A Rule-Based Design Approach to Generate Mass Housing in Rural Areas of the North China Plain

Abstract

Affected by the development strategy of Rural Space Reconstruction in China, the demand for rural mass housing has peaked in the North China Plain in the past 20 years. However, due to the inefficiency of conventional design methods, the rural houses built appear to have a noticeable trend of urbanization and homogeneity. To propose a more effective design approach to change the hitherto unsuccessful homogenized phenomenon of rural design, the study is based on investigating the composition, configuration and characteristics of the dwellings in some traditional villages of the eastern Shandong Province, and it compares and analyzes the differences between conventional methods and generation methods through three design tests: Test 1 is for the reappearance of a general mode of planning, Test 2 is based on the definition of shape-grammar-based rules and Test 3 is mainly used for the optimizing and programming of rules. Furthermore, based on the three prototypes of homestead combination, three-level rules are determined through the three tests mentioned above: Level-1 describes the housing prototype consisting of four homesteads, which generates a variety of spatial relations through the translation of homesteads. Level-2 describes a neighborhood prototype consisting of 16 homesteads, which generates various samples through splitting prototypes and expanding homesteads. Level-3 describes a block prototype consisting of 64 homesteads, which controls open space and identifies a given base during sample filling. Through the analysis of the tests results, the rationality and feasibility of the generative design approach are verified, proving that this approach effectively solves the design monotony problem that commonly exists in rural mass housing in the North China Plain.

1. Introduction

Since 2000, the Chinese government has implemented the development strategy of Rural Space Reconstruction. Through planning and control, the government has led and planned rural settlements and promoted the moderate concentration of the rural population. On this basis, it is necessary to allocate the rural infrastructure appropriately, improve the quality of human settlements in rural areas, and form a spatial structure conducive to urban–rural integration [1]. To make more compact and efficient use of rural land resources in the North China Plain, some local governments have mandated that architects design mass rural housing, which may take the form of either urban multi-story houses with upstairs living spaces or single-door, single-courtyard row houses. However, upon the completion of certain projects, the homogenized and dull image clashed with the original rural aesthetic, leading to scrutiny from many architectural critics. This unified construction of mass rural housing appears to be evolving into a newly emerging architectural typology, thereby impacting the rural construction process in China. Because this type of building belongs to rural projects, its low price makes it difficult for architects to devote a lot of energy, and many projects are designed using simple row houses. Furthermore, in the past, rural houses were typically constructed by individual families, leading to diverse and evolving spatial relationships among households through diachronic evolution. In a unified construction cycle, if the traditional spatial relationships are to be reconstructed simultaneously, the overall planning of these small body and large quantity houses should align with the fundamental elements of traditional styles. This calls for the establishment of a set of efficient design methods, abandoning the conventional blueprint-based design approach.

With this scenario in the backdrop, the current study employed a rule-based approach to the design of rural housings and homesteads to generate a variety of layout patterns that represented the traditional Chinese rural settlement. The results of this study showed pre-defined, phased, hierarchical, and differentiated characteristics. The findings are expected to provide a foundation for the industrialized methods of construction which may further support automated design and mass customization.

2. Literature Review

With the advent of digital design tools, designers’ awareness and handling of complex spatial issues improved significantly. Not only can a digital design produce a personalized response to real-world design issues, but it can also provide a meaningful alternative to repetitive design logic. Based on Gilles Deleuze’s philosophical theory, Oxman [2] summarized five digital design models, viz. the CAD model, digital formation model, generative design model (GDM), performance model, and compound model. Among these models, the GDM is used to define the calculation rules of the formal generative mechanism, and the results obtained are more flexible and can maximize the implementation of interactional design. It has been widely used in the field of digital architecture design recently.

The GDM framework consists of several different design approaches or models, such as cellular automata (CA), genetic algorithms (GA), L-systems (LS), shape grammars (SG), and swarm intelligence (SI). Most of them, like SG, are branches of the rule-based system, which is one of the most basic automation design techniques. Although there are some similarities and overlaps between them, each approach deals with different design issues [3]. For instance, SG is generally used when there is a strong domain of knowledge [4], such as in the study of mass housing layout.

