Roger J. Jiao

Manufacturing systems in the industry 4.0 era are becoming increasingly complex and large-scale, creating growing demand for efficient and reliable failure recovery systems. However, traditional manual failure recovery processes often suffer from cognitive overload, inconsistent recovery decisions, and limited scalability. To address these challenges, this paper proposes a retrieval augmented generation (RAG) enhanced failure recovery framework that integrates large language model (LLM) with hierarchical knowledge graphbased knowledge management. The proposed system organizes machine failure data into a structured knowledge graph and performs semantic retrieval over failure metadata and domain specific knowledge to generate recommended recovery solutions. The framework enables scalable and context-aware troubleshooting for manufacturing failure recovery. Keywords Large-language model, Knowledge graph, Retrieval-augmented generation, Failure recovery, Industry 4.0

Design research aspires to be both scientifically credible and practically meaningful. Verification and validation (V&V) are central to these dual goals: verification demonstrates that we do things right, whereas validation establishes that we do the right thing. In design research, particularly in computational and AI-enabled strands, validation of problem context, stakeholder needs, and decision-making is often assumed rather than explicitly demonstrated. Using a What → How → Use lens, this article clarifies the distinct roles of verification and validation, exposes recurring pitfalls in current practice, and envisions a risk-informed V&V pipeline for design research. We argue that evidence should be proportional to the decisions a method is intended to support and the consequences of being wrong, and that validation must be treated as a core methodological concern alongside verification. Design research attains methodological rigour when problem validation is pursued with the same care and discipline as solution verification.

Problem formulation is arguably the most consequential determinant of rigour and originality in engineering design research, yet it remains inconsistently systematized across the literature. Although academic critiques often focus on methods and results, many recurring weaknesses originate upstream, in research questions framed as descriptive prompts rather than operational, analytical engineering problems. This article advances a constructive agenda for strengthening research framing by grounding problem formulation in design–theoretic requirements analysis. Drawing on design theory, requirements engineering, and validity principles, four complementary scaffolds are introduced to operationalise this approach: (i) a taxonomy of common pitfalls to diagnose framing errors early; (ii) a customer–needs–to–functional–requirements analogy for translating broad motivations into operational research requirements; (iii) a Metric-Measure-Method structure to ensure that constructs and evidence are auditable; and (iv) coherence mechanisms that explicitly link research questions to hypotheses, study design, and Type I/II validation. By treating problem formulation as a design artefact, these tools help clarify constructs, surface hidden assumptions, and align methods with specific decision contexts. Ultimately, this framework supports claims that are both falsifiable and transferable, raising the standard of rigour across the field.

Literature reviews in design research are meant to consolidate knowledge, clarify methodological patterns, and guide future inquiry through synthesis rather than compilation. Yet many published reviews remain largely descriptive – dominated by inventories, procedural narration, and bibliometric counting – and consequently fail to provide a coherent technical framework or actionable insight. AI–generated text further exacerbates this issue by automating surface–level summarisation while introducing risks of fabricated or erroneous citations when outputs are not rigorously verified. This perspective argues that review scholarship must shift from AI–like ‘secretary work’ to genuine knowledge construction, emphasising framework–first synthesis, defensible cross–study comparison, explicit validity regimes and failure modes, benchmark–aligned evaluation, and auditable provenance practices. Building on systematic reporting principles (e.g., PRISMA) while treating reporting as a floor rather than the endpoint, the article outlines a synthesis–driven, auditable review pipeline and calls for coordinated action among authors, reviewers, editors, and the broader community to elevate standards, strengthen integrity, and restore literature reviews as central, high–impact contributions to design research.

