Reviewer has chosen not to be Anonymous

Overall Impression: Good

Content:
Technical Quality of the paper: Good
Originality of the paper: Yes, but limited
Adequacy of the bibliography: Yes

Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Good
Organization of the paper: Satisfactory
Level of English: Satisfactory
Overall presentation: Good

Detailed Comments:

This paper presents a survey of neuro-symbolic AI approaches for advanced product design, covering applications in CAD generation, topology optimization, manufacturing planning, assembly, quality assurance, and sustainability. The topic is timely and relevant, as product design increasingly requires systems that combine data-driven learning with explicit reasoning, constraint satisfaction, and explainability. The manuscript is generally well organized, clearly written, and provides a useful overview of a fragmented research area. The taxonomy based on neuro-symbolic integration types provides a coherent framework for organizing the literature, and Table 3 serves as a valuable reference for researchers entering the field.

The paper's primary strength is its systematic organization of the literature and its attempt to connect general neuro-symbolic paradigms with specific stages of the product design lifecycle. The discussion of symbolic substrates, neural functions, and application domains provides readers with a useful conceptual map of the field. The figures and tables are generally clear and support the narrative effectively.

However, the manuscript would benefit from a deeper level of analysis and synthesis. Much of the discussion remains descriptive, focusing on summarizing individual papers rather than extracting broader lessons from the literature. For example, while the paper identifies the prevalence of Types I–III integrations and notes the relative scarcity of tighter integrations (Types IV–VI), the reasons behind these trends are only briefly discussed. A stronger comparative analysis of the advantages, limitations, and practical trade-offs associated with different integration paradigms would significantly strengthen the survey.

Similarly, the review would benefit from more quantitative synthesis. The manuscript states that 55 papers were included in the review, but it does not provide aggregate statistics regarding the distribution of integration types, symbolic substrates, application domains, or evaluation methodologies. Summary figures illustrating such trends would help readers better understand the current state of the field and identify research gaps.

The evaluation section correctly highlights the lack of standardized benchmarks and heterogeneous evaluation protocols. However, this discussion remains relatively high level. Given the increasing importance of benchmarking within the neuro-symbolic community, the paper would benefit from a more detailed discussion of benchmark requirements, evaluation dimensions, and recommendations for future benchmark development in product design.

Some sections, particularly the discussion of ethical, regulatory, and deployment challenges, are informative but somewhat generic and could be more closely tied to the specific characteristics of product design and manufacturing applications. Several observations in these sections are broadly applicable to AI systems in general and would be strengthened through more domain-specific examples and analysis.

Overall, this is a useful and timely survey that addresses an emerging area at the intersection of neuro-symbolic AI and engineering design. The manuscript provides a solid foundation for researchers entering the field and is suitable for publication after revision. Strengthening the synthesis, increasing the level of critical analysis, and providing more quantitative insights into the reviewed literature would further enhance its impact and long-term value to the community.

Reviewer has chosen not to be Anonymous

Overall Impression: Good

Content:
Technical Quality of the paper: Good
Originality of the paper: Yes, but limited
Adequacy of the bibliography: Yes, but see detailed comments

Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Good
Organization of the paper: Satisfactory
Level of English: Satisfactory
Overall presentation: Good

Detailed Comments:

This paper presents a comprehensive survey of NeSy AI methods applied to advanced product design, manufacturing, CAD, topology optimization, assembly planning, and sustainability-aware engineering workflows. The topic is timely and relevant, particularly given the increasing interest in explainable and constraint-aware AI systems for engineering design automation.

The paper’s main strength lies in its structured synthesis of the literature. Using a PRISMA-inspired protocol, the authors screened 1,095 records and synthesized 55 papers. The authors organize the field using Kautz’s six integration paradigms and provide a coherent mapping between NeSy integration types and stages of the product design lifecycle. The inclusion of Tables 2 and 3, along with the lifecycle-oriented Figure 3, substantially improves readability and helps position existing approaches within a unified framework. The discussion of symbolic substrates (ontologies, constraints, logic, knowledge graphs) and neural functions (generation, optimization, surrogate modeling, perception) is also useful for researchers entering the field.

