Pith. sign in

REVIEW 2 major objections 5 minor 59 references

Transformer-Based Approach to Optimal Sensor Placement for Structural Health Monitoring of Probe Cards

T0 review · 2 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A hybrid CNN-Transformer trained on finite-element-simulated frequency response functions distinguishes healthy, cracked, and loose-screw probe-card states at 99.83% accuracy, and its attention weights single out a small set of critical sen

desk verdict A careful simulation study with honest limitations, but sample-level cross-validation leaks correlated scenario variants across folds, so the headline accuracy and the attention-based sensor ranking are not yet proven. read the letter →

arxiv 2509.07603 v1 pith:WPQEQ735 submitted 2025-09-09 cs.LG cs.AI

Mehdi Bejani , Marco Mauri , Daniele Acconcia , Simone Todaro , Stefano Mariani This is my paper · ORCID
classification cs.LGcs.AI
keywords structuralhealthmonitoringprobecardtransformerattentionmechanismoptimalsensorplacementfrequencyresponsefunctionfiniteelementsimulationfailuredetection
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 claims that a hybrid CNN-Transformer model can classify the health of a semiconductor probe card—healthy, loose screw, or cracked substrate—from simulated frequency-response data with 99.83% accuracy and 99.86% balanced accuracy, including near-perfect recall for cracks. It further claims that the model's attention weights identify which of 28 candidate sensor locations carry the most diagnostic information, singling out sensors 16, 7, and 9. If true, this would give manufacturers a data-driven way to choose a small, cost-effective set of vibration sensors for continuous monitoring, catching damage early instead of during visual inspections or after test failures. The claims rest on a finite-element-generated dataset whose physical realism is expanded by varying material properties, temperature, and loading, and on a training protocol that applies augmentation only after splitting to avoid leakage.

What carries the argument

TransformerSHM: a hybrid model in which each of the 28 sensor channels is first encoded by a stack of 1D convolutions into a 128-dimensional embedding; a two-layer Transformer encoder with four-head self-attention models inter-sensor relationships; and a final multi-head attention layer aggregates sensor embeddings into a classification, emitting per-sensor attention weights. Those weights—regularized with L1 to encourage sparsity—are the mechanism that doubles as a sensor-importance ranking, i.e., the optimal sensor placement output.

What would settle it

Run leave-one-scenario-out cross-validation, holding out all variants of a given crack or screw configuration together. If accuracy stays near 99.8% the generalization claim holds; if it drops, the reported numbers are inflated by scenario leakage. Alternatively, mount microphones at the attention-flagged locations on a real probe card, introduce a known crack and a loosened screw, and check detection rate against the simulated predictions.

Watch

Extended reading notes

Core claim

The central discovery is that a per-sensor CNN encoder followed by a Transformer encoder, with a final attention layer trained under L1 sparsity, both separates the three probe-card health states and produces a stable per-sensor importance ranking across 30 cross-validation runs. On 3,750 simulated FRF samples, aggregate accuracy is 99.83% and balanced accuracy 99.86%, with crack recall 0.997 and loose-screw recall 0.999; the only notable imperfection is baseline precision of 0.959. Averaging attention weights over all validation samples and models ranks sensor 16 first, with sensors 7 and 9 also high and low-variance, suggesting these locations near ceramic plate interfaces and screw attach

Load-bearing premise

The paper treats the 125 physics-variant simulations of each crack or screw scenario as independent examples, so test folds can contain near-duplicates of training scenarios; if the variants are correlated, the reported accuracy overstates true generalization to new damage states.

Editorial extensions

If this is right

  • A probe card monitoring system could be built with the few high-attention sensors rather than all 28 candidates, simplifying hardware and cutting cost without, the paper argues, sacrificing detection capability.
  • Crack and loose-screw conditions can be distinguished from the healthy baseline with near-perfect recall, so the approach supports early, automated warning rather than reactive maintenance.
  • The attention-based importance ranking offers a data-driven alternative to classical OSP criteria such as the Fisher information matrix or MAC-based methods.
  • The dataset-generation pipeline (FE simulation plus physics-informed scenario expansion) plus the split-then-augment protocol provides a reusable template for SHM research on other probe card designs.
  • The high ROC-AUC of 0.9999 indicates the ranking survives threshold changes, which matters when missed failures are expensive.

