1 Introduction

Coastal erosion is a major issue in coastline areas, caused by natural or human sources. Sediments are removed and dragged away by oceanic forces, such as waves, currents, and tides, which causes the coastline to retreat and land to be lost. In coastal urban areas, the swell and varying tides can cause damage or losses near buildings, so maritime protection structures are used to minimize such risks. The construction of breakwaters and seawalls has been done for centuries to break and weaken ocean waves, preventing coastal erosion and protecting coastal areas from the oceans. Wave energy dissipation is typically achieved by pilling thousands of large natural stone or concrete elements in a chaotic distribution to form voids or gaps. Fishermen, hikers, and tourists often use these constructions for their work or leisure, but they can be dangerous because of their irregular shapes and difficulty of access.

Building any type of structure over these breakwaters is a complex challenge. However, recent advancements in technology, such as computational design, drone photogrammetric surveys, and 3DCP manufacturing, are allowing the creation of complex forms, previously unimaginable with traditional techniques (Craveiro et al., 2019). Complexity and customization are achieved without the need for formwork, cutting labor costs and reducing material waste, a prime example of mass customization. Yet, two decades after the introduction of the technology is still hard to find this kind of Additive Manufacturing (AM) processes in real construction projects.

This paper proposes using 3DCP to create a system of platforms which can be overlaid on breakwaters or pontoons, providing leisure and social spaces, fishing spots, pathways with viewpoints or urban furniture (Fig. 1). Currently, photogrammetric technology enables the accurate digitization of complex geometries like those of the pontoons. While 3DCP can be leveraged to obtain a perfect fit to a complex topography, such that each segment replicates the digitalized geometry of its location, supporting itself and reducing the need for mechanical connections to the site. Furthermore, since the structure will not be permanently attached to the site, it can be partially retrieved where and if needed during the wintertime, when the sea is stronger, and put back in place during the summer months.

Fig. 1
figure 1

Scheme of occupation of the rocky jetties, with custom-fit prefabricated concrete components, which expand public use and marine biodiversity

Full size image

Another important issue to consider in interventions in this context is weighting the benefits and impacts in sea life, particularly focusing on how the intervention may improve the conditions for survival of species threatened by the rising water levels and temperatures in the oceans. Coastal areas, as spaces of transition between land and sea, are also affected. They are home to important marine and wetland species whose diversity must be conserved. However, in today's rushed world, this link between the constructed and natural environments is often ignored or broken. In such sense, we believe AM processes can offer solutions of added value regarding the creation of new micro-habitats. In our project, we configured the infill of our slabs with two main objectives in mind: increasing the strength of the prints to resist sea waves and creating complex shapes with various textures to emphasize the local marine life.

In Sect. 2, the contribution of the current research is compared with the state-of-the-art of similar efforts using 3DCP. In Sect. 3, our survey-to-production workflow is detailed. Section 4 presents the application of our workflow to the case-study of the pontoon of Póvoa de Varzim, detailing how each of the steps was implemented. The paper concludes with a discussion and the key findings on the application of 3DCP to complex sites.

2 State of the art

3DCP has been used before to create coastal protection structures. Experiments carried out by WinSun have tested the prototyping of an element similar to the "Core-Loc®" typology, by using 3DCP techniques for the production of an integrated formwork (Winsun, 2017). Nevertheless, we believe that the role of this technology does not match the scale and number of elements required for the formation of a breakwater structure. However, 3DCP has the potential to offer customized solutions that guarantee the resolution of specific problems.

