Academia.edu no longer supports Internet Explorer.

To browse Academia.edu and the wider internet faster and more securely, please take a few seconds to upgrade your browser.

Log In
Sign Up

Figure 7 – uploaded by Iris D. TOMMELEIN

See full PDF downloadDownload figure

Table 2.3: Warehouse Sizing Estimates for Non-Nuclear Power Plants (Table from [Stone& Webster 79]) Descriptions of reference plants consist of sketches of the building volume, coarse layouts, and a plausible milestone schedule. Supplements to these documents are construction field manuals which summarize company practice. Such manuals usually have a section on temporary facilities: they provide rules to estimate the needs for temporary facilities based on any one or several of the features of the reference plant. Table 2.3, an excerpt from Stone & Webster’s field construction manual, shows how one could obtain a first-order estimate on the size of warehouses needed for construction of non-nuclear power plants [Stone& Webster 79]. Note the very simple linear relationship between unit size and warehouse area estimates. Site weather conditions are crudely taken into account, but the time frame within which each project might be built is omitted entirely. into account, but the time frame within which each project might be built is omitted — Table 2 3: Warehouse Sizing Estimates for Non-Nuclear Power Plants (Table from [Stone& Webster 79]) Descriptions of reference plants consist of sketches of the building volume, coarse layouts, and a plausible milestone schedule. Supplements to these documents are construction field manuals which summarize company practice. Such manuals usually have a section on temporary facilities: they provide rules to estimate the needs for temporary facilities based on any one or several of the features of the reference plant. Table 2.3, an excerpt from Stone & Webster’s field construction manual, shows how one could obtain a first-order estimate on the size of warehouses needed for construction of non-nuclear power plants [Stone& Webster 79]. Note the very simple linear relationship between unit size and warehouse area estimates. Site weather conditions are crudely taken into account, but the time frame within which each project might be built is omitted entirely. into account, but the time frame within which each project might be built is omitted

Related Figures (90)

operation, and some permanent facilities could be redesignated temporarily to Support get involved in it. In this way, layout design becomes an iterative process, construction. The interaction between temporary and permanent facilities can indeed be

Figure 2.2: Principles, Considerations, and Criteria that Affect Site Layout hen describe different physical and computer models that can assist people with this task. Having defined the site layout problem, several factors that affect it are shown in Figure 2.2. With all its complexity, the problem may appear virtually impossible to solve. Yet, field practitioners routinely lay out construction sites. To examine their process, | will sketch field layout practice: who does this task, what information is available, wha methods are used, when is the problem solved, and what do the result looks like? I wil then describe different physical and computer models that can assist people with this task.

Figure 2.3: Schematic Drawing of Layout Design Procedure (Figure from [Rad 83]. Note that the “labor force requirements” in my schema are called “manpower requirements” in Rad’s.) or not particular arrangements will work. ‘ine tolowing examples show how site managers May simply their task. Managers divide available space up into smaller sub-areas, and delegate the task of laying ou contractor areas to the contractors themselves. Once they commit major facilities tc a location, they do not modify or even question these locations when positioning les: important facilities. Managers adapt layouts from other sites to fit the new projec: conditions. They group and arrange facilities separately, before positioning the group intc the overall arrangement. They sketch in the trial-and-error process of generating the arrangement, before drawing the final plan, and move around templates to test whethe or not narticular arrancements will wort.

satisfying or optimal solutions. However, the two types of work cannot unequivocally bt separated; for example, each contains so-called heuristic methods that could be classifies either way. This large spectrum includes project studies and field manuals describin; concerns in site layout [Tatum 81; Neil 82], checklists to help select temporary facilitie {Handa 87; Rad 83], criteria to assist field managers in evaluating site condition [Dressel 63; Handa 88], approaches to step-wise construct layouts [Eastman 72 Hamiani 87, 88], and optimization routines [Warszawski 73a, 73b; Rodriguez-Ramos 82 837. [Handa 87; Rad 83], criteria to assist field managers in evaluating site conditions

