Upcycling CLT production waste: a computational design and fabrication framework and a proof-of-concept demonstrator

Increasing the use of wood as a construction material and implementing circular economy principles are possible ways of reducing the environmental impact of the building construction sector. However, with an increasing use of timber and mass timber products in buildings, the amount of wood production waste is increasing as well. Wood production waste is typically downcycled and not considered as a resource for construction. This is particularly questionable in the case of Cross-Laminated Timber (CLT). Compared to other types of engineered wood products, its production requires large amounts of raw material and results in large amounts of waste. In contrast to downcycling, upcycling CLT production waste bears the promise of preserving or even increasing the material’s value, which makes it economically more compelling. In this paper, we present a prototypical computational design and digital fabrication workflow that enables the use of CLT production waste in prefabricated assemblies, potentially to be used as wall or façade elements, ideally with modular qualities. We elaborate on how the application of this workflow enabled the design and construction of a full-scale demonstrator, which serves as a proof-of-concept, and share insights gained during and after its construction. Finally, the paper provides an overview of possible applications and discusses the adjustability of the workflows for other design principles and construction systems.

Reducing the environmental impact of the building construction sector has become one of the prime objectives for today’s practitioners and researchers in the Architecture, Engineering, and Construction (AEC) sector. This development is spurred by an intensifying climate crisis and an increasing sense of responsibility of architectural designers and engineers. According to the International Energy Agency and the United Nations Environment Programme (United Nations Environment Programme, 2022), the global share of buildings and constructions regarding the final energy demand is 34%. The report also shows that construction, renovation and demolition approximately cause 100 billion tons of waste globally, with about 35% sent to landfills. Moreover, the report states that construction materials, such as concrete, steel, aluminum, glass and bricks, are estimated to represent around 9% of overall energy-related CO 2 emissions. Implementing Circular Economy (CE) principles (Stahel, 2016) in the building construction sector and turning buildings into carbon sinks by increasing the use of wood-based building materials, are possible strategies for addressing the above-stated issues.

One approach that recently emerged in the context of circular construction is to work with material stocks, sourced from building deconstruction or from production waste. Typically, these materials end up either as landfill or, in the case of wood-based materials, are grinded down and/or burnt. When looking at wood production waste, Cross-Laminated Timber (CLT) is of particular interest. Compared to other Engineered Wood Products (EWP), its manufacturing process is less resource efficient, meaning that it results in a relatively large amount of wood waste (Pramreiter et al., 2023). Moreover, inserting openings for doors and windows in CLT panels results in 5%–10% percent of waste (Ottenhaus et al., 2023) in the form of cut-outs and offcuts. This is concerning, especially when considering the cost of CLT, which can vary depending on the surface quality, the thickness, the number of layers, and the country of purchase. For example, 1 m 2 of three-layered CLT with a nominal material thickness of 90 mm costs approximately EUR 82.50 in Austria (Zimmereibedarf Lagerliste, 2019). A comparable product would cost CHF 95 per m 2, or about EUR 100.75, in Switzerland (Plattenwerkstoffe Preisliste, 2023). Another aspect supporting the upcycling of CLT production waste is the carbon storage and emissions related to wood and CLT production processes. As stated in (Younis and Dodoo, 2022), trees sequester carbon, meaning they “[…] absorb and store atmospheric CO 2 as carbon in wood fibers–which is also known as ‘biogenic carbon’“. Embodied or upfront carbon, on the other hand, results from the processes involved in the production of CLT. Compared to other construction materials, the amount of embodied carbon in CLT is relatively low, namely 0.25 kgCO 2 e/kg (Place et al., 2021). Upcycling CLT production waste both keeps the biogenic carbon from being released and reduces the carbon emissions needed to produce new CLT products.

A growing number of digital tools, such as Circularity Indicators and Material Passports (Heisel et al., 2024), support the implementation and assessment of circularity in building construction. Stock-based design contributes to the circular transition by enabling upcycling, typically utilizing computational design methods. Using non-virgin material such as CLT production waste has a positive impact on Circularity Indicators. Linking components made of upcycled production waste with a Material Passport could facilitate their use in yet another building in the future, or in a cascading manner, where the embedded material would be used for building components with lower structural or aesthetic requirements.

Implementing circular construction and stock-based design has the potential for producing buildings and construction systems with distinct aesthetic qualities, which could translate into added value for investors as well as for users and society as such. Initiatives such as the New European Bauhaus (New European Bauhaus, 2025) show that politicians and other decision makers have understood that focusing on technological solutions alone is not enough.

In this paper, we present an integrated computational framework that facilitates the upcycling of CLT production waste into building assemblies, such as façade modules. The versatility of the framework is achieved by a combination of multiple features. It adds to the individual contributions of previous studies in the domain of stock-based design with wood waste by integrating stock input, geometry generation, joint definition, fabrication logic, and assembly sequencing within a single computational workflow, enabling full-scale realization. More specifically, using a genetic algorithm, it enables the design of building assemblies made of interlocking CLT elements sourced from production waste. The generation of an assembly sequence as well as the joint geometry are integral parts of the workflow. The operability of the framework is demonstrated by the design, fabrication, and assembly of a proof-of-concept prototype. The demonstrator also served as an exhibition piece on two occasions.

The paper is structured as follows. Section 2 provides a state-of-the-art overview on Circular Economy in the building construction sector, timber construction and CLT production waste, as well as digital tools and computational design methods related to stock-based design. Section 3 presents the methodology, including the development of Design Algorithm 1, Design Algorithm 2, the fabrication framework, and the assembly sequencing method. Section 4 provides more detailed information on the fabrication and assembly of a proof-of-concept demonstrator, and the observations made throughout this process. Section 5 is dedicated to the discussion and conclusions. Here, we address the advantages, limitations, and potential of the presented approach, as well as future work.

