Developing a fabrication workflow for irregular sawlogs

Context of research and material

The research described in this article strongly builds on a growing agenda within digital architectural research – in-depth wood understanding combined with digital tools, design methods and fabrication. The inspiration and knowledge for this particular research expand on admiration and understanding of former and ongoing research and seek to contribute to new perspectives in this evolving research context.

Projects like the Ratatosk Pavilion by Helen Hard ¹ that engages the natural tree with three-dimensional (3D) scanning and digital fabrication, the wood fibre behaviour explorations conducted by the ICD/ITKE research through the HygroScope project, ² the following HygroSkin pavilion, ³ the recent Urbach Tower, ⁴ as well as the work performed at Architectural Association Hooke Park ⁵ are examples of state-of-the-art research that has directly influenced this project.

As a part of the preparation for this research, Hooke Park was visited to learn and understand from their expertise. Likewise, the research team travelled through Finland and visited all parts of the Finish wood industry from harvest to sawmill, engineered wood plants, large-scale digital fabrication facilities and pre-build wood component–based housing factories.

As the research context suggests, the point of origin for this project is an interest in wood on a material behavioural level combined with an ambition to utilise this knowledge in more component-scaled fabrication prototypes. Therefore, the research initiated parallel trajectories of investigations to explore both the interiority of wood (see Figure 1) and development of digital fabrication workflows for machining. Digital machining methods, metrology and scanning techniques were used for a series of preliminary explorative studies (see Figures 1–4). The length, directionality, strength and elasticity of the grains are particular to each wood species. By cutting, splitting, bending, sawing, milling, scanning and tomography, each initial experiment pays attention to material and processing in a different way. These studies act as both hands-on material experiments and ways to adapt and acquire further existing knowledge. Books like ‘Understanding Wood’, ⁶ ‘Wood and Wood Joints’ ⁷ and ‘Roar Ege’ ⁸ mixed with professional assistance from cabinet makers and researchers from DTU Imaging and Niels Bohr Institute helped inform and conduct these early experiments. More architectural research from especially the books ‘Advancing Wood Architecture’, ⁹ ‘Advanced Timber Structures’ ¹⁰ and ‘Digital Wood Design’ ¹¹ has informed and framed the more component-oriented studies and helped to establish various robotic machining processes. These machining processes partners with bespoke digital analysis and modelling tools to perform introductory explorations that later would inform more focused experimentation (see Figures 2–4).

OPEN IN VIEWER

Figure 1.

Top: Macro photos showing internal patterns of pine, beech and ash. Mid: X-ray scanning of spruce clearly shows the orientation of grain structure and the location of internal knots (done in collaboration with Niels Bohr Institute). Bottom: CT scans are used to construct voxel and mesh representations of the internal wood structure (scanning is done in collaboration with DTU Imaging).

OPEN IN VIEWER

Figure 2.

Different machining techniques are tested out both to establish digital processing workflows and to explore material behaviour.

OPEN IN VIEWER

Figure 3.

Logs processed with a milling tool mounted on a six-axis robot, demonstrating the exact coherence between the geometry of the scanned logs and the milling tool paths. Left: A 3D model with isocurves. Right: Real log with milled isocurves.

OPEN IN VIEWER

Figure 4.

Crooked oak log cut into curved boards that follow grains direction.

Crooked sawlogs in constructions

Besides its practical and tactile properties, wood in construction has a positive impact on the carbon dioxide emission level enabling carbon sequestration and replacing building materials with higher levels of carbon dioxide emission.^12,13 In other words, swapping steel and concrete with wood in construction will generally improve the environmental cost of buildings.

In today’s wood industry, actions are taken to optimise tree growth, sawing techniques and production strategies in order to supply the construction sector with standardised products. The industry is highly developed in terms of using advanced technology, such as x-ray scanning, 3D analysis, exterior scanning and customised sawing to optimise timber yield (see Figure 5). The research team has experienced this hands-on in Finland. The scanning methods are used for sorting the logs when they first enter the sawmill, for optimising the division of each log into different types of timber and for quality control. While these methods are in indeed highly effective, they still fall short. Strongly crooked timber is not used in today’s sawmill, and while slightly curved wood can be cut and processed into engineered wood products, this comes at the cost of vast and energy consumption. Furthermore, the processing of trees into standardised shapes actually devalues the strength since cutting wood angled to grain direction will break the fibre continuity. ⁶

OPEN IN VIEWER

Figure 5.

