2510.0013 A Review of Intelligent Rock Mechanics: From Methods to Applications v3

🎯 ICAIS2025 Submission

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, a domain traditionally reliant on empirical methods and numerical modeling. The authors meticulously trace the evolution of AI methodologies, from early approaches like backpropagation and support vector machines to modern deep learning architectures such as convolutional neural networks (CNNs) and transformers. They highlight the diverse applications of AI in rock mechanics, including property estimation, image-based modeling, and constitutive modeling, demonstrating how AI is transforming the way we understand and predict complex geological behaviors. The paper also addresses the challenges and limitations of current approaches, such as data quality, model generalization, and interpretability, and proposes future research directions, emphasizing the need for standardized datasets and interdisciplinary collaboration. The authors emphasize the potential of AI to enhance our understanding and prediction of complex geological behaviors, offering new paradigms for addressing traditional challenges in rock mechanics, such as anisotropy, discontinuities, and multiphysics coupling. The paper's core contribution lies in its systematic synthesis of recent progress in AI applications within rock mechanics, bridging foundational AI methods with their practical deployment in engineering. It provides a valuable resource for researchers and practitioners, offering a clear overview of the current state of the field and identifying key areas for future development. The inclusion of a detailed appendix with datasets, code, and case studies further enhances the paper's value and reproducibility. The paper concludes by outlining a forward-looking perspective on developing next-generation intelligent frameworks capable of coupling physical knowledge, spatial reasoning, and adaptive learning, thereby advancing rock mechanics from empirical modeling toward fully intelligent, autonomous systems. Overall, this paper serves as a crucial resource for understanding the current landscape of AI in rock mechanics and provides a roadmap for future research in this rapidly evolving field.

✅ Strengths

This paper presents a thorough and well-structured review of the current state of intelligent rock mechanics, effectively synthesizing a wide range of topics from foundational AI methodologies to practical applications. The authors have successfully traced the evolution of AI in this field, highlighting key milestones and advancements. The paper's organization is commendable, making it accessible to both experts and non-experts in the field. The authors provide a critical assessment of the current limitations and challenges in the field, offering valuable insights for future research. The inclusion of a detailed appendix with datasets, code, and case studies further enhances the paper's value and reproducibility, which is a significant strength. The paper effectively highlights the role of AI in addressing traditional challenges in rock mechanics, such as anisotropy, discontinuities, and multiphysics coupling, through data-driven and hybrid approaches. The discussion of emerging techniques, including physics-informed neural networks and graph-based learning, that bridge data-driven inference with physical interpretability is particularly insightful. The paper also identifies key challenges, such as data quality, model generalization, and interpretability, and proposes future research directions for developing next-generation intelligent frameworks. The paper's ability to provide a comprehensive overview of the current research status of intelligent rock mechanics, covering a wide range of topics from the basic concepts and development process of AI to the application of AI in the field of rock mechanics, is a significant strength. The detailed appendix, including data sets, code, and case studies, is also a valuable resource for readers seeking to further understand the content of the paper. The paper is well-written and easy to understand, using clear and concise language to explain complex concepts and technologies, and the structure of the paper is logical and clear. The paper's ability to bridge foundational AI methods with their mechanistic integration and engineering deployment is a significant contribution, providing a valuable resource for researchers and practitioners in the field. The paper's systematic analysis of the evolution of AI in rock mechanics, highlighting key milestones and advancements, is a notable strength. The paper's identification of key challenges and proposed future research directions for developing next-generation intelligent frameworks is also a valuable contribution, providing a roadmap for future research in the field.