Background and state of the art

Circular economy principles and carbon sequestration in the building construction sector

Transitioning from a linear to a Circular Economy (CE) is one way of reducing the building construction sector’s large amount of resource consumption and waste production. Implementing CE principles, respectively circular construction, could eventually result in what some scholars refer to as Circular Built Environment (van Stijn and Gruis, 2019; De Wolf et al., 2024). Among other things, this is encouraged by new standards, such as the recently released DIN SPEC 91484, (DIN SPEC 91484:2023-09 D/E, 2023), and the recent addition of LCA Module D, introduced as part of the EN 15804+A2 2019 update (EN, 2019).

The European Parliament (European Parliament, 2023) defines Circular Economy (CE) as follows: “The circular economy is a model of production and consumption, which involves sharing, leasing, reusing, repairing, refurbishing and recycling existing materials and products as long as possible. In this way, the life cycle of products is extended.” The Ellen McArthur Foundation (Ellen MacArthur Foundation, 2023) provides a similar definition. One goal that is articulated here is the circulation of products and materials at their highest value.

This can be achieved by means of recycling, or more specifically, upcycling. The Merriam Webster dictionary (Merriam-Webster, 2023) defines upcycling as follows: “to recycle (something) in such a way that the resulting product is of a higher value than the original item,” and/or “to create an object of greater value from (a discarded object of lesser value).” Upcycling is also defined as the process of upgrading value and/or quality of materials or products in their second life and beyond (Sung, 2015).

An emerging concept that is relevant in the context of upcycling is stock-based design. In (Brütting et al., 2019), stock-based design is described as follows: “Designing structures from reused elements is unlike conventional structural design because stock element availability is a design input. […] In this new paradigm structural form follows availability.” Stock-based design can enable the upcycling of waste wood into larger timber assemblies, and can thus contribute to carbon sequestration, which is defined as “[…] the process of capturing and storing atmospheric carbon dioxide” (What is carbon sequestration, 2023). This is discussed in more detail in Section 2.2. Computational design methods are an integral part of stock-based design. An overview of the key methods, based on (Sun et al., 2026) is provided in Section 2.3.

One challenge regarding circular construction and stock-based design is how to collect information on waste or surplus materials, and how to make this information available for potential users. Reclaimed material needs to be registered via scanning or photogrammetry, unless a BIM model of the source building is available. Production waste on the other hand could potentially come along with digital data, if the source material is processed via CNC.

Timber construction and CLT production waste

Wood is a renewable resource, derived from trees, which sequester carbon whilst growing (They can capture more carbon, 2021). Moreover, timber buildings can serve as carbon sinks (Heräjärvi, 2019; Churkina et al., 2020; Mishra et al., 2022), and, by doing so, contribute to climate change mitigation (Climate change mitigation, 2023). Reversibility, adaptability and modularity can be found in some traditional timber construction methods, such as timber framework (Bygningsbevaring Raadvad, 2018), and more generally, in frame-based and mechanically connected assemblies. Using timber in a circular manner is also a means to support sustainable forestry, and a way of acknowledging that it’s not an infinite resource.

Considering the potential advantages of building with wood, it is encouraging to see that more and more buildings are being realized in timber, including projects using Mass Timber construction. The latter is reflected in a notable increase in the production of CLT –which some consider as an emerging modern construction material (Harte, 2017) – over the past two decades (Kurzinski et al., 2022). The annual European output of CLT in 2023 was about 1.7 million m 3, with a projected Compound Annual Growth Rate (CAGR) of 9.6% between 2024 and 2032 (European Cross-Laminated Timber Market Report by Application et al., 2024). Other sources equally expect an increasing demand in the future (Waugh Thistleton Architects, 2025).

What is often overlooked is that timber construction also generates considerable amounts of wood waste, respectively waste wood, which typically is downcycled, or ends up as energy wood. This waste is generated both during the construction stage on site, but also during the manufacturing of timber elements off-site. The focus of this work is on the latter type, which is also referred to as production or fabrication waste. “Currently, less than half of European waste-wood is being recycled, contributing to an estimated 50 million tonnes of waste annually,” according to (Waugh Thistleton Architects, 2025). The Swiss Federal Office for the Environment (FOEN) recommends making use of waste wood before incineration and distinguishes between four types of wood waste: natural untreated wood, scrap wood, used wood and problem wood waste (FOEN, 2023). FOEN defines scrap wood as production waste from sawmills, carpenters’ workshops or furniture factories and untreated wood off-cuts from building sites, such as scaffolding planks and struts, as well as off-cuts and sanding dust from joineries.

The potential of using waste wood in architectural and structural applications is explored, for example in (Browne et al., 2022; Ruan et al., 2022; Bergsagel and Heisel, 2023). These studies involve salvaged, reclaimed and recovered wood, all of which have the issue that the elements’ remaining mechanical properties are unknown. Using Non-Destructive Testing (NDT) (Tamke et al., 2025) can help addressing this challenge. In order to use reclaimed timber as part of stock-based design, it needs to be digitally registered, using, for example laser scanning, photogrammetry, and/or computer vision. Wood production waste, on the other hand, can be considered as quasi-virgin material, if not exposed to outdoor storage for a longer period of time. Hence, this would make additional processing for quality assurance unnecessary. In principle, obtaining information on the elements’ dimensions should become easier here as well, if the cutting process is CNC driven.

Making use of CLT production waste is of particular relevance, as the manufacturing of CLT consumes a large number of resources and produces considerable amounts of wood waste, as shown in (Pramreiter et al., 2023). CLT production waste is typically the result of inserting window and door openings, or results from trimming panels, which come with standardized width, to the required size. According to Ottenhaus et al. (Ottenhaus et al., 2023), the cut-offs related to door and window openings correspond to 5%–10% of the virgin panels. The increasing interest in using CLT production waste in building construction is reflected in several recent research projects and articles. (Poteschkin et al., 2019) proposes combining large CLT wall cut-outs with natural and glue-laminated timber for standardized ceiling and wall construction systems, in which the CLT elements are connected with loose tenons made of beech veneer plywood. (Robeller and Haaren, 2020) shows how CLT door- and window cut-offs can be employed in the construction of a segmented timber shell. (Vessby et al., 2023) features three cases: finger-jointed CLT cut-offs as transverse layers in CLT floor and roof slabs, CLT cut-offs as rails in timber frame structures, and as integrated laminations to enhance the capacity of glue-laminated timber beams for carrying concentrated loads. Although the references listed in this section are valuable examples of how CLT production waste can be upcycled, they are lacking the implementation of computational design methods, which is a key characteristic of stock-based design.