Advanced scanning technology is an integrated part of the wood industry. From left to right: Intelligent, semi-automated harvesting, 3D x-ray scanning and computerised quality analysis.

While the above description applies to the current industry, greater attention to natural trees was paid in earlier times. The Vikings refined their shipbuilding through understanding and processing of natural wood and transferred that knowledge directly into their building.^8,14 The use of specific types of tree growths like curves, twist or branches became an important element in shipbuilding in general. ¹⁵ Likewise, several construction types have benefitted from curved trees. This is seen both in the cruck frame buildings where split curved logs are used to define the special volumes of the construction ¹⁶ and in the Haubarg constructions where crooked logs are used in combination with straight logs to define the interior of the farm buildings.^17,18

Today’s fact is that some high-quality timber ends up as firewood due to its non-standard shape and, in general, little attention is paid to the potentials of the individual tree. The sawmills use interior scanning (see Figure 5, second photo) on every single log before any processing is performed. Consequently, the sawmills have digital representations of both the straight trees and the curved discarded logs. The first is used for optimisation of sawing patterns and the second discarded together with the log.

In recognition of this fact, the research discussed in this article grows from the following hypothesis: The natural forms of crooked sawlogs can be handled, processed and utilised in construction if computational tools, digital fabrication and existing knowledge from the industry are explored in an interconnected workflow. The benefit of such a workflow is of both sustainable and economic relevance because better utilisation of grown trees would improve both carbon footprint and profit.

The core of the research thereby builds on investigations of workflows that tie together the non-uniform reality of the logs with precise digital machining. To explore this interconnected handling of digital data and physical material, a testing scenario is created: By using procedures for scanning and analysis, similar to those used in modern sawmills on discarded crooked timber, the naturally grown shape of the material itself is used to inform machining. These shapes are organised parametrically into constructions, thereby creating a workflow where the shape and properties of the natural tree inform its machining. For the testing scenario, a system based on lamella roof construction has been chosen. Previous research on lamella roof constructions demonstrates workflows that integrate computational design and robotic fabrication. ¹⁹ However, these projects make use of standardised timber rather than wood in its natural shape. In this testing scenario, only exterior scanning is utilised because the research team do not have regular access to the computed tomographic (CT) scan and x-ray equipment used in the preliminary tests. It is, however, a future ambition to combine the developed workflow with interior scans to better accommodate internal cracks and fibre directionality through the developed digital tools and processing methods.

Workflow

It is the goal to establish completely integrated design-to-production workflows through the use of bespoke computational design tools and digital technology for analysis and fabrication. The method described here makes use of a Faro laser scanner, positioning with an OptiTrack system, analysis and form generation based on Rhino/Grasshopper with Python and Volvox, and machining is done with an ABB six-axis robot arm mounted with, at different times, a band saw and a spindle. The current workflow for machining crooked saw logs is based on a combination of local resource arrangements and available fabrication facilities. The workflow, however, seeks to include scalability and elements that could be relevant in a more industrial context, including future use of interior scannings instead of exterior scannings.

The sawlogs used for this experiment is sourced from a local sawmill. The logs are branches and trunks from oak trees that are too crooked for use at the sawmill. Straight oak is being used for furniture or flooring, and semi-curved pieces are being utilised in making of more rustic furniture and other arts and craft interiors. The remaining material, which is too crooked, irregular or inconvenient to process at the sawmill, is sold cheaply as firewood. That is the wood we utilise in this project. The handling of the oak logs can be described as being in three phases: stockpile, design and fabrication.

Stockpile

The oak logs are picked up at the sawmill when the stock has reached a suitable amount (see Figure 6). The logs are then cut up in lengths appropriate for the subsequent machining. This step is necessary for this specific setup but could be avoided with larger fabrication facilities. The logs are laid out systematically, labelled for later referencing and tagged each with three red reference sticks in predrilled holes. The total array of oak logs is then scanned to get a digital representation of the stockpile. The scanning is done near our fabrications facilities but could just as well take place at the sawmill. This option would be convenient when sourcing logs with specific properties and managing data from several sawmills. The digitisation of logs has been tested using different types of equipment including Faro Arm laser scanner, photogrammetry and Faro Focus lidar scanner. For larger quantities, the Focus lidar solution has proven most efficient (see Figure 7). However, scan data could also be obtained using, for example, drones or existing exterior/interior scanning solutions in the industry.