❌ Weaknesses

While this paper provides a comprehensive review of AI applications in rock mechanics, several limitations warrant attention. First, the paper's focus is primarily on the application of AI within rock mechanics, with limited discussion of the broader implications and challenges of AI in other fields. While the applications in rock mechanics are well-covered, a more comprehensive discussion of AI's impact and challenges across various domains would provide a broader context for the reader. For instance, the ethical considerations, data bias issues, and the need for explainable AI are crucial aspects that are not addressed in the context of rock mechanics. The paper's introduction mentions the broader impact of AI but quickly narrows down to rock mechanics, and the discussion of challenges is also specific to the field. This narrow focus might limit the reader's understanding of the wider context and challenges of AI. Second, the paper lacks a detailed comparison with other review papers on the application of AI in rock mechanics. A more thorough comparison with existing literature would help to highlight the unique contributions of this paper and its position within the broader research landscape. It is unclear how this review differentiates itself from other similar reviews in terms of scope, depth, and the specific aspects of AI in rock mechanics that it emphasizes. The paper presents a comprehensive review but lacks a section explicitly comparing itself to other reviews, making it harder for the reader to understand the novelty and specific contributions of this review. Third, the paper does not provide a detailed discussion of the limitations and challenges of the proposed methods and approaches. While some challenges are mentioned, a more in-depth analysis of the limitations of specific AI techniques, such as the computational cost of training deep learning models or the sensitivity of physics-informed neural networks to hyperparameter tuning, would be beneficial. Furthermore, the practical challenges of deploying these models in real-world rock mechanics scenarios, such as data acquisition and model validation, are not fully explored. The paper acknowledges some limitations but lacks a detailed, method-specific analysis of challenges like computational cost and hyperparameter sensitivity. The paper mentions computational efficiency as a challenge and the sensitivity of PINNs, but a more in-depth analysis of these limitations is missing. Fourth, the paper is a review paper, and its innovation is limited. It does not propose new research ideas or methods, but simply summarizes and analyzes the existing research results. The paper's stated goal is to review and synthesize existing work, and it does not present new experimental results. The lack of novel methods is inherent to the nature of a review paper. Fifth, the paper does not deeply explore the application of AI in other fields. This limits the comprehensiveness and universality of the paper. The paper's scope is deliberately limited to rock mechanics, which might limit the broader perspective for some readers. Sixth, the paper does not provide a detailed discussion of the limitations and challenges of intelligent rock mechanics. Although it briefly mentions some problems and challenges, it does not give a detailed analysis and discussion. The paper touches upon limitations but lacks an in-depth analysis, and the relatively brief discussion of limitations in Section 6 does not provide a nuanced understanding of the field's current state. Seventh, the paper lacks detailed practical examples and case studies that demonstrate the real-world application of these methods. Including more concrete examples of how AI has been successfully applied in rock mechanics projects would enhance the paper's practical value and demonstrate the tangible benefits of AI in this field. The paper focuses on summarizing methods and citing examples rather than providing detailed, self-contained case studies within the main text. The absence of detailed, self-contained case studies within the main body of the paper might make it harder for readers to understand the practical implementation and benefits of AI in rock mechanics. Eighth, the paper acknowledges the challenge of data quality and availability in rock mechanics but does not provide a detailed discussion of potential solutions or strategies for addressing this issue. The paper identifies data limitations but doesn't elaborate on specific solutions within the main text, and the brief mention of potential solutions without detailed discussion in the main text is a limitation. Ninth, the paper mentions the need for model interpretability and trust but does not delve into specific techniques or approaches for achieving this in the context of rock mechanics. Providing more details on how to improve the interpretability and trustworthiness of AI models in this field would be valuable for practitioners. The paper acknowledges the need for interpretability but doesn't provide a detailed discussion of specific techniques, and the brief mention of potential approaches without detailed discussion in the main text is a limitation. Finally, the paper lacks a thorough discussion of the ethical considerations associated with using AI in rock mechanics. This should go beyond simply mentioning the potential for bias and delve into specific examples of how bias can manifest in AI models, such as through biased training data or flawed model assumptions. The paper also lacks a critical analysis of the existing literature, highlighting not only the successes but also the limitations and failures of previous studies. The paper lacks a thorough discussion of ethical considerations and a critical analysis of the literature, focusing more on successes. These limitations, while not invalidating the paper's contributions, highlight areas where further research and discussion are needed.