Some examples of the use of technology for the manufacture of coral reefs are also known. One of them is the X-Reef project developed by the consortium of XtreeE (2017) and Seaboost (2017). XtreeE is a specialist in large scale 3D printing, while Seaboost is an expert in maintaining and creating marine habitats. These companies aimed to replicate natural coral formations, which take centuries to form. 3DCP was chosen for this undertaking due to its capability to craft intricate voids within the structure, bearing biomimetic shapes of intricate geometry and diverse curvatures. Such intricacies would prove impractical with conventional concrete fabrication methods. Another benefit is the cost of production, which is smaller than methods like formwork. Our approach differs in that we focus also on creating recreational spaces and public areas in hard-to-use places, instead of objects completely submerged underwater. However, we think that the space between layers of infill at the base of the modules can be also used to create colonies of microorganisms like algae or small crustaceans, as proposed by the X-Reef project.

Typically, when a designer wishes to construct something upon uneven terrains or rock formations, the conventional method is to flatten the area (Wibranek & Tessmann, 2019). This strategy will be costly and damaging to the environment, as the land cannot be restored. The SDU CREATE research group has shown the potential of 3DCP compared to conventional construction in similar cases with the Sense-ENV project. The project aimed to create a manufacturing process where the design is adapted instantly to the surface it will be applied to (Naboni et al., 2022). A depth sensor-equipped camera was used to track the printing surface, enabling the extrusion path to be adjusted to the surface's topography. The research presented in this paper proposes to use 3D scanning for automated custom-fit modeling of 3DCP prefabricated components for real contexts.

The use of 3DCP for prefabricating a system of parts is a widely researched strategy, which demands considering transport, handling, and onsite assembly requirements during the design phase (Mechtcherine et al., 2019). This involves the design of effective connections between elements and the calculation of the size of the discretized components according to the handling, transport, and assembly capacities of the in-situ structures.

The most common strategy is to subdivide the structures according to the normal of the printing plane of the parts. In structures with discontinuous supports, such as bridges, the subdivision is frequently combined with post-tension strategies (Salet et al., 2018; Vantyghem et al., 2020; Xu et al., 2020). Other approaches involve creating interlocking features between 3d printed parts to obtain structurally cohesive systems, i.e., the Cohesive Pavilion (Grasser et al., 2020) or the research by Shaker et al (2021). However, for the application presented in this paper we propose a hybrid strategy combining post-tension and interlocking between parts.

It was also interesting for this paper's framework to consider some of the work of Darmstadt Technical University's Digital Research Unit (DDU). This group acquires 3D data of irregular stones using photogrammetry and laser scanning, then creates digital twins of the rocks to integrate them in a project of assembly (Wibranek & Tessmann, 2019). This methodology is similar to the project in this paper, the main difference being the scale of production and the new uses proposed.

3 Survey-to-production workflow

Our workflow has six parts, as illustrated in Fig. 2. First, a photogrammetric survey of the intervention area is conducted. The second step is the post-processing of the 3D model. Design issues are addressed in step three. A parametric workflow (in Grasshopper) was used to generate solutions based on manufacturing rules (constraints such as maximum inclination angles) and parameters set by the user (such as desired function). Step four deals with pre-production for 3DCP, comprising the generation of the print path for each individual slab and the addition of its internal structure. Then, the fifth step comprises robotic manufacturing in a laboratory environment. We used a Kuka KR120 robotic arm as manipulator of 3DCP process. Finally, the transport and post-tensioning in-situ to complete the final aggregation. However, the intervention's cycle will only be finished when humans use and adapt it regularly, and local fauna and flora create new micro-habitats inside.

Fig. 2
figure 2

Scheme of the working methodology

Full size image

4 Póvoa de Varzim pontoon case study

Following a comparison of urban seafronts along the northern coast of Portugal, the lighthouse pontoon of Póvoa de Varzim was chosen as a case study. Its proximity to urban areas, tourist activity and fishing heritage made it the ideal choice (Fig. 3). A specific intervention area was proposed based on fishing activities observed on-site, the uneven terrain, and the risk it represented in terms of access.

