1. Introduction
Its economic and social relevance notwithstanding, the construction sector in the European Union (EU) consumes a great amount of resources, with figures up to 50% of all extracted materials [
1], and produces vast quantities of waste, accounting for 37.5% of the total [
2]. Moreover, the built environment is responsible for 40% of the total energy consumption and 36% of all CO
2 emissions in the EU [
3].
Among some of the most commonly employed construction materials are concrete and steel. Understanding the properties and constraints of materials is essential to both the design and future construction and use; however, evaluating the environmental impact has become equally critical as climate change and sustainability concerns increase. Thus, the use of life cycle assessment (LCA) databases allows for the assessment and comparison of construction materials’ environmental performance through the use of indicators such as embodied energy or CO
2 emissions. For instance, the Ecoinvent database [
4] indicates that steel has an embodied energy of 27.90 MJ/kg and emissions of 1.71 kg CO
2/kg. Similarly, values of 0.618 MJ/kg and 0.112 kg CO
2/kg are reported for concrete [
4].
Nowadays, similarly to other economic sectors, construction is veering towards practices within material selection, construction methods, operational efficiency, and end-of-life strategies that regard social and environmental well-being to further sustainability. In this regard, the European Green Deal [
5] promotes the use of timber as a means to reduce the environmental pressure of the construction industry.
Although the use of timber is not new, approaches to alleviate some of the inherent limitations of wood (i.e., dimensions, strength, instability, fire resistance, susceptibility to biodegradation, etc.) through the development of wood-engineered products [
6,
7] and various technological advances (e.g., chemical, thermal, or mechanical treatments) [
6,
8,
9] are propelling its current popularity. Moreover, another factor driving the increase in use is the comparatively better environmental profile with respect to other construction materials [
10]. As a renewable resource, timber can be continually sourced from sustainably managed forests, which contributes to the maintenance and expansion of forested areas, thereby mitigating deforestation and promoting ecological quality [
11]. Since it originates from a photosynthetic organism, uptake of carbon occurs during its growth due to biomass conversion, with approximately 1.5 t CO
2/m
3 of wood [
12]. Thus, sequestration or carbon storage could be considered in the manufacture of timber, provided that the harvested tree is used in long-life products such as construction materials. Moreover, the use of timber in substitution of other more environmentally damaging materials constitutes an additional benefit in CO
2 mitigation, since the manufacturing emissions of the replaced construction material are avoided. After sawing, lumber is naturally or force-dried to achieve dimensional stability and planed according to use or processing into other wood-engineered products. Undoubtedly, these operations pose a negative impact on the environment. Although precise figures vary for the different wood products, both carbon emissions and embodied energy are significantly lower in timber construction [
10,
13,
14,
15,
16].
Among the wood-engineered materials, this research focuses on glued laminated timber (also known as glulam). This construction material is composed of layers of sawn lumber bonded together, with the grain running parallel to the length of the structural element. This manufacturing allows for large spans and variable cross sections, as well as high strength-to-weight ratios. Nevertheless, the different strength classes (e.g., from GL 20h to GL 32h for homogeneous glulam [
17]) exhibit specific mechanical performances relating to the wood species. For instance, lower strength classes could be associated with softwoods (
Thuja plicata,
Picea sitchensis,
Abies magnifica,
Abies grandis,
Abies concolor,
Abies procera,
Abies amabilis, etc.), whereas greater strength classes are connected to
Pinus sylvestris,
Larix decidua,
Pseudotsuga menziesii, etc. [
18].
In this regard, the scientific community has predominantly centered its attention on the study of a single strength class, and the interest has not been uniformly distributed among them. Conversely, it is worth mentioning the research by Baranski et al. [
19], who addressed the optimization of different beam geometries as well as different qualities of glulam (i.e., GL 22h, GL 24h, GL 26h, GL 28h, GL 30h, and GL 32h).