💡 Suggestions

To enhance this paper, several improvements could be made. First, the authors should expand the discussion to include a broader perspective on the challenges and implications of AI, drawing parallels with other fields where AI is being applied. This could involve discussing the ethical considerations of using AI in decision-making processes related to rock mechanics, such as the potential for bias in training data leading to inaccurate predictions that could have safety implications. Furthermore, the authors could explore the limitations of current AI models in handling the inherent uncertainty and variability in geological data, and how these limitations might impact the reliability of predictions. A discussion on the need for explainable AI in rock mechanics, where the reasoning behind model predictions is crucial for trust and adoption, would also be valuable. This would provide a more comprehensive understanding of the broader context of AI in rock mechanics and its potential impact on the field. Second, the authors should provide a more detailed comparison with existing review papers on the application of AI in rock mechanics. This comparison should not only highlight the unique contributions of this paper but also identify the gaps in the existing literature that this review aims to fill. The authors could create a table that summarizes the scope, methodology, and key findings of other relevant review papers, and then clearly articulate how this paper differs in terms of its focus, the specific AI techniques it covers, and the depth of its analysis. This would help to establish the paper's position within the broader research landscape and demonstrate its value to the research community. Furthermore, the authors should discuss the limitations of existing reviews and how this paper addresses those limitations, providing a clear justification for the need for this new review. Third, the authors should provide a more detailed discussion of the limitations and challenges of the proposed methods and approaches. This should include a thorough analysis of the computational cost associated with training complex AI models, the sensitivity of these models to hyperparameter tuning, and the challenges of deploying these models in real-world scenarios. The authors should also discuss the limitations of physics-informed neural networks, such as their reliance on accurate physical models and the difficulty of incorporating complex physical constraints. Furthermore, the authors should address the practical challenges of data acquisition and model validation in rock mechanics, such as the scarcity of high-quality data and the difficulty of obtaining ground truth for model training and evaluation. This would provide a more realistic assessment of the current state of AI in rock mechanics and highlight the areas where further research is needed. Fourth, to enhance the practical impact of this review, the authors should include more detailed case studies that illustrate the successful application of AI in rock mechanics. These case studies should not only describe the problem being addressed and the AI methods used but also provide a thorough analysis of the results, including a comparison with traditional methods. For example, a case study could focus on the use of convolutional neural networks for automated rock mass classification, detailing the specific data used (e.g., images of rock faces), the architecture of the network, the training process, and the achieved accuracy compared to manual classification by geologists. The case study should also discuss the limitations of the approach and potential areas for improvement. This would provide readers with a concrete understanding of how AI can be applied in real-world rock mechanics projects and the challenges involved. Fifth, the paper should delve deeper into the issue of data quality and availability, providing specific strategies for addressing this challenge. The authors could discuss techniques for data augmentation, such as generating synthetic data through transformations or simulations, and methods for handling noisy or incomplete data, such as imputation techniques or robust learning algorithms. The paper should also explore the potential of transfer learning, where models trained on one dataset can be adapted to another dataset with limited data. For example, a model trained on a large dataset of rock images from one region could be fine-tuned on a smaller dataset from another region, leveraging the shared features of rock textures. The authors should also discuss the importance of data standardization and the development of open-source datasets to facilitate the sharing of data within the rock mechanics community. Sixth, the paper should provide a more detailed discussion of model interpretability and trust, focusing on techniques that are relevant to rock mechanics. The authors could explore methods for visualizing the internal workings of complex models, such as saliency maps or feature importance analysis, which can help users understand which features are most influential in the model's predictions. The paper should also discuss techniques for quantifying uncertainty in model predictions, such as Bayesian neural networks or ensemble methods, which can provide a measure of confidence in the model's output. Furthermore, the authors should emphasize the importance of validating model outputs against physical principles or expert knowledge, ensuring that the model's predictions are consistent with established understanding of rock mechanics. This would help build trust in AI models and facilitate their adoption in practical applications. Finally, the authors should provide a more critical analysis of the existing literature, highlighting not only the successes but also the limitations and failures of previous studies. This could involve a more detailed discussion of the specific assumptions and limitations of different AI models, as well as the challenges in validating and verifying AI-driven predictions. The authors should also discuss the need for more rigorous benchmarking and comparison of different AI approaches, as well as the development of standardized datasets and evaluation metrics. This would provide a more balanced and nuanced perspective on the current state of the field and help guide future research efforts. The inclusion of case studies that highlight both successful and unsuccessful applications of AI in rock mechanics would further enhance the paper's value.

❓ Questions

Several questions arise from my analysis of this paper. First, how do the authors envision the integration of AI and ML techniques into the daily practice of rock mechanics engineers? This question seeks to understand the practical implications of the research and how it can be translated into real-world applications. Second, what are the key challenges in validating and verifying the predictions made by AI-driven models in rock mechanics? This question aims to explore the limitations of current validation methods and the need for more robust approaches. Third, how can the authors ensure the reproducibility of their results and facilitate the adoption of AI techniques by other researchers and practitioners in the field? This question focuses on the importance of open science and the need for standardized protocols and resources. Fourth, could the authors provide more detailed case studies or examples of how AI has been successfully applied in real-world rock mechanics projects? This would help illustrate the practical benefits and challenges of using AI in this field. Fifth, what are some potential solutions or strategies for addressing the issue of data quality and availability in rock mechanics? How can researchers and practitioners ensure that they have access to high-quality data for training and validating AI models? This question seeks to explore the limitations of current data resources and the need for more robust data collection and sharing practices. Sixth, how can the interpretability and trustworthiness of AI models be improved in the context of rock mechanics? Are there specific techniques or approaches that can help users understand and trust the predictions made by these models? This question aims to explore the limitations of current AI models and the need for more transparent and explainable approaches. Finally, how does this paper compare with other review papers on the application of AI in rock mechanics? What are the unique contributions of this paper? This question seeks to understand the paper's position within the broader research landscape and its specific contributions to the field. These questions are crucial for further understanding the current state and future directions of AI in rock mechanics.