Reading between the lines

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

  • The 125 variants of each base scenario are perturbed versions of the same crack or screw configuration, so a sample-level split may let the model memorize scenario patterns; a scenario-grouped split is the test that would reveal whether the reported accuracy holds for truly unseen damage configurations.
  • If the simulated FRFs faithfully represent a physical probe card, then mounting microphones at sensors 16, 7, and 9 should reproduce the high detection rates; this is directly testable on hardware with implanted cracks and loosened screws.
  • The same attention-weight sparsification could be applied to other multivariate sensor fusion tasks to prune redundant channels, though the physical meaning of the weights would need task-specific validation.
  • The paper's baseline precision (0.959) with perfect recall suggests a conservative bias toward flagging damage; tuning the decision threshold could trade away some false alarms at the cost of slightly lower recall.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper proposes TransformerSHM, a hybrid CNN-Transformer model for classifying probe-card health states (baseline, loose screw, crack) from simulated FRF data collected at 28 candidate sensor locations. The training data are obtained from FE simulations and expanded by a physics-informed scenario expansion (5 material cases × 5 temperatures × 5 loads) applied to one baseline, 21 screw-loosening scenarios, and 8 crack configurations, yielding 3,750 samples. The model uses per-sensor 1D CNN encoders, a two-layer Transformer over sensor embeddings, and an attention-based classification module. Evaluation uses 3 repetitions of 10-fold stratified cross-validation, with SMOTE and physics-aware augmentation applied only to training folds. The paper reports 99.83% accuracy, 99.86% balanced accuracy, 99.7% crack recall, and identifies sensors 16, 7, and 9 as most important via attention weights, which is presented as a basis for optimal sensor placement.

Significance. If the reported generalization is valid, the paper would be a useful demonstration of attention-based deep learning for simulated structural-health monitoring and sensor selection, with careful data-handling practices: augmentation after splitting, class-weighted loss, SMOTE confined to training data, and repeated cross-validation. The physics-informed scenario expansion is a genuine effort to enrich a small FE dataset. However, the headline accuracy depends on treating 125 variants of each base failure scenario as independent samples, which is not justified; and the sensor-importance claim is not validated by ablation or external physical ground truth. The contribution is promising, but the central claims currently outrun the evidence.

major comments (2)
  1. [Sec. 3.2.3 / Sec. 3.4 / Table 3 / Fig. 8] The cross-validation protocol splits at the sample level, not by the 30 base failure scenarios. Each of the 21 screw and 8 crack configurations is expanded into 125 variants (5 material cases × 5 temperatures × 5 loads) that are correlated perturbations of the same underlying damage configuration. Under a random 10-fold split, roughly 90% of variants of a given base scenario appear in training while the remaining 10% appear in validation, so test samples are near-duplicates of training instances. The reported 99.83% accuracy and the near-perfect confusion matrix therefore do not establish generalization to an unseen screw/crack position. Please add grouped or leave-one-scenario-out cross-validation with the scenario identity as the grouping factor, and report metrics for configurations fully held out. If those metrics drop materially, the abstract and Section 5 must be re-scoped to avoid
  2. [Sec. 4.3 / Eq. (2)] The sensor-importance conclusion (sensors 16, 7, 9) is based on attention weights from a model trained on the same simulation data, with an L1 penalty explicitly applied to the attention weights. This penalty is designed to drive many weights toward zero, so the resulting sparsity is partially engineered rather than discovered from the data. The paper does not retrain the classifier on the top-k sensors, compare accuracy against the full 28-sensor set, or relate the ranking to a physically grounded damage-sensitivity analysis. Please add an ablation study (e.g., training with only sensors {16,7,9} and evaluating with grouped CV) or otherwise validate that the selected sensor set retains the classification performance claimed. Without this, the 'critical sensor locations' claim is not load-bearing evidence for optimal sensor placement.
minor comments (5)
  1. [Sec. 3.1.1] The sentence 'Based on the failure analysis and investigation into their root causes...' is duplicated.
  2. [Fig. 8] The labels contain typos: 'Baseli e', 'Co fusio', and 'Overall Confusion' should be corrected.
  3. [Sec. 3.1.2] Typo: 'focuseed' should be 'focused'.
  4. [Eq. (2) / Sec. 3.4] Eq. (2) says 'model weights' but the regularization is applied to attention weights; please clarify. Also, λ = 1e-4 is a free hyperparameter with no sensitivity analysis reported.
  5. [Sec. 4.1 / Sec. 3.4] Table 3 reports results 'averaged over the 30 models,' but the text describes an ensemble of 10 models per fold within a triply repeated 10-fold CV. Please clarify how the 30 models relate to the 10 folds and 3 repetitions, and how predictions are aggregated.

