Abstract

We describe here methodologies for embedding geometric and topological operations in contemporary Building Information Modeling (BIM) Computer Aided Design (CAD) software, along with the computational implementation of domain specific evaluations; this is done to achieve the integration between spatial based spatial properties evaluations and architectural design workflows. This research describes methodologies and computational technologies used for the development of a computational design assistant in the context of laboratory buildings in Preliminary Concept Design (PCD), the geometric operations used for the creation of a prototype software are explained in their operation, and the architecture of said software is detailed.

1. Introduction

This paper describes research in the area of automatic derivation of computational building models spatial properties for the purpose of automating ventilation systems engineering in laboratories, which is an important assessment parameter when performing design decisions and impact the performance of the building in early stages of design. Our research concentrates in developing computational technologies to allow computer software to derive spatial relationships. Some of these can be considered critical to support designer’s capabilities to perform design decisions or evaluations in many areas of automation of building assessment. We describe methodologies for embedding geometric and topological operations in contemporary Building Information Modeling (BIM) Computer Aided Design (CAD) software, along with the computational implementation of domain specific heuristic tests; this is done to achieve close to real time design feedback, therefore, facilitating the integration between morphological properties evaluations and architectural design workflows. This research has been developed in the context of laboratory buildings and the relation between design layout and Heating Ventilation and Air Conditioning (HVAC) systems, since for these the correct design and operation of HVAC systems is critical for the adequate operation of the facility. This applies not only to the environmental conditions inside the facility, also the airflow patterns designed for laboratories determine the security conditions for building occupants and given the operational requirements of 24 hours a day 7 days a week for these types of systems, the operation of ventilation systems can account for about 50% of the energy usage of the facility ¹.

Ventilation systems engineering is intrinsically connected to the design of laboratories, different types of these are applied at almost every step of the design process from Preliminary Concept Design (PCD) to Design Development (DD). Even though collaboration between architects and engineers is a common practice in later stages of design, this has not been the case for PCD workflows, this is mainly the result of the speed in which design alternatives are produced during PCD, the complexity of traditional simulation tools, and the fact that most of these tools require complex data modeling, before any feedback can be provided to designers ². Design decisions taken during PCD can affect the performance of laboratory facilities and the ventilation system itself, most of the time these are made mainly by the architectural designer, and when a design decision is based on ventilation engineering is mostly based on non-rigorous rules of thumb. We will explore how to automate and improve on what is considered traditional practices on the estimation of ventilation systems engineering during PCD, and how the engineering feedback involved in these practices can be produced by a domain specific BIM computer software. For this purpose we developed a prototype named Laboratory Ventilation Design Assistance (LVDA) which is designed to be used by both Architects and Engineers (AE’s) when evaluating the PCD of ventilation systems engineering in laboratories.

2. Methodology

This research explores the automation of spatial properties for building design assessment. Therefore, for the adequate implementation of it is required to define a domain in which spatial properties play a role in building performance. For this our implementation is based on an extensive literature review dealing with the design and engineering of ventilation systems for laboratories, the current state of computer software used to support the design of these and the traditional design practices and workflows for laboratory design. We investigate domain expert knowledge used for ventilation systems engineering in laboratories and the development of new approaches for computational support of these. With the knowledge acquired we have developed the LVDA prototype. We evaluate here the capabilities of the prototype contrasting its performance to commonly accepted computational tools for the engineering of ventilation systems.

Currently there is lack of computational support for close to real time engineering of ventilation systems in early stages of design; most of the computational systems for this purpose are designed to operate in late stages of design such as Final Concept Design (FCD) or Design Development (DD). In these stages design changes occur at a very slow pace but might be costly in nature. This is unlike very early stages of design, where design decisions happen at a very fast pace, revisions are less costly and design decisions have big impact on building performance. even more the correct design an estimation of ventilation systems can positively affect the energy consumption of laboratories.

