Predictive Algorithms for Structural Integrity in Sculptures

 

Dr. Pragati Pandit ¹, Damanjeet Aulakh ², Dr. S Nithya ³, Sudhanshu Dev ⁴, Neha ⁵, Dr. Hradayesh Kumar ⁶

 

¹ Assistant Professor, Department of Information Technology, Jawahar Education Society's Institute of Technology, Management and Research, Nashik, India

² Centre of Research Impact and Outcome, Chitkara University, Rajpura- 140417, Punjab, India

³ Assistant Professor, Department of Computer Science and Engineering, Presidency University, Bangalore, Karnataka, India

⁴ Chitkara Centre for Research and Development, Chitkara University, Himachal Pradesh, Solan, 174103, India

⁵ Assistant Professor, School of Business Management, Noida International University 203201, India

⁶ Assistant Professor Department of Arts, Mangalayatan University, Aligarh, India

 

 

1. INTRODUCTION

Sculpture conservation has been a fragile and interdisciplinary practice, and it needs an interpretation of artistic intentions, material behavior, and environmental forces that determine long-lasting stability. The inherent vulnerability of sculptures to structural deterioration due to weathering, stress due to mechanical forces, pollution, biological growth, and fatigue of inner material, make sculptures, whether implemented of stone, casted in metal, modeled in clay or made of modern composite materials, prone to deterioration by structural factors Aziz et al. (2025). Those conventional technique of conservation are more based on visual observation, hand examination and periodic condition evaluation, which, still useful, tend to fail in detecting subtle or internal damage until it is in a critical phase. These responsive methods restrict the conservation of the conservators to act in a strategic manner, often leading to expensive restorations or permanent losses Jayasinghe et al. (2025). To these challenges, the predictive modeling has been introduced as some kind of revolution in preservation of art. The accuracy of predictive algorithms, especially those that integrate machine learning, predictive technology, and real-time sensing technologies, is unprecedented to predict the vulnerability of structures before they actually occur physically Yang and Huang (2025). These models can be used to proactively, data-driven address sustainability of sculpture by analyzing complex stress distributions, simulating the crack propagation, and adding environmental and historical data to sculpture conservation and improve safety and conservation planning. Nevertheless, the conventional forms of assessment remain predominant, as access to digital infrastructure, lack of technical training of conservators, and apprehension about the intrusiveness or cost-effectiveness of monitoring technologies still mitigate the potential of other technologies Latif et al. (2022). The proposed research will fill this gap by describing a unified paradigm of predictive algorithm development that is highly specific to structural integrity assessment of sculptures. The research is centered on three fundamental aims: enhancing the accuracy of the forecasting mechanical and environmental failure modes; creating the workflows that combine the 3D scanning, sensor data, and the material behavior in the past; and assessing the performance of the algorithms using the case studies of various materials and environmental conditions. Moreover, the study aims at providing the principles of the working implementation so that the predictive modeling instrument would be non-invasion, accessible, and ethically in line with conservation principles. Offering an introduction to modern predictive algorithm methods such as FEA-based simulation methods, unsupervised and supervised machine learning methods and hybrid systems that adjust on-the-fly via sensor feedback this paper highlights the sheer importance of computational methods when protecting the heritage of sculptures Rashidi Nasab and Elzarka (2023). With the cultural sector becoming more and more open to the digital technologies, predictive algorithms can be seen as the future of conservation as it becomes less of a reactive practice and more of a scientifically grounded, proactive practice.

2.  Literature Review

Traditional methods of structural analysis of art objects have always been based on qualitative observation, material knowledge, and physical investigation, and the conservators have used empirical knowledge to diagnose the object, determine patterns of stress distribution, and prescribe restoration strategies Rashidi Nasab and Elzarka (2023). The early conservation practices did not use standardized tools of analysis, so their evaluations were usually subjective, extremely variable, and lacked predictive power, particularly when applied to sculptures with complicated internal geometries or internal defects of structure Gao and Elzarka (2021). The introduction of scientific methodologies into the conservation sector saw the scholars starting to use engineering-oriented methods of analysis in the conservation field, and quantitative analysis of the structural system became a fundamental transition to a qualitative assessment of the conservation problem. Among them, one of the most powerful methods to calculate stress distribution, load responses, and deformation and fracture behavior of cultural heritage artifacts was the so-called finite-element analysis (FEA) Hu et al. (2023). FEA was demonstrated to be useful in conservation of sculptures and other objects of interest to conservators by studies that used FEA to determine the mechanical stability of stone statues, bronze casts, wooden carvings, and modern installations, and provided conservators with a more insight into the effects of long-term environmental influences on structural integrity (humidity changes, thermal expansion, wind load, vibration, etc.). Although the FEA has its advantages, it needs specific material data and finer geometries, which may be hard to get on old or heterogeneous objects, and scientists have sought alternative computational methods to FEA.