Fig. 3
figure 3

Localization of the case study – Póvoa de Varzim lighthouse pontoon, Portugal

Full size image

4.1 Digitization

Laser, photogrammetry, and structured light scanning techniques were considered to capture the intervention area of the pontoon. To produce a suitable context mesh, photogrammetry was chosen due to its high resolution, feasibility in outdoor environments, and cost-effective equipment and post-processing. A single person surveyed the area with a mid-range digital camera, taking about 120 pictures (Fig. 4). Each image is taken by moving around the selected area of interest in circular paths, varying the distance from the rocks and the camera's angle. It is important to avoid self-shadowing and wet conditions that create reflective surfaces, as these create defective artifacts in the final mesh. After evaluating available photogrammetry software such as RealityCapture, Meshroom and KIRI Engine, we got the fastest and best quality reconstruction through PhotoCatch, computing in 30 min for an area of 3.5m2.

Fig. 4
figure 4

Example of four pictures used for site photogrammetry (left) and generated mesh (right)

Full size image

The resulting mesh was then imported into MeshLab (Cignoni et al., 2008) to clean and simplify high-poly meshes, which is necessary for the efficiency of the following design steps. We used a quadric edge collapse decimation algorithm to reduce the mesh size by 40%, with a deviation of less than 3 mm. This adjustment ensured that the degree of similarity to the original rocks necessary for a custom-fit print remained uncompromised.

4.2 Parametric design

A parametric design system automates the generation of optimal solutions for the slab modules. The system can be applied to any section of the pontoon intervention area (Fig. 5). It considers the delimitation of an area of interest and overlaps a rationalized grid where each cell represents one module of the discretized platform. Different topological grids were studied to find the best configuration that balances local support, a good interlocking behavior and a convenient geometrical outcome for fabrication in 3DCP.

Fig. 5
figure 5

Generative schema of a platform

Full size image

A genetic algorithm is used to arrange an 80 × 80 cm rectangular grid (available print area and weight limits affect the size) in place so that modules have local support from the rocks below (Fig. 5a). The vertical edges are then broken in the middle point to create an interlocking effect amongst neighbors (Fig. 5b). Finally, each strip is subdivided by the same type of algorithm to ensure the slope is controlled, thus guaranteeing the fabrication of each piece (Fig. 5c). This produces pieces of different sizes that do not exceed maximum dimensions allowed by our printing process.

The final grid informs the perimeter of each slab. To obtain the surface model, we project points from each module along the z-axis with 1 cm resolution onto the surveyed mesh. Then we use Delaunay triangulation and conservative quadratic remeshing to polish off imperfections. The projection of the points was limited to a depth of 1 m to guarantee the necessary fit and support of the structure, while avoiding its excessive extension into the voids between the rocks. Finally, this mesh is extruded to reach the intended height onsite, according to its functional role at the top. This process was chosen instead of a straightforward boolean operation between the extrusion and the surveyed mesh to avoid the custom-fit mesh becoming tangled, with holes and unprintable.

Figure 6 shows a part of the designed platform. Each element is unique, serving a specific purpose in the intervention, satisfying four objectives: (1) conformity to the geometry of the supporting rocks that latches it in place; (2) interlock with neighboring modules; (3) meet a functional necessity between horizontal circulation, vertical circulation, and resting; (4) create a habitat for local fauna by variations on the infill pattern.

Fig. 6
figure 6

3D simulation of part of the platform to be produced

Full size image

4.3 Slicing

A pre-production phase is needed to develop an effective production strategy for robotic manufacturing. Due to the complex and unique geometry of each slab contact surface with the rocks of the pontoon, this phase involves developing a custom slicer with unique infill patterns. Additionally it entails deciding the piece printing position/orientation. After some preliminary tests, we defined that the parts should be printed laterally relative to their position in-situ.

The purpose of the custom slicer is to process each module and extract the toolpath that guides the robotic arm equipped with the extruder. Initial inputs are the parts to be produced and a boolean value per part determining whether the part is reversed to facilitate production. This decision is based on an initial assessment of the surface slopes and the need for supports. The algorithm first considers horizontal planes spaced by the height of the printed layer (10 mm). It then calculates intersections between each plane and the geometry of the slab, resulting in a collection of closed curves that define its exterior boundary. These curves are then offset by half the print layer thickness to ensure a correct contact with the rocks.