SG can be both descriptive and generative. It can be used both as an analytical tool to describe the formal structure of a corpus of existing designs, and as a synthetic tool to create new languages of design. Initially, SG was used for the generative specification of painting and sculpture [5]. Stiny and Mitchell [6] defined SG that constituted the style of Palladio’s villas and translated them into concrete methods and a generative process. Chiou and Krishnamurti [7] were the first to use SG to study folk houses in Taiwan, which established the fact that SG can be applied to study Chinese traditional architecture. Duarte [8] described the grammar of Alvaro Siza’s houses at Malagueira, and applied an interactive computer system to design customized mass housing. In recent years, several computational design methods have been applied on a large scale. Beirão et al. [9] defined Urban Grammar formally and developed a digital design tool to start the research of a customized city. Therefore, SG is being increasingly used to clarify any design rules of complex architectural designs and urban planning, leading to the building of a multi-level design corpus [10].

Considering Duarte [11] and Wang et al. [12], the residential design world can be described as a hierarchical structure comprising materials, components, parts, rooms, dwellings, housings, neighborhoods, blocks, settlements, and towns, of which the housing, neighborhoods, and blocks have been the focus of the present study (Figure 1). Regarding housing, the individual buildings which comprised similarly shaped elements [13] were analyzed, such as in the case of Erem et al. [14] who added Turkish rural housing grammars to the new rules, which can improve the adaptability of traditional housing after reconstruction. With regard to neighborhood, combinations of similar housing were considered, such as in the case of Herbert et al. [15]. With respect to blocks, mainly the roads, layout, and scale were paid attention to, such as in the case of Dias [16].

3. Methodology

To propose a more effective method, selecting an actual base for design research is necessary. Therefore, Zhuo village in the North China Plain was selected as the target, and an investigation was conducted on it and some surrounding villages to clarify the composition, configuration and characteristics of typical rural housing in the local area. And, then, we qualitatively proposed a housing prototype, core housing, and their relative relationship with homesteads. In the subsequent research on the rule-based approach, on the one hand, some design tests were conducted using the base of Zhuo village. On the other hand, a series of applicable rules have been proposed, some of which come from the investigation results and some from the professional experience of the researchers.

3.1. Investigation and Analysis Approach

The first method was a field investigation. In this method, 16 villages were selected which are distributed in the plain areas of the Wulong River, adjacent to Laiyang City of the Jiaodong Peninsula (Figure 2). These villages are all urban fringe villages with a history of more than 200 years and comprising 100 to 2000 households. Due to the rapid surge in demand for urban entertainment and rural tourism in recent years, these villages have gradually transformed their traditional farming to experiential agriculture. In our initial investigation, Google Earth images were used to analyze the boundary, road, layout and scale of each village. Next, through field research, particularly the texture of the Zhuo village was focused to systematically analyze the three levels—block, neighborhood and housing. At the block level, the roads were divided into main roads with a width of about 6 m and alleys with a width of about 3–4 m. The main road is used for household and agricultural vehicles, and the alleys are primarily used for walking. Tortuosity and cul-de-sacs are the two basic forms of the roads; the homesteads adjacent to the main road are larger because they are often used for commercial services. With regard to the neighborhood aspect, most of the blood-related families were found to live next to each other. This is due to the influence of the traditional family culture. Later, as the economy developed and population mobility increased, the closely related families split into single-family households with a small courtyard. At the housing level, five typical fragments consisting of four detached houses were selected for diagram and transformation to designate a housing-level prototype (Figure 3). In addition, the core housing faced south with a maximum of 2 floors and rooms having widths between 3 and 3.3 m. The shape of the homestead mainly included two types: long rectangle in a vertical direction and approximate square. Furthermore, the gable of the core housing and the northern boundary of the homestead were usually not provided with any gates.