Over the past few years, machine learning has diversified the scope of tools of analysis that can be used in material science and structural monitoring to allow automatic pattern recognition, anomaly detection, and predictive modeling using past data and sensor data streams Siahpour et al. (2022). The support vector machines, neural networks, and clustering techniques are machine learning algorithms that have been effectively utilized to forecast the crack propagation, identify the early signs of corrosion, identify the patterns of deterioration, and determine the material fatigue in the field of engineering and adaptation to the conservation domain is increasing Zhao et al. (2024). These methods are used to address certain limitations of deterministic models since they can learn incomplete or uncertain data hence are especially helpful in artworks with material properties which change with age, craftsmanship, and environmental exposure Ao et al. (2025). In line with machine learning innovations, sensor-based monitoring systems have been on the rise as viable tools to conduct real-time structural evaluation. Strain gauges, RFID-based stress sensors, accelerometers, fiber optic sensors, and environmental temperature, humidity, and air quality modules are some of the technologies that give continuous data that enable conservators to identify emerging risk factors and confirm the simulations models Esteghamati and Flint (2021). Sculpture conservation Sculpture conservation has been applied to monitor micro-cracks, stress caused by vibration, internal moisture migration, and surface displacement, and has been highly successful in both indoor museums and outdoor heritage environments Ko et al. (2021). Sensors networks combined with machine learning and FEA can also be utilized to enable the hybrid predictive frameworks, which change with time to enhance the accuracy and reliability of structural predictions Abdelmalek-Lee and Burton (2023). Taken together, the current literature has shown considerable improvements in the field of computational conservation, but there are still issues with implementation cost, interoperability of data and interdisciplinary expertise is required. With the further development of research, the intersection of engineering simulation, artificial intelligence, and sensor technology is an attractive trend to create powerful predictive systems that will be able to provide preservation of sculptural artworks to future generations Jin et al. (2023).

Table 1

 

3. Methodology

3.1. DATA ACQUISITION

3.1.1. 3D Scanning Techniques

The basis of the development of predictive models with integrity assessment of sculptures is the scanning in 3D. Laser scanning involves high precision LiDAR beams to scan the geometry of the sculpture in dense point clouds with accuracy down to sub-millimeters. This method is particularly useful to record complicated surface data, complicated carvings, and uneven shapes that affect the distribution of stress. Laser scanners can also be acquired at high rates and this makes them suitable in the indoor and the outdoor environment with minimum physical contacts. On the other hand, photogrammetry is based on the acquisition of high-resolution images in different angles and processing them with computer vision algorithms to create a three-dimensional representation of the sculpture. Photogrammetry is flexible, less expensive and highly movable thus suitable to remote location or whereby the conventional scanning devices cannot be moved to. Whereas laser scanning is more geometrically accurate, photogrammetry is better at textural fineness, color gradient, and surface damage, and material heterogeneity. All these techniques form a rich digital twin of the sculpture, which is necessary to make structural simulations reliable and predictive modeling.