📊 Scores

Soundness:3.0
Presentation:3.0
Contribution:2.75
Rating: 5.75

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 AI methods into rock mechanics, tracing developments from early neural networks and SVMs to CNNs/transformers, generative models, and physics-informed approaches. It organizes the field into four pillars: (i) data-driven estimation of rock properties (strength, stiffness, velocities) (Section 2); (ii) image-based modeling and fracture detection, including 3D reconstruction and segmentation from images and point clouds (Section 3); (iii) AI-assisted constitutive modeling and physics-aware solvers (PINNs, hybrids) (Section 4); and (iv) applications, including rock mass classification, rockburst/geohazard prediction, tunneling/boring operations, slope stability, and emerging use cases (Section 5). Section 6 synthesizes challenges—data scarcity and standardization, interpretability/trust, computational efficiency, workflow integration, and lifecycle maintenance—and Section 7 offers an outlook toward hybrid physics–data models and digital-twin workflows. The paper claims a bibliometric survey across 17 journals (Section 1) and references an appendix with datasets and tooling (Appendix A.1, A.3).

✅ Strengths

  • Broad and up-to-date synthesis that spans the full rock mechanics pipeline: property inference (Section 2), imaging and fracture detection (Section 3), constitutive modeling and PINNs (Section 4), and field applications (Section 5).
  • Clear identification of practical challenges and actionable directions (Section 6), including data standardization, interpretability, and workflow integration; this is valuable for bridging research and practice.
  • Forward-looking discussion of physics-informed approaches and the nascent role of LLMs for code generation/decision support (Abstract; Section 6), which is relatively novel in this domain-focused review landscape.
  • Solid coverage of recent literature (including 2023–2025), with specific technique-to-application mappings and many domain-relevant examples (e.g., rockburst, tunneling, slope stability).
  • References an appendix that curates datasets and tools (Appendix A.1, A.3), signaling an intent to improve reproducibility and benchmarking.

❌ Weaknesses

  • Methodological transparency: Although Section 1 mentions a bibliometric overview of 17 journals with defined search terms, the review does not state explicit inclusion/exclusion criteria, screening stages, inter-rater procedures, or a PRISMA-style flow. This undermines claims of being "systematic" and makes the synthesis non-reproducible (Rigor report).
  • Lack of quantitative synthesis: No standardized comparison tables, effect sizes, or cross-study meta-analyses for key tasks; limited critical appraisal of evidence quality or biases across studies; difficult to assess robustness and generalization claims (e.g., in Sections 2–5).
  • Some conceptual claims are under-specified: e.g., the proposed framework to "categorize AI ... according to its cognitive resemblance and level of abstraction" (Section 1) is not formally defined or operationalized; readers are left without a concrete taxonomy or criteria.
  • Scope creep and heterogeneity: The paper mixes rock mechanics with adjacent geotechnical topics (soils, foundations, landslides) without a crisp definition of inclusion boundaries, which further stresses the need for a transparent screening protocol.
  • For a top ML venue, the novelty is limited: primarily a narrative review without new algorithms, datasets/benchmarks, or rigorous meta-analytic contributions; significance to the general ML community is modest despite relevance to geomechanics.
  • Reproducibility of the bibliometric claims is not ensured: no released corpus, code, or spreadsheet listing included/excluded records, and no assessment of retrieval/selection bias.