3.2. Rule-Based Approach

The second method was the automation of rule-based design, which was divided into three stages (Test-1, 2, 3) of progressive implementation. Specifically, the second method included the definition of rules, implementation of rules, comprehensive analysis and the encoding of rules. (1) In the field investigation, we used SG-based rules to derive the different housing prototypes. We also defined the rules for non-SG (non-Euclidean transformation) at all the 3 levels—the housing, neighborhood and block levels (homesteads from 4 to 16, 64). Moreover, the combination rules and filling rules were also defined. (2) Both qualitative and quantitative analyses were performed to verify the results of every test. In the qualitative analysis, the results of the tests were compared with traditional settlements, which showed whether the test results had the characteristic form and organization of a traditional residential space. In the quantitative analysis, Space Syntax was applied. The Space Syntax can be used to quantitatively analyze the spatial relationships implied in environmental design, interpret the impact of urban form and land use for pedestrians and bicycle transportation, and assist in decision-making in the planning and design of future environments. The integration value of the tests results was measured, and the high integration value represented high reachability and great potential to become a center [17]. (3) Furthermore, using the preliminary results, the filtration rules, classification rules, and adjustment rules were set to optimize the whole design process. The rules were coded via algorithms using Python and applied in a computer program to develop the design corpora. Subsequently, we selected a real base from the Zhuo village which needs to be renewed and used Pyautocad to identify its boundaries and roads. Samples from the corpora were then randomly and automatically filled into the base following the developed rules in order to realize design automation (Figure 4).

4. Predefinition

A basic module of 3.3 m was used to divide the grid of the homestead; the widths of the main road and the alley were considered as 6.6 m and 3.3 m, respectively. The length of the standard homestead (H0) in the north–south and east–west directions were 13.2 m and 9.9 m, respectively, and the area was 130.68 m². The given base, as shown in Figure 5, comprised an abandoned gated factory which was surrounded by walls. Since this land is not residential, the scope of this was excluded from the design tests. It should be emphasized that centrality is an important factor in traditional Chinese rural settlement. Therefore, in the analysis of Space Syntax, the results of tests which had a higher integration of the central area were better.

5. Test 1 (Reappearance of General Mode of Planning)

Test 1 aimed to make the prevailing rural housing planning methods in China reappear. This process only involved the housing-level-based design rules, mainly by arranging and combining standard homesteads horizontally and vertically for layout.

5.1. Filling Rules

For a and b, starting from the left border, copy three (Figure 6a) or four (Figure 6b) of them horizontally to the right side and add a vertical road; repeat until the given base is filled. For c, starting from the upper border, copy two of them vertically to the lower side and add a horizontal road; repeat until the given base is filled (Figure 6c).

5.2. Implementation and Analysis

The filling rules were applied to the given base, generating three different plans to build the 3D models. Qualitatively, Figure 7a,b show the orderly form of row housing and homogeneous landscape, which displayed strong contrasts with the different styles and features of traditional villages. From a practical point of view, although this approach was conducive to mass production, the residential demands of each homestead could not be maintained as completely consistent, thereby losing the features of traditional villages. Quantitatively, while using Space Syntax to measure the integration value of the three results, the integration value showed an increase from the outer area to the central area, which indicated that the results were good with respect to integrity, centrality and enclosure (Figure 7c).

6. Test 2 (Definition of SG-Based Rules)

More complicated rules were defined in Test 2 to address the problem of homogeneous design in Test 1. In order to improve the flexibility of the results, we not only defined the SG-based rules for housing level, but also defined the combination rules and filling rules for neighborhood and block level (Figure 8).

6.1. Housing Level: SG-Based Rules

In the housing prototype, the homesteads can be shifted horizontally or vertically by one module to form new shapes. Therefore, we used SG to define the rules to generate all the samples at the housing level. The main shape rules used are as follows (Figure 9):

1. Label rule. Labeling homesteads of the initial shape indicated, on which the transformation rules could be applied.

2. Axis rule. This rule placed a grid over the initial shape and controlled the application of transformation rules, along with Euclidian transformations when translating the labelled (·) homesteads.

3. Transformation rules. The labels only allow for each of the homesteads to be acted on and can only be applied once to a selected homestead. Therefore, each homestead in the initial shape can only be manipulated once with the application of the four transformation rules.