From the literature analysis, GL 24h stands out as the most evaluated class, with a greater incidence of studies on bending elements such as beams. In the context of structure optimization, two investigations can be highlighted. Firstly, Jelušič and Kravanka [
20] assessed the optimization design of timber floor joists for a given imposed load and span of the structure. Then, De Vito et al. [
21], who developed a topology optimization of Douglas fir GL 24h beams, considered the orthotropic nature of wood and the layering manufacture. Kilincarslan and Turker [
22] considered the strengthening of the GL 24h spruce timber column–beam connection with carbon fiber-reinforced polymer through experimental evaluation. Wang et al. [
23] evaluated the stiffness of GL 24h timber from Scotch pine beam–column joints with bolted connections and introduced a stiffness prediction method. Moreover, in the current state of transition to more sustainable materials, there are also some examples of the hybrid use of GL 24h with inert materials. Fu et al. [
24] studied the optimization of the bonding performance between prefabricated concrete and GL 24h timber from spruce. Similarly, Giv et al. [
25] studied the effect of adhesive type on the bending behavior of the GL 24h timber–concrete composite panel. Ferrara et al. [
26] conducted real-scale experiments on the mechanical performance of a GL 24h timber–concrete composite floor supported at two edges. Gomez-Ceballos et al. [
27] presented a numerical method for the analysis of the mechanical behavior of GL 24h Douglas beams reinforced with ultrahigh-performance fiber-reinforced concrete. For the timber–steel conjunction, Ching et al. [
28] developed a topology optimization framework for trusses made of GL 24h timber and steel aimed at reduce global warming potential.
Several authors have also assessed the GL 28h class, mostly in the context of reinforced elements such as tendons or steel bars. For instance, De Luca and Marano [
29] tested the failure of prestressed GL 28h spruce timber reinforced with steel bars. Also for European spruce, McConnell et al. [
30] tested GL 28h timber with steel tendons in post-tensioning conditions. In regard to optimization, Mam et al. [
31] focused on GL 28h bracing structures. For a timber–timber composite with glulam ribs and cross-laminated timber flanges, Suárez-Riestra et al. [
32,
33] tested the GL 28h from
Picea abies and proposed an estimation model for the long-term behavior of the composite.
Amid the less studied strength classes, an investigation carried out by Jelušič [
34], who proposed an optimization approach to variable cross-section beams of GL 30h timber should be mentioned, as well as the research performed on GL 32h timber by Šilih et al. [
35] for timber trusses and Simón-Portela [
36] for an entire timber roof structure.
Despite the current clear focus on specific strength classes, the complete consideration of the strength class range within the material selection could benefit the optimization of the timber volume required for a specific structure, which is one of the objectives of this research. In this regard, there is extensive literature on reducing material consumption in the design of steel and concrete structures; some examples can be found in [
37,
38,
39]. Albeit more limited, optimization approaches have also been made for timber construction, such as beams [
19,
34,
36] and trusses [
20,
35,
40,
41].
Additionally, the consideration of the use of the strength classes could also result in advantages from a sustainability standpoint, since a greater variety of wood species would be considered for construction purposes. It should be noted that besides their performance and classification within a strength class, the selection of available tree species would also be motivated by their intrinsic characteristics (e.g., climatic adaptation, growing rate [
34], etc.) as well as the final acquisition cost of the material. In this regard, research shows that tree species richness can enhance wood productivity while maximizing ecosystem functioning [
42,
43,
44]. Notwithstanding, to adequately supply the increasing demand for wood as a construction material, the source material must originate from responsibly managed forests.
Therefore, the present investigation assesses the influence of glulam strength classes in timber construction, specifically, on the design of a timber roof structure consisting of timber double-tapered beams and purlins according to Eurocode 5 [
45] requirements. To that end, an optimization tool based on genetic algorithms has been developed, and based on the results, several equations have been proposed to determine the optimal geometry (width and height) of the structural elements (beams and purlins) as well as their spatial configuration given the roof length, span, snow load, and strength class. Similarly, a general equation has been derived to predict the optimum volume required for the roof structure considering the different strength classes, which would promote efficient use of resources and economic advantages while complying with structural and safety requirements.
2. Materials and Methods
This section presents the materials and methods employed throughout the research. Firstly, to establish a comprehensive framework of construction solutions relevant to the European context, the range of study for the different design parameters (material properties, geometric dimensions, and load scenarios) is considered. Subsequently, the specifics regarding the development of the optimization tool are detailed. On the one hand, the Eurocode 5 [
45] equations utilized to evaluate the potential of each proposed design solution are mentioned. On the other hand, the concrete parameters within the genetic algorithm, as well as the objective function and the associated penalization criteria implemented, are defined. Finally, the statistical analysis conducted on the optimal solutions to derive the optimization equations intended to serve as a decision tool is described.
2.1. Design Parameters: Material, Dimensions and Loads
A roof structure comprised of timber double-tapered beams and purlins has been examined in this research. For the assessment, six homogeneous glued laminated (glulam) timber strength classes, as defined in EN 14080 [
17], have been considered: GL 20h, GL 22h, GL 24h, GL 26h, GL 28h, GL 30h and GL 32h (
Table 1). This broad selection of strength classes would allow for consideration beyond the traditionally used wood species and thus the promotion of previously unexploited or underutilized species in timber construction.