❓ Questions

  • Please provide a transparent, reproducible review protocol: databases searched, exact query strings, time window, language restrictions, gray literature policy, deduplication, inclusion/exclusion criteria, screening stages, and whether you followed PRISMA or similar guidelines.
  • Can you release the bibliometric corpus and a screening spreadsheet (with reasons for exclusion) to substantiate the Section 1 claims and enable reproducibility?
  • Can you formalize the proposed categorization framework based on "cognitive resemblance and level of abstraction" (Section 1)? For example, provide a taxonomy (with criteria) and map representative methods to the taxonomy.
  • Across the core tasks (e.g., UCS/moduli estimation, fracture detection, rockburst prediction, slope reliability), can you add standardized comparison tables summarizing datasets, features/modalities, metrics, sample sizes, and cross-validation protocols, and provide aggregated statistics (median/variance) by method family?
  • How did you assess study quality and risk of bias (dataset representativeness, leakage, hyperparameter search fairness, cross-site generalization)? If not done, can you introduce a quality appraisal rubric and apply it to key studies?
  • For physics-informed solvers (Section 4), can you delineate known failure modes (e.g., boundary enforcement, stiffness, scaling) with quantitative benchmarks and include guidance for practitioners on hyperparameter scaling and residual balancing?
  • The paper mentions Appendix A.1 (datasets) and A.3 (LLM-assisted tooling). Can you provide persistent links, licenses, and minimal documentation so others can replicate the workflows and evaluate LLM utility in practice?
  • Can you discuss OOD and cross-site generalization explicitly for field applications (Sections 5.1–5.4)? What protocols (leave-one-site-out, temporal holdouts) are recommended, and what does the literature show?
  • For the image-based reconstruction and fracture detection section (Section 3), can you contrast deep learning vs. conventional pipelines under class imbalance, low-contrast conditions, and 3D topological fidelity, with quantitative evidence?
  • Please clarify the exact scope: what geotechnical subdomains were included/excluded (soils vs. rocks, foundation vs. underground, hazard types), and why?

⚠️ Limitations

  • Selection bias and reproducibility: Without explicit screening criteria and released corpora, the synthesis risks selection bias and cannot be replicated; addressing this is essential.
  • Heterogeneity and generalization: Many cited studies are site-specific or small-sample; cross-site generalization and OOD robustness are insufficiently quantified; emphasize protocols (e.g., leave-one-site-out) and uncertainty quantification.
  • Over-reliance and safety: In high-consequence decisions (e.g., rockburst, slope stability), over-trusting black-box models can lead to harm; model cards, uncertainty estimates, and human-in-the-loop procedures are necessary.
  • Data governance: Privacy/ownership concerns in industrial monitoring data; clarify licensing and anonymization for shared datasets (Appendix A.1).
  • Compute and environmental cost: Physics-informed and large models can be compute-intensive; discuss model compression, energy footprints, and practical deployment constraints.
  • Interpretability: Expand concrete XAI tooling for the featured applications (e.g., feature attributions for rockburst predictors, saliency for fracture segmentation) and discuss failure analysis.

🖼️ Image Evaluation

Cross-Modal Consistency: 30/50

Textual Logical Soundness: 23/30

Visual Aesthetics & Clarity: 10/20

Overall Score: 63/100

Detailed Evaluation (≤500 words):

Image-first understanding (visual ground truth)

  • Figure 1: Regional geologic profile and tunnel cross-section. Cues: lithology patterns, faults, boreholes, “rockburst” callouts with depths, red “high geo-stress hazards” zone. Trend: multiple rockbursts clustered within the high-stress section.
  • Figure 2: Bar/line plot vs Time categories (4BR, 3BR, 2BR, 1BR, R, 1AR). Left axis: “Number of daily microseismic events.” Right axis: “S Value.” Bars colored by dates (12.17*R, 12.29*R, 1.14*R). Trend: events peak near 1BR–R; S-value varies non-monotonically.
  • Figure 3: Bar/line plot vs Time categories (3BR–1AR). Left axis: “Daily cumulative energy (×10^7).” Right axis: “Incidence of high- and medium-energy events (%)”. Trend: cumulative energy and incidence peak near R, decline at 1AR.

1. Cross-Modal Consistency

• Major 1: Figures are not referenced or captioned in the manuscript, blocking claim verification and context. Evidence: No “Figure” callouts in Sec. 5.2 despite discussing microseismic/rockburst precursors.

• Major 2: Ambiguous mapping in Fig. 2 between legend and axes (line labeled “S Value” but no color/series mapping; categories undefined). Evidence: Fig. 2 legend shows “S Value” without a series key; X-axis uses 4BR…1AR with no definition.

• Major 3: Illegible elements at print size in Fig. 1 (lithology legend, borehole IDs, many labels), impeding interpretation. Evidence: Fig. 1 lithology patterns and city/borehole text are too small to read at ≈100%.