Circularity Check

1 steps flagged · score 4.0 of 10

Classification accuracy is self-contained; OSP sensor ranking partly reduces to L1-regularized attention weights.

  1. fitted input called prediction [Sec. 3.4 (L1 penalty) and Sec. 4.3 / Fig. 11 (sensor importance)]
    "Additionally, L1 regularization was applied to the attention weights from the Multihead Attention layer of the TransformerSHM model. ... By penalizing the absolute magnitude of the attention weights, L1 regularization encourages sparsity, meaning it drives many of the weights to become exactly zero. ... In the context of our model, this promotes sensor selectivity. ... Sensor 16 consistently exhibits the highest attention weight, with sensors such as 7 and 9 also demonstrating significant contributions and low variability, suggesting they can play a pivotal role in failure detection."

    Sec. 3.5 defines sensor importance as the attention weights averaged over heads and validation samples. These weights are fitted parameters of the network, and Sec. 3.4 adds an L1 penalty that intentionally drives most of them to zero. Therefore the 'discovery' that only sensors 16, 7, 9 are critical is partly engineered by the loss: the model is forced to be selective, and the reported ranking is a read-out of the trained weights. The paper does not retrain on the selected sensors or compare against a physical ground truth, so the OSP claim is a renamed description of the fitted attention vector rather than an independent prediction. The classification accuracy is unaffected by this issue, since it is measured on held-out folds of the simulated FRFs.

full rationale

The main classification result (99.83% accuracy, 99.86% balanced accuracy) is not circular: the health-state label is not encoded in the FRF input, and the model is evaluated on held-out folds under 3x10-fold stratified CV. The physics-informed expansion creates correlated variants, so sample-level splits may allow scenario leakage; this is a validity/generalization risk, not a logical circularity. The self-citations [25,53] are methodological and not load-bearing. The one partly circular element is the OSP interpretation: sensor importance is defined as the model's L1-regularized attention weights, so reporting sensors 16, 7, 9 as 'critical' partially restates the fitted weights. The paper's own limitation ('Experimental validation of these OSP findings represents an important next step') reduces the severity but does not remove the presentation of attention weights as an OSP result.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claims rest on simulation fidelity, on the independence of the 125 scenario variants, and on several hand-chosen expansion and augmentation parameters. No new physical entities are introduced. The most consequential assumption is that sample-level cross-validation on correlated variants measures generalization to new failure states.

free parameters (4)
  • L1 attention regularization coefficient lambda = 1e-4
    Hand-chosen in Sec. 3.4; directly shapes the sparsity of attention weights and therefore the sensor importance ranking in Sec. 4.3.
  • Material property variation ranges = SiN: E +/-10%, density +/-5%; MeAl: E +/-5%, density +/-2%
    Chosen by authors as 'reasonable estimates' in Sec. 3.2.3; these bounds define the scenario expansion and hence the dataset.
  • Augmentation hyperparameters = Gaussian noise sigma = 10% of class std; jitter <=3 points; amplitude scaling +/-10%
    Chosen in Sec. 3.2.4; these parameters generate most of the training variance and affect generalization estimates.
  • Temperature and load grids = Temps: 25, 100, 150, 200, 250 deg C; loads: 0.1, 0.5, 1, 1.5, 2 MPa
    Discrete levels selected in Sec. 3.2.3, defining the 125-way scenario expansion.
assumptions (5)
  • domain assumption The FE model is a faithful digital shadow of the physical probe card; simulated FRFs match real structural response.
    Invoked in Sec. 3.2.1; no experimental validation of the FE model is presented.
  • domain assumption Crack and screw-loosening failures produce FRF changes in the 400-4000 Hz band that are detectable and classifiable.
    Supported only by the same simulations (Sec. 3.1.1, Figure 3).
  • domain assumption Physics-informed scenario expansion (material, temperature, load variations) samples the space of realistic operating conditions.
    Sec. 3.2.3; the five discrete cases are not a random sample and are chosen by hand.
  • domain assumption Augmented samples (noise, jitter, scaling) are physically plausible and do not distort class boundaries.
    Sec. 3.2.4; the authors later acknowledge in Discussion that the physical realism of augmented samples 'cannot be definitively guaranteed'.
  • ad hoc to paper Sample-level cross-validation provides unbiased generalization estimates.
    Sec. 3.4 splits samples, not scenarios; the 125 variants per scenario are correlated, so test folds overlap in scenario identity. This is a load-bearing independence assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transformer-Based Approach to Optimal Sensor Placement for Structural Health Monitoring of Probe Cards." pith.science (2026). https://pith.science/paper/WPQEQ735