3.1.2. Material Characterization and Historical Degradation Data

Characterization of materials is also important in developing the right predictive algorithms because sculptures are usually made of heterogeneous materials which change with time. This step implies determining the main composition of the sculpture, i. e. stone, bronze, marble, wood, or composite materials, and examining its mechanical, thermal and chemical characteristics. Micro-XRF, ultrasound testing, and micro-indentation are some of the techniques used to measure density, porosity, elasticity, hardness, and internal flaws. The parameters are directly into simulation models so that predicted responses to stress are realistic of material behaviour in the real world. Besides the current characterization, historical degradation data will give an understanding of the deterioration trends over a long period of time. The historical and past restorations, climatic records, and photographic documentation of the sculpture indicate how it reacted to aging, exposure to the environment, and the mechanical loads over the decades. This kind of data increases the predictive accuracy of such data by enabling algorithms to learn trends of deterioration as opposed to using only short term data. The influence of previous environmental conditions like freeze-thaw or exposure to pollution or a seismic event could also be captured with historical data and this could have created hidden structural weaknesses. Material analysis can be utilized together with historical degradation datasets to provide a robust model calibration that can provide predictive algorithms with high-precision crack propagation, deformation and material fatigue. This is a holistic solution so that the modeling framework only reflects not the current state of the structure but the dynamic ageing process of the sculpture.

 

3.1.3.  Environmental and Mechanical Load Data Collection

Environmental and mechanical loading information is necessary to know the external forces which cause structural degradation. Parameters of environmental data are temperature changes, humidity cycles, solar radiations, precipitation, and wind pressures which determine diffusion of moisture, thermal expansion and surface erosion. Mechanical load data, in its turn, records vibrations, constant loads, human interaction dynamic forces and installation-induced stresses. Accelerators, hydro sensors, thermocouples, and load sensors can be used to monitor continuously, which is a valuable time-series data to be used in predictions. These inputs are used to test the simulated stress conditions of the real world and validate long-term predictive algorithms of the structural integrity.

 

3.2. ALGORITHMIC FRAMEWORK

3.2.1. fFINITE-ELEMENT MODELING PROCEDURES

1)    Geometry Preparation: Import of the 3D-scanned model, and simplification of it to eliminate noise and unwarranted detail.

2)    Mesh Generation: Subdivision of the geometry into small finite elements (tetrahedral or hexahedral), which is highly mesh-dense in stress prone areas.

3)    Material Assignment: Consenting characterized material characteristics including the Youngs modulus, the Poisson ratio and density.

4)    Boundary Condition Set-up: The support points, fixed constraints, and free movement areas.

5)    Load Application - Subjecting environmental and mechanical load environments, such as temperature change, vibration and gravity.

6)    Simulation Run: Static, Dynamic and Thermal stress Simulations are run to monitor deformation, strain distribution and possible areas of failure.

7)    Results Interpretation Visualization of stress maps, risk region identification and export data to integrate with machine learning models.

                                                                         Figure 1

Figure 1 Finite-Element Modeling Cycle

 

The Figure 1 shows the work flow of finite-element modeling sculpture integrity analysis. It starts with the preparation of geometry then mesh generation and assigning the material. Limit conditions and loads are introduced and the simulation is simulated. The cycle is completed with the help of stress visualization and data export. The color scheme and symbols make it easier to read and comprehend, and difficult stages of the simulation become understandable to both researchers and conservators.

 

3.2.2. Machine Learning Models

Machine learning improves predictive accuracy by distilling patterns of data which are perhaps not entirely represented by the traditional models.

Linear regression, random forest regression and gradient boosting predictors have been used as regression tools to predict quantitative measures of deterioration such as crack growth rate, moisture accumulation or structural displacement. These models are effective in the estimation of long-term material behavior utilizing both historical and sensor based data.

Isolation forests and autoencoders are anomaly detection models that are specialized at detecting unusual deviation of normal structural behavior and can be used to detect micro-fractures or abrupt environmental stress early. Such models have an added advantage when the number of labeled datasets is scarce.

Deep learning and their neural networks are effective at handling nonlinear associations and high dimensional data. Surface imagery (convolutional neural networks or CNNs) can be used to identify early signs of structural instabilities such as erosion or corrosion, whereas time-serial sensor readings (recurrent neural networks or RNNs and LSTMs) are used to predict them. Collectively, these models form a holistic system of analysis that is able to predict various modes of failure with a high level of accuracy.