Commonly known as ventilation driven facilities, laboratories demand higher numbers of air exchanges than other building types and well-planned directional air flows ^{2, 3, 4}, this becomes of extreme importance when dealing with high levels of Biosafety Level (BL) laboratory spaces. Also, lab- oratories require 100% fresh air since in many cases their equipment exhaust cannot be recirculated, therefore more air needs to be brought in to the building to make up for the exhaust. These requirements translate to higher energy consumption. Also, the design of the air distribution network in a laboratory is commonly considered an environmental safety measure, since it is designed to reduce the possibility of cross contamination in case of a chemical spill, directional airflow patterns in laboratory layouts are designed to be negative towards all laboratory spaces, this condition must be kept at all times. In traditional practices the design of the directional airflow structure is made explicit on floor plans by placing arrows pointing the direction of the flow along with the flow rate (Figure 1 directional air flow mapping in cubic feet per minute).

Traditional practices in the design of directional air flow networks has been usually based on the pressure differential between adjacent spaces, this type of structure is commonly constructed by increasing the amount of air flow exhaust within the laboratory space ^{4, 5}, thus increasing the energy consumption of the facility. Bennet ⁵ has pointed recently that inward air flow can be more accurate for controlling the air flow patterns between two adjacent spaces. The design of the inward based directional air flows is directly connected to the layout’s spatial adjacency structures and the air pressurization of the different spaces in it, making this a valid approach for the automated evaluation of spatial properties. The spatial arrangement of the layout defined by the architect also affects the performance of ventilation systems regarding the required air pressure of the ventilation branches. The locations of served spaces within the layout and the typology of layout have direct impact in both the length and number of turns that the ventilation branches must have to reach all the required spaces.

The level of design abstraction of a design might not be critical when dealing with paper-based representations and human based data derivation, such as in traditional practices for the estimation of PCD ventilation engineering, but it becomes relevant when the design representation is a computational model for supporting automated design assessment. Although there is not a widely adopted standard for the level of completeness of computational models in architecture and engineering (AE), recent efforts have tried to define the different levels of it for computational models ^{6, 7}. Among these the General Services Administration (GSA), Facilities Standards for the Public Building Service (P100) clearly establishes design standards for new buildings, among these; the levels of design development are defined as: Preliminary Concept Design (PCD), Late Concept Design (LCD), Concept Design (CD), and Design Development (DD). In the P100, the semantic content for each of these categories is clearly defined making it a suitable for the research developed here. In it PCD projects are defined in terms of the content as follows: placement and massing of the building are defined; pro- gram spaces are identified only at a departmental level, circulation spaces both human and vehicular are identified, no internal partition walls or wall openings, basic definition of building boundary surfaces (Figure 2 preliminary concept design diagram (NIH design policies and guidelines)). Later stages of design, such as Late Concept, Design Development and Construction Documents follow. During these the information contained in the design will continuously gain both in definition and content.

Few objects usually are included in PCD BIM model besides space objects; among these; building envelope needs to be identified, partitions among spaces are represented either with wall objects or virtual walls. These PCD BIM models are usually developed for the purpose of massing and spatial layout studies. In the case of laboratory buildings, the main spatial referent for spatial layout programming is the laboratory module. The sizing of these allows AE’s to define the PCD layout of the building’s structural grid, and to have a clear approximation to the expected occupancy loads of the facility. Depending on design requirements the laboratory module can include two other program components besides the laboratory space: scientist office and lab support space. The spatial relations among these will affect both human circulation conditions and dimensioning of the building’s ventilation systems engineering.

The PCD of laboratories and engineering assessment

In the case of laboratory design, there is no precise framework for how engineering expertise is brought in to PCD and particularly how spatial properties of the layout play a role in the estimation of ventilation systems, most of the time engineers will get involved once the massing of the building and internal layout has been completed by the architect. Often the PCD architecture tends to optimize the spatial adjacency of the facility in terms of spatial relationships, but in terms of ventilation engineering there is no specific optimization but the application of engineering knowledge at the level of rules of thumb regarding the floor to floor clearance required building systems ⁸. In traditional PCD of laboratories (Figure 1: Business Process Model for traditional PCD of laboratories (author)), after the PCD model is completed by the architect, it is handed to the engineer who will extract its geometric spatial properties, add to it his or her expertise in terms of code requirements and best practices to produce a ventilation engineering data model. Then after the cooling loads have been estimated the results are returned to the architect, who based on the results might explore different design alternatives. If new alternatives are explored by the architect, a new cycle of engineering estimation is conducted. The overall time required for each of these iterations might be hours in the best of cases. After analyzing the Business Process Model (BPM), please see Figure 3, is easy to infer the reasons for the inefficiency of the process, only in data exchanges/inputs, there are at least 9 steps, even more; some of these exchanges rely on manual extraction/manipulation of data. Therefore, they are susceptible to error.