• Minor 1: Fig. 3 unit ambiguity; left axis shows “×10^7” without explicit unit (likely J); “incidence” definition unspecified.

• Minor 2: Acronyms BR/R/AR not defined in visuals or text near first reference.

2. Text Logic

• Major 1: Bibliometric claims lack shown evidence/plots, weakening a central thread of the Introduction. Evidence: “A bibliometric overview… The results highlight that AI has taken root…” (no accompanying figure/table).

• Minor 1: Typographical breaks/formatting errors disrupt flow (e.g., “compu tational”; truncated citation “Li et al., 2025)”). Evidence: Introduction contains split words and a dangling citation.

• Minor 2: Several strong generalizations (e.g., “AI has transcended empirical prediction”) lack quantitative support within the body.

3. Figure Quality

• Major 1: Fig. 1 contains dense symbology and small fonts; critical legend items unreadable at print size.

• Minor 1: Inconsistent legend design and missing series keys in Figs. 2–3.

• Minor 2: Axes labels and acronyms not self-explanatory; no captions to guide interpretation.

Key strengths:

  • Comprehensive, well-structured survey spanning properties, imaging, constitutive modeling, and applications.
  • Balanced discussion of PINNs’ limitations and hybrid/uncertainty-aware trends.
  • Strong, up-to-date bibliography across subdomains.

Key weaknesses:

  • Missing figure callouts/captions; ambiguous/illegible visuals block verification.
  • Bibliometric narrative lacks presented evidence.
  • Several notation/typo issues and undefined acronyms in figures.

Recommendations:

  • Add formal captions and in-text references for Figs. 1–3; define BR/R/AR and S-value; fix legend/axis mappings and units.
  • Provide a bibliometric figure/table supporting trend claims.
  • Increase font sizes and simplify Fig. 1 legend; add call-outs and scale bars.
  • Tighten prose and resolve formatting/citation glitches.

📊 Scores

Originality:2
Quality:2
Clarity:3
Significance:2
Soundness:2
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 presents a comprehensive review of the application of artificial intelligence (AI) and machine learning (ML) techniques in the field of rock mechanics. The authors meticulously trace the evolution of AI methodologies, from early neural networks to modern deep learning architectures, and their integration into various aspects of rock engineering. The paper is structured around the flow of information in rock engineering projects, beginning with data acquisition and progressing through property estimation, fracture detection, constitutive modeling, and finally, to practical applications. The authors highlight the increasing adoption of AI/ML methods for tasks such as rock mass classification, rockburst prediction, and tunneling operations, demonstrating the potential of these techniques to enhance efficiency and accuracy in geotechnical analysis. The paper also acknowledges the limitations of current approaches, particularly concerning data quality, model interpretability, and generalization. The authors advocate for the development of standardized datasets and interdisciplinary collaboration to address these challenges. The paper's significance lies in its attempt to synthesize a large body of literature into a coherent narrative, providing a valuable resource for researchers and practitioners in the field. While the paper does not present novel research findings, it offers a well-organized and insightful overview of the current state of AI/ML in rock mechanics, highlighting both the achievements and the areas that require further investigation. The authors' emphasis on the need for standardized datasets and interdisciplinary collaboration underscores the importance of addressing the practical challenges that hinder the widespread adoption of AI/ML in this domain. The paper's focus on the flow of information in rock engineering projects provides a useful framework for understanding how AI/ML can be integrated into the various stages of analysis and design. Overall, the paper serves as a useful starting point for researchers and practitioners interested in exploring the application of AI/ML in rock mechanics, providing a broad overview of the field and highlighting the key challenges and opportunities.