 

3.3. MONITORING SYSTEM DESIGN

3.3.1. Types of Sensors

The range of sensors used in monitoring systems is immense and each sensor has a particular diagnostic use. Strain gauges detect micro-deformations and determine the location of stress concentrations which could be the precursor of crack formation. Accelerometers record vibration patterns due to the wind, traffic or a mechanical interaction, and so are vital in the dynamic load analysis. Humidity and temperature sensors monitor changes in the environment that can cause expansion and contraction of materials as well as weathering on the surface. These sensors form a continuous feedback system together which captures the response of the sculpture to environmental exposure as well as mechanical forces.

3.3.2.  Data Transmission, Sampling Frequency, and Storage

The transmission of the data is usually based on the wireless protocols: Wi-Fi, LoRaWAN, or the Bluetooth Low Energy, based on the local conditions and power supply. The sampling frequency depends on the parameter to be measured: Vibration measurements can be sampled with high frequency whereas thermal measurements can be sampled with low frequency. Information is stored on edge devices or sent over cloud servers where it is analyzed over a long period, thus providing safety in long-term archiving and accessibility to predictive modelling.

3.3.3.  Real-Time Processing Pipelines

The real-time pipelines is used to filter the incoming data, look at the anomalies, and update prediction models in real-time. Edge computing devices are used to have rapid initial analysis solving the latency and bandwidth consumption. At the same time, centralized processing systems summarize long-term data sets, improve machine learning forecasts and send notifications in case of the critical thresholds. This real time responsiveness is an improved measure of early intervention and proactive measures in conservation.

                                                                         Figure 2

 

4. Predictive Modeling and Analysis

4.1. STRESS DISTRIBUTION AND FAILURE MODE SIMULATIONS

The key part of a predictive modeling process is represented by stress distribution and failure mode simulations that allow assessing the load behavior in the sculpture structure. Stress maps under different conditions of environmental and mechanical loads are produced using the finite-element analysis, such as gravitational stress, thermal expansion, moisture swelling, and vibrational forces. These simulations show areas of critical stress concentration, which points to areas prone to cracking, deformation, or fatigue of material. Differences in temperature, e.g., generate expansion-contraction cycles that cause concentration of stresses in joints and thin-sections, whereas mechanical vibrations cause temporary stress peaks which can build up with time. Possible failure modes including buckling, surface flaking, shear cracking or tensile rupture are determined at a very early stage through iterative simulations and therefore the conservators can prioritize interventions. Changes in environmental conditions or installation support also can be analyzed using the simulation platform to determine the effect of changes on stress behavior of a particular system on a what-if basis. The given predictive ability allows developing a more advanced strategy of conservation planning and offer quantitative resources to justify structural reinforcements, material stabilization, or display changes.

5.  Result and Analysis

The crack propagation and deformation prediction outcomes refer to how the micro-cracks that are present develop with time as a result of an interaction between the environmental and mechanical effects. The length of the initial crack as indicated in the table is a significant factor in the propagation rate as well as the long-term failure time. Cases whose initial cracks are larger like Case 3 (3.1 mm) have significantly higher propagation rates (0.40 mm/month) resulting in higher ultimate deformation and reduced time to failure. This trend shows that crack detection at an early-stage is very important since the rate of crack propagation increases as stress is built up at the end of the crack. The risk index is highly related with the propagation rate and the magnitude of deformation: the cases in which the tendency towards deformation is higher (1.12 to 1.47 mm) possess a higher level of risk that surpasses 50 percent meaning that the structural deterioration is imminent. There is also the phenomenon of environmental load interactions enhancing crack evolution with swelling due to humidity augmenting the displacement of crack openings, and vibration accelerated propagation due to fatigue. The model is used to determine these interactions with time-dependent simulations and a clear insight into nonlinear degradation patterns is realized. The deformation values in the table can be used to show how the displacement is spread beyond the immediate area in the crack, which influences the stability of the sculpture as a whole. In high-risk cases final deformation exceeds 1 mm which is a level that may be linked with visible instability in heritage material.

The predictions of failure time demonstrate the actual applicability of this model. The highest rate and risk of case 3 indicate that the failure window is only 11 months, which is the reason why an urgent intervention is necessary. On the other hand, less hazardous cases such as Case 2 offer greater timeframes, and the conservation can be planned and preventive. Taken together, this analysis provides a predictive perspective in a wholesome manner so that conservators can group the threats, ranking the intervention areas, and institute specific reinforcement measures in line with the available empirical evidence.