@misc{pith2026250907603,
  author       = {Pith},
  title        = {Pith review of: Transformer-Based Approach to Optimal Sensor Placement for Structural Health Monitoring of Probe Cards},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WPQEQ735}},
  note         = {Machine review of arXiv:2509.07603}
}
read the original abstract

This paper presents an innovative Transformer-based deep learning strategy for optimizing the placement of sensors aiming at structural health monitoring of semiconductor probe cards. Failures in probe cards, including substrate cracks and loosened screws, would critically affect semiconductor manufacturing yield and reliability. Some failure modes could be detected by equipping a probe card with adequate sensors. Frequency response functions from simulated failure scenarios are adopted within a finite element model of a probe card. A comprehensive dataset, enriched by physics-informed scenario expansion and physics-aware statistical data augmentation, is exploited to train a hybrid Convolutional Neural Network and Transformer model. The model achieves high accuracy (99.83%) in classifying the probe card health states (baseline, loose screw, crack) and an excellent crack detection recall (99.73%). Model robustness is confirmed through a rigorous framework of 3 repetitions of 10-fold stratified cross-validation. The attention mechanism also pinpoints critical sensor locations: an analysis of the attention weights offers actionable insights for designing efficient, cost-effective monitoring systems by optimizing sensor configurations. This research highlights the capability of attention-based deep learning to advance proactive maintenance, enhancing operational reliability and yield in semiconductor manufacturing.

Figures

Figures reproduced from arXiv: 2509.07603 by the authors.

Figure 1
Figure 1. Probe Card Structure and Associated Failure Modes [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. PH model in ANSYS, with a crack in the plate. 3.1.2 Proposed SHM Solution for Mechanical Failures By prioritizing the most critical failure modes and operational challenges of the PCs, this study identifies a tailored set of sensors to enable real-time SHM and predictive failure detection. Specifically, the following sensors are proposed for integration [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Comparison of PH FRF amplitudes for three different scenarios. • Accelerometers: for counting probe touchdowns to estimate contact fatigue and optimize cleaning intervals, thereby supporting predictive maintenance strategies. • Microphones: For capturing acoustic/vibrational signatures indicative of structural anomalies such as cracks or screw loosening, making them highly suitable for FRF-based SHM applications. • … view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Sketch of the proposed sensor network. 3.2.2 Candidate Sensor Locations For OSP, 28 potential sensor locations were identified in the FE model, see [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Representation of the 28 locations on the lower side (left) and upper side (right) of the PH, where FRFs were extracted. 3.2.3 Data Challenges and Physics-Informed Scenario Expansion A significant challenge in the initial simulated dataset was the substantial imbalance…
Figure 6
Figure 6. Figure 6: Adopted coding for the 21 screws used to connect the lower frame (left) to the housing (right) of the PH. screw loosened in each scenario; 8 crack scenarios, 4 in the upper plate and 4 in the lower plate of the PH; and only one single undamaged baseline case. This imba…
Figure 7
Figure 7. Figure 7: Scheme of the DL architecture with the attention mechanism [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Overall confusion matrix, as aggregated from 3 repetitions of the 10-fold cross-validation. M       [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Accuracy distribution across the 30 models [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Model learning curves averaged over the 30 models. 4.3 Sensor Importance Analysis A key advantage of the TransformerSHM model is its attention mechanism, which provides interpretable insights into sensor contributions. Figure 11a displays the mean attention weight per…
Figure 11
Figure 11. Figure 11: Sensor importance analysis: (a) overall sensor importance; (b) sensor importance variability. 5 Discussion This study has addressed PC SHM and failure detection via a hybrid CNN-Transformer DL framework, specifically working on limitations highlighted in prior researc…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 50 canonical work pages