✅ Strengths

The paper's primary strength lies in its comprehensive and well-structured review of AI and ML applications in rock mechanics. The authors have successfully synthesized a large body of literature into a coherent and accessible narrative. I found the organization of the paper particularly effective, as it follows the natural flow of information in rock engineering projects, making it easy to understand how AI/ML can be applied at different stages. The paper's attempt to bridge the gap between theoretical AI/ML concepts and practical rock mechanics problems is commendable. The authors provide concrete examples of how different AI/ML techniques, such as CNNs for image-based modeling and LSTMs for time-series prediction, are used in various rock engineering tasks. The inclusion of a forward-looking perspective is another strength, as it highlights the potential of emerging techniques like physics-informed neural networks (PINNs) and graph-based learning to address current limitations. The paper also acknowledges the challenges associated with data quality, model interpretability, and generalization, demonstrating a balanced and critical approach. The authors' call for standardized datasets and interdisciplinary collaboration is a crucial contribution, as it highlights the practical steps needed to advance the field. The paper's emphasis on the need for transparent and reproducible AI workflows is also important, as it promotes best practices in research and development. The inclusion of a case study in Appendix A.2, while brief, provides a concrete example of how deep learning can be applied to rockburst prediction, further enhancing the paper's practical relevance. Finally, the paper's attempt to provide a comprehensive list of references, although not exhaustive, demonstrates the authors' commitment to providing a thorough overview of the field. Overall, the paper's strengths lie in its comprehensive coverage, clear organization, and insightful discussion of both the achievements and the challenges in applying AI/ML to rock mechanics.

❌ Weaknesses

Despite its strengths, the paper exhibits several weaknesses that warrant careful consideration. A recurring theme across multiple reviews is the lack of a comprehensive literature search. While the paper includes a substantial number of references, several reviewers have pointed out specific omissions, particularly in the area of physics-informed neural networks (PINNs). For instance, Reviewer 4 provided a list of relevant PINN publications that are not cited in the paper, indicating a gap in the coverage of this specific area. This lack of exhaustiveness undermines the paper's claim to be a comprehensive review and raises concerns about the authors' awareness of the latest developments in the field. The paper's structure, while generally well-organized, is also a point of concern. Reviewer 1 notes that the paper's organization, while following the flow of information in rock engineering, might not be the most intuitive for a review paper, as it mixes methodological descriptions with application areas. This structure makes it difficult to get a clear overview of the different AI/ML techniques used in rock mechanics and their specific applications. Furthermore, the paper's discussion of interpretability is limited. While the authors acknowledge the "black box" nature of some AI models and the need for interpretability, they do not delve into specific techniques for achieving this in the rock mechanics context. Reviewer 1 suggests discussing methods like LIME or SHAP, which are not mentioned in the paper. This lack of detail limits the paper's practical value, as interpretability is a crucial requirement for the adoption of AI/ML models in engineering practice. The paper's treatment of uncertainty quantification is also superficial. While the authors mention the need for uncertainty quantification, they do not provide a detailed discussion of specific techniques or their application in rock mechanics. Reviewer 1 suggests discussing Bayesian neural networks or ensemble methods, which are not explored in depth. This omission is significant, as uncertainty quantification is essential for reliable decision-making in geotechnical engineering. The paper's discussion of data quality is also limited. While the authors acknowledge the challenges associated with data scarcity and quality, they do not provide a detailed analysis of the types of data used in rock mechanics or the specific challenges associated with each type. Reviewer 1 suggests discussing the impact of noise, outliers, and missing data, which are not adequately addressed in the paper. This lack of detail limits the paper's practical value, as data quality is a critical factor in the success of AI/ML models. The paper's discussion of model generalization is also insufficient. While the authors acknowledge the risk of overfitting, they do not provide a detailed analysis of the factors that affect generalization or the specific techniques used to mitigate this risk. Reviewer 1 suggests discussing cross-validation and domain adaptation, which are not explored in depth. This omission is significant, as model generalization is essential for the reliable application of AI/ML models to new datasets. Finally, the paper's lack of a dedicated discussion on the limitations of current AI/ML approaches is a significant weakness. While the authors acknowledge some limitations, they do not provide a comprehensive analysis of the potential pitfalls of relying solely on data-driven methods. Reviewer 1 suggests discussing the limitations of current AI/ML approaches and the need for hybrid approaches, which are not adequately addressed in the paper. This omission limits the paper's critical perspective and its ability to guide future research in the field. In summary, while the paper provides a useful overview of AI/ML in rock mechanics, its lack of exhaustiveness, limited discussion of interpretability, uncertainty quantification, data quality, model generalization, and limitations significantly weaken its overall contribution.

