When engineering drawings are regenerated across design iterations or reused across part families, subtle discrepancies often arise, making manual comparison error-prone and labor-intensive. This paper presents Drawing-Checker, a vision Retrieval-Augmented Generation (RAG) framework that automates drawing comparison using reference standards. The system integrates YOLO-based view detection, CLIP-based embedding retrieval, and a state-of-the-art multimodal Large Language Model (LLM) reasoning to identify annotation and geometric discrepancies between reference and target drawings. By decomposing full drawings into semantically matched view pairs and flagging unmatched views, Drawing-Checker achieves higher precision, substantially higher recall, and better overall F1 than a full-image prompting baseline, without increasing false positives, across 500 customized test cases. These gains confirm the effectiveness of view-level prompting in surfacing localized differences (particularly missing, added, or altered features) while maintaining precision. At the same time, the study highlights persistent challenges in geometric reasoning, localization accuracy, and hallucination control, underscoring the need for enhanced prompt design, view detector generalization, and domain-specific model adaptation to achieve robust and scalable deployment.

Recent advances in generative Artificial Intelligence (AI)—particularly Large Language Models (LLMs)—offer a new paradigm for CAD interaction by enabling natural and intuitive input through texts, images, and context-aware selections. In this study, we present CADialogue, a multimodal LLM-powered conversational assistant to enable intuitive parametric CAD modeling through natural language, speech, image, and selection-based geometry interactions. Built on a general-purpose large language model, CADialogue translates user prompts into executable code to support geometry creation and context-aware editing. The system features a modular architecture that decouples prompt handling, refinement logic, and execution—allowing seamless model replacement as LLMs develop—and includes caching for rapid reuse of validated designs. We evaluate the system on 70 modeling and 10 editing tasks across varying difficulty levels, assessing performance in terms of accuracy, refinement behavior, and execution time. Results show an overall success rate of 95.71%, combining a 91.43% baseline under Text-Only input with additional recoveries enabled by Text + Image input, with robust recovery from failure via self-correction and human-in-the-loop refinement. Comparative analysis reveals that image input improves success in semantically complex prompts but introduces additional processing time. Furthermore, caching confirmed macros yields over 85.71% speedup in repeated executions. These findings highlight the potential of general-purpose LLMs for enabling accessible, iterative, and accurate CAD modeling workflows without domain-specific fine-tuning. The source code and dataset for CADialogue are available at https://github.com/Hiram31/CADialogue.

Procedural CAD models are the primary source of detailed mechanical designs as they are editable, can convey design intent and be enriched with design and manufacturing information. However, many companies still spend significant time modifying low-quality procedural CAD models, which are often poorly structured or become unusable when geometric changes are made. Current best practices focus mainly on "how to do it well," which can frustrate users who struggle to identify the sources of errors and determine how to resolve them, leading to issues that often persist when integrating models into other applications. In contrast, we propose practices based on the principle of “how to avoid doing it wrong” derived from analyzing negative knowledge in CAD modeling. This paper contributes to advancing the quality of 3D procedural models. Our framework adopts the three levels of quality identified in the literature: usability, re-usability, and design intent richness, while focusing specifically on deriving best prac-tices to prevent failures, based on negative knowledge related to reusability. We build on the premise that quality issues hindering reusability can be categorized into three interrelated but largely independent types, each affecting the sketches, datums, or features within a procedural model’s tree structure. In this work, we address the study of sketches. The negative impact of typical failures, identified through representative case studies that illustrate both design and redesign errors, is analyzed to extract best practices that ensure CAD model reusability. While this study results in a practical guide for avoiding sketch-related errors that compromise the reusability of CAD models, the main contribution lies in demonstrating a framework that transforms negative knowledge into effective best practices.

Oftentimes, designers need to reverse engineer detailed design drawings to create 3D models. These drawings are commonly accepted as complete information sources in design datasets but may contain errors or ambiguities and lead to misinterpretation, resulting in digital CAD models that inherit such ambiguities, or modelling situations with no solution. We propose a classification of errors based on the type of ambiguity they produce (redundancy, omission, contradiction, and polysemy) and discuss the results of an experiment designed to evaluate the effectiveness of a strategy based on this classification. The classification represents an essential step for developing a checklist to ensure drawing ambiguities are fully resolved before the corresponding 3D models are built. In the long term, the proposed classification can guide a divide-and-conquer strategy for training artificial intelligence systems in the detection and repair of these ambiguities.

The quality of mechanical 3D models generated with Computer-Aided Design (CAD) systems is a key aspect of the product development process. Evidence shows that formal modeling methodologies contribute to creating higher-quality 3D models that improve design performance, but little attention has been drawn to actual practices in industrial companies. Some of the challenges involved in studying these strategies include lack of access to what is often considered an intellectual asset for organizations and thus kept secret, as well as the variety of industrial products and different levels of automation for a particular design within the organization. In this paper, we describe the process of creating a comprehensive assessment instrument for diagnosing the level of awareness of CAD quality among the different stakeholders involved in the product development process and the extent to which organizations have implemented strategies to leverage and promote the creation and use of high-quality parametric CAD models. The survey instrument contains questions about key aspects of 3D CAD model quality, from methodologies to training. A formal survey validation process was conducted using the expert judgement technique. The modified Kappa coefficient was used to identify questions that needed improvement. The final version of this instrument is provided in the appendix.

