2510.0013 A Review of Intelligent Rock Mechanics: From Methods to Applications v7

🎯 ICAIS2025 Submission
Back to Paper List

Paper Content

📄 Open in New Tab

AI Review from DeepReviewer

AI Review available after:
--d --h --m --s

📋 AI Review from DeepReviewer will be automatically processed

📋 Summary

This paper provides a comprehensive review of the integration of artificial intelligence (AI) and machine learning (ML) techniques into the field of rock mechanics. The authors trace the evolution of AI applications in this domain, starting from traditional methods like backpropagation and support vector machines to modern deep learning architectures such as convolutional neural networks (CNNs) and transformers. The paper highlights the diverse applications of AI in rock mechanics, including microstructure reconstruction, mechanical parameter estimation, constitutive modeling, and real-time hazard prediction. The authors emphasize the potential of AI to address key challenges in rock mechanics, such as the complexity of geological materials, the difficulty in obtaining high-quality data, and the need for more accurate and efficient predictive models. The review also discusses emerging techniques like physics-informed neural networks (PINNs) and graph-based learning, which aim to bridge the gap between data-driven models and physical interpretability. Furthermore, the paper explores the potential of large language models (LLMs) in automating code generation and decision support for geotechnical analysis. Despite the progress made, the authors acknowledge ongoing challenges such as data quality, model generalization, and interpretability. They call for the development of standardized datasets, interdisciplinary collaboration, and transparent AI workflows to overcome these issues. The paper concludes with a forward-looking perspective on the development of intelligent frameworks that couple physical knowledge, spatial reasoning, and adaptive learning, aiming to transform rock mechanics from empirical modeling to fully autonomous systems. The authors advocate for a shift towards more integrated and intelligent approaches that leverage the strengths of both AI and traditional rock mechanics principles. The paper serves as a valuable synthesis of the current state of AI in rock mechanics, highlighting both the achievements and the remaining challenges in this rapidly evolving field. It underscores the potential of AI to revolutionize rock mechanics, while also cautioning against the need for careful consideration of the limitations and challenges that come with its application. The paper's emphasis on the need for standardized datasets and interdisciplinary collaboration highlights the importance of a holistic approach to the successful integration of AI in rock mechanics. The authors' vision of a future where rock mechanics is transformed into a fully autonomous science is both ambitious and compelling, and the paper provides a solid foundation for future research in this area.

✅ Strengths

This paper offers a thorough and well-structured review of the integration of AI and ML in rock mechanics, effectively covering a wide range of applications and methodologies. The authors demonstrate a clear understanding of the historical context, tracing the evolution of AI techniques in the field from traditional methods to modern deep learning approaches. This historical perspective is crucial for understanding the current state of the field and the potential for future advancements. The paper excels in highlighting the practical relevance of AI and ML in rock mechanics, showcasing its importance in addressing key challenges such as microstructure reconstruction, mechanical parameter estimation, and real-time hazard prediction. The authors effectively demonstrate how AI can be used to improve the accuracy and efficiency of these tasks, which are critical for the safe and efficient design and operation of rock engineering structures. The discussion of emerging techniques like PINNs and graph-based learning is another strength of the paper. These techniques offer promising avenues for enhancing the interpretability and physical relevance of AI models in rock mechanics, addressing a key limitation of many purely data-driven approaches. The paper also acknowledges the potential of LLMs in automating code generation and decision support, although this aspect could be further developed. The authors' call for standardized datasets, interdisciplinary collaboration, and transparent AI workflows is a significant contribution, highlighting the need for a more systematic and rigorous approach to the application of AI in rock mechanics. The paper's forward-looking perspective on the development of intelligent frameworks that couple physical knowledge, spatial reasoning, and adaptive learning is both ambitious and compelling. The authors effectively synthesize the current state of AI in rock mechanics, providing a valuable resource for researchers and practitioners in the field. The paper's emphasis on the need for standardized datasets and interdisciplinary collaboration highlights the importance of a holistic approach to the successful integration of AI in rock mechanics. The authors' vision of a future where rock mechanics is transformed into a fully autonomous science is both ambitious and compelling, and the paper provides a solid foundation for future research in this area. The paper's ability to synthesize a large amount of information into a coherent and accessible narrative is a significant strength, making it a valuable resource for both experts and newcomers to the field.