BPS Building Performance Simulation (BPS) integration to laboratory design processes

In general terms Ventilation system engineering using BPS’s is usually performed after the spatial arrangement of the facility is consolidated; the material specification and configuration of the building envelope is well known by the design team. In traditional building design practices is during CD or FCD, where the HVAC engineer will conduct a detailed analysis of the heat loads in the facility, define the ventilation rate per space, determine the air- flows, and propose the duct layout including the location of vertical drops as well as other pieces of equipment, all this based on the spatial organization of the layout. Results of the CD or FCD HVAC engineering analysis can generate a set of design revisions in order to properly fit the HVAC components and their requirements ^{9, 10}, which due to the state of completeness of the design might produce costly revisions and time delays.

The previously described process is supported by the use of BPS by the building design team, many of these tools features makes it hard for any implementation of multidisciplinary or collaborative design environments in early stages of design ¹. Some efforts have been made to automate some areas of the BPS to make them more suitable to early stages of design, still many of these require for a certain level of semantic content in the PCD model, such is the case of; material definition for walls, sizing and placement of doors and windows. This content, usually available in CD or FCD is not commonly part of the design semantics of PCD. Other reasons can also be pointed as to why BPS’s are not suitable for PCD such as; the speed in which the de- sign changes happen during this phase ¹¹, the time required to prepare and complete BPS assessment, among others clearly limits the application of these, since once the BPS analysis has been completed the entire design might have changed making the analysis results obsolete ¹². Both Holzer ¹¹ and Chaszar ¹³ indicate that software’s results might not enable interdisciplinary collaboration and that different domain semantics can create friction among AE design teams. Holzer ¹¹ also points among the issues limiting interdisciplinary collaboration, the need for team members to reflect in privacy regarding the proposed solutions.

Current trends in engineering assistances

The development of computational BPS tools has been going on for over 40 years. The range of these varies greatly from; excel based to special purpose highly advanced software. Trˇcka and Hensen ¹⁴ identify three generations of BPSs; the first was based on analytical formulations and simplified assumptions, the second one based on numerical methods, provided partial integration of performance aspects of buildings, the current generation of BPS can capture reality better and are fully integrated regarding different performance aspects. In the area of HVAC there are roughly four categories of BPS’s, these are based in the problem they are trying to analyze:

• Equipment sizing: Carrier HAP, Trane Trace, Energy Plus, Design-Builder, MC4suite etc.

• Energy performance: Carrier HAP, Trane Trace, Energy Plus, DOE-2, Equest, ESP-R, IDA ICE, Trnsys, Hvacsim+, VA114, Simbad, Building Energy Analyzer, Design Builder, etc.

• System optimization and controls: Genopt (generic), Contam, Energy Plus, ESP-R, Trnsys, Dymola, etc.

• Duct sizing: AFT Fathom, Dolphin, Duct Calculator, Duct size, Pipe-Flo, Python, Indus, Cymap, etc.

• Source: https://apps1.eere.energy.gov, Trˇcka and Hensen ¹⁴.

Most of the previously listed have been developed for the purpose of HVAC engineering design; therefore, they require the construction of engineering domain specific semantics, have complex user interfaces, also the feedback produced might be hard to understand by non-domain experts ¹⁵.

Current efforts in BPS development concentrate on improving the integration of these to the overall building design process ¹⁶. Three main areas are being researched: the simplification of either the calculations being performed ¹⁷, the simplification of the simulation data model being used, and the automating generation of simulation models for the execution of BPS ^{18, 19}. This research takes on these trends and goes a step further in the effort of integration to design process by embedding engineering estimation within CAD software.