OPEN IN VIEWER

Figure 6.

A pile of discarded crooked oak logs is inspected and picked up at the sawmill.

OPEN IN VIEWER

Figure 7.

Logs are scanned near the fabrication facilities. A combination of lidar scanning and reference sticks enables scanning of larger quantities of logs to be scanned directly on the ground without further preparation.

The point cloud is processed into individual point clouds for each log. The reference sticks in each log are identified using an RGB filter and extracted as features. Each point cloud is analysed, and a lightweight NURBS representation of the log is created (see Figure 8). The analysis is a combination of Grasshopper components mainly build around components from the Volvox-plugin to handle import, sampling and sectioning of the point clouds and as a series of custom Python scripts. The custom scripts are specifically made for analysing logs. One script finds the centre curve of each log by iterating through evenly distributed sections of the point cloud. Another script adjusts the end sections to match the logs end planes defined during the scanning process. A third script evaluates each point cloud section to detect an accurate and distributed edge description, and finally, a fourth script creates a closed NURBS polysurface that is very lightweight and low in resolution, but sufficiently accurate due to the precise selection and evaluation of the scan data. These custom tools are made in-house and highly specific for this workflow. They are, however, shared with partners and fellow researchers at other institutions and now used in other wood research projects.

OPEN IN VIEWER

Figure 8.

Scanned data are transformed into NURBS geometry, which functions as a lightweight representation of the digital stockpile. The translation is done using a custom script optimised for crooked log: The overall shape of the log is precisely described using simply geometry, without paying attention to smaller surface variations.

A combined data set of point cloud, NURBS polysurface, reference points and labelling is created for each log. The series of data sets now becomes a digital stockpile of available logs. This accumulated digital representation can now be used for a generative design process where the dimensions and curvatures of the digitised logs are considered in a computational negotiation between overall design and shape of the individual building components.

Design

The physical and digital stockpile functions as an amassing of resources that can dynamically shrink or grow. At any time, the stockpile can be reached, and available resources assigned to a design. At the moment, the stockpile is used for several experiments aiming at exploring the material for architectural elements. This article focuses on the design workflow for a lamella roof construction, but the workflow for both design and fabrication could be applied to other types of constructions that involve other shapes and principles. The lamella roof construction is chosen as a test scenario for two main reasons. First, a lamella roof structure could benefit from having curved members to avoid over-shooting situations in the joints, especially in relation to high curvatures. Second, the lamella roof construction consists of multiple smaller elements, which fits very well with the available robotic setup and the possible reach during machining.

A core agenda of the research is grounded in the natural properties of wood. This includes both the internal structural principles of wood and the outer shapes. The specific experiment seeks at maintaining unbroken fibre directionality and curviness of the logs. This maintains both natural strength,^6,15 and mode of expression. The arrangement of the logs into a larger geometrical constellation must, therefore, be a dialogue between the available stock material and architectural design intent. This objective is approached using a design method where a provided geometrical solution is modified in negotiation with the saw logs – and the sawlogs are processed to perform with the designed geometry (see Figure 9).

OPEN IN VIEWER

Figure 9.

Diagram showing how logs are positioned according to their curvature in the topological relation. An algorithmic dialogue between overall and component geometry is established, allowing both parts to influence the overall design and the precise processing of the components.

The negotiations between available oak logs and overall design follow a sequence (see Figure 10): A reference NURBS surface is defined, and a series of isocurves matching the resolution of the lamella construction systems are extracted. The isocurvature of each individual segment is analysed, and information of curviness, orientations and tangencies at endpoints is obtained. This information is then matched with corresponding parameters from the stockpile. Using the best match of dimension, curviness and surface warp, the saw logs are placed on the control surface. The custom-made algorithm seeks to find a way to distribute the wood elements, so they meet the intended form, as designed by the architect, and allows this shape to be modified to a certain degree. The length and ends of the stock logs are adjusted to follow the surface, but their natural curviness is preserved when placing their inherent centreline at the position of the isocurve segment. The log is rotated to match the surface geometry as close as possible. Based on the centreline of the logs and the shape of the main surface at the log end positions, a ruled surface is generated. The ruled surface informs the saw path that divides the logs into halves. The two halves are positioned simultaneously in the overall topology, thereby helping the material efficiency of the system. This serves as a negotiating action where both natural appearances of the logs are maintained on one side, and a limited adaptation to the designed surface is accommodated on the other side. Because the original centreline is maintained, the logs preserve original fibre directionality through the design phase. Furthermore, the computational design method includes the placement and positioning of joints between the wood members. It is important to note that the deviation between the original design surface and the outcome is a combination of several factors. First and foremost, the size of the stockpile plays a role. The larger the stockpile, the better the match. Second, the original curvature of each individual logs is sought to be maintained, for both aesthetic and strength reasons. At the same time, joints need to perform together with other logs. This means that the algorithm will prioritise original log geometry where the log is floating freely, but force a shape closer to the input surface in and near joint situations. The deviation will therefore vary at different places in the geometry, simply because the final shape is a negotiation between the local log geometry and global input surface.