  1. [1]

    T. H. V an Nguyen and C.-F. Chien. Semiconductor probe card proactive maintenance using graph self- supervised learning and an empirical study. Computers & Industrial Engineering , 203:110955, 2025. doi:10.1016/j.cie.2025.110955

    arXiv 2025

  2. [2]

    Chien and H.J

    C.F. Chien and H.J. Wu. Integrated circuit probe card troubleshooting based on rough set theory for ad- vanced quality control and an empirical study. Journal of Intelligent Manufacturing , 35:275–287, 2024. doi:10.1007/S10845-022-02042-8

    work page doi:10.1007/s10845-022-02042-8 2024

  3. [3]

    W. Fu, C.F. Chien, and L. Tang. Bayesian network for integrated circuit testing probe card fault diagnosis and troubleshooting to empower Industry 3.5 smart production and an empirical study. Journal of Intelligent Manufacturing, 33(3):785–798, 2022. doi: 10.1007/S10845-020-01680-0

    work page doi:10.1007/s10845-020-01680-0 2022

  4. [4]

    B. Shin. High-Speed Probe Card Analysis Using Real-time Machine Vision and Image Restoration Technique . Master’s Thesis, University of Waterloo, Waterloo, Ontario, Canada, 2013

    work page 2013

  5. [5]

    C. F. Chien, Y . S. Lin, and S. K. Lin. Deep reinforcement learning for selecting demand forecast models to empower Industry 3.5 and an empirical study for a semiconductor component distributor. International Journal of Production Research, 58(9):2784–2804, 2020. doi: 10.1080/00207543.2020.1733125

    arXiv 2020

  6. [6]

    J. M. Nawaz, M. Z. Arshad, and S. J. Hong. Fault Diagnosis in Semiconductor Etch Equipment Using Bayesian Networks. Journal of Semiconductor Technology and Science, 14(2):252–261, 2014. https://www.dbpia.co. kr/journal/articleDetail?nodeId=NODE02405138

    work page 2014

  7. [7]

    B. Li, T. Han, and F. Kang. Fault diagnosis expert system of semiconductor manufacturing equipment using a Bayesian network. International Journal of Computer Integrated Manufacturing , 26(12):1161–1171, 2013. doi:10.1080/0951192X.2013.812803

    work page arXiv 2013

  8. [8]

    Rostami, J

    H. Rostami, J. Blue, and C. Y ugma. Automatic equipment fault fingerprint extraction for the fault diagnostic on the batch process data. Applied Soft Computing , 68:972–989, 2018. doi: 10.1016/j.asoc.2017.10.029

    work page doi:10.1016/j.asoc.2017.10.029 2018

Show all 59 references
  1. [9]

    Y . M. Bae, S. H. Lee, K. S. Lee, J. H. Lim, K. B. Kim, and B. H. Lee. Detecting abnormal behavior of automatic test equipment using autoencoder with event log data. Computers & Industrial Engineering , 183:109547, 2023. doi:10.1016/j.cie.2023.109547

    2023

  2. [10]

    Y . T. Jou, H. M. Wee, H. C. Chen, Y . H. Hsieh, and L. Wang. A neural network forecasting model for consumable parts in semiconductor manufacturing. Journal of Manufacturing Technology Management, 20(3):404–412, 2009. doi:10.1108/17410380910936828

    2009 doi

  3. [11]

    K. C. C. Cheng, C.-J. Su, T.-C. Chang, C.-J. Du, S. C. Hsieh, and Y .-F. Chen. Machine Learning-Based Detection Method for Wafer Test Induced Defects. IEEE Transactions on Semiconductor Manufacturing , 34(2):161–167,

  4. [12]

    M. S. A. Sulaiman, M. A. Y unus, A. R. Bahari, and M. N. Abdul Rani. Identification of damage based on frequency response function (FRF) data. In MATEC Web of Conferences , volume 90, page 01025, 2016. doi:10.1051/matecconf/20179001025

    2016

  5. [13]

    R. P . B. Kocharla, M. Kolli, and M. Cheepu. Real-Time Detection of Faults in Rotating Blades Using Frequency Response Function Analysis. Applied Mechanics, 4(1):356–370, 2023. doi: 10.3390/appmech4010020

    2023 doi

  6. [14]

