medium ( Fan et al., 2021 ; Malekloo et al., 2021 )	Medium ( Blakseth et al., 2022 ; Arias Chao et al., 2022 )	Medium
Interpretability	Low ( Azimi et al., 2020 ; Nicassio et al., 2025 )	High ( Fan et al., 2021 ; Malekloo et al., 2021 )	Low–Medium ( Blakseth et al., 2022 ; Arias Chao et al., 2022 )	High

Table 7: Qualitative comparison of impact identification strategies under limited and noisy data with incomplete physics.

Structural Health Monitoring

Structural Health Monitoring uses sensor data and computational models to detect, locate, and assess damage in infrastructure and engineered systems.

See all 80 papers

DebrisTracer: Reliable Tracking in Hypervelocity Impact Fast Imaging
Jul 17, 2026Théophane Loloum, Fabien Vivodtzev, David Hébert +3Visual Object Tracking Multi-Object Tracking
CNRS and Sorbonne University
Randomized Atomic Feature Models for Physics-Informed Identification of Dynamic Systems
May 14, 2026Rajiv Singh, Mario Sznaier, Lennart LjungRandom Feature Methods System Identification
∗The MathWorks Inc., 1 Apple Hill Drive, Natick, MA 01760 USA · ∗∗ECE Dept., Northeastern University, Boston, MA 02115, USA · ∗∗∗Div. of Automatic Control, Linköping University, Sweden
Decision-Aware Evaluation of Physics-Informed Surrogates
Jun 5, 2026Daniel Cieślak, Andrzej CzyżewskiSurrogate Modeling Benchmark Design
Gdańsk University of Technology, Gdańsk, Poland

Jul 17, 2026J/K move · Enter open · S save

DebrisTracer: Reliable Tracking in Hypervelocity Impact Fast Imaging

Théophane Loloum, Fabien Vivodtzev, David Hébert, Baptiste Reynier+2

CNRS and Sorbonne University

Visual Object Tracking Multi-Object Tracking

Open paper AskPDF arXiv

This application paper presents DebrisTracer, a framework for the reliable tracking of debris in hypervelocity impact fast imaging. These noisy and highly specific datasets capture the ejection of a large number of debris fragments after the impact of a projectile launched at hypervelocity into a target material. The reliable estimation of debris mass and speed distributions is of major importance in aerospace applications. We document how to extend an off-the-shelf topology tracking framework based on critical point extraction and matching, in order to incorporate domain knowledge and physical assumptions. Our approach automatically produces an accurate and reliable debris tracking, enabling an interpretable visual analysis of this complex space-time phenomenon. Extensive experiments demonstrate the accuracy improvements provided by our approach over established tools used by domain experts in terms of physical validation, specifically via the prediction of the experimental ejected mass and crater depth profiles. We illustrate the utility of our approach across several use cases (with varying impact angles and physics). We show that our statistical summaries enable the visual identification of distinct regimes within the debris population, corroborating and refining prior expectations of domain experts. Our database and our C++ implementation are available at this address: https://github.com/tloloum/DebrisTracer.

Abstract

Structural Health Monitoring

Explore similar work

DebrisTracer: Reliable Tracking in Hypervelocity Impact Fast Imaging

Randomized Atomic Feature Models for Physics-Informed Identification of Dynamic Systems

Decision-Aware Evaluation of Physics-Informed Surrogates