⚡ electrical engineering

Capability-based domain assignment for patent-based convergence analysis in classification-dispersed fields: an ISRU space-construction case

This paper proposes a Work Breakdown Structure–Integrated Technology Classification (WBS–ITC) framework to map dispersed lunar in-situ resource utilization (ISRU) space-construction patents by construction capability rather than traditional classification, revealing distinct technology concentrations and functional adjacencies that conventional patent mapping overlooks.

Original authors: Tai Sik Lee, Yoonsun Lee

Published 2026-08-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Tai Sik Lee, Yoonsun Lee

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a future where humanity builds permanent outposts on the Moon, not by hauling every brick from Earth, but by using the lunar soil itself. This concept, known as in-situ resource utilization, involves turning the dusty surface material, called regolith, into roads, landing pads, and living shelters. It is a massive engineering challenge that requires a mix of skills: processing raw dirt, mixing it into strong concrete-like materials, printing structures layer by layer, and using robots to do the heavy lifting. Because this field is so new and complex, the inventions designed to make it happen are scattered across many different categories. Some inventors file their patents under rules for construction, others under rules for space travel, and others under rules for robotics or chemistry. There is no single label that captures the entire effort, making it difficult to see the full picture of what is being invented and how these different pieces fit together.

A team of researchers set out to solve this problem of scattered information. They wanted to create a clear map of the inventions related to building on the Moon, but they knew that simply searching for keywords or grouping patents by their official classification codes would miss the mark. Instead, they developed a new way to organize these patents based on the actual job the invention does. They started by breaking down the entire process of lunar construction into four main stages: preparing the raw materials, manufacturing the building components, using robots to execute the work, and assembling the final structures and systems. They then created fifteen specific categories within these stages to act as a guide.

Using this guide, the researchers examined nearly 450 patent families that had been gathered from public databases. They did not just look at the titles or the official codes attached to the patents. Instead, they read the detailed descriptions of what the inventors claimed their machines or methods could actually do. They compared these claims against their four-stage guide to decide which category each patent truly belonged to. This process allowed them to keep patents that fit the description, add patents that were missed by standard searches, and remove patents that looked relevant but did not actually perform the specific construction task.

The resulting map revealed a surprising concentration of activity. The vast majority of the patents focused on three areas: the materials used to build, the manufacturing processes to shape them, and the final structures themselves. Specifically, the most common inventions involved creating habitats, developing composite or ceramic materials, and using extrusion-based 3D printing to build layer by layer. Another significant area was the monitoring of these processes to ensure quality. In contrast, while robots are essential to the vision of lunar construction, the patents specifically claiming autonomous robotic execution were far fewer. The researchers found that robotics appeared as a narrower, more specialized layer within the broader construction effort, rather than the dominant theme one might expect.

The study also showed how much the picture changes depending on how you look at it. When the researchers compared their detailed, capability-based map with a simpler map based only on standard patent codes and keywords, the results were quite different. The standard approach missed many relevant inventions and included many that were not actually about construction. For example, the connection between 3D printing and process monitoring appeared much stronger in the new map than in the old one. Similarly, the link between habitat structures and deployable systems became much clearer once the patents were sorted by what they actually did rather than how they were labeled.

This work provides a new tool for anyone trying to understand the state of lunar construction technology. By organizing patents based on the specific construction capability they offer, the researchers created a clearer view of where the technology is strong and where it might be lacking. They identified that while materials and manufacturing are well-represented, other areas like excavation and site preparation are not yet fully captured in the current patent landscape. This map does not predict the future or guarantee that these technologies will work, but it offers a reliable way to see the current state of invention. It helps mission planners, companies, and researchers see exactly which building blocks are being developed and how they connect, providing a solid foundation for future exploration and development on the Moon.

Technical Summary: Capability-Based Domain Assignment for Patent-Based Convergence Analysis in Classification-Dispersed Fields

Problem Statement
The paper addresses a critical methodological gap in patent-based convergence analysis within "classification-dispersed fields." In emerging technological domains such as Lunar In-Situ Resource Utilization (ISRU) space construction, relevant patent families are scattered across numerous Cooperative Patent Classification (CPC) classes and diverse title/abstract vocabularies. No single CPC class or keyword set adequately defines the field. Consequently, conventional patent mapping approaches—relying solely on keyword retrieval, CPC aggregation, or citation networks—fail to reveal the underlying construction-capability structure. The admitted corpus in such analyses often becomes a "hidden denominator," where the choice of field delineation method dictates the resulting portfolio shares, concentration measures, and co-tagging indicators, potentially obscuring the true technological landscape.