💡 Suggestions

To address the identified weaknesses, I recommend several concrete improvements. First, the authors should conduct a more comprehensive literature search, particularly focusing on physics-informed neural networks (PINNs) and other hybrid approaches. This would involve including the specific references provided by Reviewer 4, as well as other relevant publications in this area. The authors should also expand their search to include journals and conferences that focus on the intersection of AI/ML and physics-based modeling. This would ensure that the paper provides a more complete and up-to-date overview of the field. Second, the authors should restructure the paper to improve its clarity and accessibility. I suggest organizing the paper by AI/ML technique, rather than by the flow of information in rock engineering projects. This would make it easier for readers to understand the different techniques and their specific applications. For example, there could be separate sections on neural networks, support vector machines, and physics-informed neural networks, each with its own subsections on specific applications in rock mechanics. This would also allow for a more focused discussion of the strengths and limitations of each technique. Third, the authors should significantly expand their discussion of interpretability. This should include a detailed explanation of techniques such as LIME and SHAP, and their application in the rock mechanics context. The authors should also discuss the importance of feature importance analysis and sensitivity analysis in understanding the behavior of AI/ML models. This would provide readers with practical guidance on how to interpret the results of AI/ML models and make informed decisions. Fourth, the authors should provide a more detailed discussion of uncertainty quantification. This should include a detailed explanation of techniques such as Bayesian neural networks and ensemble methods, and their application in rock mechanics. The authors should also discuss the importance of confidence intervals and prediction intervals in assessing the reliability of AI/ML models. This would provide readers with a better understanding of the limitations of AI/ML models and how to account for uncertainty in their predictions. Fifth, the authors should significantly expand their discussion of data quality. This should include a detailed analysis of the types of data used in rock mechanics, as well as the specific challenges associated with each type. The authors should also discuss the impact of noise, outliers, and missing data on the performance of AI/ML models, and the techniques used to mitigate these issues. This would provide readers with a better understanding of the importance of data quality and how to ensure the reliability of their AI/ML models. Sixth, the authors should provide a more detailed discussion of model generalization. This should include a detailed analysis of the factors that affect generalization, such as model complexity, data size, and feature engineering. The authors should also discuss the techniques used to improve generalization, such as cross-validation and domain adaptation. This would provide readers with a better understanding of how to ensure that their AI/ML models generalize well to new datasets. Finally, the authors should include a dedicated section on the limitations of current AI/ML approaches. This should include a critical analysis of the potential pitfalls of relying solely on data-driven methods, and the need for hybrid approaches that combine data-driven models with physics-based understanding. This would provide readers with a more balanced perspective on the strengths and weaknesses of AI/ML in rock mechanics and guide future research in the field. By addressing these weaknesses, the authors can significantly improve the quality and impact of their paper.

❓ Questions

Several questions arise from my analysis of the paper that I believe warrant further consideration. First, given the paper's focus on the application of AI/ML in rock mechanics, how can we ensure that these models are not merely fitting to noise or spurious correlations in the data, but are actually capturing the underlying physical processes? This question is particularly relevant given the complexity of rock mechanics and the potential for data quality issues. Second, the paper acknowledges the "black box" nature of some AI models. However, how can we develop more interpretable AI models that provide insights into the underlying mechanisms of rock behavior, rather than just providing predictions? This question is crucial for building trust in AI/ML models and ensuring their adoption in engineering practice. Third, the paper highlights the need for standardized datasets. However, what specific steps can be taken to create such datasets, and how can we ensure that they are representative of the diverse range of rock mechanics problems? This question is important for addressing the data scarcity issues that hinder the development of robust AI/ML models. Fourth, the paper discusses the potential of physics-informed neural networks (PINNs). However, how can we effectively integrate physical constraints into neural networks, and what are the limitations of this approach? This question is important for understanding the potential and challenges of using PINNs in rock mechanics. Fifth, the paper mentions the use of AI/ML for tasks such as rock mass classification and rockburst prediction. However, how can we validate the performance of these models in real-world scenarios, and what are the risks associated with relying solely on AI/ML for these critical tasks? This question is important for ensuring the safe and reliable application of AI/ML in rock engineering. Sixth, the paper does not provide a detailed discussion of the computational cost of training and deploying AI/ML models. How can we develop more efficient AI/ML algorithms that can be used in resource-constrained environments? This question is important for making AI/ML more accessible to practitioners in the field. Finally, the paper does not explicitly address the ethical implications of using AI/ML in rock mechanics. What are the potential ethical concerns associated with the use of AI/ML in this field, and how can we ensure that these technologies are used responsibly? These questions are important for guiding future research and ensuring the responsible development and application of AI/ML in rock mechanics.

📊 Scores

Soundness:2.75
Presentation:3.0
Contribution:2.25
Rating: 4.25

Keywords

Click the button to extract keywords

Insights

Click the button to extract insights
Version 3 ⚠️ Not latest
Citation Tools

📝 Cite This Paper