Methodology for acquiring ventilation systems engineering in laboratories

We have conducted an extensive research dealing with widely recognized compilations of best practices and normative calculations applied to the engineering of ventilation systems in laboratory design, these range from energy standards ²⁰, design requirements ², design guidelines ^{4, 21, 22, 23, 24}, and HVAC engineering ^{25, 26}. From these we have extracted provisions dealing with the following issues regarding engineering of ventilation systems in the following areas:

• Recommended design practices in terms of operational procedures

• Code compliance for the design of ventilation systems

• Minimum ventilation requirements for the operation of the facility

• Best practices for the safety conditions for the facility

• Systems serviceability provisions

Capturing formulaic and domain expert data

Methodologies for capturing expert data vary depending on the area of expertise being processed. For instance, in some areas of architecture it might be the size of service areas in a building regarding the usable square footage of the layout, in engineering it might be the types of connectivity that a pre-cast concrete beam requires when installed under particular conditions, or the result of a combination of multiple forms of expertise data which represent complicated areas of design knowledge ²⁷. In the case of ventilation engineering for laboratories, the expertise data is in most cases based on the relation between spatial organization of the layout, space usage and attributes of the space instance in terms of: environmental requirements, scientific processes, internal equipment or mechanical system requirements. In this research it is of extreme importance the capturing of spatial components in the layout, and the space instance properties that play a role in the automated derivation of spatial properties in it. For this purpose, we have developed a comprehensive space instance classification. This has been collect in a human readable input file to support customization.

The input file developed for the LVDA incorporates the following space types:

• General chemistry

• Radio chemistry

• Research

• Hospital or clinical

• Biological containment

• Animal research

• Isolation/clean rooms

• Materials testing

• Electronics/instrumentation

• Teaching

• Laboratory Support

• Offices

• Toilet

• Lockers/showers

• Conference/ Break rooms

• Corridor

• Service Corridor

• Elevators

• Loading docks

• Housekeeping closets

• Mechanical, electrical, and telecommunication areas

• Service Shaft

• Interstitial Space

• Stairs

Each of these space types is explicitly associated to a set of attributes; these along with their values have been compiled from domain specific guidelines ^{2, 4, 21, 22, 23, 24}. When processed by the system these attributes are embedded by the LVDA in the BIM database, enhancing the semantics of the BIM model to both; support LVDA proprietary computations, and other types of BPS assessment that might happen downstream in the design process. The attributes embedded take as reference those identified in the indoor climate simulation to HVAC design model view definition, developed by Voulle, Hanninen, Berard, and Lehtinnen.

Documenting and interpreting provisions behavior

In this research we have documented provisions behavior directly in to the algorithms composing the LVDA prototype, it is understood here that the capability of decision trees available in computational algorithms suits well the translation of provisions behavior E.g. “temperature controlled rooms shall be lockable, and all mechanical components shall be accessible and serviceable form outside the room” ³. In this example the provision is translated to an algorithm that estimates the ventilation system routing and constraints the geometry of the route so it never passes through a serviced space to supply another serviced room, the described provision behavior is implemented in the LVDA Ventilation Routing Estimator Module (VREM).

Implementation technologies

For the implementation of the LVDA two pieces of contemporary technologies have been selected; Firstly, the rich objects semantics provided by BIM data bases; Secondly, the estimation of engineering data using normative calculations instead of traditional BPS. Lee and Eastman ²⁶ demonstrated how semantically rich environments in BIM can be used for the derivation of spatial relationships embedded in the building design. Their research also enhanced the decision process for very early stages of design by operating within the reduced semantics typical of these, their assessment structure was based on a standalone rule-based BIM checker. Park and Augenbroe ^{17, 27} demonstrated the viability of using normative calculations for energy consumption estimation; they also pointed that normative calculations are well suited for sensitivity/feasibility studies for buildings in design stages.

Implementation of the LVDA

During the initial stages of this research it was identified the need for the LVDA to provide close to real time user feedback with a limited number of inputs, and to structure the system operation to suit the characteristics of laboratory PCD workflows. Therefore, the software prototype was developed requiring minimal interaction from users. There are two stages of operation with the second being the airflow estimator, which serves a a wrapper for other pieces of the software; the air pressure structure analyzer, the routing estimator, and the airflow calculator. This class will take care of all the operations needed to derive information related to the spatial properties of the layout. Another aspect identified early in the development of the LVDA was the necessity for it not to disrupt the flow of the design process. For this reason, instead of developing a standalone application, all of the modules in the LVDA have been embedded in the back end of CAD BIM software in the form of a plug-in.

The LVDA prototype has been implemented in Autodesk Revit. This has been chosen because of its popularity, almost 70% of the market in the US ²⁴ has adopted it. The user interface designed for the LVDA, please see Figure 5, is based on the concept of simplicity; therefore, it requires from end users the least possible number of inputs. There are only two buttons in it. Within the LVDA system architecture, these two modules take on the responsibility of controlling the execution of all other modules in the LVDA. If data needs to by dynamically loaded in to the system users might be required to point the location an input file or connect the computer to the World Wide Web.