Table 2

Table 2 Sensitivity Analysis Results
Parameter Tested	Variation (%)	Impact on Max Stress (%)	Impact on Deformation (%)	Stability Rating
Young’s Modulus	±10	18	14	Medium
Density	±8	6	9	High
Thermal Expansion Coefficient	±12	22	27	Low
Humidity Absorption	±15	25	31	Low
Boundary Constraints	±5	11	8	Medium

 

 The results of the sensitivity analysis are shown in Table 2, and they demonstrate the effect of the changes in the main parameters on the results of the stress and deformation. The modulus of Young and boundary constraints has moderate effects so it is possible to conclude that the mechanical stiffness and support conditions have a strong impact on stress distribution. Density has small effect, since it has constant effects on structural response as is shown in Figure 3 in sensitivity analysis. On the other hand, the effect on the thermal expansion coefficient and the ability of the substance to absorb humidity are the most powerful, which proves that the environmental factor has an enormous effect on deformation and long-term weakening. The stability ratings also point to the parameters that need closer characterisation in order to gain predictive modeling reliability and sound structural simulations.

                                                                         Figure 3

Figure 3 Sensitivity Analysis – Impact on Max Stress

 

Table 3 is the evaluation of predictive model performance based on accuracy, RMSE, prediction intervals and metrics of error reduction. The regression and anomaly detection models have a good performance and have more scattered prediction ranges due to less trust in small scale deterioration predictions. The CNN and LSTM models perform better because they can capture both spatial and time patterns leading to low errors.

Table 3

Table 3 Predictive Model Evaluation Results Analysis
Model Type	Accuracy (%)	RMSE	Prediction Interval (95%)	Error Reduction (%)
Regression Model	87	0.42	±0.88	12
Anomaly Detection	82	0.51	±1.02	9
CNN Surface Model	90	0.36	±0.74	15
LSTM Time-Series Model	93	0.29	±0.62	18
Hybrid Model (ML + FEA)	96	0.21	±0.48	24

The hybrid ML+FEA model is the best as it has the highest accuracy and minimum RMSE. Its large rate of error minimization shows the usefulness of integrating physical simulation and machine learning, providing the most accurate structural integrity forecasts, example of accuracy in Figure 4.

                                                                         Figure 4

Figure 4 Predictive Model Evaluation Comparison of Accuracy

 

6. Case Studies

6.1. APPLICATION TO HISTORICAL OR CONTEMPORARY SCULPTURES

Both past and present sculptures were fed in predictive modeling to prove the flexibility of the algorithmic structure. In ancient stone carvings especially those that have been exposed to changes in the environment over centuries the models put emphasis on long-term stress corrosion and the development of micro-cracks caused by the variation occurring in humidity and thermal contraction. A large number of these sculptures had areas of critical stress around the joints, excavation cuts, and slender projections, and this evidence supports the usefulness of digital simulations in identifying concealed weaknesses. In comparison, the modern metal and composite sculptures had various degradation characteristics with vibrational fatigue, welding residual stress, and corrosion-based thinning as the predominant. Predictive algorithms were able to distinguish these material responses and provide custom information on the predicted deformation, fatigue life and risk of failure. The procedure confirmed the usefulness of a 3D scan, FEA, and machine learning model integration to produce trusted structural predictions on artistic periods, fabrication styles, and conditions.

 

6.2. MODEL PERFORMANCE COMPARE THE PERFORMANCE OF MATERIALS (STONE/METAL/WOOD/COMPOSITE)

The analysis of the performance of various materials indicated that there were unique computational behaviors related to the material stiffness, density, and degradation mechanisms. In the case of stone sculpture, there was high conformity in simulation and measured deformation, because of the elastic-brittle behaviour of mineral substrates which are predictable. The responses to vibrations were more dynamic with metals where algorithms were required to be trained with vibration-awareness to gain an appropriate interpretation of fatigue patterns.

                                                                         Figure 5

 

The sculptures made in wood yielded more complicated forecasts due to anisotropic grain patterns and sensitivity to moisture and needed hybrid sensor-ML combination to make accurate forecasts. Real sculptures showed a consistent performance because of homogeneous properties, which permit machine learning models to provide a high accuracy with little tuning as presented in Figure 5. In general, the predictive framework exhibited a strong cross-material generalization, and hybrid FEA-ML systems performed better in comparison to single-model based approaches.