Methodology: The WBS–ITC Framework
The study proposes a capability-based domain assignment procedure using a Work Breakdown Structure–Integrated Technology Classification (WBS–ITC) framework. This approach decouples the initial retrieval of candidate patents from the final assignment of construction capabilities. The methodology proceeds through three distinct analytical stages:

Retrieval and Relevance Assessment: A broad query combining CPC anchors and keyword expansions is used to retrieve candidate patent families (initially 1,093 records, reduced to 443 active families after deduplication and relevance filtering).
Deterministic Rule Hits: A transparent layer of CPC prefixes and title/abstract keywords generates "rule hits" to identify candidate domains. This serves as a first-pass, deterministic signal.
Claim-Based Capability Fit: This is the core differentiator. The representative abstract and the first independent claim of each patent family are manually assessed against a predefined WBS–ITC specification. This specification, derived from public mission architectures (ISECG, NASA, ESA, etc.) and construction taxonomies, defines four WBS layers (Materials, Manufacturing, Robotics, Structures/Systems) and 15 specific ITC domains.
- Keep/Add/Remove Logic: Based on the claim-level assessment, domain tags are either kept (if the rule hit aligns with the claimed capability), added (if the capability is present in the claims despite no rule hit), or removed (if the rule hit was a false positive based on lexical proximity but not actual capability).

The final output is a patent-family $\times$ ITC-domain assignment matrix used to compute portfolio distributions and pairwise Jaccard co-tagging indices.

Key Contributions

WBS–ITC Specification: The authors define a specific capability architecture for Lunar ISRU space construction, translating external mission requirements into 15 operational patent-assignment domains.
Transparent Assignment Mechanism: The study releases the full assignment table, detailing the "keep/add/remove" decisions for every patent family. This allows for the independent inspection of how deterministic rule signals diverge from capability-based assignments.
Methodological Separation: The paper explicitly separates the retrieval set, the rule-hit layer, and the final capability assignment, demonstrating that the choice of domain assignment method significantly alters convergence indicators.

Results

Classification Dispersion: Within the active set of 443 patent families, relevant patents span 166 distinct four-character CPC prefixes, with no single prefix covering more than 21% of the families. This confirms the classification-dispersed nature of the field.
Capability Distribution: The resulting capability map shows concentrations in Structures and Systems (35.5% of tags), Manufacturing (31.8%), and Materials (28.1%). Robotics appears as a narrower, claim-supported execution layer (4.7%), suggesting that while robotics is essential, the current patent corpus focuses more heavily on the materials and structural systems it manipulates.
Divergence from Rule Signals: There is a 53.0% overlap between the final WBS–ITC tags and the initial rule hits, calculated as the proportion of shared cells (rule-hit present and assignment present) out of the total 795 assignment tags. Significant "add" and "remove" actions occur; for instance, many patents related to "Process Monitoring" were removed because they lacked specific in-process monitoring claims, while "Habitat Structures" gained tags based on claim content despite lacking specific keyword hits.
Co-Tagging Shifts: The assignment method drastically changes co-tagging rankings. For example, the pair "Extrusion AM" and "Process Monitoring" rises from rank 4 (under CPC rules) to rank 1 (under WBS–ITC) with a Jaccard similarity of 0.341. Conversely, "Autonomous Construction" and "Habitat Structures" drops from rank 1 to rank 52. This indicates that deterministic signals often misrepresent the actual functional integration found in patent claims.
Boundary Evidence: 70 patent families were identified as "domain-external," possessing construction relevance but not fitting the current 15-domain specification. These point to candidate extensions such as excavation, foundation engineering, and site preparation.

Significance and Claims
The paper claims that assigning patents by construction capability, rather than by CPC or title/abstract signals alone, fundamentally alters the reported technology-domain map. The study does not claim to validate engineering maturity or predict market success; rather, it provides a patent-based capability map that makes construction functions and their co-tagging relationships visible for subsequent analysis.

The primary significance lies in the reusable reporting structure it offers for classification-dispersed fields. By documenting the "keep/add/remove" logic and releasing the assignment table and scripts, the authors enable the reconstruction of the capability map and provide a transparent method for researchers, firms, and agencies to scope technology portfolios, identify gaps in the regolith-to-infrastructure chain, and understand the true adjacency of construction capabilities without the distortion of classification artifacts.

More like this