Shuo Jiang (姜朔)

Summary: This paper introduces Intelligent Design 4.0 (ID4.0), a new paradigm enabled by foundation model-based agentic AI, and reviews the evolution of intelligent design from rule-based systems to multi-agent collaboration. It proposes an ontological framework for ID 4.0 and discusses its potential for end-to-end design automation, along with key challenges and opportunities in data, agent coordination, and problem formulation.

Summary: This paper explores the use of LLMs for engineering combinatorial optimization by leveraging their reasoning capabilities and contextual knowledge. It proposes an LLM-based framework for optimizing Design Structure Matrix (DSM) sequencing and demonstrates faster convergence and higher solution quality than benchmark methods, with further gains from incorporating domain context.

Summary: This paper introduces AutoTRIZ, an AI-based ideation system that integrates LLMs to automate and enhance the TRIZ methodology for engineering innovation. AutoTRIZ takes a user-defined problem statement as input, conducts the TRIZ reasoning process automatically, and generates structured solution reports, with its effectiveness demonstrated through textbook cases and a real-world battery thermal management system design.

Summary: This paper presents TechDoc, a multimodal deep learning architecture for scalable and automated technical document classification. By integrating textual content, document images, and inter-document associations using convolutional, recurrent, and graph neural networks, the approach achieves higher classification accuracy than unimodal and state-of-the-art benchmark methods.

Summary: This paper constructs a technology fitness landscape using deep neural embeddings of patent data to characterize technological change across 1,757 domains. The landscape provides a bird’s-eye view of the technology space, revealing heterogeneous improvement rates and offering a biologically inspired way to interpret technology evolution and future innovation directions.

Summary: This paper surveys and categorizes research that uses patent data for engineering design, organizing the literature by contributions to design theories, methods, tools, and strategies, as well as by data types and analytical approaches. It highlights how advances in AI and data science create new opportunities for patent-driven design research and outlines promising future research directions.

Summary: This paper surveys data-driven design-by-analogy (DbA) research enabled by large-scale design databases and advances in data science and AI. It categorizes existing studies into analogy encoding, retrieval, mapping, and evaluation, and identifies promising future research directions, including a conceptual integrated data-driven DbA system.

Summary: This paper proposes an automated approach to derive design feature vectors from patent images using a convolutional neural network that captures visual characteristics and technology-related information. By embedding visual design stimuli into a vector space, the method supports near-field and far-field design-by-analogy retrieval and is validated through classification accuracy and a case study.