The LVDA has been developed using to two high level software modules which encapsulate other operations, therefore reducing the required user interaction to a minimum, the first of these modules controls the operations related to the computations of ventilation semantics related to the model, the second module, controls the operations related to evaluation the layout properties and to test the possible geometrical solution for ventilation system routing, we have called this module the Air distribution Routing Estimator Module (ADREM).

Development of the Air Distribution Routing Estimator Module (ADREM)

The second phase of ventilation engineering estimation produced by the LVDA is the ADREM it verifies the spatial properties of the layout and generates a routing solution based both on domain heuristics and the layout itself, it also analyzes the routing and performance of the air distribution system. This capabilities are not commonly available in traditional engineering of ventilation systems in PCD. This represents a mechanism for integrating the estimation of building systems performance to the properties of the architectural layout. During this phase, the LVDA derives the following from the BIM; building morphology, airflow pressure structure, and spatial adjacencies. These provide the LVDA with the information required for the estimation of the properties and performance of the air distribution systems that would better suit the PCD layout.

Morphology derivation Module (MDM)

The first step during the execution of the ADREM is the derivation of the building layout morphological features; this has a direct impact on the order in which the ADREM algorithms are executed. The MDM first analyzes the PCD BIM looking for spatial properties which indicate the design being either an interstitial or a service shaft type of facility. The building level morphology derivation informs the LVDA prototype about the behavioral constraints to be applied for the ventilation system engineering estimation. We deal with the derivation of laboratory building morphology with the application of three different approaches; firstly, the verification of the existence of interstitial spaces in the BIM, this is done by querying the BIM for spaces which long name is interstitial space, this is understood by the system as a dedicated space capable of servicing process driven spaces located directly above or below it. Secondly searching for the spatial properties indicating the presence of process driven spaces with floor to floor height capable of hosting interstitial spaces, and thirdly for those layouts in which none of the previous indicators can be identified, this is understood as a layout which belongs to a service shaft or service corridor building typology, please see Figure 6.

Interstitial space routing

The existence of interstitial space objects in the BIM is interpreted by the LVDA prototype as the intent of having re-configurable mechanical and support systems for the laboratory. It also defines a specific relationship between service spaces and serviced spaces; in this case the building system connectivity is constructed in the vertical plane, the vertical adjacency be- tween service and serviced spaces is analyzed by the LVDA system in order to estimate the routing of the ventilation system. We propose here that such relationships can be extracted from the vertical adjacency, which lies implicit in the BIM data structure. The vertical adjacency is derived by analyzing the vertical overlapping between service space and the serviced spaces. The LVDA prototype identifies the relationship of the conditioned spaces above or below the interstitial space and computes the distribution system vertical drops in accordance to the best practices guides. Deriving the vertical adjacency in laboratory layouts Interstitial typologies require from the LVDA prototype to analyze the vertical properties of the laboratory design, during this process the LVDA retrieves the boundary geometry in the serviced spaces and evaluates their relation to the boundaries of the interstitial space. In this structure is important to note, that besides the explicit flexibility provided by interstitial typologies, the operational constraints and best practices for laboratories remain. Therefore, practices such as placing the insertion point for the HVAC close to the space occupant’s entry/exit point still is considered a good practice. Given this provision, vertical connections to service spaces should have a very specific location. Unlike later design phases, where doors can be utilized to point to the entry/exit of the space, in PCD the location of the entry point of each space is derived by the system, and explicitly associated to the interstitial space. This is then used as target point for each of the branches of the distribution system (Figure 11 derivation of intake points in interstitial typologies) within the interstitial space itself. Then entry point is estimated by setting it at the midpoint of the common boundary between the serviced space and the circulation space, assuming the door object is not available during PCD but this will be placed somewhere along this common bound. This approach still needs for the definition of a start space for the branch, such as a location of the fan, therefore in this model end users are required to provide the location of the fan, please see Figure 7.