Previous research in human-computer interaction (HCI) has focused relatively little attention on making parametric Computer-Aided Design (CAD) tools user-friendly.This study aims to address key research questions about the difficulties beginners encounter when using parametric CAD tools, the reasons behind these challenges, and the principles of interface design that can improve their understanding.To answer these questions, a comprehensive three-stage design framework for CAD Graphical User Interfaces (GUIs) is proposed.The first stage involves analyzing user experience (UX) within the context of the CAD interface, followed by developing customized solutions to meet specific requirements in the second stage.Finally, the framework includes rigorous testing and evaluation of the CAD GUI solutions against the identified requirements.Statistical analysis was used to validate the improved usability perception of the new interfaces.This framework leads to the creation of rules supporting the design of understandable GUIs for parametric CAD tools.Ultimately, this research contributes to advancing comprehensible GUIs by shedding light on the challenges beginners face, offering practical recommendations to enhance their experience, and facilitating a better understanding of tools to increase their efficiency.

Sketched drawings sometimes include non-solid lines drawn as sets of consecutive strokes. They represent dashed lines, which are useful for various purposes. Recognizing such dashed lines while parsing drawings is reasonably straightforward if they are outlined with a ruler and compass but becomes challenging when they are hand-drawn. The problem is manageable if the strokes are drawn consecutively so we can leverage the entire sequence. However, it becomes more challenging if they are drawn unordered, and/or we do not have access to the sequence (like in batch vectorization). In this paper, we describe a new approach to identify groups of strokes as depicting single hand-drawn dashed lines. The approach does not use sequence information and is tolerant with irregularities and imprecisions of the strokes. Our goal is to identify hidden lines of sketched engineering line-drawings, which would enable the interpretation of line-drawings with hidden edges, which currently cannot be efficiently vectorized. We speculate that other fields like hand-drawn graph interpretation may also benefit from our approach.

Digital product data quality and reusability has been proven a critical aspect of the Model-Based Enterprise to enable the efficient design and redesign of products. The extent to which a history-based parametric CAD model can be edited or reused depends on the geometric complexity of the part and the procedure employed to build it. As a prerequisite for defining metrics that can quantify the quality of the modeling process, it is necessary to have CAD datasets that are sorted and ranked according to the complexity of the modeling process. In this paper, we examine the concept of perceived CAD modeling complexity, defined as the degree to which a parametric CAD model is perceived as difficult to create, use, and/or modify by expert CAD designers. We present a novel method to integrate pair-wise comparisons of CAD modeling complexity made by experts into a single metric that can be used as ground truth. Next, we discuss a comprehensive study of quantitative metrics which are derived primarily from the geometric characteristics of the models and the graph structure that represents the parent/child relationships between features. Our results show that the perceived CAD modeling complexity metric derived from experts’ assessment correlates particularly strongly with graph-based metrics. The Spearman coefficients for five of these metrics suggest that they can be effectively used to study the parameters that influence the reusability of models and as a basis to implement effective personalized learning strategies in online CAD training scenarios.

Describing and supplementing geometric shapes (parts) and layouts (assemblies) with relevant information is key for successful product design communication. 3D annotation tools are widely available in commercial systems, but they are generally used in the same manner as 2D annotations in traditional engineering drawings. The gap between technology and practices is particularly evident in plain text annotations. In this paper, we introduce a functional classification of text annotations to provide an information framework for shifting traditional annotation practices towards the Model-Based Definition (MBD) paradigm. In our view, the current classification of dimensions, tolerances, symbols, notes, and text does not stress the inherent properties of two broader categories: symbols and text. Symbol-based annotations use a symbolic language (mostly standardized) such as Geometric Dimensioning and Tolerancing (GD&T) to provide precise information about the implications of geometric imperfections in manufacturing, whereas notes and text are based on non-standardized and unstructured plain text, and can be used to convey design information. We advocate that text annotations can be characterized in four different functional types (objectives, requirements, rationale, and intent), which should be classified as such when annotations are added to a model. The identification and definition of a formalized structure and syntax can enable the management of the annotations as separate entities, thus leveraging their individual features, or as a group to gain a global and collective view of the design problem. The proposed classification was tested with a group of users in a redesign task that involved a series of geometric changes to an annotated assembly model.

Our goal is to automatically generate CAD 3D models from 2D sketches as part of a design chain where models should be procedural, containing features arranged in a model tree and linked to suitable datums. Current procedural models capture much about the design intent and are easy to edit, but must be created from scratch during the detailed design state-given conceptual sketches as used by designers in the early part of the design process, current sketch-based modeling approaches only output explicit models. Thus, we describe an approach to extract high-level information directly from 2D engineering wireframe sketches and use it to complete a CSG feature tree, which serves as a model tree for a procedural 3D CAD model. Our method extracts procedural model information directly from 2D sketches in the form of a set of features, plus a set of datums and relationships between these features. We detect and analyze features of 2D sketches in isolation, and define the CSG feature tree by the parent-child relationships between features, and combine this information to obtain a complete and consistent CSG feature tree that can be transferred to a 3D modeler, which reconstructs the model. This paper focuses on how to extract the feature datums and the extrusion operation from an input 2D sketch. (c) 2021 The Authors. Published by Elsevier Ltd.