❌ Weaknesses

While the paper provides a comprehensive overview of AI applications in rock mechanics, several weaknesses limit its overall impact and practical value. One significant limitation is the lack of concrete examples and detailed explanations regarding the application of large language models (LLMs) in the field. The paper mentions LLMs in the introduction and in Appendix A.4, stating that they are beginning to facilitate automated code generation and decision support. However, the main body of the paper lacks specific examples of how LLMs are currently being applied in rock mechanics beyond this brief mention. The example in Appendix A.4, which describes the use of a LLM (ChatGPT) to generate and refine a Python script for solving the steady-state seepage equation, is presented as supplementary material and lacks detailed explanation of the LLM's role, architecture, or training. This absence of specific examples makes it difficult to assess the practical impact and feasibility of LLMs in this field. The paper does not discuss specific LLM architectures or training methodologies that have been explored for code generation in geotechnical analysis, nor does it analyze their performance and limitations. Furthermore, the paper does not address the critical challenges associated with ensuring the accuracy and reliability of code generated by LLMs, which is a major concern in engineering applications. This lack of detail and critical analysis undermines the paper's discussion of LLMs and their potential in rock mechanics. My confidence in this assessment is high, as the absence of concrete examples and discussion of challenges is clearly evident in the paper's content. Another significant weakness is the paper's call for standardized datasets without providing specific recommendations or guidelines on how to create or curate such datasets. The paper acknowledges the need for open, standardized databases for rock mechanical properties, case histories, and monitoring data, citing Xiao et al. (2022). However, the paper does not delve into the specific characteristics that a standardized dataset should possess. For example, it does not specify the types of rock properties, the range of geological conditions, or the variety of loading scenarios that should be included. The paper also fails to discuss the potential challenges in creating such datasets, such as the cost and time required for data collection and the need for international collaboration to ensure broad applicability. This lack of specific guidance limits the practical value of the paper's call for standardized datasets. The paper's failure to provide concrete recommendations on dataset creation is a significant oversight, given the importance of high-quality data for the successful application of AI in rock mechanics. My confidence in this assessment is high, as the paper's call for standardized datasets is not accompanied by any specific recommendations or discussion of challenges. The paper also identifies data quality, model generalization, and interpretability as ongoing challenges but does not offer concrete solutions or strategies to address these issues. While the paper suggests methods to augment data, such as using physics-based simulations or generative networks to create plausible synthetic data, it does not delve into specific techniques for improving data quality, such as data augmentation or noise reduction methods. The paper also does not explore strategies for enhancing model generalization, such as domain adaptation or transfer learning. Furthermore, the paper does not discuss methods for improving model interpretability, such as feature importance analysis or model-agnostic interpretability techniques, and how these methods can be applied to the specific challenges of rock mechanics. The paper mentions the use of simpler surrogate models and explainable AI techniques to improve interpretability, but it does not provide a detailed discussion of these methods or their application in the context of rock mechanics. This lack of concrete solutions limits the paper's ability to address the identified challenges effectively. My confidence in this assessment is high, as the paper's discussion of solutions is limited and lacks specific details. Finally, the paper's contribution is limited by its nature as a review rather than a presentation of novel research. While the paper provides a comprehensive overview of the field, it does not present any new methodologies or offer any new insights into the application of existing methods. The paper reads more like a catalog of applications rather than a critical analysis of the field. This lack of novel contribution limits the paper's impact on the field of rock mechanics. My confidence in this assessment is high, as the paper is explicitly presented as a review and does not introduce any novel research.

💡 Suggestions

To enhance the paper's discussion on large language models (LLMs), it would be beneficial to include specific examples of how LLMs can be applied in rock mechanics beyond just code generation. For instance, the authors could explore the potential of LLMs in tasks such as automated literature review and summarization, which could help researchers stay abreast of the latest developments in the field. Furthermore, the paper could discuss the use of LLMs for generating hypotheses or suggesting experimental designs based on existing data. A detailed analysis of the limitations of current LLMs in the context of rock mechanics, such as their inability to understand complex geological processes or their tendency to generate inaccurate or nonsensical code, would also be valuable. This would provide a more balanced perspective on the potential and challenges of using LLMs in this field. The authors could also explore the potential of combining LLMs with other AI techniques, such as physics-informed neural networks (PINNs), to create more robust and reliable models for rock mechanics. This would demonstrate a more critical and nuanced understanding of the potential of LLMs in this domain. Regarding the need for standardized datasets, the paper should provide more specific guidance on the characteristics that such datasets should possess. For example, the authors could propose a framework for categorizing rock properties, such as mechanical, physical, and chemical properties, and specify the range of values that should be covered for each category. The paper could also discuss the importance of including data from different geological settings and loading conditions to ensure the generalizability of AI models. Furthermore, the authors could explore the potential of using synthetic data generation techniques to augment existing datasets and improve model performance. The paper should also address the practical challenges of creating and maintaining standardized datasets, such as the need for international collaboration and the development of data sharing protocols. A discussion of the ethical considerations associated with data sharing, such as data privacy and ownership, would also be relevant. This would provide a more practical and actionable approach to the creation of standardized datasets. To address the challenges of data quality, model generalization, and interpretability, the paper should provide more concrete solutions and strategies. For example, the authors could discuss the use of data augmentation techniques, such as adding noise or applying transformations to existing data, to improve data quality and model robustness. The paper could also explore the use of domain adaptation techniques to improve model generalization across different geological settings. Furthermore, the authors could discuss the use of model-agnostic interpretability techniques, such as SHAP values or LIME, to provide insights into the decision-making process of complex AI models. The paper should also emphasize the importance of incorporating domain knowledge into the development of AI models to improve their interpretability and reliability. Finally, the paper could discuss the need for rigorous validation and testing of AI models in real-world scenarios to ensure their safety and effectiveness. This would provide a more comprehensive and practical approach to addressing the identified challenges. To improve the paper's overall contribution, the authors should focus on providing a more in-depth analysis of the current state of ML in rock mechanics. Instead of simply listing applications, the review should categorize the different types of problems being addressed (e.g., prediction of rock failure, characterization of rock properties, optimization of mining operations) and discuss the specific ML techniques used for each category. For each category, the review should analyze the strengths and weaknesses of the applied ML methods, and identify the key challenges that remain. For example, in the context of rock failure prediction, the review could discuss the limitations of current models in capturing the complex physics of fracture propagation and suggest potential avenues for improvement, such as incorporating physics-informed neural networks or graph neural networks to better model the interactions between rock fragments. This would provide a more critical and analytical approach to the review, making it more valuable to the field. Finally, the paper should provide a clear roadmap for future research in this area. This roadmap should not only highlight the current limitations but also propose concrete research directions that could lead to significant advancements. For example, the review could suggest the development of new ML algorithms that are specifically tailored to the challenges of rock mechanics, or the creation of standardized datasets that can be used to benchmark the performance of different ML models. The review should also emphasize the importance of interdisciplinary collaboration between geotechnical engineers, data scientists, and domain experts to ensure that ML is applied effectively and responsibly in rock mechanics. By providing a more critical and forward-looking analysis, the review can make a more significant contribution to the field.