When designing buildings or building components with Design for Disassembly (DfD) and Design for Reuse (DfR) in mind, the design of the connections or joints between the individual elements is of major importance. The potential of different connection types for disassembly and reuse is studied in Ottenhaus et al. (2023). Due to their plate-shaped geometry, CLT cut-outs enable the implementation of interlocking joints, which can be described as linear slots that enable the sliding together of several elements. Due to the simplicity of this joining principle, the elements can be dis- and reassembled without suffering relevant damage, depending on the tolerances and the number of joints. Several recent studies investigate the potential of these connections (Hudert and Kotnik, 2017; Hudert et al., 2018; Aranha et al., 2021) and their integration in digital design and fabrication workflows (Mangliár and Hudert, 2022; Atanasova et al., 2022).

Digital tools and computational design methods for CE and stock-based design

Digital tools have the potential of enabling circular economy principles in the building construction sector and could thus establish a digital circular built environment (Augustynowicz and Aigner, 2023). Circularity Indicators (CI), such as the ones developed by the Ellen McArthur Foundation and Madaster (Material Circularity Indicator MCI, 2025; Madaster Services et al., 2018), are one of these tools. They can facilitate the measuring of a design’s environmental impacts in the context of circularity (Heisel et al., 2024), for example by measuring the ratio of virgin to non-virgin materials in a building. Such indicators might soon become crucial, as future building regulations might ask for a minimum percentage of non-virgin material in new buildings. Material Passports (Material Passport, 2025) are another relevant tool. They are intended for “recording the quantity and quality of each material in a building to prevent it from becoming ‘unidentified waste’” (Mao and Cao, 2025).

A comprehensive overview of relevant digital tools and technologies is provided in (Çetin et al., 2021): “(1) additive/robotic manufacturing, (2) artificial intelligence, (3) big data and analytics, (4) blockchain technology, (5) building information modelling, (6) digital platforms/marketplaces, (7) digital twins, (8) the geographical information system, (9) material passports/databanks, and (10) the internet of things.” Digital platforms and marketplaces are particularly relevant for stock-based design and upcycling, as identifying a stock is a first and fundamental step in this approach. Several of such platforms, focusing specifically on construction products and materials, have emerged over the past few years. Some examples are Precycle, GreenDozer, Circue, formerly known as Circle Bank, and Concular. While offering waste and reclaimed materials online, none of these platforms provides digital models of the materials on sale.

The above-mentioned digital tools and platforms can facilitate the implementation of circular construction principles and allow for measuring the circularity potential of buildings. However, for upcycling via stock-based design, computational design methods are equally, if not more, relevant.

The origins of stock-based design via computational methods can be traced back to two pioneering studies. Firstly, Certain Measures ’ project Mine the Scrap (Mine the Scrap, 2023), which highlights the potential of computational design processes for upcycling construction waste into architectural elements, such as building envelopes. Secondly, the work carried out at the Structural Exploration Lab (SXL) (Brütting et al., 2021), which presents a novel computational design approach that can enable the reuse of structural components for multiple service lives. Here, a base structure is optimized under stock constraint, guided by structural performance. The workflow uses linear programming, where the multi-objective functions are element assignment, topology, stock, structural and geometry optimization, and life-cycle assessment. The method can be used for generating 2D and 3D trusses and can be applied in two scenarios: for the reuse or integration of elements of a given stock, or to design a kit-of-parts that can be used to form different structural configurations.

More recently, there has been an increasing use of computational design workflows in the context of circular construction with non-uniform and non-standard materials, including reclaimed materials and production waste (Sun et al., 2026). As shown in (Sun et al., 2026), one can identify six computational design methods within these workflows, including heuristics, metaheuristics, mathematical, rule-based, and physics-based methods, as well as machine learning.

In view of their role in our framework, meta-heuristics are explained in more detail here. Metaheuristic algorithms are widely recognized as one of the most practical approaches for solving combinatorial optimization problems. The category of metaheuristic algorithms includes genetic algorithms, simulated annealing, Tabu Search, and others (Schrijver, 2003). These algorithms are based on the iterative execution of the following two steps: First, search new solutions on the basis of the previous history, and second, evaluate the solutions generated in the first step and extract necessary information for the subsequent search (Yagiura and Ibaraki, 2001). In (Kumar Thakkar et al., 2022), genetic algorithms are explained as follows: “The genetic algorithm is an evolution-based algorithm inspired by natural selection. […] These algorithms take a set of candidate solutions and then keep iteratively evolving these solutions to give optimal or near-optimal solutions. The correctness of the solution increases with iteration. The primitive components of this algorithm are the initial population, fitness function, selection, crossover, and mutation. The initial solution consists of components like gene and chromosome; each solution given in the set of initial solutions is called a chromosome, while parts of the chromosome are known as a gene.”

Related studies

In the following, four related studies using stock-based design are discussed, taking into consideration the implemented computational design method, the material used, the integration of joint generation, and the type of structures designed and/or realized.

Augustynowicz and Aigner (2023) employs a rule-based method, respectively shape grammar, based on a 3D reciprocal node. The material is sourced from production waste and consists of Oriented Strand Board (OSB) and three-layered timber panels, with a material thickness ranging between 18 and 42 mm. The elements are connected via orthogonal interlocking, for which slots are milled with a robotic arm for half of the panels. The approach is used for the design and fabrication of pavilion-scale structures that can serve as seating area or as climbing installations for children.

Tuksam and Pihlak (2024) likewise employs a rule-based method. Here, CLT offcuts are matched to a non-orthogonal polyhedral grid. The method is applied for the design and fabrication of a shelter structure. The panels are connected with concealed spline joints, enabled by automatically fabricated grooves in all edges of a panel.