4. Subtraction rules. As the purpose of the grammar was to facilitate samples of housing level development, the generation of overlapping or embedded shapes should be avoided. This is also supported by the application of subtraction rules which eliminate a complete permutation if overlapping or embedded shapes occur with the transformation of the smaller rectangles.

As permutations are progressively developed, they were categorized according to the number of applications of each of the four transformation rules. In the present study, the application of the shape rules and the subtraction rules generated 79 different samples of housing level—Set T (Figure 10).

6.2. Neighborhood and Block Level: Non-SG-Based Rules

1. Combination rules. The samples in Set T were randomly extracted and combined, like the neighborhood prototype, to form a new sample Set N. The samples in Set N were randomly extracted and combined, similarly to the block prototype, to form a new sample Set B.

2. Filling rules. The samples in Set B were collected and filled on the given base, and each sample was separated by a road. Moreover, during filling, 9 of the 64 homesteads were randomly transformed into open space.

6.3. Implementation and Analysis

We collected the samples from Set B and implemented the filling rules on the given base. Any extra homesteads beyond the boundary were removed to generate three different plans and build 3D models. Qualitatively, Figure 11a, b displays a repetitive and different form which has the traditional rural style and features. This shows good similarity and continuity in the results. Some shapes were identified from the traditional housing layout; however, there was only one size of homestead which did not meet the different residential demands. The results obtained were contingent, and reflected some errors that the gate could not be set up at the homestead (red-marked in Figure 11a). Quantitatively, the integration value of Test 2 was unevenly distributed within the given base, which means that the results did not show good centrality and enclosure, and the space was comparatively negative (Figure 11c). This is because some samples lacked the definition of restrictions or rules related to road connectivity in the combination process.

7. Test 3 (Optimization and Programming of Rules)

7.1. Predefinition

Based on the results of Test 1 and Test 2, the rules were optimized and coded using a computer program using Pyautocad. The samples in Set T and the housing and neighborhood prototypes were retained. Furthermore, to meet different residential demands, five sizes of homestead were predefined, including one standard type (H0), one small type (H1) and three large types (H2, H3, H4) (Figure 12a). Additionally, the block prototype was modified by adding north–south and east–west roads at the center (Figure 12b).

7.2. Level-1 Rules (Housing Level)

1. Filtration rules. Limited by factors such as walking accessibility and gate position, Set T did not support the survey conclusions fully. Therefore, some samples were removed by setting new filtration rules, which are as follows: (1) as long as there is a homestead not close to an internal alley, filtering was necessary, (2) if the gate can only be opened on the north side of the homestead or the gable wall of the core housing, it must be filtered and (3) delete the sample of housing prototype. After filtering, 16 samples constituted the level-1 corpus (T0) (Figure 13a).

2. Classification rules. The level-1 corpus was divided into four quadrants for combination (ZONE-A1/A2/A3/A4), following which the four corner points could be accessed by the internal alleys. The samples that had access to the southeast corner point were placed in ZONE-A1, the ones with access to the southwest corner point were placed in ZONE-A2, while the ones with access to the northeast and northwest corner points were placed in ZONE-A3 and ZONE-A4, respectively. Since some samples’ internal alleys could reach two corner points, they qualified to be placed in two quadrants equally. Also, to enrich the styles and features, the samples in each quadrant were divided into subsets M and N based on whether there were alleys running through the samples vertically (Figure 13b).

7.3. Level-2 Rules (Neighborhood Level)

1. Combination rules. First, in the neighborhood prototype, the level-1 samples of the four quadrants (ZONE-A1/A2/A3/A4) were extracted to fill. To avoid the monotonic problem caused by combinations, the following constraints were imposed during filling: (1) no adjacent ZONEs would have a mirror combination of the same samples and (2) no samples from M subsets of ZONE-A1 and ZONE-A3 can be adjacent to each other. Similarly, no samples from M subsets of ZONE-A2 and ZONE-A4 can be adjacent to each other. After imposing the above constraints, 148 combination shapes were obtained (Figure 14a).

2. Filtration rules. To improve accessibility, as long as one side of the neighborhood samples could not open the entrance to access the center, filtration became necessary. After filtering, 102 level-2 shapes were generated (Figure 14b).