The survey methodology is generally well described. The PRISMA-style workflow, inclusion/exclusion criteria, and coding strategy improve transparency and reproducibility. The companion ORKG and GitHub resources are valuable additions that increase the long-term usefulness of the work.

However, several limitations should be addressed:

Novelty and analytical depth:
Although the survey is well organized, much of the paper remains descriptive rather than deeply analytical. The manuscript summarizes existing works effectively, but stronger critical comparison across methods is needed. For example, the discussion rarely quantifies trade-offs between different NeSy integration types in terms of "scalability, computational cost, interpretability, or deployment feasibility." Those are the important factors for any system to be used in industrial settings.

Limited engineering evaluation discussion:
The paper repeatedly highlights the need for standardized benchmarks and lifecycle-aware validation, but it does not propose concrete evaluation protocols or benchmark recommendations. A more rigorous comparative discussion of evaluation metrics across CAD, manufacturing, and RL-based systems would strengthen the contribution.

Missing industrial perspective:
The survey focuses heavily on academic prototypes and proof-of-concept systems. More discussion of industrial adoption barriers, integration with commercial CAD/PLM pipelines, and practical deployment challenges would improve the relevance for engineering audiences.

Coverage imbalance:
Types I–III architectures are discussed in greater detail than Types IV–VI. While the authors acknowledge that tighter integrations are less mature, the survey would benefit from a more detailed discussion of differentiable symbolic reasoning approaches and their future potential in engineering design automation.

Writing and presentation issues:
Overall English quality is acceptable, but there are several grammatical inconsistencies, formatting artifacts, and missing references throughout the manuscript. For example:
There are occasional spacing and typography issues (e.g., “NeuroSymbolic” vs. “Neuro-Symbolic”).

Some section references appear incomplete (e.g., “a gap revisited in section .”).

Several citations and figure references could be integrated more smoothly into the narrative.

Bibliography:
The bibliography is broad, recent, and generally adequate. The authors include important foundational and recent NeSy works from both AI and engineering domains. However, a few additional industrial AI/CAD automation references and benchmark-oriented studies could further strengthen the review.

The sustainability and lifecycle assessment (LCA) literature is mentioned but poorly cited. Bougzime et al. (2025) carries significant weight in the sustainability section, but this is a conference proceedings preprint. More established LCA-AI integration references would be appropriate.

Overall, this is a useful and timely survey paper that provides a strong organizational framework for NeSy AI in product design. The paper requires moderate revision, particularly to improve critical analysis, strengthen evaluation discussion, and correct presentation inconsistencies.

Reviewer has chosen not to be Anonymous

Overall Impression: Good

Content:
Technical Quality of the paper: Good
Originality of the paper: Yes
Adequacy of the bibliography: Yes

Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Good
Organization of the paper: Satisfactory
Level of English: Satisfactory
Overall presentation: Good

Detailed Comments:

This survey paper compiles and reviews over 20 years of research exploring the integration of Neurosymbolic AI (NS-AI) within advanced product design, generative CAD, manufacturing optimization, and sustainability-driven design. It successfully bridges a literary gap by mapping diverse AI methodologies directly onto practical stages of the engineering product lifecycle—specifically CAD generation, topology optimization, factory manufacturing planning, robot assembly sequence planning, and quality assurance. The authors frames the core problem by contrasting traditional Neural AI (which excels at data-driven pattern recognition and creative generation but lacks formal rule adherence) with Symbolic AI (which provides rigorous constraint verification and explicit logic but cannot learn from raw data). Neurosymbolic AI is presented as a hybrid solution where the neural component explores design spaces while the symbolic component enforces engineering and physics-based boundaries. Using a PRISMA-inspired methodology, the authors filtered an initial pool of 1,095 papers down to a final corpus of 55 articles. To categorize these approaches, the review uses Kautz’s six neurosymbolic integration types (Types I–VI) as an architectural backbone. It examines how loose coupling (Types I–III) vs. tight architectural coupling (Types IV–VI) are implemented across different engineering workflows.