    J. E. T. Penny, M. I. Friswell, and S. D. Garvey. Automatic choice of measurement locations for dynamic testing. AIAA journal, 32(2):407–414, 1994. doi: 10.2514/3.11998

    1994 doi

  7. [15]

    Y . K. Zhou and Z. D. Duan. Optimum Sensor Placement for Structural Damage Detection. Journal of Engineer- ing Mechanics, 126(11):1173–1179, 2000. doi: 10.1061/(ASCE)0733-9399(2000)126:11(1173)

    2000 doi

  8. [16]

    Worden and A

    K. Worden and A. P . Burrows. Optimal sensor placement for fault detection. Engineering Structures, 23(8):885– 901, 2001. doi: 10.1016/S0141-0296(00)00118-8

    2001 doi

  9. [17]

    Capellari, E

    G. Capellari, E. Chatzi, and S. Mariani. Cost-Benefit Optimization of Structural Health Monitoring Sensor Networks. Sensors, 18(7):2174, 2018. doi: 10.3390/S18072174. Transformer-Based Approach to Optimal Sensor Placement A PREPRINT

    2018 doi

  10. [18]

    Capellari, E

    G. Capellari, E. Chatzi, and S. Mariani. Structural Health Monitoring Sensor Network Optimization through Bayesian Experimental Design. ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, 4(2):04018016, 2018. doi: 10.1061/AJRUA6.0000966

    2018 doi

  11. [19]

    Malings and M

    C. Malings and M. Pozzi. V alue-of-information in spatio-temporal systems: Sensor placement and scheduling. Reliability Engineering & System Safety , 172:45–57, 2018. doi: 10.1016/j.ress.2017.11.019

    2018 doi

  12. [20]

    Kamariotis, E

    A. Kamariotis, E. Chatzi, and D. Straub. V alue of information from vibration-based structural health moni- toring extracted via Bayesian model updating. Mechanical Systems and Signal Processing , 166:108465, 2022. doi:10.1016/j.ymssp.2021.108465

    2022

  13. [21]

    Huang, J

    M. Huang, J. Li, and H. Zhu. Optimal sensor layout for bridge health monitoring based on dual-structure coding genetic algorithm. In 2009 International Conference on Computational Intelligence and Software Engineering (CiSE), pages 1–4, 2009. doi: 10.1109/CISE.2009.5366481

    2009

  14. [22]

    T. H. Yi, H. N. Li, and M. Gu. Optimal sensor placement for structural health monitoring based on multiple opti- mization strategies. Structural Design of Tall and Special Buildings , 20(7):881–900, 2011. doi: 10.1002/tal.712

    2011 doi

  15. [23]

    T. H. Yi and H. N. Li. Methodology Developments in Sensor Placement for Health Monitoring of Civil Infras- tructures. International Journal of Distributed Sensor Networks , 2012:612726, 2012. doi: 10.1155/2012/612726

    2012 doi

  16. [24]

    T. G. Carne and C. R. Dohrmann. A modal test design strategy for model correlation. Technical Report, Sandia National Labs., Albuquerque, NM, 1994. doi: 10.2172/10114604

    1994 doi

  17. [25]

    J. W. Kim, M. Torzoni, A. Corigliano, and S. Mariani. Attention Mechanism-Driven Sensor Placement Strategy for Structural Health Monitoring. Engineering Proceedings, 27(1):43, 2022. doi: 10.3390/ECSA-9-13354

    2022 doi

  18. [26]

    ANSYS Mechanical, Release R2 2024

    ANSYS, Inc. ANSYS Mechanical, Release R2 2024 . Pittsburgh, PA, 2024

    2024

  19. [27]

    Y . J. Cha, R. Ali, J. Lewis, and O. Büyüköztürk. Deep learning-based structural health monitoring. Automation in Construction, 161:105328, 2024. doi: 10.1016/j.autcon.2024.105328

    2024

  20. [28]

    Jia and Y

    J. Jia and Y . Li. Deep Learning for Structural Health Monitoring: Data, Algorithms, Applications, Challenges, and Trends. Sensors, 23(21):8824, 2023. doi: 10.3390/S23218824

    2023 doi

  21. [29]

    Zhang, Q

    Y . Zhang, Q. Zhou, K. Zhou, and J. Tang. Damage Detection of a Pressure V essel with Smart Sensing and Deep Learning. IF AC-PapersOnLine, 56(3):379–384, 2023. doi: 10.1016/j.ifacol.2023.12.053