❓ Questions

Given the paper's mention of LLMs, could you provide specific examples of how these models are currently being applied in rock mechanics, beyond just their potential for automating code generation and decision support? What specific LLM architectures or training methodologies have been explored for these applications, and what are their performance limitations? Furthermore, what measures are being taken to ensure the accuracy and reliability of code generated by LLMs in this context? The paper calls for standardized datasets to improve model generalization and interpretability. What specific recommendations or guidelines do you propose for creating or curating such datasets? What specific rock properties, geological conditions, and loading scenarios should be included, and what are the practical challenges associated with creating such datasets? The paper identifies data quality, model generalization, and interpretability as ongoing challenges. What concrete solutions or strategies do you suggest to address these issues? What specific techniques for improving data quality, such as data augmentation or noise reduction methods, should be employed? How can model generalization be improved across different geological settings, and what methods can be used to enhance model interpretability in the context of rock mechanics? Finally, considering the paper's nature as a review, what specific steps could be taken to provide a more in-depth analysis of the current state of ML in rock mechanics? How can the paper move beyond a simple listing of applications to provide a more critical and analytical perspective on the strengths, weaknesses, and remaining challenges in this field? What specific research directions could be proposed to address the identified limitations and advance the field of AI in rock mechanics?

📊 Scores

Soundness:3.0
Presentation:2.75
Contribution:2.5
Rating: 5.25

AI Review from ZGCA

ZGCA Review available after:
--d --h --m --s

📋 AI Review from ZGCA will be automatically processed

📋 Summary

This paper reviews the integration of artificial intelligence into rock mechanics, tracing historical developments from early ANN/SVM methods to modern deep learning (CNNs, transformers), generative models, and physics-informed frameworks. It surveys AI-enabled advances across: (i) data-driven estimation of rock properties (Section 2), (ii) image-based modeling and fracture detection, including generative and segmentation methods (Section 3), (iii) AI-assisted constitutive modeling and PDE solvers with PINN variants and hybrid FEM/DEM surrogates (Section 4), and (iv) applications in rock engineering such as rock mass classification, rockburst/geohazard prediction, tunneling control, slope stability, and other emerging use cases (Section 5). A bibliometric overview of 17 journals frames the diffusion of AI across the field (Section 1). Section 6 outlines challenges around data scarcity, generalization, interpretability/trust, computational efficiency, workflow integration, and lifecycle maintenance, and offers a forward-looking perspective toward hybrid, physics-aware, and autonomous systems.

✅ Strengths

  • Broad and coherent synthesis across key subareas: property estimation (Section 2), imaging/fracture analysis (Section 3), constitutive modeling/PDE surrogates (Section 4), and engineering applications (Section 5).
  • Useful domain contextualization: connects classical rock-mechanics challenges (heterogeneity, anisotropy, multiphysics) to specific AI families (e.g., ANFIS/SVM for small data, CNN/UNet for fracture segmentation, LSTM for history dependence, PINN/PIRBN/PI-TCN for PDEs).
  • Highlights hybrid and physics-informed directions with concrete exemplars (e.g., PIRBN and PI-TCN in Section 4; physics-guided velocity inversion and tunneling deformation prediction in Sections 2 and 5.3).
  • Identifies salient challenges and practical needs (Section 6), including data standards, uncertainty quantification, interpretability, integration into engineering workflows, and model maintenance.
  • Rich referencing with specific models and case studies (e.g., FraSegNet for fracture extraction in Section 3; LSTM-based rockburst early warning in Section 5.2; CNN/RF surrogates for slope reliability in Section 5.4).

❌ Weaknesses

  • Lack of systematic comparative analysis: the review enumerates methods but does not provide cross-method comparisons on shared tasks, standardized metrics, or quantitative trade-offs (e.g., accuracy vs. data requirements vs. interpretability) that would guide practitioner choice (analysis reports’ core critique).
  • Bibliometric methodology is only partially specified (Section 1): journals and generic keywords are listed, but inclusion/exclusion criteria, databases beyond selected journals, time windows, de-duplication, screening procedures, and PRISMA-style documentation are missing, limiting reproducibility and coverage assessment.
  • No consolidated taxonomy or guidance matrix mapping rock-mechanics problem types to recommended AI methods with indicative data regimes, features, and evaluation metrics; no benchmark task definitions or dataset catalog with standardized splits.
  • Limited treatment of uncertainty quantification and validation protocols across safety-critical applications (e.g., rockburst, tunneling control, slope reliability), beyond scattered mentions; little discussion of calibration, OOD detection, and reliability under domain shift.
  • Speculative components (e.g., LLMs for code generation/decision support) are discussed without concrete, domain-validated workflows, datasets, or failure modes, making these parts less actionable.
  • Clarity could be further improved by summary tables/figures (e.g., datasets, methods, metrics, typical performance ranges) and by distilling key takeaways per subdomain into concise, comparable artifacts.