3. Adjustment rules. Homesteads facing the main road often have a larger area, mainly used for selling daily necessities. There are also a small number of large homesteads in the inner areas not facing the road. Therefore, we used three larger homesteads (H2, H3 and H4), to accommodate the above possibilities. At this level, it is necessary to discuss the two cases of facing road and non-facing road. In advance, we should consider the following conditions: (1) a homestead at a corner point which did not face the main road should be selected as a labelled homestead; (2) the level-2 shapes with three or four sides facing the main road are not discussed. Furthermore, six types of adjustment rules were employed to generate the level-2 corpus, as described below.

• Adjustment rules—T1. When only one side faced the main road, five standard homesteads turned into large homesteads, as shown in Figure 14c;
• Adjustment rules—T2. When two sides faced the main roads, seven homesteads became large homesteads, as shown in Figure 14d;
• Adjustment rules—T3. If no side faced the main road, three homesteads became large homesteads, as shown in Figure 14e;
• Adjustment rules—S1. Based on T1, the samples should be cut in half and the parts with the labeled homestead should be retained, as shown in Figure 14f;
• Adjustment rules—S2. Based on T3, the samples should be cut in half and the parts with the labeled homestead should be retained, as shown in Figure 14g;
• Adjustment rules—S3. Based on T3, the samples should be cut in quarters and the parts with the labeled homestead should be retained, as shown in Figure 14h.

Among these, the adjustment Rules S1, S2 and S3 are able to generate smaller samples of the neighborhood for filling the narrow or patchy space of a given base.

7.4. Level-3 Rules (Block Level)

The level-3 rules are the control rules for filling a given base to generate a block, so no level-3 samples were generated. We filled the block prototype facing the main road with the corpus of T1, T2 and T3 to verify how the samples of the level-2 corpus recognized the main road. Then, the following rules of reduction and adjustment control were set to recognize the number and area of homesteads in the block, and form of the block’s open space.

1. Filling rules. When one side is facing a main road, the corpus of T1 and S1 can be used. When two sides are facing the main roads, the corpus of T2 and S1 can be used. When no side is facing the main roads, the corpus of T0, T3, S2 and S3 can be used. To improve the accessibility of traffic inside and outside the block, roads should also be reserved between the level-2 samples (Figure 15a).

2. Reduction rules. Open space is necessary for a block, so some homesteads must be removed. This includes the following two cases: (1) homesteads that are facing the main road could be block prototype entrances, and (2) homesteads that are not facing the main road could be open spaces for landscaping. Therefore, reduction rules are an effective method for dynamically adjusting the number of homesteads according to specific situations (Figure 15b).

3. Adjustment rules. To further increase the type of homesteads, standard homesteads (H0) at the center of a block can be transformed into small homesteads (H1) (Figure 15c).

7.5. Implementation and Analysis

By analyzing Test 1 and Test 2, more reasonable rules of the three levels can be redefined in Test 3 and coded using a computer program to generate the planning solutions. These rules and the resulting corpus are more flexible and can adapt to more complex situations of construction land. For a given base, samples are collected from T0, T1, T2, T3, S1, S2 and S3 for filling according to these rules, and three different results (2D and 3D) were generated randomly (Figure 16a,b). We used Pyautocad to automate the design in AutoCAD. The specific operating principles of the program designed in Test 1-3 are as follows:

1. Sample representation. As the module system was predefined in advance, a matrix of numbers can be used to represent the positional relationships between the homesteads, alleys and roads. For example, the level-1 samples were represented by a 9 × 7 matrix, where 1 represents the homestead and 0 represents the alley.

2. Sample filtration. In terms of filtration rules of level-2 rules, the sample matrices were accessed one by one using Depth-First Search (DFS), complying to the rules which allowed us to identify the unreasonable samples, such as one side of the neighborhood samples cannot open the entrance to access the center.

3. Sample combination and classification. During the combination process, the low-level samples were classified first via DFS, and the prerequisites for the high-level combined samples were analyzed using algorithms. The low-level samples were then combined with the high-level ones following matrix combinations.