Space based system routing

After the decision tree has traversed through the steps dealing with different types of interstitial spaces in the BIM, the MDM verifies the existence in the model of service corridors; this is interpreted by the LVDA prototype as the designer’s intention of having the ventilation system running through them. The MDM identifies then all the spaces which could be serviced by routing the ventilation system through the service corridor, and proceeds to estimate the adequate path. The routing functions used here are based in on the explicit definition of a vertical drop adjacent to a service corridor space.

Start space-based routing

The start space-based routing is built on the idea of tagging the space re- quiring the most ventilation in the entire layout. This is done automatically by the LVDA, the system considers this space, as suitable for the location of the system’s vertical drop, and this approach is taken whenever the layout does not contain definition for shafts and vertical drop spaces. When the lay- out contains service shafts and no service corridors; then the LVDA routing algorithm uses the vertical shafts as vertical drops for the ventilation system ducts, and the system assumes that the designer’s intention is to host the ventilation branch within the circulation area. The derivation of the morphology of each ventilation branch is responsibility of the routing algorithm, particularly to the Spatial Adjacency Analysis Module (SAAM).

Spatial Adjacency Analyzer Module (SAAM)

An important functionality developed for the LVDA is the analysis of spatial adjacencies, since this allows it to f ante rooms are part of the layout, such as lobbies or sound locks or layouts including laboratory spaces with extreme requirements due to their of Biosafety level classification (BSL), such as BSL-3 and BSL-4. The SAAM allows the LVDA to build space sets including the potential target spaces, the SAAM analyzes each of the branches of the graph looking for the following scenarios;

Target spaces for the ventilation system not directly adjacent to the service space but, that can be reached by the system by going through a non-process driven space such as custodial closet or lobby.

Non-serviced spaces directly adjacent to the service space that might serve as anteroom for process driven spaces. Such as locker rooms.

The SAAM adds to the space set list all suitable spaces directly adjacent to the service corridor, then recursively analyzes each space and all the spaces adjacent to it, this is described here as second level spatial adjacency analysis, please see Figure 8. The SAAM verifies the second level spatial adjacency and the space classification included in the LVDA prototype which indicates if the space must be supplied by the ventilation system, all spaces requiring ventilation are then added to the space set. The SAAM iterates through the space list until it runs in to; a previously visited target, a service space or a space for which there is an operational constraint for running the ventilation system through it. If the module identifies target spaces through the SAAM these and their ventilation requirements are added in to the ventilation branch properties and flagged as already included in a ventilation system branch.

SAAM Implementation structure

Directional Airflow Structure Analyzer (DASA)

One of the extended capabilities of the LVDA analyzes the directional airflow within the layout, this functionality is based in safety guidelines found in best practice compilations, usually constructed by the engineer, it helps to identify the compliance of negative air pressure airflows towards process driven spaces. This is usually incorporated in to the design documentation later during the design by the engineers. The DASA retrieves from the BIM all those spaces classified by the system as process driven spaces, the DASA then labels all of these as targets. The DASA then proceeds to interrogate each target in regards of its spatial adjacencies, it checks for the directional air flow among the target and all its surrounding spaces, please see Figure 11. If during the analysis an airflow pattern which might allow for air to escape the target (process space) is detected, a warning is generated by the system. In this warning, the error space the target and the building location information are identified in an error list. After all spaces in the building are analyzed the list is saved as document (.txt), please see Figure 10 and the end user is informed about the existence of errors in the BIM and location of the error file.

Interstitial space routing

Ventilation Routing Estimator Module (VREM)

Another major function carried by the LVDA prototype is the VREM, during its execution the actual building morphology is evaluated, this is done to provide an accurate estimation for the ventilation system routing, based both on layout design and ventilation system properties. At this stage the routing of air distribution ducts is derived, duct geometry, and ventilation system attributes are estimated by the LVDA prototype. The potential locations for the vertical drops are defined and informed to the end user through Revit’s interface.

Routing estimation derivation

BIM technologies have been used to automate several aspects of building design assessment, Lee and Eastman ²⁸ utilized neutral format BIM data for a variety of design assessments during PCD, among these the circulation and security validation. In it multiple circulation paths were analyzed regarding rules extracted from design guidelines. Although the implementation of these has been done for a different building type and for a different type of engineering this demonstrates how building data can be used to infer the performance of circulation paths regarding design guidelines rules. In their work, Lee and Eastman ²⁸ used graphs traversing all the possible circulation paths between what they call start and target spaces, the validation of these was done by checking the attributes of the different spaces along a potential paths. An extrapolation of this approach is used in this research for the estimation of HVAC air distribution layouts. Here we identify what spaces in the model have the required conditions to host the distribution ducts and derive the apace adjacencies between this and all the serviced spaces. We propose the representation of the service-serviced spatial adjacency structure in a Service Adjacency Graph (SAG). In the SAG structure where the service space (S) acts as the root and all the conditioned spaces (C) are nodes of the graph, the construction of SAG is constrained by the adjacency relationship existing among them, please see Figure 12.