❓ Questions

  • Can you provide a detailed, reproducible bibliometric protocol (databases searched, exact keyword strings, time window, inclusion/exclusion criteria, screening/selection process, PRISMA-style flow, and inter-rater agreement if applicable)?
  • For each major task family (e.g., UCS/elasticity estimation; fracture segmentation; constitutive modeling; rockburst prediction; slope reliability), can you include a comparative table summarizing representative methods, datasets (size, modality), metrics, typical performance ranges, data requirements, and computational cost?
  • Could you propose or curate a minimal set of standardized benchmark tasks and datasets (with recommended data splits and metrics) to enable fair comparison across methods within intelligent rock mechanics?
  • What guidance can you provide on uncertainty quantification and calibration (e.g., probabilistic models, conformal prediction, ensembling, Bayesian neural nets) for high-consequence applications discussed in Sections 5.2–5.4?
  • Can you present a task–method guidance matrix mapping geotechnical problem attributes (data size/modality, heterogeneity, need for interpretability, physics constraints) to recommended AI families and training recipes?
  • For physics-informed approaches in Section 4, can you offer practical recipes for stabilizing training in 3D and multiphysics cases (loss design, normalization, residual weighting, curriculum, domain decomposition) and when to prefer PINN variants (e.g., PIRBN, PI-TCN) over traditional solvers or surrogates?
  • Could you elaborate specific, field-validated workflows for integrating AI into existing software used by practitioners (e.g., FEM packages, GIS pipelines), including interfaces, data exchange formats, and validation steps?
  • How do you recommend handling domain shift and transfer across sites/materials (e.g., pretraining on synthetic/experimental corpora, domain adaptation, physics-constrained augmentation)?
  • For LLM-based tooling (mentioned in Section 6 and Appendix A.4), can you delineate concrete, auditable use cases (e.g., code generation with unit tests, report parsing with human-in-the-loop QA) and guardrails for safety-critical contexts?
  • Will you release any structured resources (e.g., curated reference lists by task, dataset index, code links) to improve reproducibility and community uptake?

⚠️ Limitations

  • Safety-critical deployment risk: Overreliance on black-box predictors in rockburst/tunneling/slope contexts without calibrated uncertainty and robust validation can induce false confidence and hazardous decisions.
  • Data bias and domain shift: Site-specific data and heterogeneous measurement protocols can bias models; generalization beyond the training domain is uncertain without transfer/robustness strategies.
  • Reproducibility gaps: The review lacks standardized benchmarks, dataset catalogs, and protocol details; without public resources and clear evaluation standards, results in the literature are hard to compare.
  • Computational burden: Physics-informed models in complex 3D/multiphysics cases may be prohibitively expensive; practical guidance on efficiency (e.g., domain decomposition, multi-fidelity, adaptive sampling) is needed.
  • Interpretability/trust: Many cited successes rely on opaque models; more emphasis on explainability, physical consistency checks, and validation against governing constraints is needed.
  • Workflow integration and maintenance: Integration into practitioner toolchains and long-term model maintenance (drift detection, retraining policies) remain under-specified.

🖼️ Image Evaluation

Cross‑Modal Consistency: 33/50

Textual Logical Soundness: 24/30

Visual Aesthetics & Clarity: 9/20

Overall Score: 66/100

Detailed Evaluation (≤500 words):

1. Cross‑Modal Consistency

Visual ground truth (image‑first):

• Figure 1: Blue/yellow workflow chart linking geology→stress→mechanical behavior→construction/monitoring/design; arrows show data/knowledge base.

• Figure 2/(a): Longitudinal geological profile (elevation vs distance, lithologies/faults). (b): Excavation timeline bar (monitored/unmonitored). (c): Tunnel cross‑section (7.2 m×6.2 m). (d): Face photo with “rockburst area.” (e): LSTM for regression (events/energy/volume). (f): LSTM for multi‑class intensity.

• Figure 3: Matrix of blasting cycles; colored/hatched cells for warning vs actual intensity.

• Figure 4/(a): Physics/data/coordinate loss DPNN schema for tunneling. (b): Sampling grid and PLAXIS model; color map. (c): 3D surface of predicted ground deformation. (d): 2D heatmap of deformation.

• Figure‑level synopses: Fig.2–3 depict a rockburst case and LSTM workflows; Fig.4 illustrates physics‑guided tunneling deformation prediction.

• Major 1: Only Figure 1 is cited; subsequent multi‑pane figures (Figs. 2–4) appear without numbering or references, breaking traceability. Evidence: “Figure 1: Systematic composition…” (Sec 1); no other figure numbers in text.

• Major 2: LSTM architecture panes (Fig. 2e–f) are not tied to the Sec 5.2 LSTM claim, leaving ambiguity about dataset/features. Evidence: Sec 5.2 “Hu et al. … an LSTM‑based framework… accuracies exceeding 70%.”

• Major 3: The warning matrix (Fig. 3) is shown, but the text’s performance claim lacks a matching metric/legend mapping. Evidence: Sec 5.2 “accuracies exceeding 70%” with no figure reference.