OPEN IN VIEWER

Figure 10.

Diagrammatic illustration of construction in progress; stockpile is matched with surface geometry. The visualisation shows a digital representation of logs scanned in Figure 6. Actual, fabricated subelement of generated construction can be seen in Figure 13.

Since often multiple solutions are possible, a single unique output is not provided. Instead, various assemblies of selected parts are generated, and the final choice is entrusted to the architect. This ensures that the design becomes a result of a sensitive professional process that truly mediates between the materials at hand and the design intent.

Fabrication

When the distribution of logs in the structure has been defined, visual representation and a list of required material from the stockpile are provided. The realisation of the designs explored in this project consists of three steps: sawing, joint milling and assembly. All three require different types of matching of digital data and physical material.

The first step is the splitting or trimming of the logs. The logs are split along or parallel their centreline, with a rotating cut that meets the surface geometry as close as possible. Here, the sawing is done with a custom build band saw mounted on a six-axis industrial robotic arm (see Figure 11). The setup provides high freedom of cutting, precise control of the cut surface and is deeply integrated with the design software. The toolpath and robot code are generated directly from the digital geometry and pairs uniquely with each saw log – both digital and real. Before the robot code is executed, the real oak log needs to be positioned in a corresponding position that also respects the robot setup’s machining limits. This is done using the three reference features extracted from the associated data set. The robot cell is equipped with an OptiTrack motion capture setup that is used to locate the orientation of the logs by probing the three reference sticks (see Figure 12). The tool paths are then reoriented using the corresponding digital features before the log is sawed using the band saw mounted on the robotic arm.

OPEN IN VIEWER

Figure 11.

A band saw on a six-axis robot arm helps to produce curved pieces of a crooked log.

OPEN IN VIEWER

Figure 12.

The orientation of the saw log is located using an OptiTrack motion capture system. Three reference points are probed using a custom probe tool. Tool paths are reoriented to match the placement of the saw log.

The sawing is followed by a joint milling process. Each half log is machined to have half-lap joint details in the ends and on the sides (see Figures 13 and 14). The orientation and position of the connections are created in accordance with the chosen design. Again, the workpiece is placed in the robot cell, and this time, each corner of the half log is probed. The points are used for the location and orientation of the toolpath needed for the joint milling. When the matching of physical half-log and its digital representation is achieved, the joint connections are milled using the robotic setup. The tool paths for milling the joints are, as with the sawing, automatically generated from the digital representation of the log. Assembly is done at the fabrication facilities in this setup (see, however, the parts could also be transported to another location for assembly and montage.

OPEN IN VIEWER

Figure 13.

Joining of the curved half-logs is done using a milled joint aligned carefully on the surface defining the overall design.

OPEN IN VIEWER

Figure 14.

Two curved half-logs joined into a lamella subelement. Note the different orientations of the end joints.

Future industrial implementation

The workflow explained in this article is developed and refined in a lab environment. The stockpile of scanned oak logs is limited to a smaller portion, and many logs are used for basic experimentation and equipment calibration, leaving only a reduced amount of scanned wood to be available for the actual surface matching. The scenario is not ideal but exemplifies the scope of the research. If the research is to be seen in a larger context and in a more industrial framework, the potentials of upscaling must be addressed. In order to approach industry directly, the research team of this project visited the Finish wood industry and experienced most of the processes behind the scenes. Therefore, most of the steps and methods used in the research workflow are either almost identical or strongly resembles methods actively used in today’s sawmill and wood processing industry.