4. Homestead identification and filling. When the rules were applied to a given base, the level-2 corpus should be used first, identified and filled according to the corresponding layers (base boundary and main road). Then, the level-1 corpus can be called to fill the remaining small space, in order to complete adaptive filling at the given base. Moreover, in response to the design goals, some parameters were set to control the results. For example, to render the main road accessible, the filling probability of the S1 and S3 samples can be increased to identify the boundary of the base and exclusive layer; the filling probability of the T3 and S2 samples can be increased.

Qualitatively, the results of Test 3 were similar to those of Test 2, showing the appearance of traditional rural settlements. Quantitatively, the results of Test 3 were similar to those of Test 1. The integration value of these results was higher for the central area, which meant that the central area had high accessibility (Figure 16c). In general, the results of Test 3, based on automatic design, better represented the characteristics of traditional rural self-organization and this is more reasonable in road planning.

8. Discussion

The three tests, T1, T2 and T3, applied to the given base, with a small area and relatively regular shape, were found to be suitable for the rapid qualitative and quantitative analysis of the results of reappearing the characteristics of traditional rural settlements. However, due to the regular shape and small size of the given base, it could not be proved that a feasible layout could be generated when dealing with a larger and more irregular site. In addition, due to a small number of samples collected at all levels, it was impossible to show the mutual influence of a large number of samples at all levels during filling. As a result, to further verify the adaptability of the application of this design approach, we selected a new base (Figure 17a) for a test which had more complex site conditions. The corpus and rules of Test 3 were used. The main road and site boundary in AutoCAD were automatically identified by the computer program (Figure 17b), and the layout of the homestead was also predefined and randomly generated (Figure 17c). Both qualitatively and quantitatively, the results of the test successfully retained the traditional Chinese rural style and characteristics (Figure 17d, e). Moreover, the random filling and mutual influence of a large number of samples at all levels created a flexible and diverse road network at the block level (Figure 17b). Therefore, the test proved that the corpus and rules can reasonably and accurately generate the housing layout in accordance with the traditional rural characteristics, even on a large-sized irregular site. However, due to the size limitation of the samples, some small areas could not be filled (Figure 17b, c). In addition, since the samples have regular rectangular boundaries, they may not be well adapted to sites with various actual complex conditions, such as curved or oblique boundaries. In future work, we intend to improve the three-level rules, enrich the corpus and filling rules and make them adaptable to more complex topographic conditions, such as altitude difference and irregular site boundary. Moreover, since the main roads and alleys are only considered in this work, we plan to improve the communal-facilities-based rules corpus at the settlement level and town level, with a view to developing the generative approach.

9. Conclusions

This study explores a rule-based design approach for rural mass housing, mainly based on the results of three tests. The optimized rules and condition restrictions were coded, and a relatively reasonable result was obtained via the automatic operation of the program. Through a comprehensive analysis of the tests, the following conclusions can be drawn. First, this rule-based generative design approach can help architects solve the design problem of the homogenization of rural mass housing, and the design results have “harmony in diversity” characteristics. The method keeps architects’ role in the housing design process. It acts as a research booster, helping architects systematically explore the logic and process of generating research objects. Secondly, compared with traditional design methods, it can achieve the “emergence” and “unpredictability” of design results through rule-based reasoning, which can better meet the needs of multiple parties, especially associating architects and users. It can not only help architects use computer-aided tools to select and optimize the generated results and better control the geometric size, entrance direction and other parameters to realize the traceable and targeted research process, but also allow users to adjust the layout of their houses and their built-in building information according to their needs, depending on the family’s life cycle. Thirdly, as a visual design tool based on shapes, SG has the advantages of processing multiple complex rules simultaneously, directly applying rules to functional geometry and better perceiving and expressing complex shapes. An effective relationship between abstract “spatial demands” (traditional architectural features, user design demands) and specific design rules can be established. Architects can apply this approach to control the design generation process, optimizing the design quality and improving the design efficiency.

10. Patents

The following Chinese invention patent was produced: “A Generative Design Method for Mass Housing Layout in Rural Areas of the North China Plain”, with the authorization number “ZL202010327152.8”.