A Machine Learning and Computer-Vision Framework for Real-Time Control in 3DCP - Layer Morphology as a Design Feature - Speed Control

Aug 5, 2026

3D Concrete Printing (3DCP) demands precise tuning of multiple parameters to ensure print quality, yet current operation often relies on frequent manual intervention, causing inconsistencies and sensitivity to environmental and material variations. To approximate an autonomous self-correcting workflow, machine learning computer-vision models analyze captured images or video to identify defects and geometric deviations, then update key print settings. This paper reviews related real-time and offline approaches and introduces a computer-vision system for robotic 3DCP, emphasizing an ML Speed Control module.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages7

Reading Time18 min

AccessFree to download

Introduction

Challenges of concrete in 3D printing

Speed and flow rate effects on layer morphology

Aim: layer morphology control via machine learning system

Data and sensing approach for real-time speed control

A MACHINE LEARNING AND COMPUTER-VISION FRAMEWORK FOR REAL-TIME CONTROL IN 3DCP: LAYER MORPHOLOGY AS A DESIGN FEATURE João M. Silva 1 Rafael Macedo 3 António Morais 1 João Ribeiro 1, 2 Sacha Mould 3 Paulo J.S. Cruz 1, 2 Bruno Figueiredo 1, 2 1 EAAD – School of Architecture, Art and Design, University of Minho | Portugal 2 Lab2PT – Landscape, Heritage and Territory Laboratory | Portugal 3 DTx – Digital Transformation CoLab | Portugal Corresponding author: [bfigueiredo@eaad.uminho.pt](bfigueiredo@eaad.uminho.pt) Keywords 3DCP; Additive manufacturing; Robotic fabrication; Machine learning; Computer vision Abstract 3D Concrete Printing (3DCP) is a fast-paced process that requires multiple parameters to be accurately tuned in order to guarantee a high-quality print. Despite the technological advances in 3DCP, the control of this process still requires frequent manual intervention, which can lead to error, inconsistency under different executions and the reliance on human expertise to accommodate changes in the characteristics of the printing environment. In order to bypass these issues and approximate an autonomous self-correcting process, machine learning vision models have been applied, particularly in polymer extrusion, to extract and analyse information from captured images or video - colour and texture, geometric deviation, defect recognition, amongst others-and consequently introduce corrections to the print settings - motion path, speed, acceleration, material flow or temperature - to improve the print quality. In this paper, we first review related techniques, which include both real-time and offline correction approaches. We then present a comprehensive computer vision system for real-time control suited to the characteristics of robotic 3DCP. Within this scope, we focus on a particular ML component of this system-Speed Control -that manages layer width through direct access to robot motion speed or material flow rate. The proposed framework has three main components: (1) a data acquisition and processing pipeline for extracting printing parameters and build a synthetic training dataset, (2) a machine learning model for tuning parameters in real time, and (3) a depth camera mounted on a custom 3D-printed rotary mechanism for close-range monitoring of the printed layer shape. 1. INTRODUCTION Concrete offers very particular challenges within the context of 3D printing [1] . Its material properties make for a fast-paced process that requires frequent adjustments to the printing parameters during fabrication. This is especially applicable when producing architectural components that require a very high print resolution, much greater than what is usually necessary for on-site construction. Variables such as environmental conditions, mixing time, concrete batch changes and the geometry of the print path all affect the end quality. Our experience with robotic 3DCP has shown that a close monitoring of the extruder´s motion speed and flow rate is necessary to minimize the impact of such variables and achieve accurate and high-quality results. Both parameters primarily influence the shape of the printed layer and adjustment of its size. While higher values of robot motion speed decreases the layer width, the opposite is also true for the extrusion flow. However, varying either has its own distinct effects. Motion speed affects print duration, which should be minimized when printing concrete. Variations in extrusion flow rate cause changes to material behaviour, related to the kinetic energy provided to the mix by the rotation of the spindle, and higher values are also responsible for pump heating. Therefore, in the context of our setup, we have found that it is better to have a constant flow rate and adjust the motion speed instead. The present work on layer morphology control is part of a broader machine learning system for 3DCP currently being developed at ARENA – Digital Fabrication Lab at the E

FAQs

Why is real-time control important in 3DCP?

Concrete 3D printing requires frequent parameter adjustments during fabrication, since environmental conditions, mixing time, batch changes, and path geometry affect output quality. Real-time control reduces inconsistency and improves print accuracy.

How do robot motion speed and extrusion flow rate influence layer width?

Higher robot motion speed reduces layer width, while extrusion flow rate increases it. In the proposed setup, the flow rate is kept constant and speed is adjusted to achieve the target layer shape.

What components make up the proposed ML computer-vision framework?

The framework includes (1) a data acquisition and processing pipeline with a synthetic training dataset, (2) a machine learning model for tuning parameters in real time, and (3) a depth camera on a custom 3D-printed rotary mechanism for close-range monitoring of the printed layer shape.

Aug 5, 2026

Upload by Ava Thompson