Sketch-Based Modeling (SBM) is a field of study that focuses on the automatic creation of 3D models from freehand drawings. Today, two related branches coexist: one aimed at facilitating input for 3D content creation, and the other aimed at routing engineering designs into CAD/CAM/CAE. The latter is the goal of this position paper. Early attempts concentrated on the problem of line-drawing interpretation, but efforts switched toward geometric reconstruction as the “inverse projection” problem became the most challenging step to produce 3D models from 2D line-drawings. The term SBM was popularized when sketches began to displace line-drawings as main input. In the context of engineering design, interest in SBM has somewhat decreased, as studies have shown that current SBM tools are not as usable as paper and pencil sketches, nor do they yet provide any additional value to traditional media. Furthermore, engineers feel reasonably comfortable with current Mechanical CAD (MCAD) paradigms based on parametric modeling, and fail to recognize the potential benefits of interacting with computers via sketches. However, new technological trends such as personal fabrication and the democratization of CAD and manufacturing can significantly benefit from improved SBM tools. In this paper, we conduct a meta-review of the SBM literature which we view as a combination of three elements: geometry (shape), psychology (perception), and engineering (function). We advocate for a new approach to SBM based on reformulating the weights of these elements as an approach for searching the set of intentions in sketches conveyed through cues which, when perceived, reveal regularities and features of the object. Finally, we consider quality of CAD models not just as error-free models, reusable models, or even models that convey design intent, but models that consider these interrelated aspects as a whole. B-Rep models produced by current SBM approaches are “dumb models” without parametrization or procedures required to enable reusability and ensure that the design intent of the sketch is properly conveyed. We advocate for improved analysis approaches aimed at revealing higher-level design information embedded in engineering sketches, as a critical stage to generate richer 3D MCAD models.

In the Model-Based Enterprise (MBE) paradigm, digital product models are the primary source of information to support the design, analysis, and manufacturing of products. The master geometry of these product models is generally built procedurally in a CAD system, typically as history-based parametric geometry. However, some model conversion and exchange processes imply losing procedural information (i.e. the internal structure that represents how the model was built). In this position paper, we review the state of the art in procedural CAD model exchange and discuss an approach for conveying procedural information during model exchange. Although not available in practice, the approach is currently theoretically feasible, as the exchange of procedural information is already supported in modern neutral formats. Finally, we develop our position by advocating for defining Conformance Classes to facilitate the practical development of the ISO 10303 AP 242 standard, thus enabling official implementations that can provide full exchange capabilities for procedural CAD models.

The ability to recognize misleading data visualizations is a key aspect of visualization literacy. In this article, we argue that learning to successfully identify a deceptive graphic requires strategies that deliberately force learners to take an active role in the visualization process. We describe a series of experiments where three groups of learners were shown various deceptive graphics and asked to answer a series of questions. Three different interventions were analyzed to compare the educational effectiveness of the strategies used to engage learners into the process of identifying deceptive visualizations. Our results suggest that the ability to identify deceptive visualizations must be explicitly taught as a core element of visualization literacy. Although both traditional and self-learning approaches are beneficial, the more active the intervention, the higher its educational effectiveness.

Strategies for design reuse play a fundamental role in the development of products that change and evolve over time. Design changes often involve modifications in the geometry of parts and assemblies, which are driven by changes in the digital representation of the product, i.e. the procedural and parametric CAD model. Consequently, constraint redundancies in the two-dimensional profiles that build the parametric model can significantly hinder alterability and reusability. This paper argues that constraint redundancy conditions are not solely a computational problem but a more complex issue that involves the interaction with the sketch throughout the modeling process and the modeling scheme used to convey a specific design intent. We analyze a representative group of 3D MCAD systems and the interaction with their corresponding geometric kernels and Geometric Constraint Solvers (GCS) to determine how constraint redundancy is managed at the profile level. Next, we report the results of a series of experiments to evaluate the influence of redundant constraints on model editing and reusability tasks and describe the development of a new software tool for identifying and parsing constraint redundancy conditions. We conclude that constraint redundancy in profiles significantly and unnecessarily compromises model conciseness, robustness, and the overall model quality, which negatively affects user productivity and downstream processes. The implications of constraint redundancy conditions for CAD training are emphasized. Our experiments also demonstrate the value of our parsing tool to assist users in maximizing model reuse (by removing redundancies) and the communication of design intent (by proposing an optimal set of non-redundant constraints), but further development is necessary to use it as a practical tool for engineering analysis. (c) 2021 Elsevier B.V. All rights reserved.

Departament d'Enginyeria Mecanica i Construccio. Codis assignatura: ET1009, EM1009, EQ1009, EE1009, ET1028, EM1025, DI2028, SDI122