 

7. Discussion

7.1.  MERITS AND WEAKNESSES OF PREDICTIVE ALGORITHMS.

Predictive algorithms have significant value in conservation of sculptures due to the ability to identify structural weaknesses early, measure stress behavior, and predict the dynamics of deterioration, which is impossible to track manually. Their capacity to incorporate environmental information, the material properties and historical degradation trends generate a comprehensive information about the structural health. Nevertheless, there are some disadvantages especially, the requirement of quality datasets, expert calibration of simulation parameters and the cost of computation with high density 3D-meshes. Also, machine learning models can have difficulties in extrapolation when historical data is rare and lumpy or inconsistent and sensor networks can create noise or gaps in long-term measurements.

 

7.2. IMPLICATIONS TO PRACTITIONERS, I.E. CONSERVATORS AND ENGINEERS.

Predictive modeling integrated into the processes of conservation enables the conservator and structural engineers to shift operations of restoration back to maintenance. These algorithms are evidence-based reinforcement strategy recommendations, environmental control modification, display configuration modification and restoration priority. The conservation teams can more effectively distribute resources by determining the timing when the failure will take place. The better load simulations have been useful to engineers to design safer mounting systems, supports and display structures (particularly to fragile or full-sized artworks).

 

7.3.  ETHICAL AND PRESERVATION-RELATED ISSUES.

Predictive modeling should be morally justified so that it does not affect the integrity of art. The excessive use of intrusive sensors or unneeded support may change the physical or aesthetic nature of a piece of art. The interpretation of data should also be open and not based on automated research options without the supervision of conservators. Ethical principles also favor minimal intervention, i.e. predictive insights ought to be used to create preventive actions instead of taking excessive restoration. Fair access to technology is another issue: smaller museums might have financial limitations, and this will establish an imbalance in the ability to conserve.

 

7.4. COOPERATION WITH MUSEUM AND DISPLAY OUTDOOR PROTOCOLS.

The predictive insights can be appropriately incorporated into museum and outdoor display procedures by informing the decision-making process of light deployment, air conditioning, and physical handling and installation procedures. In the case of the indoor environments, models can be used to minimize the microclimate settings, as well as the isolation of vibration to increase the lifewell of the delicate sculptures. In the out-of-doors environment, the forecasted outcomes guide preventive actions against weather, pollution and temperature extremes in the choice of coating, sheltering buildings or seasonal movement. Algorithms can be used to create alerts that are part of the daily monitoring of museums and allow the staff to react to the noticed threats. This unification guarantees the long-term preservation management that is driven by data and is sustainable.

 

8. Conclusion

The advent of predictive algorithms is a paradigm shift in the paradigm of structural integrity measurement as well as maintenance of sculptures, and a scientific and proactive addition to the classic forms of conservation undertaking. This multidisciplinary framework can elucidate the current and future structural behavior with high precision and accuracy due to the combination of high-resolution 3D scanning, detailed material characterization, environmental observations, the finite-element modeling, and sophisticated machine learning models. Simulation of stress response, prediction of crack propagation and quantification of deformation patterns help conservators and engineers to identify emerging risk early before they are apparent, eliminating the invasive solutions and prolonging the lifespan of artworks. The case studies reveal how these predictive tools can be used in a wide range of various material, including brittle stone and anisotropic wood to dynamic metals and modern composite materials, which is indicative of the flexibility and consistency of hybrid FEA-ML modeling strategies. Despite the fact that some of these issues like high-quality datasets, computational requirements, and ethical considerations that are to be maintained on an ongoing basis persist, the prospects of predictive modeling provide a strong basis of data-driven conservation decisions. With the further adoption of sensor-based monitoring systems in museums and outside heritage sites, some real-time data streams will enhance the accuracy of algorithms and risk prediction. Finally, predictive algorithms not only allow contributing to the scientific integrity of conservation policies but also assist in protecting cultural heritage by making sure that sculptures are preserved in the least possible way and according to their historical and artistic potential.