• Minor 1: Bibliometric findings referenced (keyword network shift) lack a corresponding figure/table. Evidence: Sec 1 “keyword network analysis reveals… ‘machine learning’ now surpasses…”

2. Text Logic

• Major 1: No Major issues found.

• Minor 1: Timeframe inconsistency (“all articles published before 2025”) while also citing 2025 works without clarifying inclusion cut‑off. Evidence: Sec 1 “all articles published before 2025…”

• Minor 2: Bibliometric method lacks reproducible details (exact journal list, query strings, counts). Evidence: Sec 1 describes 17 journals, but not fully enumerated.

3. Figure Quality

• Major 1: Many panes illegible at print size (tiny text/legends), especially Fig. 2a–d and Fig. 3 grid; critical labels cannot be read. Evidence: Fig. 2a axis ticks/legend glyphs and Fig. 3 headers are unreadable at provided size.

• Major 2: Missing captions for Figs. 2–4; readers cannot infer variables, units, or datasets. Evidence: Only Fig. 1 has a caption near images; later images lack numbered captions.

• Minor 1: Similar color hues/hatching in Fig. 3 impede class differentiation without a legend. Evidence: Fig. 3 uses red/blue hatches with no visible legend.

Key strengths:

  • Comprehensive, well‑cited review spanning properties, imaging, constitutive ML, and applications.
  • Balanced emphasis on physics‑aware ML and uncertainty.
  • Useful synthesis of application‑specific models (rockburst, tunneling, slopes).

Key weaknesses:

  • Figure–text alignment gaps (uncited/unlabeled figures) and illegible visuals impede verification.
  • Bibliometric claims lack visual/tabular evidence and reproducibility details.
  • Several core visuals fail the “figure‑alone” test; add captions, legends, and readable fonts.

Recommendations:

  • Number and caption all figures; reference each where discussed.
  • Provide a bibliometric figure/table (trend lines, co‑occurrence map) with methods appendix.
  • Enlarge fonts, add legends/units, and annotate key ROIs for case‑study figures.

📊 Scores

Originality:2
Quality:3
Clarity:3
Significance:2
Soundness:3
Presentation:3
Contribution:2
Rating: 4

AI Review from SafeReviewer

SafeReviewer Review available after:
--d --h --m --s

📋 AI Review from SafeReviewer will be automatically processed

📋 Summary

This paper provides a comprehensive review of the integration of artificial intelligence (AI) and machine learning (ML) techniques into the field of rock mechanics. It meticulously traces the historical development of AI applications in this domain, starting from early methods like backpropagation and support vector machines to more recent advancements in deep learning, including convolutional and transformer architectures. The authors highlight the transformative impact of these technologies on various aspects of rock engineering, such as microstructure reconstruction, mechanical parameter estimation, constitutive modeling, and real-time hazard prediction. A key emphasis is placed on the shift towards data-driven and hybrid approaches that combine data-driven models with physical principles, exemplified by the use of physics-informed neural networks (PINNs) and graph-based learning. The paper also acknowledges the emerging role of large language models (LLMs) in automating code generation and decision support within geotechnical analysis. Despite the progress, the authors identify persistent challenges, including data quality, model generalization, and interpretability. They advocate for the development of standardized datasets, fostering interdisciplinary collaboration, and establishing transparent and reproducible AI workflows to address these issues. The paper concludes with a forward-looking perspective, envisioning next-generation intelligent frameworks that integrate physical knowledge, spatial reasoning, and adaptive learning, ultimately propelling rock mechanics from empirical modeling towards fully autonomous, intelligent systems. The authors aim to provide a roadmap for the future of AI in rock mechanics, emphasizing the need for robust, reliable, and interpretable AI-driven solutions.

✅ Strengths

I found several strengths in this paper. Firstly, the paper provides a well-structured and comprehensive overview of the historical development of AI in rock mechanics. The authors effectively trace the evolution from early methods like backpropagation and support vector machines to the current state-of-the-art deep learning techniques, including convolutional and transformer architectures. This historical perspective is crucial for understanding the current landscape and future directions of the field. Secondly, the paper does an excellent job of highlighting the diverse applications of AI in rock mechanics. The authors discuss how AI is being used for microstructure reconstruction, mechanical parameter estimation, constitutive modeling, and real-time hazard prediction, demonstrating the broad impact of these technologies on various aspects of rock engineering. The emphasis on the shift towards data-driven and hybrid approaches, combining data-driven models with physical principles, is particularly insightful. The inclusion of physics-informed neural networks (PINNs) and graph-based learning as examples of this trend showcases the authors' awareness of cutting-edge research. Furthermore, the paper acknowledges the emerging role of large language models (LLMs) in automating code generation and decision support, indicating a forward-thinking approach. Finally, the paper's identification of key challenges, such as data quality, model generalization, and interpretability, is a significant contribution. The authors' call for standardized datasets, interdisciplinary collaboration, and transparent AI workflows demonstrates a practical and solution-oriented mindset. Overall, the paper provides a valuable synthesis of the current state of AI in rock mechanics and offers a clear roadmap for future research.