Parigi (2021) makes use of a Strength Pareto Evolutionary Algorithm II (SPEA-II) and uses linear reclaimed timber elements to produce wall and slab components based on a reciprocal logic. The joints are not specified.

Dupas and Hudert (2024) expands on the method proposed in (Parigi, 2021). The material used is CLT offcuts, and the workflow generates reciprocal configurations in which the elements are not perpendicular to each other. The rotation of the elements reduces the need for adjusting their length. The resulting structures can be used for roofs. The joints between the elements are not specified. We will show that the workflow presented in this paper goes beyond each of these studies in one or more aspects.

While the above discussed studies demonstrate the potential of upcycling CLT waste and applying computational design methods, they typically address isolated aspects of the workflow. Previously studied approaches often focus on geometry generation or structural arrangement, but do not integrate stock input, joint design, fabrication logic, and assembly sequencing within a unified process. As a result, the link between computational design and full-scale realization remains limited. The below presented workflow aims to address this gap by integrating design, joint generation, and fabrication-aware assembly into a continuous, end-to-end process. Moreover, the work challenges established assembly methods, aiming to facilitate design for disassembly by using non-orthogonal, timber-only slotted connections, resulting in potentially load-bearing structures with unique aesthetic qualities.

Based on the above-stated background, challenges, and goals, this work sets out to develop a novel computational design and digital fabrication framework, enabling circular construction via reuse and upcycling of CLT cut-outs, and implementing the concept of stock-based design. The research employs empirical as well as numerical methods as well as a research-by-design approach. In this section, we elaborate on the steps and details of the computational workflow, and the design solutions for the interlocking structure.

The proposed framework and its workflows, as illustrated in Figures 1, 2, are developed with the overall goal of enabling circular construction through digital technologies, which also other scholars promote and work towards. Following the idea of stock-based design, we work towards a solution that enables the upcycling of wood production waste via digital tools. We envision that Building Information Modelling (BIM), digital platforms and marketplaces, material passports, and databases can facilitate the use of production waste as a resource for new buildings and the modification of existing ones. At the end of a building’s service life, components made of upcycled materials can equally become part of this process, which keeps the material, and the biogenic carbon stored in it, in use even longer. An essential requirement for handling such complex, synchronized, and circular supply chains is the connection of digital data bases with computational design tools, and partially or fully automated digital fabrication facilities.

Architecture and implementation of the framework

In the following, we present the considerations and iterations that led to the prototypical workflows. Their functionality was later tested by applying it for the design and fabrication of a demonstrator, the Re:Shuffle pavilion (Mangliar and Hudert, 2022). The computational design workflow is developed to generate potentially load-bearing and structurally stable assemblies made of wood production waste, as previously shown in Figure 2. Starting point of the digital workflow is the definition of the available materials and the design parameters, including design goals such as transparency and views, spatial boundaries, and machining constraints. Following that, a fabrication ready design is generated using metaheuristic optimization based on the defined optimization criteria.

The digital framework is implemented in the visual programming environment Grasshopper 3D, an integrated extension of the CAD software Rhinoceros. It also makes use of Galapagos, the native evolutionary solver in Grasshopper3D, and Karamba3D for structural analysis.

The overall workflow is divided into three partially nested and interconnected processes, as shown in Figure 3. The basic module, a computationally lightweight parametric algorithm, Design Algorithm 1 (DA-1), generates design instances based on input parameters. An evolutionary solver (Galapagos) is then used to explore the design space. The optimization considers multiple performance criteria, which are aggregated into a single weighted fitness function. The optimization is thus formulating a single-objective problem. Upon reaching optimum, based on stagnation-based convergence criteria, a high-performing solution is reinstated, and a second, more detailed algorithm, Design Algorithm 2 (DA-2), generates a refined parametric model based on this geometry. It should be noted that the converged solution represents one optimum with respect to the defined fitness function, which depends on the chosen weighting of criteria. Consequently, alternative high-performing solutions remain available within the explored population, allowing the user to manually select among near-optimal design variants according to additional preferences not explicitly captured in the fitness formulation.

Key workflow parameters and components

Each design variation, or specimen, is described by its defining parameters. These fall into three subcategories: design boundary, optimization criteria, and specimen chromosome. The first two categories define the design problem, while the third parameter defines the given solution from the design space, as shown in Figure 4.

The design boundary parameters are defined by a collection of constant variables describing the design space, boundary conditions and constraints, and most importantly the material-stock as well as the boundary space. Other criteria are the loading and support areas, structural loads, and machining constraints.

The design goal is expressed by a combination of optimization criteria, to evaluate the performance of one specimen regarding structural performance, design, and fabrication. The third parameter, the specimen chromosome, defines the configuration of the parts in space by one float value between 0 and 1, using chromosome encoding, explained in Section 3.6.

Optional additional parameters could be the further segmentation, or sub-division, of the elements of the given stock, the vector of insertion, and the inclusion of new parts made of virgin material. However, including virgin material widens the design space with more dimensions, exponentially increasing the number of solutions. While the feature of expanding the design space might be reasonable in many cases, this study only considers a fixed and pre-defined material stock.

Design Algorithm 1 – Initial gene-based model generation

Design Algorithm 1 is a traceable parametric model, and the function on which the Generative Algorithm (GA) operates. Since the GA iterates over hundreds of populations of hundreds of specimens, this initial algorithm is set up to be as computationally lightweight as possible, while still being sufficiently accurate.

This algorithm generates a model of the structure based on a specific set of genes and the defined boundary parameters, by distributing the individual panel elements within the boundary geometry, setting up a preliminary model for structural evaluation. Using Karamba 3D, a low-fidelity finite element model is created with 2D shell elements representing the panel geometry, with coarse mesh and linear elastic orthotropic material properties. The interlocking slotted joints are represented as semi-rigid connections using linear rotational and translational springs with calibrated axial and in-plane rotational stiffness.

