Constraint specification in architecture : a user-oriented approach for mass customization

The last several decades, particularly those after the end of World War II, have seen an increasing industrialization of the housing industry. This was partially driven by the large demand for new houses that resulted from the baby boom and the destruction caused by the war. By adopting mass production the building industry was able to meet the demand. A downside of this development is a decrease in the amount of input buyers have in the design of their house. Due to the economies of scale that are the basis of mass production, there is a tendency to make houses identical as much as possible, reducing the price at the cost of flexibility. This decrease in consumer choice is common when an industry shifts to mass production, as illustrated by Henry Ford’s famous quote about the Model T Ford: "Any customer can have a car painted any colour that he wants — so long as it is black." Currently, people buying a new house typically have to either accept the architect’s design, or they are offered a limited amount of alternatives to choose between. The limited number of these choices and the fact that in most cases they are not based on input of the buyers, however, mean that the alternatives do not match the buyers’ needs as well as they might. This in contrast to many other industries, such as the food, fashion and automotive industries where customers are presented with a wide range of alternatives. These industries prove that mass production does not necessarily result in less flexibility. This mixture of mass production and customization is called mass customization. In chapter 2, mass customization is explored in more detail by looking at the various categorizations that have been made of mass customization. Different subcategories can be identified based on criteria such as the degree of standardization or the extent to which the client is involved in the design process. Subsequently, the use of mass customization in the building industries is examined. Although not particularly common, there have been studies into this phenomenon, among others by famous architects such as Le Corbusier and John Habraken. In order to make the adoption of mass customization in the building industry more feasible from a design perspective, it is preferable to use a design representation that allows for a more high-level view of the design than the traditional collection of two-dimensional lines. The ability to refer to building elements such as walls and dormers as a whole makes modifications easier to perform and greatly simplifies automated model verification. For this reason, a discussion of Building Information Models in general and IFC, as the de facto standard for BIMs in the architecture domain, in particular is included in this chapter. As mentioned, BIMs facilitate automated model verification. This is particularly useful in relation to mass customization, since part of the design process will now be performed by non-expert users. It is to be expected that the designs made by the buyers will not be in full accordance with either the building codes or the architect’s intentions. Checking the designs for problems can be done manually, but this is a laborious and error-prone process. It would be preferable to be able to automate this checking process and ideally to perform it in real time while the buyer is modifying the design. This way the architect can be guaranteed that the resulting designs have no trivial problems, freeing him up to look at more complicated criteria that are difficult or impossible to automate, such as aesthetics. Automated model verification is performed by establishing constraints that the design must satisfy. In chapter 3 the use of constraints in other industries, such as software and electrical engineering, is discussed. So far, use of constraints in the building industry has been limited. Some projects in which they have been used are listed and some possible explanations for this slow adoption are presented. Constraints can be used to either check existing designs or to automatically generate new ones. In this thesis, only the former is explored. The goal in this research process was to examine the use of constraint checking in the building industry with the aim of supporting mass customization and, more specifically, to develop a method for specifying these constraints. To this purpose, several prototypes were developed to explore various ideas. Two of the prototypes are described in chapter 4. The first of these two prototypes was used to test the viability of performing constraint checking on building models. No fundamental technical problems were discovered. The prototype fulfilled the intended function of accepting allowed designs and rejecting designed that violated any of the constraints. Having concluded that constraint checking in the building industry is possible, the next step was to devise a method for entering constraints into the system. Since constraints such as building codes are currently predominantly stored in natural language, allowing architects to specify the constraints in natural language was one of the more obvious approaches. This approach, however, presents a significant technical challenge, since interpreting natural language is very complicated. After an analysis of possible alternatives (ranging from programming languages to visual constraint entry), the first attempt at a constraint entry system was based on the same basic idea as natural language, but with some modifications to reduce the technical complexity. Rather than allowing free-from natural language input, sentences are instead constructed using puzzle pieces. Each puzzle piece contains a single word or short sentence fragment, which are linked together to form the full constraint. This has the effect of limiting the range of grammars that can be used, making the system fairly simple to implement. User testing conducted on architects revealed that although the prototype performed reasonably well in terms of usability, the process of constructing the sentences was deemed to be too laborious to be used in practice. After rejecting the puzzle piece-based method, the decision was made to see if it was possible to approach natural language parsing to a sufficient degree to be used for constraint entry. This method allows architects to simply type in the constraint, which solves the problem of the entry method being too laborious. In chapter 5, an architectural language parsing algorithm is described where the syntax tree of the constraint is modified using a very simple grammar in order to fix problems such as redundant or omitted words. The advantages of this method are that it is very flexible with regards to the grammar used, that it doesn’t require a full grammar of the language being used, that little training is required in order to use it and that it can theoretically be used in multiple (Western European) languages. Additionally, the general nature of the algorithm means that it is not applicable solely to the building industry, but to any industry in which constraints need to be specified. To test the algorithm, two tests were conducted, one testing the system’s performance on legal constraints such as building codes and one testing the performance on design constraints, i.e. the constraints written by architects. Chapter 6 describes the results of these two tests. The legal constraints were drawn from the building codes regarding dormers of the municipality of Rotterdam. For the design constraints, a user test was conducted in which architecture students were presented with building designs that were in some way flawed and asked to provide objective constraints to prevent these problems in the future. This was done in order to get a representative sample of the type of language used by architects. In both cases, the desired resulting syntax tree was specified for each constraint to test how many constraints were interpreted correctly by the system. Despite being a very early prototype, the system is already able to correctly interpret about half of both the legal and design constraints. Chapter 7, finally, provides discussion and indentifies potential avenues for future research.