### Weaknesses:

While the paper presents a comprehensive overview, I have identified several weaknesses that warrant attention. Firstly, the paper's focus is primarily on the application of existing AI and ML techniques to rock mechanics problems, rather than introducing novel AI methodologies. While the authors discuss the use of physics-informed neural networks (PINNs) and graph-based learning, these are presented as examples of existing techniques applied to the domain, rather than new methodological contributions. This is not inherently a flaw, but it does limit the paper's novelty in terms of AI methodology. The paper synthesizes existing work and does not present new experimental results or propose new AI algorithms. This is evident in the paper's description of its contributions, which focuses on synthesizing progress and tracing the evolution of AI in rock mechanics, rather than introducing new AI methods. Secondly, the paper lacks a detailed discussion of the specific challenges and nuances of applying AI to rock mechanics. While the authors mention the heterogeneity, anisotropy, and discontinuities of rock materials, they do not delve deeply into how these properties specifically affect the performance of different AI models. For example, the paper does not discuss how the non-linear stress-strain behavior of rocks, which is often influenced by factors such as confining pressure and rock type, affects the training and generalization of machine learning models. This lack of in-depth discussion limits the paper's ability to provide practical guidance for researchers and practitioners in the field. The paper does mention these challenges, but does not explore them in detail. Thirdly, the paper does not provide a detailed comparative analysis of the performance of different AI models in the context of rock mechanics. While the authors mention various AI techniques, such as CNNs, LSTMs, and GANs, they do not compare their performance in terms of accuracy, computational cost, and data requirements for specific rock mechanics tasks. This lack of comparative analysis makes it difficult for readers to assess the relative strengths and weaknesses of different AI approaches. The paper mentions various AI techniques but does not provide a detailed comparison of their performance. Fourthly, the paper does not adequately address the practical challenges of implementing AI models in real-world rock mechanics scenarios. While the authors mention the need for standardized datasets and interdisciplinary collaboration, they do not discuss the practical difficulties of deploying these models in field conditions, such as the robustness of AI models to noisy or incomplete data, the computational resources required for real-time predictions, and the integration of AI models with existing engineering workflows. This lack of practical considerations limits the paper's immediate applicability. The paper mentions the need for standardized datasets but does not delve into the practical challenges of real-world deployment. Fifthly, the paper lacks a detailed discussion of the interpretability of AI models in rock mechanics. While the authors acknowledge the importance of interpretability, they do not provide concrete examples of how to achieve this in practice. The paper does not discuss the limitations of current interpretability techniques, such as feature importance or saliency maps, in the context of complex rock mechanics problems. This lack of practical guidance limits the paper's ability to address the trust and reliability issues associated with AI models. The paper mentions the importance of interpretability but does not provide concrete examples or discuss limitations. Finally, the paper does not provide a detailed discussion of the limitations of current AI approaches in rock mechanics. While the authors mention the need for standardized datasets, they do not discuss the specific challenges associated with creating such datasets, such as the variability in rock properties, the difficulty of obtaining high-quality data, and the ethical considerations of using AI in geotechnical engineering. This lack of detailed discussion limits the paper's ability to provide a balanced perspective on the current state of AI in rock mechanics. The paper mentions the need for standardized datasets but does not discuss the challenges in creating them. These weaknesses, while not invalidating the paper's contributions, highlight areas where further research and discussion are needed.

### Suggestions:

To address the identified weaknesses, I recommend several concrete improvements. Firstly, the authors should include a more detailed discussion of the specific challenges and nuances of applying AI to rock mechanics. This should include a thorough exploration of how the heterogeneity, anisotropy, and discontinuities of rock materials affect the performance of different AI models. The authors should discuss how the non-linear stress-strain behavior of rocks, influenced by factors such as confining pressure and rock type, impacts the training and generalization of machine learning models. This could involve providing specific examples of how these factors affect model performance and discussing potential mitigation strategies. Secondly, the authors should provide a more detailed comparative analysis of the performance of different AI models in the context of rock mechanics. This should include a comparison of accuracy, computational cost, and data requirements for specific rock mechanics tasks. The authors should discuss the trade-offs between different AI approaches and provide guidance on selecting the most appropriate model for a given task. This could involve presenting case studies or examples where different AI models have been applied to similar rock mechanics problems and comparing their performance. Thirdly, the authors should address the practical challenges of implementing AI models in real-world rock mechanics scenarios. This should include a discussion of the robustness of AI models to noisy or incomplete data, the computational resources required for real-time predictions, and the integration of AI models with existing engineering workflows. The authors should provide practical guidance on how to overcome these challenges and ensure the reliable deployment of AI models in field conditions. This could involve discussing specific techniques for handling noisy data, optimizing models for real-time predictions, and integrating AI models with existing engineering software. Fourthly, the authors should provide a more detailed discussion of the interpretability of AI models in rock mechanics. This should include concrete examples of how to achieve interpretability in practice and a discussion of the limitations of current interpretability techniques. The authors should explore methods for explaining the predictions of AI models and ensuring that these models are trustworthy and reliable. This could involve discussing specific techniques for visualizing model behavior, extracting feature importance, and validating model predictions against physical principles. Fifthly, the authors should provide a more detailed discussion of the limitations of current AI approaches in rock mechanics. This should include a thorough exploration of the challenges associated with creating standardized datasets, such as the variability in rock properties, the difficulty of obtaining high-quality data, and the ethical considerations of using AI in geotechnical engineering. The authors should discuss potential solutions to these challenges and provide a balanced perspective on the current state of AI in rock mechanics. This could involve discussing specific strategies for data collection, data sharing, and ethical AI development. Finally, the authors should consider including a section that explicitly outlines the limitations of current AI approaches in rock mechanics. This could include a discussion of the challenges associated with creating standardized datasets, the difficulties in interpreting complex models, and the limitations of current algorithms in handling the inherent variability and uncertainty in geological data. By acknowledging these limitations, the authors can provide a more balanced and realistic perspective on the current state of AI in rock mechanics. Furthermore, the authors should consider including a discussion on the ethical implications of using AI in this field, such as the potential for bias in training data and the impact of AI-driven decisions on safety and environmental sustainability. This would demonstrate a responsible and forward-thinking approach to the application of AI in rock mechanics. By addressing these points, the paper can be significantly strengthened and provide more practical guidance for researchers and practitioners in the field.