Figure 2.6: 600-900 megawatts Configuration of a Reference Fossil Power Plant Model (Figure from [Stone& Webster 78]) For example, in the case of power plant construction, large engineering firms that had built several projects of a similar nature developed so-called reference plants based on well-established plant concepts and system configurations and did so for typical ranges of plant capacities (for example, Stone & Webster has 350-600 megawatts and 600~900 megawatts Reference Fossil Power Plant Models [Stone& Webster 78], see Figure 2.6). megawatts Reference Fossil Power Plant Models [Stone& Webster 78], see Figure 2.6). plant capacities (for example, Stone & Webster has 350-600 megawatts and 600-900

construction support buildings, parking areas, and Jaydown areas. Due to the fact that the data in the table were derived by factoring out many project-specific attributes, it may have only limited relevance to project managers who lay out a given site. Conversely, paper: that describe individual projects [Tatum 81; Weidemier 86] and shed light on many mor attributes than those pertaining to the reference plant, may provide data that lacks generalih and that may not be directly applicable to other sites. It is thus up to the people wh develop reference plants to do case-based reasoning by choosing relevant attributes for | type of project. These attributes should be neither too general nor too specific to be usefu to layout designers. (Case-based reasoning is an active area of current research in artificia intelligence and design.) Table 2.4: Area Estimates of Temporary Facilities (Table from [Rad 83]) The practice of conceptualizing reference plants appears to be common in the power The practice of conceptualizing reference plants appears to be common in the powe! industry, and reference plant designs and construction field manuals are generally available at large engineering firms. This suggests that the payoff for using these documents must oe high enough to warrant their generation. But, though they could constitute a good starting point for site layout design, the information in them is often so deeply buried in these documents that it becomes hard to retrieve. A more accessible way of storing this information, for instance in an easy-to-query computer system, would therefore be mos) useful. In addition to this kind of generally applicable rules of thumb, however the reference plant description needs to be complemented with project-specific information. What the project-specific layer of knowledge entails is described in the following section.

[able 2.5: Ranges on Area Estimates (Table from [Rad 83]) ‘ihree iayers OF Knowledge Can classify known facts pertaining to consiruction site layout. Though these layers may not be defined uniquely, there seems to be a natural division between them. It is definitely true, though, that the knowledge they contain is not agreed upon industry wide. [Rad 83] surveyed 36 construction companies on their approach to site layout. His work revealed that companies had similar practices, but when asked about specific data, /arge variations in actual numeric values became apparent. Rad provides minimum, maximum, and average values, but did not say to which extent the data collected from different companies were comparable. Table 2.5 shows the results of Rad’s survey.

Table 3.1: 13 Topological Relations Between Two Intervals (Figure from [Allen 84])

relations such as “to the right of” are approximated as strictly binary relations. “A round apple:” This is a simplification of the apple’s shape. after a human’s internal representation. What that human representation is, however

Any point on the circle is a position for the white rectangle at a distance d from the grey rectangle. Any point on the circle is a position for the white rectangle at a distance d from the grey rectangle.

symbolically, using geometry as just one attribute of each component [Howard 89]. SHecause a geometrical representation 18 only an aosiraction, Ww exciudes much information. Therefore, when used in knowledge-based systems, it must join with other forms of knowledge representation that capture missing data. Because geometrical representation is so restricted in data content, it may not generally be the right choice to underlie a system that must later annotated with other data. As opposed to such “decorated geometry” (a word coined by Smithers in [Smithers 89]), a knowledge-based system must flexibly allow changing representations. Among others, Howard made this same criticism of CAD systems used as design tools rather than drafting tools. Evolving design tools will model the structure as well as the behavior of a component symbolically, using geometry as just one attribute of each component [Howard 89], 3.3 Design Problem-Solving Methods

Figure 4.1: Layers of the SightPlan System oightrian 1s an expert system that assists construction managers in laying out the temporary facilities on a construction site. The system consists of knowledge specific to construction site layout, but it is layered on top of a domain-independent blackboard architecture named BB1, and uses information represented in other layers, such as the one specific to the class of problems solved by the constructive assembly method. Figure 4.1 shows the different layers of SightPlan: LISP is the programming language in which the entire system is implemented; BB1 is a domain-independent blackboard architecture; ACCORD is a language for use in the constructive assembly method; PROTEAN is an application that elucidates protein structures; GS is its 3-d geometry system; SightPlan is an application that lays out construction sites; GS2D is its 2-dimensional geometry system. Both PROTEAN and SightPian make use of ACCORD.