Modern sawmills, for example, make use of advanced laser and x-ray scanning for each individual log. The scanned data are then used to categorise the logs and plan efficient sawing methods or to discard logs that do not meet the standards, for example, if they are too crooked. This also means that each log, including each discarded log, has a 3D surface, and often also interior scan data, attached. An obvious potential here is to pass the log data into a shared cloud-based database and thereby easily combine discarded wood resources from multiple sawmills. Matching algorithms, like the ones created for this project, could then make use of thousands of logs to find the best solutions. Information like species, age and location would then also be available, and the selection of logs could be limited or expanded to include local or regional logs depending on the type of construction. Likewise, a sourcing optimisation could ensure that logs are selected from either a limited number of or only neighbouring sawmills.

A reference platform could be the harvestmap.org ²⁰ project that is based on re-distributing surplus or redundant building materials. Here, architects, builders and others can source cheap leftover materials from industry, construction sites or redundant buildings and let those materials define or influence a new building. While a platform like this is not planned development work within this research project, the intention of the research is pointing in this direction. The intermingling of physical and digital resources has reached a level today where it often does not make any sense to separate the two. A quality of a material is no longer just defined by the material properties, but just a much through its ability to be handled, specified and processed within a digital environment. On-demand and customised manufacturing is highly reliant on informed workflows where architectural specifications and material properties and availability are always connected. In such workflows, intelligently optimised and sustainable architectural solutions, like the ones suggested in the article, could exist.

Conclusion

This article and its associated trajectory of experimentation suggest a flexible and scalable approach to include natural irregular geometry in a digital design-to-production workflow. Through the use of scanning, bespoke computational design tools and robotic fabrication, the irregular shape of natural wood is embedded in the realisation of a geometrically complex structure. This type of design mostly entails mass-customisation in the sense that each component has to be manufactured individually, where this project seeks to benefit from the fact that the material from the beginning has large geometric variation. Knowledge and acknowledgement of material characteristics can, in this way inform the resulting architecture, enriching the physical environment with some of the variation and tactile qualities found in nature.

The suggested workflow includes several approaches where natural physical appearance and digital representation, or information needs to pair and cooperate. The workflow aims at minimising resource demanding processes and instead, when possible, relies on simple matching of physical and digital references. Initially, a full point cloud representation of each log is produced, and, for establishing a versatile digital stockpile, these heavyweight pieces of information are reduced to rationalised geometry and a minimal amount of data. Besides, allowing the development of this project more efficiently, this approach also points towards upscaling in an industrial context. While industry today does utilise computational power, data abundance can easily be a bottleneck in a high-volume processing plant.

While the experimentation itself does not solve a resource problem, it points towards an issue in today’s material and building industry; economic optimisation of industry workflows entails standardisation of materials. The current material industry does not accept irregular raw material inputs. The research in this article uses current technology to up-cycle rejected material. The methods presented enable a close coherence between material, design intent, fabrication process and the realised result. A large part of the material, as found in natural form, is used in the construction, thereby optimising the carbon sequestration. As the industry already is utilising advanced technology, like intelligent harvesting, x-ray scanning, sawing optimisation and so on, the proposed hardware setup is within the realm of possible implementation at the industry level. Most of the in-lab workflows correspond with industrial-scale solutions already in use, though currently only with automation, efficiency and high-volume production as a result.

Instead of using standardised materials for non-standard designs, this article promotes matching of occurring natural shape and architecturally designed geometry as a way to optimise material use and induce rationality to more freely formed architecture. The added time, attention and knowledge required to utilise these diverging materials are not considered an obstacle, but instead regarded a direct way to obtain constructions with freeform appearances.

The architectural potential of the experimentation is still to be determined through more extensive and more complex fabrications. Already at this stage of the research, the proposed inclusion of natural wood in architectural elements is promising on a workflow development basis where varying natural shapes seamlessly blend with computational design and digital fabrication routines. The custom-made digital tools, scripts and algorithms presented here are at the time of writing being further developed into a coherent system, but already show robustness and originality in terms of demonstrating this promising workflow. Work is currently in progress to explore elements and assemblies on an architectural scale (see Figure 15), based on revised tools and a much larger stockpile.

OPEN IN VIEWER

Figure 15.

Screenshot of computationally generated lamella roof structure sketch for exploration of architectural potential. The logs are automatically selected and oriented where they fit best in the structure. The logs are trimmed parallel to their centreline, while the angle of the cut follows the curvature of the overall surface geometry. In this example, the cut surfaces of the logs are visible from the sides and the rough bark pointing down and up to achieve an interesting and coherent spatial aesthetics. A subelement of a similar roof system is planned for physical production.