### Questions:

I have several questions that arise from my analysis of the paper. Firstly, given the paper's focus on synthesizing existing work, what specific criteria were used to select the AI and ML techniques discussed in the paper? Were there any specific AI or ML techniques that were considered but ultimately excluded, and if so, what were the reasons for their exclusion? This question aims to understand the selection process and the scope of the review. Secondly, the paper mentions the use of physics-informed neural networks (PINNs) and graph-based learning. Could the authors provide more specific examples of how these techniques have been applied to rock mechanics problems, and what were the key findings and limitations of these applications? This question seeks to understand the practical application of these techniques in the domain. Thirdly, the paper identifies data quality, model generalization, and interpretability as key challenges. Which of these challenges do the authors consider to be the most significant barrier to the widespread adoption of AI in rock mechanics, and what are the most promising avenues for addressing this challenge? This question aims to understand the authors' perspective on the most pressing issues in the field. Fourthly, the paper calls for the development of standardized datasets. What specific characteristics should these datasets have to be truly useful for the rock mechanics community, and what are the practical challenges associated with creating and maintaining such datasets? This question seeks to understand the authors' vision for standardized datasets and the practical considerations involved. Finally, the paper concludes with a forward-looking perspective on next-generation intelligent frameworks. What are the key components of these frameworks, and what are the main technological hurdles that need to be overcome to realize this vision? This question aims to understand the authors' vision for the future of AI in rock mechanics and the challenges that need to be addressed to achieve this vision. These questions are intended to clarify the authors' choices and perspectives and to encourage further discussion on the key issues in the field.

❌ Weaknesses

While the paper presents a comprehensive overview, I have identified several weaknesses that warrant attention. Firstly, the paper's focus is primarily on the application of existing AI and ML techniques to rock mechanics problems, rather than introducing novel AI methodologies. While the authors discuss the use of physics-informed neural networks (PINNs) and graph-based learning, these are presented as examples of existing techniques applied to the domain, rather than new methodological contributions. This is not inherently a flaw, but it does limit the paper's novelty in terms of AI methodology. The paper synthesizes existing work and does not present new experimental results or propose new AI algorithms. This is evident in the paper's description of its contributions, which focuses on synthesizing progress and tracing the evolution of AI in rock mechanics, rather than introducing new AI methods. Secondly, the paper lacks a detailed discussion of the specific challenges and nuances of applying AI to rock mechanics. While the authors mention the heterogeneity, anisotropy, and discontinuities of rock materials, they do not delve deeply into how these properties specifically affect the performance of different AI models. For example, the paper does not discuss how the non-linear stress-strain behavior of rocks, which is often influenced by factors such as confining pressure and rock type, affects the training and generalization of machine learning models. This lack of in-depth discussion limits the paper's ability to provide practical guidance for researchers and practitioners in the field. The paper does mention these challenges, but does not explore them in detail. Thirdly, the paper does not provide a detailed comparative analysis of the performance of different AI models in the context of rock mechanics. While the authors mention various AI techniques, such as CNNs, LSTMs, and GANs, they do not compare their performance in terms of accuracy, computational cost, and data requirements for specific rock mechanics tasks. This lack of comparative analysis makes it difficult for readers to assess the relative strengths and weaknesses of different AI approaches. The paper mentions various AI techniques but does not provide a detailed comparison of their performance. Fourthly, the paper does not adequately address the practical challenges of implementing AI models in real-world rock mechanics scenarios. While the authors mention the need for standardized datasets and interdisciplinary collaboration, they do not discuss the practical difficulties of deploying these models in field conditions, such as the robustness of AI models to noisy or incomplete data, the computational resources required for real-time predictions, and the integration of AI models with existing engineering workflows. This lack of practical considerations limits the paper's immediate applicability. The paper mentions the need for standardized datasets but does not delve into the practical challenges of real-world deployment. Fifthly, the paper lacks a detailed discussion of the interpretability of AI models in rock mechanics. While the authors acknowledge the importance of interpretability, they do not provide concrete examples of how to achieve this in practice. The paper does not discuss the limitations of current interpretability techniques, such as feature importance or saliency maps, in the context of complex rock mechanics problems. This lack of practical guidance limits the paper's ability to address the trust and reliability issues associated with AI models. The paper mentions the importance of interpretability but does not provide concrete examples or discuss limitations. Finally, the paper does not provide a detailed discussion of the limitations of current AI approaches in rock mechanics. While the authors mention the need for standardized datasets, they do not discuss the specific challenges associated with creating such datasets, such as the variability in rock properties, the difficulty of obtaining high-quality data, and the ethical considerations of using AI in geotechnical engineering. This lack of detailed discussion limits the paper's ability to provide a balanced perspective on the current state of AI in rock mechanics. The paper mentions the need for standardized datasets but does not discuss the challenges in creating them. These weaknesses, while not invalidating the paper's contributions, highlight areas where further research and discussion are needed.