The overall performance of one specimen design is expressed in a fitness value, the weighted sum of the optimization criteria, which is to be minimized by the evolutionary solver. The interlocking structure and its components are evaluated based on geometric properties and structural performance, respectively. The selection of these criteria, their calculation, and their weighting is critical, and it required several iterations to develop the optimal formula, see Figure 5.

To direct the evolutionary solver to converge faster, prior criteria, defining structural performance (internal elastic energy, displacement, and ‘stable’ head and base), are accounted for with an exponential factor. This steers the solver to find structurally sound assemblies first (time stamps a and d in Figure 6). Later, fitness values, such as joint axis proximity, joint angle limit, and minimum intersection line length, start to take effect, optimizing the design towards feasible machining, fabrication, and assembly. Finally, criteria with lower importance, such as utilized material area and the number of disconnected parts, with lowest weights, push for maximizing material stock utilization. The evolution of the design is captured at six stages in Figure 6, based on the time stamps a to f, indicated in Figure 7.

The population size is implicitly defined by the material stock size, e.g. stock-size multiplied by position coordinates, plus insertion vector coordinates. This equals 31 genes for the final demonstrator with 14 panels. These genomes are then parametrized to the domain of the respective design parameters. The solver terminates upon convergence based on stagnation rate, which is set to 75%; the rate of genomes showing no improvement over successive generations. The solver settings were tested and refined iteratively during development, until reaching consistent results.

Design Algorithm 2 – Detailed geometry generation

The next, separate phase of the computational workflow is the generation of the detailed geometry of the components by an extensive parametric design algorithm. Once the solver converges to a final stage, DA-1 returns rectangular reference surfaces representing the panels in their final position. This becomes the geometric input of Design Algorithm 2 (DA-2), aside of the main boundary parameters. This transition phase creates an opportunity for the user to make minor edits in the final design, by adjusting the position of the reference surfaces according to aesthetic preferences or goals, while getting constant feedback on the modified structure’s performance. In the case of the demonstrator, panels were shifted slightly to adjust the positions of the openings, and the lowest, almost horizontal panel was moved up vertically, to achieve a more balanced visual appearance.

Design Algorithm 2 generates the final geometry, including the detailed geometry of the interlocking panels, based on the previously generated simple model and the further pre-set parameters.

The slot joints are generated based on the assembly order. The outer parts of the panels can be trimmed with a user-defined bounding box, upon choice. The algorithm outputs data for multiple purposes, such as a detailed fabrication-ready digital model of the elements with the joinery design including CNC machining toolpath, a 3D model for visual representation with automatic realistic CLT texture orientation, a 3D print-ready model for physical representation, of the whole structure and of the individual parts in given scale, as well as the Finite Element model implementing the method used in DA-1 (Figure 8). To illustrate that the evolutionary workflow equally functions without user intervention, an alternative design, solely produced by the algorithm, is shown in Figure 9. While working with the same design goal settings and boundary parameters, a different material stock was used here.

Assembly design–global insertion direction

As a result of the expanded design space it causes, the free orientation of the interlocking planes increases the complexity of the assembly. Specifically, each newly placed element must be slid into other elements oriented on different planes. As shown by other scholars, this can be addressed in several ways, for example on joint detail level, by widening the slot angle to enable the insertion in multiple angles simultaneously, or on the global level with element placement and orientation optimization, as discussed previously by (Cignoni et al., 2014) and (Schwartzburg and Pauly, 2013). Possible downsides of these methods are overall weaker assemblies or over-optimization.

To overcome this complexity, we implemented a simple design rule, the global insertion direction. This restricts the orientation of an element’s base plane by removing one degree of rotational freedom, thus each panel’s normal must be perpendicular to the global insertion vector, as shown in Figure 10. This solution ensures the possibility of the assembly of any multi-parts combination in such a way that the plane orientations are still relatively flexible. As a parameter in itself, the global insertion vector could be part of the specimen chromosome, or as in the case of this study, a predefined boundary condition. Aside from ensuring a buildable design, the insertion vector also determines the main see-through and shading direction of the structure, in dependence of the global orientation and location.

Chromosome encoding

It is necessary for the genetic algorithm to relate the specimens’ fitness to a single parameter. Therefore, specimens of the population are described in a binary form, called chromosome or string, representing each possible variation of the parameter variations. The chromosome is the list of gene parameters, which is encoded into a single value, called string, using combinatorics.

The function of the chromosome encoding can be described as follows: the list of chromosome parameters [C] is encoded into the chromosome string [S]. S refers to the n th item from all possible variations of C, by remapping S along the number of all variations. Thus, C is a variation of objects, or more specifically, one permutation of combinations of objects with repetitions allowed, where the objects are values in the range of 0 to 1, in D number of decimals. Eventually, this arrangement of float values will be remapped along the respective parameter range in the parametric model. Therefore, the two parameters affecting the number of parameter variations, and hence the number of feasible specimens, are D and the size of the material stock, since the material stock defines the length of C, while D defines the spatial ‘resolution’ of the structure.

Chromosome decoding–plane definition

To specify the position of the elements in the boundary geometry, C is segmented into lists with six items [Px, Py, Pz, vx, vy, vz], the components of the planes’ origin ‘P’ and their main axis ‘v’.

The planes are oriented along their two main axes, ‘v’ and the insertion vector ‘i’, presuming that ‘v’ is not equal with ‘i’. Thus, one panel’s orientation is expressed in a list of six numbers [Px, Py, Pz, vx, vy, vz]. Finally, this data can be translated to Euclidean geometry, by remapping components of P relative to the boundary space ((L, W, H), respectively), while components of ‘v’ are remapped relative to range of (−1; 1), see Figure 11.

Generative joint design

Design Algorithm 2 generates the slot joints based on initial constant values and variables inherited from the basic model, which is the output of Design Algorithm 1. A joint’s shape and position are defined by the intersection line and by the panel angles. The orientation of the open side of a slot is determined by the assembly sequence, as explained below. One of the above-mentioned fitness values, namely the joint axis line length, is implemented to ensure a sufficient amount of surrounding material for the slot to be milled.