The next step in slicing is creating infill that can adapt to the three function typologies: (1) stairs, (2) circulation and (3) bench. The infill adopts a honeycomb-like configuration for structural performance, as well as to create pocket-like habitats. Its final design is automatically readjusted to accommodate the various functions and multiple section dimensions along the platform. In the prototype component there are three variants: the final step of a stair (Fig. 7a), the first step of a stair or bench (Fig. 7b), and a flat top surface (Fig. 7c).

Fig. 7
figure 7

Different types of infill for reinforcing printed components and to ensure support for the upper layers

Full size image

Infill A is composed of three stacked chambers of different heights. Their boundaries are aligned horizontally to guarantee continuity when overlapping the other infill types. Duplicate vertical lines are used to increase strength and guarantee a continuous printing path with aligned seams between layers. These are interrupted along their length to create voids for the steel cable post-tensioning system aggregating the modules.

4.4 Fabrication

Our 3DCP robotic printing setup comprises three sequential phases (see Fig. 8): (1) producing/mixing the material; (2) pumping it into the extrusion head; and (3) depositing the material in layers, with continuous filaments, according to a specific printing path.

Fig. 8
figure 8

3DCP setup at the ARENA Robotic Fabrication Lab in the School of Architecture, Art and Design of the University of Minho, Guimarães, Portugal

Full size image

The mixing was carried out in a planetary mixer in batches of 2 or 3 bags of Weber 160–1 mortar, a dry cement mixture specifically formulated for 3DCP processes, adding 10.8% water. Better material behavior was achieved by ensuring a minimum mixing time of 5 min.

Preliminary tests showed that for a print configuration with a 20 mm extrusion nozzle and a layer height of 10 mm, a layer width of 40 mm should be guaranteed throughout the printing process. To achieve this, a standard print speed of 100 mm/s was maintained, and the pump's extrusion flow rate was continuously adjusted to adapt the workability of the material to the defined layer width.

Prototyping showed that in a fabrication setup without added support, partial collapse may occur when the slope of the vertical face is too steep. Similar experiments of printing highly complex geometries have reported favorable results using sand as support material (Ahmed et al., 2022). Thus, we made some local support tests in the areas of greater instability, adding sand after the deposition of concrete and removing it after the start of the curing process. This technique led to a great increase in the success rate of the fabrication and minimized the geometric deviation to the digital twin.

Figure 9 on right side, shows a deviation analysis performed on a photogrammetric model from a fabrication test done inside a sand enclosure. It shows that the greatest deviations (yellow) happen when the curvature moves inwards, where there is no sand to support it. However, deviation in convex curvature is minimal.

Fig. 9
figure 9

Section print test of one of the prototypes, using sand as support material (left) and analysis of the photogrammetry model obtained with the target geometry (right)

Full size image

This principle was then adopted in the production of the components, guaranteeing bearing capacity for the surfaces in contact with the rocks. Figure 10 illustrates the production of one of the parts.

Fig.10
figure 10

Printing process of a full-scale prototype (left) and post-tensioning aggregation scheme of two printed sections (right)

Full size image

4.5 Offsite testing

To validate the assembly process of the slabs and the envisioned post-tensioning system, the site topography was reproduced in our lab. This artificial topography was obtained through a milling process of six expanded insulation cork blocks (1000 × 500x320mm each) performed with the KUKA KR120 robot equipped with an industrial spindle (Fig. 11).

Fig. 11
figure 11

Roughing operation of expanded insulation cork blocks

Full size image

The milling was accomplished in two steps: (1) a roughing operation with a 12 mm flat end mill using 8 mm stepover and 20 mm stepdown, and (2) a finishing operation with a 12 mm ball noose mill. The roughing operation was setup in Fusion 360 software, the paths were generated and imported into Grasshopper with the Kuka|PRC plugin, where the process was simulated and the KRL code generated for production. The roughing operation took around 12 h.