💡 Suggestions

To address the identified weaknesses, I recommend several concrete improvements. Firstly, the authors should include a more detailed discussion of the specific challenges and nuances of applying AI to rock mechanics. This should include a thorough exploration of how the heterogeneity, anisotropy, and discontinuities of rock materials affect the performance of different AI models. The authors should discuss how the non-linear stress-strain behavior of rocks, influenced by factors such as confining pressure and rock type, impacts the training and generalization of machine learning models. This could involve providing specific examples of how these factors affect model performance and discussing potential mitigation strategies. Secondly, the authors should provide a more detailed comparative analysis of the performance of different AI models in the context of rock mechanics. This should include a comparison of accuracy, computational cost, and data requirements for specific rock mechanics tasks. The authors should discuss the trade-offs between different AI approaches and provide guidance on selecting the most appropriate model for a given task. This could involve presenting case studies or examples where different AI models have been applied to similar rock mechanics problems and comparing their performance. Thirdly, the authors should address the practical challenges of implementing AI models in real-world rock mechanics scenarios. This should include a discussion of the robustness of AI models to noisy or incomplete data, the computational resources required for real-time predictions, and the integration of AI models with existing engineering workflows. The authors should provide practical guidance on how to overcome these challenges and ensure the reliable deployment of AI models in field conditions. This could involve discussing specific techniques for handling noisy data, optimizing models for real-time predictions, and integrating AI models with existing engineering software. Fourthly, the authors should provide a more detailed discussion of the interpretability of AI models in rock mechanics. This should include concrete examples of how to achieve interpretability in practice and a discussion of the limitations of current interpretability techniques. The authors should explore methods for explaining the predictions of AI models and ensuring that these models are trustworthy and reliable. This could involve discussing specific techniques for visualizing model behavior, extracting feature importance, and validating model predictions against physical principles. Fifthly, the authors should provide a more detailed discussion of the limitations of current AI approaches in rock mechanics. This should include a thorough exploration of the challenges associated with creating standardized datasets, such as the variability in rock properties, the difficulty of obtaining high-quality data, and the ethical considerations of using AI in geotechnical engineering. The authors should discuss potential solutions to these challenges and provide a balanced perspective on the current state of AI in rock mechanics. This could involve discussing specific strategies for data collection, data sharing, and ethical AI development. Finally, the authors should consider including a section that explicitly outlines the limitations of current AI approaches in rock mechanics. This could include a discussion of the challenges associated with creating standardized datasets, the difficulties in interpreting complex models, and the limitations of current algorithms in handling the inherent variability and uncertainty in geological data. By acknowledging these limitations, the authors can provide a more balanced and realistic perspective on the current state of AI in rock mechanics. Furthermore, the authors should consider including a discussion on the ethical implications of using AI in this field, such as the potential for bias in training data and the impact of AI-driven decisions on safety and environmental sustainability. This would demonstrate a responsible and forward-thinking approach to the application of AI in rock mechanics. By addressing these points, the paper can be significantly strengthened and provide more practical guidance for researchers and practitioners in the field.

❓ Questions

I have several questions that arise from my analysis of the paper. Firstly, given the paper's focus on synthesizing existing work, what specific criteria were used to select the AI and ML techniques discussed in the paper? Were there any specific AI or ML techniques that were considered but ultimately excluded, and if so, what were the reasons for their exclusion? This question aims to understand the selection process and the scope of the review. Secondly, the paper mentions the use of physics-informed neural networks (PINNs) and graph-based learning. Could the authors provide more specific examples of how these techniques have been applied to rock mechanics problems, and what were the key findings and limitations of these applications? This question seeks to understand the practical application of these techniques in the domain. Thirdly, the paper identifies data quality, model generalization, and interpretability as key challenges. Which of these challenges do the authors consider to be the most significant barrier to the widespread adoption of AI in rock mechanics, and what are the most promising avenues for addressing this challenge? This question aims to understand the authors' perspective on the most pressing issues in the field. Fourthly, the paper calls for the development of standardized datasets. What specific characteristics should these datasets have to be truly useful for the rock mechanics community, and what are the practical challenges associated with creating and maintaining such datasets? This question seeks to understand the authors' vision for standardized datasets and the practical considerations involved. Finally, the paper concludes with a forward-looking perspective on next-generation intelligent frameworks. What are the key components of these frameworks, and what are the main technological hurdles that need to be overcome to realize this vision? This question aims to understand the authors' vision for the future of AI in rock mechanics and the challenges that need to be addressed to achieve this vision. These questions are intended to clarify the authors' choices and perspectives and to encourage further discussion on the key issues in the field.

📊 Scores

Soundness:2.75
Presentation:2.5
Contribution:2.5
Confidence:4.0
Rating: 5.5

Keywords

Click the button to extract keywords

Insights

Click the button to extract insights
Version Management
Version 7 ⚠️ Not latest
📋 Version History
Citation Tools

📝 Cite This Paper