The proposed SAG structure requires the usage of the SAAM. The SAAM allows the LVDA prototype to evaluate domain specific constraints regarding the different types of ventilation systems accessibility and serviceability in laboratories. The system incorporates a space classification algorithm that allows for the construction of a well-defined graph structure describing a domain specific representation of the .layout. The space classification used here is as follows: Service spaces: these are understood as the spaces suitable for containing elements of the ventilation system running through them, these include interstitial space, service corridor, shaft, and corridor. Serviced spaces: these are all those spaces for which the LVDA input file defines a ventilation requirement, among these, all process driven spaces, and depending on design conditions others such as offices, and ancillary facilities. Non-serviced spaces: these are all those spaces in the BIM for which the LVDA input file defines no ventilation requirements. this type of space might not require any mechanical pressurization or might even be naturally ventilated.

The SAG is used to represent the building system morphology as space sets (S); each of these contains all the spaces requiring service from the ventilation system, please see Figure 13.

Derivation of spatial adjacency

The construction of the SAG is based on the VREM capability to retrieve the geometric properties of the spatial layout indicating both; types of spatial adjacencies; direct or second level adjacency, and the estimation of the ventilation system connection points towards the serviced space. To obtain this type of information, a number of geometric operations and tests need to be performed. Some of these such as polygon offset, and polygon intersections, are not part of the geometric operations available through Revit’s API. To enable the LVDA prototype to perform these operations it was necessary to link the LVDA to an external geometric library. This library provides the LVDA prototype access to algorithms that extend its capabilities. Many geometric libraries are available for open-source use, but based on; implementation requirements, language compatibility and overall processing performance for the LVDA prototype implementation we have chosen the Clipper geometric library (https://angusj.com/delphi/clipper.php).

Deriving special spatial adjacency

The derivation of spatial adjacency enables the VREM to identify the spatial relationship between the service space and potential service spaces. To derive the spatial adjacency the algorithm extracts geometric information regarding the boundaries of each serviced space, and then it translates the line-based representation coming from Autodesk Revit in to a Clipper polygon object, proceeds to offset the polygon by a predetermined value, which goes further than the thickness of a standard wall object, please see Figure 14.

To determine the actual adjacency the VREM places a point at the mid- point of each of the edges of the clipper polygon, after all the points are in place the VREM uses the Clipper Point in Polygon (PIP) test to check if one of these midpoints is placed inside the service space polygon, please see Figure 15. If the test returns true the serviced space is added to the target space set. If the test returns false each of the edges of the will be recursively processed and tested again in a brute force approach until either finds the adjacency information or discards the space being tested as adjacent.

The point in polygon test is based on the crossing number geometric operation, in it a ray starting in the inspected point crosses the boundary of the polygon, if the number of crossings is even then the point is outside of the polygon, of the number is odd then the point is inside the polygon, please see Figure 16. Traditional PIP algorithms perform the test by constructing a ray using the test point and extending to the right of it parallel to the X coordinate axis, then each of the edges of the polygon is tested for crossing the bounding lines of the polygon, special cases are considered when the crossing happen at a vertex or through an edge of the polygon.

Every time an adjacency is detected by the VREM, the space being tested is added to the list of spaces needing to be serviced. The properties of the ventilation branch are evaluated regarding the ventilation requirements contained in each space set which in turn determines the maximum diameter of ducts and the required space to fit these in the service space.