The finishing operation was setup in Grasshopper using parallel passages with 6 mm stepover in both orthogonal directions of the surface. The cutter paths were generated from the interpolation of points on the surface with a distance of 20 mm. For a better finish, the tool orientation was obtained from the surface normal at each of the interpolated points. Figure 12 shows the artificial topography that resulted of the milling operation.

Fig. 12
figure 12

Cork artificial topography

Full size image

The model was surveyed using photogrammetry to determine the accuracy of the artificial topography vis-à-vis the original landscape and the flattened landscape by the parametric generation. 141 photos were taken with a Nikon Coolpix A camera which were then processed with 3Df Zephyr to obtain a dense cloud. This cloud was aligned with the site survey mesh and the flattened mesh using the ICP algorithm with 20% overlap. The first pair aligned with final RMS of 1,20,646 on 49,999 points (Fig. 13 Left). The max distance in the flattened areas is 294,48 mm. Considering a 10 mm threshold for contact areas shows a mean distance of 5,535 mm with a standard deviation of 5.434 mm. The second pair aligned with final RMS of 0,234,418 on 49,999 points. Computed cloud to mesh mean distance was 0,608357 mm with a standard deviation of 2,418684 mm (Fig. 13 Left).

Fig. 13
figure 13

(Left) Artificial landscape to site survey distances with overlap of the platform (Right) Distances from the artificial landscape to flattened mesh model from the parametric workflow

Full size image

The last step of the offsite testing involved positioning the six printed components on the previously milled cork artificial topography. Transport and positioning of the printed components was done with a forklift due to their weighs, ranging between 150 and 300 kg. A rubber strip was placed on component-to-component interfaces to absorb any irregularities and prevent direct contact. After all the elements were positioned, the steel cables were pushed through the ducts and tensioned using a dynamometric wrench (Fig. 14).

Fig. 14
figure 14

Assembled 3d printed components with post-tension installed over the cork artificial topography

Full size image

5 Discussion and conclusion

In this paper we propose a survey-to-production workflow for producing 3DCP modular components for seashores with complex geometries. The workflow was implemented and tested in a case-study. We believe our workflow has major benefits when dealing with sites of complex topography. It provides a non-destructive and cost-effective way to intervene, while taking advantage of parametric systems to customize and modify the design to a wide range of contexts and goals. Furthermore, 3DCP is the only efficient way to produce prefabricated components with the scale and level of customization required to assure an adequate fit to the complex geometries of irregular rocky pontoons.

The selected surveying methods proved sufficiently precise for this application. Assembling the structure over the replica of the onsite topography demonstrated the feasibility of using 3DCP to produce custom-fit structures for this type of complex and natural features. This result is a contribution to the overall research context of robotic 3D printing strategies and 3DCP applications.

However, the current approach has a few limitations which are obstacles to full-scale manufacturing and onsite performance. The size of our components is limited by their weight and the diminishing capacity of sand to act as a support as the height of the modules increases. This might have damaging consequences to the performance in-situ, given that an increase in weight would strengthen the resistance to waves. While the post-tensioning system can effectively aggregate the modules, its onsite performance needs to be validated. On the other hand, smaller modules are more easily deployable and removeable onsite. Yet, the capability of the structure to withstand winter storm conditions should be monitored and studied.

A more in-depth evaluation will be carried once it becomes possible to install the modules onsite. Future work should focus on evaluating the aspects related with the system capacity to sustain biodiversity and form habitats for wetland species, and monitoring the stability of the structure. Specifically, it must be determined the links between the capacity to sustain biodiversity and the position and form of the structures in breakwaters. This will allow fully determining which modules may be relevant for seasonal removal. Further work can also be done on expanding the range in functional uses of the slabs, as well as experimenting with variations in the infill patterns to offer other types of habitats for marine life.