The optimization was carried out for different timber roof dimensions. The roof length was studied at 30 m, 45 m, 60 m and 75 m; the assessed span varied from 15 m to 30 m in 1.25 m increments; and inclination angles between 5° and 10° were examined for the tapered beam. The geometry of the structural elements was also considered a variable within the experimental program. The heights of beams (Hb) and purlins (Hp) ranged from 120 to 1200 mm, with 40 mm intervals reflecting the laminate thickness. Similarly, the widths for beams (Wb) and purlins (Wp) spanned from 90 to 220 mm, adjusted in 10 mm increments. Finally, regarding the arrangement of the structural elements within the roof structure, the spacing between beams and purlins was also studied as a variable. Beam spacing (Sb) was considered within a range of 3 m to 7 m, while purlin spacing (Sp) varied from 0.625 to 1.25 m, which was the maximum allowed separation.
Similarly, specific limits were established regarding the range of potential load conditions that the roof structure is designed to withstand. All contemplated loads were combined as established in Eurocode 1 [
45] to assess the worst-case scenario in the structural integrity verification. As the roof elements fall within service class 1, a 1.25 safety factor was considered for the glulam material combined with a modification factor (k
mod) of 0.9. For permanent surface loads, in addition to the self-weight of the beams and purlins that was automatically calculated during the optimization process, a dead load of 0.45 kN/m
2 was also included. For variable loads, snow loads ranging from 0.4 to 3 kN/m
2 were considered to accommodate typical values in different locations across the European Union [
46]. It should be noted that in accordance with the Spanish Technical Building Code (also known as CTE) [
47], the combination of snow and wind loads with the maintenance load is avoided due to the assumption that during extreme conditions, no one would be present on the roof, thus resulting in an overestimation of the total load on the structure. Conversely, snow load values below 0.4 kN/m
2 were excluded to ensure that in scenarios with negligible snow loads, the maintenance load is applied. Regarding the wind load, it was considered for a gable roof without openings in a building with a height of 5 to 7 m [
48], and since it has been proven not to affect the optimization results [
36], a fixed value of 0.07 kN/m
2 was employed.
2.2. Optimization
The optimization goal is to identify the most effective design for a roof structure, taking into account six different glulam strength classes that reduce timber material consumption and comply with the requirements of strength, stability, and stiffness set out in Eurocode 5 [
45]. As such, initially, this method aims to balance structural performance (i.e., function and safety) with material conservation. However, it is worth mentioning that the interpretation of the outcomes across the different strength classes should also be regarded as a sustainability equalization, e.g., the selection of an optimal result from a different strength class that could be represented by less exploited wood materials, which alleviates the demand for certain species while promoting biodiversity.
Genetic algorithms grounded in biological evolution principles were employed to execute the optimization process (
Figure 1).
To represent different potential designs for the roof structure, an initial population of 1200 random individuals was generated. The selected population size is a compromise between the computational cost and the probability of identifying the optimal solution. The possible dimensions (i.e., Hb, Hp, Wb, Wp, and angle) and arrangements (i.e., Sb, Sp) of beams and purlins were considered through specific values in their chromosomes. Each individual within the population corresponds to a potential structural solution, encoded in a chromosome comprising seven genes: Hb, Hp, Wb, Wp, Sb, Sp, and angle. Then, the evaluation of each design solution depended on the utilization ratios determined within the structural calculation program developed in Matlab to verify the Eurocode 5 [
45] criteria, which was previously collected in [
36], as follows.
Then, the modified objective function (F(x) in Equation (1)) measures the fitness of individuals in terms of volume, in m
3, and applies a penalty to those that do not meet the structural safety criteria and according to the aforementioned utilization rates:
where x denotes an individual within the study population, f(x) is the objective function in terms of volume (m
3), j is the number of variables under examination, Pj is the penalization term conforming to the restrictions imposed on each structural element, and Gj is the penalty parameter, with the same order of magnitude as the objective function, based on the maximum utilization ratio observed in each structural component, with j assuming values from 1 to 4, as follows.
G1(x) is the highest ultimate limit state utilization ratio of the beam:
0 > G1(x) > 1 then, P1(G1(x)) = 3 × 10(1−G1(x)).
G1(x) = 1 then P1(G1(x)) = 0.
G1(x) = 0 then P1(G1(x)) = 40.
G1(x) > 1 then P1(G1(x)) = G1(x) × 400.
G2(x) is the highest serviceability limit state utilization ratio of the beam:
0 > G2(x) < 1 then P2(G2(x)) = 1.8 × 10(1−G2(x)).
G2(x) = 1 then P2(G2(x)) = 0.