When defining the size of the tolerance gap, one must consider multiple parameters: the fabrication tolerances, the material imperfections, and the assembly tolerance. The size of the assembly gap is ideally determined during preliminary fabrication tests. Even if this is possible, uncertainties are hard to eliminate. The aim is to find a solution that enables both ease of assembly and stiff connections. Additionally, fabrication tolerances, based on the actual machining setup, must be considered.

As mentioned, the orientation of the open side of the slot is determined based on the assembly sequence and the assembly direction, as shown in Figure 12. In order to enable co-planar contact surfaces at the end of the slots, the inner corners of the joint are designed with a dog-bone fillet solution, achieved by extending the toolpath in the longitudinal direction of the joint.

Assembly sequencing

To ensure the buildability of complex interlocking assemblies, paying attention to design of the assembly sequencing is crucial. The assembly sequence of an interlocking structure affects the buildability of the structure, its structural strength, and informs the joint geometry. Based on the sequence of insertion, the orientation of the joints, respectively their open side is determined. Figure 13 shows three feasible (a,b,c) and one unfeasible (d) sequence variation for a configuration with five panels, and how the sequence affects the joint design. In the case of the first three variations, the assembly is buildable because the slot cut direction is defined based on the assembly order. The last variation, d, shows a sequence where the insertion of z as the last piece would not be possible, since the slots of the z element do not follow the assembly order, resulting in a ‘locking’ element.

Out of a vast number of theoretically possible ordering combinations, many are feasible. We found that this sequencing approach, sorting the panels in respect to their base planes’ origins along the insertion vector, is a reliable method, resulting in buildable and coherent structures. The assembly sequence of the realized demonstrator, involving twelve elements, is shown on Figure 14. Assembly step four (first row, fourth image) shows a typical case, which could be avoided by further optimization of the sorting method in the algorithm. Aside from eliminating non-connected placement of panels, such optimization could enhance the structural capacity of the assembly, as well.

One motivation for realizing a demonstrator was to prove the functionality of the computational workflow, the structural soundness of the underlying construction logic, and its buildability. In addition to that, the demonstrator also served as an exhibition piece during a conference and a trade show. Here, the goal was to illustrate the challenges and potential of circular construction, and the aesthetic potential and consequences it might entail.

In an initial schematic design, the interlocking wall was part of a larger overall structure, which also featured an interlocking column and two larger slabs, one forming the floor, and the other one forming a roof, as shown in Figure 15. Later, the size of the demonstrator was reduced, due to time and budget constraints, as well as logistics.

In the initial design stage, before more detailed information on the CLT cut-outs was available, assumptions regarding their size had to be made. After receiving this information, the design of the wall was minimally modified to fit the available material quantities and dimensions. The realized structure, shown in Figure 16 has a length of 2.45 m, a width of 1.2 m, and a height of 2.8 m. The boundary volume of the interlocking wall has a length of 2.45 m, a width of 0.7 m, and a height of 2.55 m. The construction drawings are shown in Figure 17.

Fabrication of the components

In addition to proving the functionality of the computational framework, the demonstrator also successfully implements the underlying ambition of upcycling production waste and thus demonstrates the framework’s ability of contributing to a circular built environment. The demonstrator’s panels are sourced from production waste that accrued during the fabrication of CLT wall elements for a residential building in Denmark, located relatively close to the venue of the first exhibition. The received panels are shown as grey rectangles in Figure 18. Two types of 5-layered CLT panels served as a source, one with 100 mm and one with 120 mm thickness. The milling of the 17 panels – 12 for the wall, 5 for the column–was carried out in parallel to the processing of the CLT elements for the residential building. The assembly of the column, shown in Figure 15, was omitted at a later point.

The panels of the final structure, their bounding box, and the original CLT cut-outs account in volume for 1.22 m 3, 1.55 m 3, and 2.1 m3, respectively, see Figure 18. After fabrication, the reusable leftover material is approximately 0.5 m 3.

It would be interesting to investigate how different design solutions vary in material usage. This project studies the example of reusing CLT cutouts, typically from larger construction projects, where it can be assumed that several standard sizes would occur within one construction project.

The assembly process

The structure was assembled horizontally, with the front face of the wall oriented to the ground, by two and sometimes three people. Weighing between 20 and 60 kg, the elements could most of the time be handled by two persons. The self-positioning property of the joints helped in the process, which was further supported by a third person, guiding the initial stage of the insertion process and adjusting the position of the receiving parts. Printed labels, provided by the supplier of the cut-outs, made it easier to identify the individual elements and find the right one for each of the assembly steps. A 3D model of the individual elements in the scale of 1:15 was also printed. It served as a small-scale physical twin and helped to simulate and guide the assembly of the full-scale demonstrator.

Only the placement of the first pieces required additional support and attention, as there were no other elements yet to keep and support it in its inclined position. Due to the large assembly gap, it was easy to adjust and position the receiving parts. Only in one or two instances was it necessary to use a large hammer to insert an element. This occurred towards the end of the process. After the assembly of the interlocking parts of the wall, the top and bottom slabs were fixed using timber construction screws, from top and bottom, respectively. Doing so, every element of the assembly was geometrically locked, and no more additional fasteners were added to the structure, except one that served as a test. Finally, the structure was tilted up and moved to its final position with a truck crane. Extra planks were mounted to the bottom slab with an additional PVC layer, to increase the friction of the structure against possible sliding induced by wind. Figure 19 documents different stages of the assembly process. In a first stage, the interlocking panels were assembled without the need for screws. The foundation and roof panels were added at a later stage, using long timber screws.