Route Estimation Module (REM)

The points constructed by the VREM during the point in polygon test, particularly those defining adjacency, are interpreted as connecting points for the ventilation systems; therefore, they are used as targets to estimate the routing of ventilation branches. The estimation of the routing involves also the extraction of the geometric properties of the space in which the ventilation branch operates (service space) and the optimization of the route in terms of ventilation system performance. Estimation of the route for the ventilation system For the estimation of the ventilation system routing the REM takes the vertex identified as the start vertex of the system and looks for connection to the closest un-visited V vertex (target), once the closest is identified an edge is constructed by the REM, this new edged is tested for possible intersection with the boundary of the service space, if intersection exists, it means the constructed edge is out of space bounds and needs to be discarded. Then the connection is tested to the following close vertex. Each time a vertex is added a new E edge is created using the new V and the previous V, each edge is tested for path self-intersection and out of service space bounds condition, if this test return true, the algorithm traces back to previous V and tries a new connection, the process is iterated until all un-visited V are added to the path. If no target space is directly visible the algorithm tests the closest vertices of the circulation space and then check for more space target vertices, this approach allows the REM to operate in both convex and concave types of service spaces.

This approach which is somewhat easy to evaluate in square shaped service spaces, please see Figure 17, might appear a little bit more convoluted when dealing with more complex layouts (Figure 18 routing of the ventilation on concave spaces).

Although the developed approach at first might appear as a non-conventional solution for the routing of the ventilation duct, please see Figure 19, it is important to remember that it provides a shortest overall ventilation route and that the approach of running the ducts closer to the bounding’s of the space might not the shortest path and it might produce a larger number of tee’s and connections which negatively affect performance of the system.

The computational implementation of the shortest path in the LVDA prototype uses a traditional computer science algorithm, developed by Robert C. Prim in 1957, Prim’s algorithm solves the minimum spanning tree in computational weighted graph structure, in basic terms it searches the shortest route between all the nodes in the graph. The algorithm starts from the source S and searches among all the adjacent nodes in the graph which have not been relaxed or un-visited. Calculates the distance to them, adds to the graph the closest node V, proceeds to flag it as visited and set the V as S and continues through the graph until all nodes are visited. Unlike the original Prim’s the LVDA algorithm starts from a defined node and every time a new node is added the edge created by connecting the previous and the new vertex is tested for its relation to the edges of the service space bounds, if intersection exist the algorithm falls back in to the previous V and searches for another node. This algorithm deals with the metric properties of the path, meaning the length of the route and the number of connecting components, both of which influence the ventilation system performance.

3. Summary

This research evaluated the possibilities of integrating spatial properties evaluation to BIM enabled laboratory PCD, the implementation was done to support the automation of domain specific knowledge-based ventilation engineering in PCD. This was done to provide better integration of ventilation engineering on to PCD design of laboratories. There are intrinsic limitations to this research that are given by the nature of BIM CAD models for laboratory PCD, some of this can be found on the level of completeness of the BIM and how this can affect the accuracy of the prototype. One of the most important aspects of this research is that of the embedding spatial properties evaluation in the form of expert knowledge and computer algorithms that allows to non-experts the evaluation of domain specific aspects of the design layout. When evaluating the types of analysis and the range of results the provided by the LVDA prototype to those of traditional processes supported by engineers and BPS, the latter covers a wider spectrum of analysis parameters which undeniably provides a more comprehensive understanding of the expected building behavior, and also provides the evaluation of “what if” scenarios. Traditional BPS might produce more accurate results, but is important to evaluate the correlation between the types of evaluation and the speed of the feedback produced by the different approaches, and how these fit the different stages of design, in other terms what is the right tool for the right design phase Different approaches have been developed in this research to extend the scope of the assessment data of laboratories PCD, among these the capability of evaluating the spatial adjacencies of the building layout in terms of building systems engineering and routing. This type of analysis has been deemed useful for several other areas of PCD feedback and is easy to foresee a wider range of application of these on either other type of assessments or different types of building systems. In terms of the efficiency of the system developed here, at this point and without extensive end user evaluation it is hard to produce hard metrics regarding its ease of use, but it must be noted the comparative efficiency in two specific areas, firstly, in the process execution speed (compare the two developed process models), secondly the amount of data items required by both the traditional approach and by the LVDA prototype. Also it is important to mention that; at the level of efficiency metrics even for traditional approaches, it is hard to evaluate their accuracy in predicting behavior when compared with the actual building operation, but it is important to notice that the estimation of systems proposed here is not meant to simulate the behavior of the building or the system associated to it, but to provide a rigorous approach for the estimation of ventilation system engineering to be used for design decision making, either for the purpose of design modifications evaluation or for best alternative selection.

References