G2(x) = 0 then P2(G2(x)) = 24.
G2(x) > 1 then P2(G2(x)) = 60 × G2(x).
G3(x) is the highest ultimate limit state utilization ratio of the purlin:
0 > G3(x) > 1 then P3(G3(x)) = 1.725 × 10(1−G3(x)).
G3(x) = 1 then P3(G3(x)) = 0.
G3(x) = 0 then P3(G3(x)) = 23.
G3(x) > 1 then P3(G3(x)) = G3(x) × 11.5.
G4(x) is the highest serviceability limit state utilization ratio of the purlin:
0 > G4(x) > 1 then P4(G4(x)) = 172.5 × 10(1−G4(x)).
G4(x) = 1 then P4(G4(x)) = 0.
G4(x) = 0 then P4(G4(x)) = 23.
G4(x) > 1 then P4(G4(x)) = G4(x) × 11.5.
Then, the individuals with the lowest values are roulette-selected to be subjected to two-point crossover (i.e., the combination of their characteristics), elitism (i.e., the retention of the fittest individuals) and mutation (i.e., the introduction of variability) operations in order to generate new individuals that replace the initial population. Values of 80%, 10% and 1% were employed for these operators, respectively. The cycle is repeated until convergence is reached, or up to a maximum set value of 50 generations. Nonetheless, convergence was always reached after 15 to 20 generations. For instance,
Figure 2 showcases the characteristic evolution of the modified objective function across successive generations.
2.3. Statistical Analysis
A total of 1792 optimal individuals were generated based on combinations of specified values for span, depth, snow load, and strength class. A comprehensive statistical analysis was performed on the attributes of these configurations to assess for each glulam strength class, the optimal geometry (width and height) of the structural elements (beams and purlins) comprising the roof structure, and their spatial configuration. Additionally, the relationships among the different considered variables (i.e., snow load, span, depth, beam height and width, purlin height and width, beam spacing, and purlin spacing) as a function of the glulam strength class were examined through multiple linear regression analysis. The analysis also focused on the comparison of the optimum timber volume across different strength classes, offering a basis for making informed decisions based on both the material consumption (i.e., type and amount) and the cost-effectiveness. Therefore, several equations have been formulated to enable reliable predictions regarding the optimal volume and geometry parameters for a given span and roof length, snow load, and timber strength class. The validity of each predictive equation was assessed through the normality and homoscedasticity of the residuals as well as the correlation value.
4. Conclusions
This research focused on the optimization of roof structures comprised of glulam beams and purlins of different strength classes and exposed to varying snow loads. The developed genetic algorithm tool, along with the structural calculation program, was employed to generate 1792 optimal roof structures with dimensions (i.e., roof length and span) and load conditions (i.e., snow loads below 3 kN/m
2) typical of European construction. From a systematic statistical analysis, several predictive equations were proposed as a reliable method for optimizing timber roof design in accordance with Eurocode 5 [
45]. Firstly, the predictive model could be used to optimize the beam width and height as well as their spatial arrangement as a function of the roof dimensions, loads, and the desired strength class. Additionally, the developed model also enabled the determination of the overall optimal timber volume required for a roof structure as a function of the strength class. Although this approach seeks material usage reduction and economic savings, it ultimately could include sustainability considerations in the decision-making process. Since an optimal solution could arise from timber pertaining to various strength classes, it would allow for diversification in the wood species selected or adaptation to locally available materials. Thus, such a strategy would assist in the promotion of the use of alternative wood species, which would further biodiversity and future resilience against climate change.
Among the findings, it was found that double-tapered beams exhibited their optimal inclination angle at 5°, leading to a 30% reduction in the total timber volume used for the roof structure compared to an inclination of 10°. The optimal spacing between purlins coincided with the maximum set value, i.e., that allowed by the roofing material. This pointing to consideration of non-traditional roofing materials as a future optimization strategy.
For structures exposed to snow loads greater than 2.5 kN/m2, a similar timber volume requirement, with at most a 3% difference, was noticed for most strength classes (GL 24h, GL 26h, GL 28h, GL 30h, and GL 32h). Thus, suggesting the existence of a broad range of wood species that effectively could rival the most commonly employed (GL 24h and GL 32h). Similarly, for snow loads greater than 1 kN/m2, GL 30h and GL 32h classes also resulted in similar consumption of timber volumes (up to 1%). Across all strength classes, the largest differences in volume occur at span values close to 15 m and for lower snow loads, with differences up to 20% for GL 20h compared to GL 32h. In any case, a strong relationship exists between the increase in the timber volume and the reduction in the strength class or the modulus of elasticity.