Tolerances and displacements

In the case of the realized demonstrator, the CNC setup used for milling the CLT pieces involved a fabrication tolerance of ±4 mm, which is relatively large. The tolerance in the material thickness was ±1 mm. Therefore, the joint assembly gap parameter was set to 4 mm in the workflow to avoid collisions or jamming. Thus, the assembly gaps between the fabricated parts, visible in Figure 20, ranged between approximately 2 and 8 mm. On the one hand, this enabled a smooth and complication-free assembly process. Moreover, although only the top and bottom plate of the wall were fixed with screws, the entire wall could be lifted and relocated by a crane and the dynamic loading involved in this process, which demonstrated the structural capacity of the interlocking assembly and the truss-like triangulations within it. On the other hand, the above-mentioned assembly gaps contributed to a low resistance against inner settlement in the final structure.

After being stored in a standing position for a period of about 9 months, the outer edge of the cantilevering part of the CLT panel on top of the wall was visibly lower compared to the state of the demonstrator during the first exhibition. The observed largest displacement of the structure was ±50 mm out of plane, measured at the top of the structure toward the cantilevering top slab, at a height of 2.9 m.

What needs to be considered here is that the cantilevering slab introduces an eccentric load. A slab symmetrically cantilevering on both sides of the wall, or a four-walled demonstrator covered by one slab might not have displayed such displacements.

Another plausible cause for this phenomenon is local creep respectively shrinkage effects that entailed a local loss of contact stresses. Although these effects could be minor for each connection, the large number of joints might have multiplied this and led to the observed displacement. At the same time, no lateral displacement was recognizable in the plane of the wall.

The observed displacements are an important insight gained in this work. However, additional studies are required to obtain a better understanding of the observed behavior and to identify effective countermeasures. This is addressed in more detail in Sub-Section 5.3, Future Work.

In this paper, we presented a framework with bespoke computational tools and workflows that enable engineers, architects, designers, and manufacturers to produce building components made of CLT production waste. The presented prototypical workflow integrates stock input, geometry generation, joint definition, fabrication logic, as well as assembly sequencing, and enables full-scale realization. The paper provides a detailed overview of the developed computational design methods, explaining how the design goals are parametrized. We elaborated on the principles behind the structural system and its underlying parametric model, as well as on how the fabrication-ready design is generated by using metaheuristic optimization based on pre-defined criteria. The functionality of the developed algorithms is demonstrated by their successful application in the design and fabrication of a full-scale demonstrator. The demonstrator, made from CLT cut-outs originating from a residential construction project in Denmark, was displayed as part of the Critical Practices exhibition during ICSA 2022, and at Building Green Aarhus 2023.

Despite having produced a self-supporting demonstrator, load-bearing applications are not a fully accomplished, generally applicable result yet. More data, also from experimental testing, needs to be collected as part of future work.

In addition to discussing the developed digital workflows and the observations made during the fabrication of the demonstrator, the paper also indicates a possible future application of such assemblies in architecture and building construction, as modular and potentially load-bearing façade elements. This is addressed in more detail in the following sub-sections, where we discuss the framework’s advantages and potentials, as well as its limitations, and provide an outlook on future work.

Advantages and potentials

The framework and workflows presented in this paper have the potential of reducing CLT production waste, as well as production waste from other wood-based panels, as indicated below. Moreover, by enabling the upcycling of production waste into larger assemblies, in can contribute to carbon sequestration by increasing the use of buildings as carbon sinks. Designing upcycled assemblies such that they can serve as modules could enable multiple cycles of usage, and thus further extend the service life of the upcycled material.

The stock input interface of the current workflow makes it possible to link to data provided by production facilities. In addition to CLT, it is already possible now to integrate offcuts sourced from LVL, OSB, and plywood. Focusing on production waste instead of reclaimed wood has the potential advantage that information on the geometric qualities of the pieces could be obtained from manufacturers even more directly, especially if fabrication is based on CNC processes. Another advantage of production waste is that the material’s mechanical properties can be expected to be the same as those of virgin material. Future work could enable the inclusion of reclaimed wood. Some of the necessary steps for achieving this are outlined in Section 5.3.

Using interlocking joints facilitates the assembly of the panels, and is expected to do the same for disassembly, if required. Moreover, the demonstrator shows that non-orthogonal interlocking cannot only strengthen the overall coherence of the structure but also produce distinct aesthetic qualities. This could result in added value, even more so if users or clients can customize their designs. To our knowledge, there are no other studies that have achieved a demonstrator with non-orthogonally interlocking panels at this scale.

The successful assembly and exhibition of the demonstrator hint at the not yet fully achieved possibility of producing structurally sound and potentially load-bearing timber assemblies, to be used, for example as façade elements. In the future, we expect the framework to be applicable for the design and fabrication of structural wall panels, or non-structural components, such as dividing walls for interior spaces, shading components for facades, or acoustic diffusers.

The here presented workflows and the demonstrator differ from existing studies, indicated in Section 2.4, as shown in the following. Our work goes beyond (Atanasova et al., 2022) by using an evolutionary algorithm, working with a material thickness between 100 and 120 mm, and by the fully integrated fabrication of non-orthogonal interlocking joints. (Kumar Thakkar et al., 2022) equally involves CLT offcuts. However, it neither features a genetic algorithm nor non-orthogonally interlocking joints. We deviate from (Tuksam and Pihlak, 2024) by using CLT cutouts, by using a different algorithm, and by integrating both joint generation and fabrication. Whereas (Parigi, 2021) involves CLT elements, the joints between the elements are not specified, and rotation of the elements only occurs in plan. The method is applied for the design of a roof structure. Neither of the latter two examples is built.

Limitations

In order to fully unfold the potential of the framework, several limitations need to be addressed. The design for the demonstrator was chosen based on its function as an exhibition piece. The selection of this particular design was based on a subjective aesthetic evaluation by the authors, taken into consideration a balanced distribution of elements and a balanced range of different angles. To allow for a wider range of applications and their implementation in actual buildings, the pattern generation should be modified such that the optimization procedure creates patterns that are more suitable for attaching glass panels and frames. Here, it would be beneficial to have some elements that are aligned vertically, making it easier to attach standardized façade systems. Moreover, optimizing towards having several elements that create a continuous connection between bottom and top could eliminate the structurally challenging displacements observed in the demonstrator, caused by an accumulation of tolerance gaps around the interlocking joints. In addition to that, the deep cuts required for these joints reduce the flexibility of reuse of individual panel pieces.

