The application of computer-optimised solutions to tightly defined design problems

Computational simulation is typically used towards the end of the building design process, serving mainly as a design validation tool. In such situations it is applied to well developed design proposals where the majority of the design parameters are known or have been determined by the design team. At conceptual design stage, where even the basic form of the building has not yet been finalised, designers typically rely more upon the guidance of experienced consultants. The sheer number of unknown parameters at this stage are considered to render detailed computational simulation of limited use. However, with the development of parametric optimisation techniques, there is now significant opportunity for the use of computational simulation during concept design. Whilst the focus of most optimisation research has been on multiple simple numeric design parameters, if these same techniques can be applied with actual building geometry as a parameter, then the potential for form generation based on performance criteria is possible. Obviously no currently available optimisation method is going to generate a viable building design from a range of performance criteria. However, such a method may be able to generate the optimum site envelope that complies with complex right-to-light restrictions, or the optimum stadium roof shape to maximise solar gains on the playing surface. The research work presented here has attempted both of these examples. This paper argues that optimised forms based on tightly defined design problems can provide critical information for the designer to integrate into their developing ideas, even at the earliest conceptual stages. INTRODUCTION Simulation has long been part of the building design process. However its role has typically been as a validation tool, used by consultants towards the end of the design process, testing highly developed proposals to ensure that performance criteria in the brief will be met. With the availability of more interactive simulation software, analysis is being increasingly used much earlier in the process, and often by the designers themselves not just the environmental and services consultants. The aim of the research work described here is to involve performance simulation and optimisation techniques much earlier in the design process, right at the most conceptual stages to guide in the development of the built form. This means devising mechanisms by which useful simulation results can be derived from relatively incomplete models and then used to generate or modify their geometry to improve performance. Resistance to generative systems has always been high within the building design industry for good reason as the issues involved are very complex and there is often no obvious solution to any particular set of design problems. Also, every building is a compromise between a vast array of competing requirements. Rarely can any building element be truly optimised for a particular use or application, but must be adaptable to many different uses and the compromise is usually the ‘least worst’ solution. However, this does not preclude the designer from at least knowing what the optimum for a particular application would be. In fact this is how most designers work, they know exactly what they would like to achieve, but then have to work within the constraints of the budget, brief and regulations to achieve the best they can. This is the primary skill of a designer assimilating a myriad of complex and competing requirements and then making the best set of compromises from a range of available options. Of most significance here is that designers can work equally well with both objective (quantifiable) and subjective (unquantifiable) constraints. In fact, at the earliest stages of design it is only really possible to work with subjective issues as there is insufficient hard information about the building to calculate many of the objective criteria. Computer systems tend to be of little use in tasks that involve subjective or unquantifiable parameters, but excel at objective tasks with clearly defined and quantifiable parameters, and highly repetitive or iterative problems. Thus, the purpose of this paper is to propose the best compromise. Computational analysis and simulation can make a significant contribution at the very earliest stages of design by generating optimal solutions to very focused and tightly defined problems. The results may not be immediately and directly applicable, but provide useful information for the designer to assimilate within the broader design context. TIGHTLY DEFINED DESIGN PROBLEMS In this paper, a problem is described as tightly defined if all its dependant parameters can be quantified and there are clear and quantifiable criteria against which possible solutions can be tested. Whilst a simple pass/fail test is the most efficient to apply, a problem can still be tightly defined if there are clear boundaries between which the criteria must fall. Therefore, a loosely defined problem has parameters whose values are not clearly defined. They may be quantifiable, but the criteria for selecting any particular value is essentially arbitrary. Even if there are quantifiable criteria against which possible solutions can be tested, the validity or applicability of the basis on which the solution was generated will always be questionable. This uncertainty can limit the usefulness and impact of the information it provides the designer. This is an important distinction in this work as the time taken to define and generate solutions for each type of problem is the same, if not longer for loosely defined problems. However, the solution to a tightly defined problem has greater potential to provide meaningful information and real design insight. OPTIMISATION AND GEOMETRY Much work is being done on design optimisation algorithms and their application to building systems. Tools such as GenOpt from Lawrence Berkeley Laboratories (Wetter, 04), and others such as DOT (Vanderplaats, 01) and SimuSolv (Stub, 99), allow for low level optimisation of multiple numeric parameters by linking and invoking different analysis tools as part of an iterative solution. As the work presented here is in its preliminary stages, we are not yet looking at the application of complex mathematical solutions or genetic algorithms. Rather our initial concern is the translation of analysis results into geometric decision-making and the computational generation of building form to meet performance criteria. The integration of more efficient optimisation techniques will follow as the work progresses, however at this stage a much simpler brute-force approach has been taken. To establish the link between analysis and geometric form, the ECOTECT software (Marsh 1997) was used as it provides an integrated modelling and analysis platform. Moreover its scripting language capabilities allow for the generation and manipulation of model geometry as well as direct access to analysis routines and their results. Thus, scripts were created that iteratively performed calculations, modified the geometry of the model based on calculation results and then repeated the process until specific criteria were met. Using such scripts, the starting assumptions and the decision-making techniques are fundamental to the result. Different starting points and decision methods will likely yield quite different solutions to the same problem, all equally valid based on the test criteria. RIGHT-TO-LIGHT: A TIGHTLY DEFINED PROBLEM In many urban sites there is a need to determine the maximum available development envelope that conforms to local ‘right-to-light’ regulations. One example of such a regulation in London states that any proposed design shall not reduce the daylight availability on existing windows within the facades of surrounding buildings to less than 80% of the existing value. This is possible for a designer to check manually if only a small number of windows are involved, however in a complex urban site, such as that shown in Figure 2 below, there may be many hundreds of windows to check. Figure 2 – Example urban site showing existing site and the windows in surrounding facades whose right-to-light must not be adversely impacted. Such a situation is an obvious application for a computationally generated optimised solution. If the maximum compliant envelope can be determined at the outset, then there will be no need to continually check each design iteration, resulting in significant time savings during the conceptual phase. Generating the optimum shape requires: 1. A method for computationally determining daylight availability for any window, and 2. A methodology by which the results of each calculation can effectively influence the generation of the next iteration in building form. Daylight Availability In this example the UK Building Research Establishment’s Vertical Daylight Factor (VSC) (Littlefair 1991) was used as the metric for daylight availability, calculated directly from the shading mask generated for each adjacent window on the site. Shading masks are calculated in ECOTECT using spherical ray-tracing from a grid of points distributed over the surface of each window. Figure 3 shows an example mask for an east-facing window, with the BRE VSC shown in the bottom-right corner. To determine the percentage change, VSC values were first calculated for each adjacent window based on the existing buildings on the site, and stored in a reference array. The results of subsequent calculations were then compared to these values to test if they were greater than 80% of the original.