    2023 doi

  22. [30]

    C. Li, J. Chen, C. Y ang, J. Y ang, Z. Liu, and P . Davari. Convolutional Neural Network-Based Transformer Fault Diagnosis Using Vibration Signals. Sensors, 23(10):4781, 2023. doi: 10.3390/S23104781

    2023 doi

  23. [31]

    R. Zhao, R. Y an, J. Wang, and K. Mao. Learning to Monitor Machine Health with Convolutional Bi-Directional LSTM Networks. Sensors, 17(2):273, 2017. doi: 10.3390/S17020273

    2017 doi

  24. [32]

    J. Deng, S. Zhang, and J. Ma. Self-Attention-Based Deep Convolution LSTM Framework for Sensor-Based Badminton Activity Recognition. Sensors, 23(20):8373, 2023. doi: 10.3390/S23208373

    2023 doi

  25. [33]

    B. R. Chang, H. F. Tsai, and Y . R. Wu. Detection and Prediction of Probe Mark Damage in Wafer Testing. Electronics, 13(20):4075, 2024. doi: 10.3390/electronics13204075

    2024 doi

  26. [34]

    Toh and J

    G. Toh and J. Park. Review of Vibration-Based Structural Health Monitoring Using Deep Learning. Applied Sciences, 10(5):1680, 2020. doi: 10.3390/app10051680

    2020 doi

  27. [35]

    Hassani and U

    S. Hassani and U. Dackermann. A Systematic Review of Optimization Algorithms for Structural Health Moni- toring and Optimal Sensor Placement. Sensors, 23(6):3293, 2023. doi: 10.3390/s23063293

    2023 doi

  28. [36]

    Konar and P

    P . Konar and P . Chattopadhyay. Bearing fault detection of induction motor using wavelet and Support V ector Machines (SVMs). Applied Soft Computing , 11(6):4203–4211, 2011. doi: 10.1016/j.asoc.2011.03.014

    2011 doi

  29. [37]

    Soualhi, K

    A. Soualhi, K. Medjaher, and N. Zerhouni. Bearing health monitoring based on hilbert-huang transform, support vector machine, and regression. IEEE Transactions on Instrumentation and Measurement , 64(1):52–62, 2015. doi:10.1109/TIM.2014.2330494

    2015

  30. [38]

    A. C. Neves, L. González, J. Leander, and R. Karoumi. Structural health monitoring of bridges: a model-free ANN-based approach to damage detection. Journal of Civil Structural Health Monitoring , 7(5):689–702, 2017. doi:10.1007/s13349-017-0252-5

    2017 doi

  31. [39]

    R. A. Osornio-Rios, J. P . Amezquita-Sanchez, R. J. Romero-Troncoso, and A. Garcia-Perez. MUSIC-ANN Analysis for Locating Structural Damages in a Truss-Type Structure by Means of Vibrations. Computer-Aided Civil and Infrastructure Engineering, 27(9):687–698, 2012. doi: 10.1111...

    2012

  32. [40]

    M. M. Alamdari, T. Rakotoarivelo, and N. L. D. Khoa. A spectral-based clustering for structural health monitoring of the Sydney Harbour Bridge. Mechanical Systems and Signal Processing , 87:384–400, 2017. doi:10.1016/j.ymssp.2016.10.033. Transformer-Based Approach to Optimal S...

    2017 doi

  33. [41]

    J. Tian, M. H. Azarian, and M. Pecht. Anomaly Detection Using Self-Organizing Maps-Based K-Nearest Neigh- bor Algorithm. In PHM Society European Conference, volume 2, 2014. doi: 10.36001/phme.2014.v2i1.1554

    2014 doi

  34. [42]

    Abdeljaber, O

    O. Abdeljaber, O. Avci, S. Kiranyaz, M. Gabbouj, and D. J. Inman. Real-time vibration-based structural damage detection using one-dimensional convolutional neural networks. Journal of Sound and Vibration , 388:154–170,

  35. [43]

    M. A. de Oliveira, A. V . Monteiro, and J. V . Filho. A New Structural Health Monitoring Strategy Based on PZT Sensors and Convolutional Neural Network. Sensors, 18(9):2955, 2018. doi: 10.3390/S18092955

    2018 doi

  36. [44]