 

CONFLICT OF INTERESTS

None. 

 

ACKNOWLEDGMENTS

None.

 

REFERENCES

Abdelmalek-Lee, E., and Burton, H. (2023). A Dual Kriging–Xgboost Model for Reconstructing Building Seismic Responses Using Strong Motion Data. Bulletin of Earthquake Engineering, 1–27.

Ao, Y., Li, S., and Duan, H. (2025). Artificial Intelligence-Aided Design (AIAD) for Structures and Engineering: A State-of-The-Art Review and Future Perspectives. Archives of Computational Methods in Engineering, 32, 4197–4224.  https://doi.org/10.1007/s11831-025-10264-1   

Aziz, M. T., Osabel, D. M., Kim, Y., Kim, S., Bae, J., and Tsavdaridis, K. D. (2025). State-of-the-Art Artificial Intelligence Techniques in Structural Engineering: A Review of Applications and Prospects. Results in Engineering, 28.  https://doi.org/10.1016/j.rineng.2025.107882  

Esteghamati, M. Z., and Flint, M. M. (2021). Developing Data-Driven Surrogate Models for Holistic Performance-Based Assessment of Mid-Rise RC Frame Buildings at Early Design. Engineering Structures, 245, 112971.

Gao, C., and Elzarka, H. (2021). The Use of Decision Tree Based Predictive Models for Improving the Culvert Inspection Process. Advanced Engineering Informatics, 47, 101203.

Hu, S., Wang, W., Alam, M. S., Zhu, S., and Ke, K. (2023). Machine Learning-Aided Peak Displacement and Floor Acceleration-Based Design of Hybrid Self-Centering Braced Frames. Journal of Building Engineering, 72, 106429.

Jayasinghe, S., Mahmoodian, M., Alavi, A., Sidiq, A., Sun, Z., Shahrivar, F., Setunge, S., and Thangarajah, J. (2025). Application of Machine Learning for Real-Time Structural Integrity Assessment of Bridges. Civil Engineering, 6(1), Article 2.  https://doi.org/10.3390/civileng6010002  

Jin, T., Cheng, X., Xu, S., Lai, Y., and Zhang, Y. (2023). Deep Learning Aided Inverse Design of the Buckling-Guided Assembly for 3D Frame Structures. Journal of the Mechanics and Physics of Solids, 179, 105398.

Ko, H., Witherell, P., Lu, Y., Kim, S., and Rosen, D. W. (2021). Machine Learning and Knowledge Graph Based Design Rule Construction for Additive Manufacturing. Additive Manufacturing, 37, 101620.

Latif, I., Banerjee, A., and Surana, M. (2022). Explainable Machine Learning Aided Optimization of Masonry Infilled Reinforced Concrete Frames. Structures, 44, 1751–1766.

Mousavi, M., Ayati, M., Hairi-Yazdi, M. R., and Siahpour, S. (2023). Robust Linear Parameter Varying Fault Reconstruction of Wind Turbine Pitch Actuator Using Second-Order Sliding Mode Observer. Journal of Electrical and Computer Engineering Innovations, 11, 229–241.

Rashidi Nasab, A., and Elzarka, H. (2023). Optimizing Machine Learning Algorithms for Improving Prediction of Bridge Deck Deterioration: A Case Study of Ohio Bridges. Buildings, 13(6), 1517.  https://doi.org/10.3390/buildings13061517  

Siahpour, S., Ayati, M., Haeri-Yazdi, M., and Mousavi, M. (2022). Fault Detection and Isolation of Wind Turbine Gearbox Via Noise-Assisted Multivariate Empirical Mode Decomposition Algorithm. Energy Equipment and Systems, 10, 271–286.

Yang, L., and Huang, E. (2025). Structural Health Monitoring Data Analysis Using Deep Learning Techniques. In Proceedings of the 2024 8th International Conference on Electronic Information Technology and Computer Engineering. Association for Computing Machinery, 913–927.  https://doi.org/10.1145/3711129.3711286  

Zhao, P., Liao, W., Huang, Y., and Lu, X. (2024). Beam Layout Design of Shear Wall Structures Based on Graph Neural Networks. Automation in Construction, 158, 105223.