Strengths:
1) Timely and High-Value Focus: The paper addresses a highly relevant research area, bridging technical AI mechanisms with practical engineering requirements like CAD workflows, manufacturing constraints, and regulatory trust.
2) Rigorous and Transparent Methodology: The PRISMA-inspired literature search is clearly defined with explicit exclusion criteria, further enhanced by providing "living resources" via an open repository on GitHub and ORKG.
3) Excellent Structural Taxonomy: Instead of an unstructured literature list, the review relies on Kautz's six integration types as a core backbone to systematically organize every reviewed method.
4) Cross-Domain Insights: It identifies "transfer-ready" architectures from adjacent domains such as optimization (OptNet), puzzle-solving (SATNet), and robotics (Shield methods) that can be directly adapted for engineering design tasks.
5) Comprehensive Risk Mapping: The survey provides a thorough, multi-faceted analysis of risks categorized into technological, ethical, and regulatory challenges.

Weakness:
1) Overly Limited Corpus and Database Selection: A final corpus of 55 papers is relatively small for such a broad and interdisciplinary field. By limiting searches to Google Scholar, IEEE, ACM, Springer, and arXiv, the authors omitted critical mechanical and industrial engineering databases like Compendex, the ASME Digital Library, and Knovel.
2) Uneven Domain Coverage and Omissions: There is a noticeable and unexplained imbalance in the paper's comparison tables. While CAD, manufacturing, and assembly feature multiple entries, Quality Assurance (QA) is discussed in the text but completely absent from Table 3. Materials and sustainability are also highly underrepresented.
3) Lack of Reproducibility Metrics: The paper fails to document reproducibility markers for the reviewed systems, such as code public availability, hardware implementation details, or dataset accessibility, reducing its utility for researchers hoping to build upon past work. 4) Absence of Visual Evidence: The text lacks screenshots or concrete visual examples of actual system outputs like generated CAD models, topology optimizations, or manufacturing sequences, making the discussion feel detached from practical engineering outcomes.
5) Inconsistent Evaluation and Weak Benchmark Depth: The reviewed literature contains highly varied evaluation metrics, creating a challenge in judging mature vs. weak methods. The authors acknowledge this limitation but fail to propose a concrete framework to detect, benchmark, or fix common issues like models "memorizing" logic patterns.
6) Lab-to-Industry Deployment Gap: The paper notes that tighter neurosymbolic integrations (Types IV–VI) remain confined to lab benchmarks rather than industrial workflows. Some included methods are general AI solutions rather than systems tested in engineering workflows, making parts of the transferability argument feel speculative.

Suggestions:
1) Expand the Literature Search: Incorporate missing engineering-specific databases (Compendex, ASME Digital Library, Knovel) to broaden the corpus and capture key mechanical design literature.
2) Resolve Table Imbalances: Add Quality Assurance (QA) to the primary comparison matrix and explicitly justify or expand upon the thin coverage of materials and sustainability.
3) Enhance Data Transparency & Reproducibility: Introduce dedicated columns in the summary tables detailing code availability (e.g., open-source vs. closed) and dataset accessibility.
4) Incorporate Visual Figures: Embed concrete qualitative examples—such as screenshots of generated CAD models, symbolic verification logic flows, or topology optimization visualizations—to ground the survey visually.
5) Propose a Standardized Benchmark Framework: Offer a concrete evaluation rubric or set of criteria to help the community standardize performance metrics and detect superficial pattern memorization.
6) Include an Industrial Case Study: Bridge the gap between theory and industry by adding at least one real-world industrial case study showing how a neurosymbolic approach functions outside of a lab environment.
7) Refine Readability: Tighten repetitive explanations across the text and optimize dense tables to make the content more accessible to both AI researchers and design practitioners.

This is a highly valuable, well-organized, and timely survey paper that brings structure to a fragmented research landscape. Its categorization using established integration types acts as an excellent roadmap for future design automation. However, to maximize its impact and practical utility, the authors must ground the paper with visual evidence, address the missing literature databases, and offer a clearer assessment of which neurosymbolic approaches are mature enough for real-world engineering deployment.