    Y . Z. Lin, Z. H. Nie, and H. W. Ma. Structural Damage Detection with Automatic Feature-Extraction through Deep Learning. Computer-Aided Civil and Infrastructure Engineering , 32(12):1025–1046, 2017. doi:10.1111/mice.12313

    2017 doi

  37. [45]

    H. D. M. Onchis. A deep learning approach to condition monitoring of cantilever beams via time-frequency extended signatures. Computers in Industry, 105:177–181, 2019. doi: 10.1016/j.compind.2018.12.005

    2019 doi

  38. [46]

    Y . Zhao, M. Noori, W. A. Altabey, R. Ghiasi, and Z. Wu. Deep Learning-Based Damage, Load and Support Identification for a Composite Pipeline by Extracting Modal Macro Strains from Dynamic Excitations. Applied Sciences, 8(12):2564, 2018. doi: 10.3390/app8122564

    2018 doi

  39. [47]

    C. Li, R. V . Sanchez, G. Zurita, M. Cerrada, D. Cabrera, and R. E. Vásquez. Gearbox fault diagnosis based on deep random forest fusion of acoustic and vibratory signals. Mechanical Systems and Signal Processing , 76-77:283–293, 2016. doi: 10.1016/j.ymssp.2016.02.007

    2016 doi

  40. [48]

    F. Yvon. Transformers in Natural Language Processing. In Lecture Notes in Computer Science, volume 13500 of Lecture Notes in Artificial Intelligence, pages 81–105. Springer International Publishing, 2023. doi: 10.1007/978- 3-031-24349-3_6

    2023 doi

  41. [49]

    Gillioz, J

    A. Gillioz, J. Casas, E. Mugellini, and O. A. Khaled. Overview of the Transformer-based Models for NLP Tasks. In 2020 15th Conference on Computer Science and Information Systems (FedCSIS) , pages 179–183,

    2020

  42. [50]

    J. Li, W. Chen, G. Fan, J. Li, W. Chen, and G. Fan. Structural health monitoring data anomaly detection by transformer enhanced densely connected neural networks. Smart Structures and Systems , 30(6):613–625, 2022. doi:10.12989/sss.2022.30.6.613

    2022 doi

  43. [51]

    Wang et al

    C. Wang et al. Learning to Optimise Climate Sensor Placement using a Transformer. arXiv preprint arXiv:2310.12387, 2023. https://arxiv.org/abs/2310.12387

    2023 arXiv

  44. [52]

    Fatima, S

    N. Fatima, S. Riaz, S. Ali, R. Khan, M. Ullah, and D. Kwak. Sensors Faults Classification and Faulty Signals Reconstruction Using Deep Learning. IEEE Access , 12:100544–100558, 2024. doi:10.1109/ACCESS.2024.3425408

    2024

  45. [53]

    Bejani, D

    M. Bejani, D. Appello, M. Mauri, E. Missaglia, and S. Mariani. Digital Twin-Assisted Optimal Sensor Place- ment for Real-Time Monitoring of Probe Cards in EWS Applications. In 2025 26th International Conference on Thermal, Mechanical and Multi-Physics Simulation and Experiment...

    2025

  46. [54]

    Karmakov and M

    S. Karmakov and M. H. Ferri Aliabadi. Deep Learning Approach to Impact Classification in Sensorized Panels Using Self-Attention. Sensors, 22(12):4370, 2022. doi: 10.3390/S22124370

    2022 doi

  47. [55]

    Zhang, P

    X. Zhang, P . Li, X. Han, Y . Y ang, and Y . Cui. Enhancing Time Series Product Demand Fore- casting With Hybrid Attention-Based Deep Learning Models. IEEE Access , 12:190079–190091, 2024. doi:10.1109/ACCESS.2024.3516697

    2024

  48. [56]

    Rosafalco, A

    L. Rosafalco, A. Manzoni, S. Mariani, and A. Corigliano. Fully convolutional networks for structural health monitoring through multivariate time series classification. Advanced Modeling and Simulation in Engineering Sciences, 7(1):38, 2020. doi: 10.1186/s40323-020-00174-1

    2020 doi

  49. [2017]

    doi: 10.1016/j.jsv.2016.10.043

    2016 doi

  50. [2020]

    doi: 10.15439/2020F20

    doi

  51. [2021]

    doi: 10.1109/TSM.2021.3065405

    2021

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.