From a technical point of view, the fabrication of the angled interlocking joints is a potentially limiting challenge in itself, although in Europe, machinery facilitating this is widely available. Ideally, the offcut elements are processed at the facility producing them, and where large 5-axis CNC machinery is typically available. This was also the case with our demonstrator. A potential risk here is that the precision of such machines can be relatively low when working with relatively small elements, such as offcuts. In our case, this was not an issue. The tolerance gaps, likely causing the displacements under an eccentric load, were designed to facilitate the assembly of the demonstrator.

In its current state, the tool is operated by a single designer, with parameters and constraints - fabrication, stock, and structural input data–sourced from specific stakeholders. For real-world applications, the tool must accommodate live data integration and support collaborative multi-actor workflows. A further limitation is the high level of expertise required to operate the custom computational design process; to overcome this, a more accessible, and robust tools with a simpler user interface is needed.

On a more fundamental level, future advances in CLT production processes might reduce or even eliminate the generation of offcuts. However, the presented procedure remains valuable for offcuts from other panel types. Finally, manufacturers and clients need to be willing to make the data on offcuts and cut-outs available digitally. In this context, legal questions regarding ownership might emerge, or questions on how to compensate the original client if the production waste is transferred to another party.

Future work

In this section we indicate future work that could help overcoming some of the above-listed limitations and thus contribute to further unfolding the potential of the proposed framework.

In order to unfold the mitigating impact of the framework, a main goal of future work is to make it easily accessible, and to increase the range of applications. Among other things, this could be achieved by enabling the generation of assemblies with a lower geometrical complexity, allowing for integration with existing and standardized construction systems, for example for facades. If loadbearing capacities are required, current displacement issues need to be addressed. This could be accomplished by alternative connection designs. Moreover, modifying the workflow such that it enforces continuous, almost vertical elements within a larger assembly (see Figure 22), could prevent the displacement observed in the demonstrator.

Future studies should investigate alternative demonstrator designs to see if similar displacement occurs. One alternative design could feature an equally cantilevering slab, avoiding eccentric loads. Another design alternative could be four-walled, covered by one slab. In addition to this, such future studies should involve monitoring, for example by regularly executed 3D scans or photogrammetry.

In a speculative study, we explored the effects of introducing some of above-stated features, such as constraining element distributions and manual placement of elements, optionally of virgin materials. As part of this study–a simplified overview of which is shown in Figure 21 – we also looked into the potential of working with different types of production waste. Figure 22 (top) displays a structure designed for a building envelope system with optimized shading capacity. The assembly is generated from a virtual stock of LVL boards. Here, the agency of the evolutionary algorithm would be limited to positioning the short members in a given range of angles, aiming towards optimal horizontal bracing between equally distributed vertical elements. Figure 22 (middle) shows a similar structure, using a stock consisting of CLT offcuts, allowing the rotation of the short members within a limited range. Figure 22 (bottom) illustrates the consequences of fitting longer members in an assembly.

In view of the observed displacement in the demonstrator and the related structural challenges, future work needs to address the connection design, for example by improving the joint design and reducing the assembly gaps. Potential directions include solutions inspired by traditional timber joinery, e.g. wedges, and tenons. To estimate the structural capacity during the design phase, an important future direction is the development of analysis approaches that enable more realistic simulation of timber-only interlocking connections. This would support improved connection design and more reliable prediction of the global structural behavior of the assembly.

As shown earlier, a simple post stiffening method was successfully tested prior to the second exhibition of the wall at Building Green Aarhus in spring 2023. Here, the assembled wall was lifted by its top slab, and thin plywood pieces (4–6 mm thick) with matching thickness were inserted and glued into the assembly gaps. This way, the deviation caused by displacement was successfully reduced to ±10 mm.

The algorithms developed as part of the overall workflow proved to be functional and versatile, with features ranging from structural evaluation to the automatic generation of fabrication data. Nevertheless, the current version is a software prototype that needs further development and fine-tuning. For example, one important future task is the implementation of material stock sourcing solutions from databases. The refinement of the assembly sequencing optimization could ensure that the assembly’s structural capacity is utilized to a higher extent. Moreover, for more controlled results, the user’s options of interacting and defining optimization targets–such as privacy, shading, and views–should be also improved. In its current form, using the workflow requires advanced knowledge in visual programming in Grasshopper 3D, as well as in handling Karamba 3D. To increase accessibility, the workflow should be developed into a standalone plug-in, compatible with established CAD software packages.

In order to enable the further processing and use of wood production waste or reclaimed wood, an efficient automated procedure for digital registration is crucial. Digital tools and workflows, as well as the use of robotic arms, could facilitate this. Relevant data for such a digital repository would among others include the geometrical and structural properties of wood production waste elements. Registering such properties could be achieved with different methods, i.e., (1) by accessing databases of physical material storages, (2) by manual measurement and registration, (3) or digital scanning technologies, such as laser scanning, point cloud processing and object recognition and measurement, which could also be implemented via computer vision. The structural properties can be assessed by combining visual and acoustic quality control, AI -enabled knot scanning, fracture detection, as well as Non-Destructive Testing (NDT) (Tamke et al., 2025). In the case of CLT production waste, the structural properties can be assumed to be identical with those indicated in the product certificate.

Employing digital fabrication and automation, for example with robots, could help making the upcycling of wood waste technically and economically feasible. To facilitate this, additional optimization for assembly and disassembly could be beneficial, especially if such building systems are expected to be more circular, which could be partially achieved by a higher degree of reversibility. Steps in this direction are presented in recent studies (Mangliár and Hudert, 2022; Kunic et al., 2021) on implementing robots for collaborative or autonomous workflows of assembly and disassembly of reversible structures. Finally, robotic automation could also help to deconstruct conventional timber buildings in order to make the embedded components available for new projects, as shown in (Bruun et al., 2022).