Industry 5.0 advances a human‑centric paradigm where human – automation symbiosis (HAS) depends on aligning operator cognition with automated precision. This paper proposes manufacturing nudging as an engineering mechanism to steer operator behavior while preserving autonomy and develops a complete pipeline for nudge design and personalization at production scale. First, a conjoint prospect‑theoretic model captures dual stakeholder valuations – operators (workload, burden) and managers (quality, cycle time) – and aggregates multiple nudging features via cumulative prospect theory to reflect realistic risk, loss‑aversion, and probability‑weighting effects. Second, hierarchical Bayesian parameterization supports customization (segment‑level) and personalization (operator‑level) with coherent prior constraints and MCMC‑based estimation. Third, an engineering‑cost model estimates operation cycle time with a neural network trained on operator – nudge – scenario data. Finally, a two‑dimensional genetic algorithm (2D‑GA) optimizes operator‑to‑nudge assignments, exploiting a matrix encoding that preserves operator × feature structure and thereby circumvents the sparsity and locality issues of 1D encodings. A jet‑engine assembly case study with AR‑based nudges demonstrates increased symbiotic value with controlled cycle‑time cost; optimizing purely for behavioral value yields only a 0.15% gain but a 3.33% higher cost, whereas the multi‑objective formulation attains near‑maximal value at substantially lower cost. The results establish a rigorous, scalable, and interpretable approach to nudging for HAS in complex

Engineering design is increasingly influenced by abundant data and advances in machine learning, yet the common narrative of “data-driven design” mischaracterizes how design knowledge works. Unlike perception-focused AI domains that rely on large labeled datasets and stable mappings, engineering design is data sparse, knowledge rich, decision centric, and context dependent. This article reframes AI for design as data informed rather than data driven, arguing that effective decisions require integrating knowledge, constraints, mechanisms, semantics, and evolving intent, which data alone cannot supply. Drawing on the classical Data-Information-Knowledge (DIK) hierarchy, the article proposes three first principles for grounding AI in design: (i) DIK distinction, preserving the differences between raw data, processed information, and actionable knowledge; (ii) Domain problems, emphasizing alignment with domain objectives, constraints, and causal mechanisms; and (iii) Design context, embedding AI within the interpretive, intent-laden nature of design reasoning. An operational framework and end-to-end pipeline are introduced to align AI tools with decision needs, avoid dataset-first framing, and evaluate systems by decision quality rather than predictive accuracy. Positioning data as evidence rather than authority, the article shows how elevating contextual grounding, feasibility logic, and designer intent enables AI to move beyond pattern recognition and support rigorous, interpretable, knowledge-aligned design decisions.

Human-subject studies are now a common vehicle for generating empirical evidence in engineering design research, spanning protocol studies, controlled experiments, surveys, interviews, and mixed–methods approaches. Yet a recurring difficulty in current practice is that many studies remain experience–informed point observations, such as single artifacts, simplified tasks, convenience samples, or narrow decision regimes, while making broad claims about engineering design knowledge or practice. This perspective argues that the central issue is not the inclusion of human participants, but rather under–specification of design context and misalignment of validation logic, i.e. importing methods and reporting conventions from adjacent disciplines can improve transparency, but does not, by itself, establish engineering–level applicability or transferability. We synthesise state–of–the–art norms from human factors and ergonomics, human-computer interaction and open science practices, social science qualitative research, and clinical and observational research reporting, and evaluate their strengths and limitations in the context of engineering design. A central organising reflection is an artifact-decision map that frames two orthogonal validity drivers regarding artifact realism and decision realism, and links a study’s position to defensible claim scope. We envision a pragmatic pipeline, including a lightweight Metric–Measure–Method (3M) discipline, and actionable guidance for researchers to strengthen context–aware validation while preserving methodological diversity.

Engineering design research has long been distinguished by its close integration with practice, context, and decision making under real constraints. Yet, an increasing body of research adopts self contained, context free formulations that appear methodologically rigorous but lack practical relevance. This perspective argues that explicit industry relevance and concrete engineering contexts are vital for conducting meaningful and rigorously sound design research. Without such grounding, methodological sophistication risks becoming illusory. It calls for renewed collective commitment to industrial grounding as a foundation for rigorous, relevant, and impactful design research.