oriented programming” as the KSs cannot directly communicate with one another.) In each step towards the solution, one—and only one—KS gets to execute its proposed action to make certain changes to the BB. In reaction to such changes on the BB, other KSs may now focus their attention or propose to take action. It is the moderator in the meeting—here called the scheduler——which, at each cycle, listens to the contributions that KSs propose and who selects the best KS, which then gets to execute its action. Thus the inference mechanism of BB1, as embodied by the scheduler, is both incremental and opportunistic. 4.2 Conceptual Graph Representation inference mechanism of BB1, as embodied by the scheduler, is both incremental and

makes it easy to build a new application by substituting the needed BBs only. The CONCEPT-BB groups the most abstract concepts from which domain-specific concepts stem. The SITE-BB groups those concepts defining objects, for instance those related to site layout and construction management, and objects on the SITE-BB, in turn. can be specialized to concepts related to power-plant construction on the POWER-PLANT-BB. Finally, those examples of specific objects that exist on one particular site—here, for instance, the Intermountain Power Project—are on the INTERMOUNTAIN-BB, those that exist on another site—for instance, the American project—-are on the AMERICAN1-BB. Solutions to the problem the system is solving are generated in terms of instances of example objects, and these instance: are on the respective solution BBs (PROTEAN-SOLUTION, IPP-SOLUTION AMERICANI-SOLUTION). In this way, the CONCEPT-BB can be specialized tc accommodate any representation of more specialized worlds. concepts stem. The SITE-BB groups those concepts defining objects, for instance those related to site layout and construction management, and objects on the SITE-BB, in turn,

adjectives (Figure 4.7 shows the large frame). The SITE-BB contains concepts that are generally talked about on construction sites. such as trailers, laydown areas, and roads. These concepts are represented by nouns in spoken language. The SITE-BB also contains adjectives that field practitioners commonly use, such as: a “wide road” as opposed to a “narrow road,” or a “large laydown area” as opposed to a “small laydown area.” Concepts are classified by /evel to make the BB more comprehensive (Figure 4.6). For example, the PHYSICAL-OBJECTS level represents the physical objects building, laydown, and road with their relevant attributes and links (Figure 4.7 shows the building frame), whereas the MODIFIER level groups the adjectives (Figure 4.7 shows the large frame). returns a numeric value.

coordinates of the upper-left corner of the rectangle representing it on that site. The CONSTRAINT-BB represents concepts used as spatial prepositions or adverbs in common English (such as adjacent-to and near). Frames that name constraint types represent these concepts expressing unary, binary, or n-ary relations between objects (Figure 4.8). These types are exemplified by the actual example constraints that involve the related example-objects. The CONSTRAINTS level on the SITE-BB contained constraint types such as closer-than, further-than; each constraint type in turn labels its separate level on the CONSTRAINT-BB that groups the constraint’s examples (Figure 4.9).

Figure 4.8: Frames Representing a Constraint Type and a Constraint Example

Figure 4.9: Excerpt from the CONSTRAINT-BB Displayed by Levels

Figure 4.15: Type Hierarchy of Domain KSs on the SIGHTPLAN-KS BB ftpthwe is-a: SIGHTPLAN-KS.SKILL.SIGHTPLAN-DOMAIN-KS implements: SIGHTPLAN-KS.SKILL.SIGHTPLAN

Figure 4.16: Cooperation of the Domain Knowledge Sources in BB1 4.3.3.2. Control Knowledge Sources

Figure 4.17: Cooperation of the Control Knowledge Sources in BB1

Figure 4.20: Type Hierarchy of ACCORD Figure 4.20 shows the type hierarchy of ACCORD.

Figure 4.25: Convention on the Representation of an Essential Area Figure 4.26: Example of an Essential Area of a Rectangle and its Graphical Display The rectangle has dimensions 60 by 30 and was positioned in an area of dimensions 300 by 100.

Figure 4.27: Application of the CLOSER-THAN Constraint to Two Objects illustrated the definition of distance with an example. Figure 4.27 shows how GS2D reduces the sets of possible positions of two objects to allow only for positions where the objects are CLOSER-THAN a given distance to each other. People unfamiliar with GS2D will need to get used to the fact that GS2D computes on rectangles representing sets of possible locations for the centerpoint of each entity, rather than on the entities themselves. Also, according to the design of GS2D, distances between entities are defined in terms of the minimum orthogonal distance between the edges of objects, and this may be counter-intuitive. As I have argued in Chapter 3, each simple definition of distance carries some disadvantage of this kind. Figure 3.3 c illustrated the definition of distance with an example.

Figure 4.28: SightPlan, the expert system for construction site layout, Interfacing with SightView, the remote graphical display.

Figure 4.29: The Communication Interface (Figure modified from one by Tony Confrey) In this setup, I removed those KSs in SightPlan that were needed to reason about when to display objects (note that J could have done this while using SightView without the CI as well). When blackboard changes occur, the system sends off those changes concurrently (in the sense of “in the same action”) to the CI, which routes it to SightView. The user of SightView then decides what to display and when. Eventually, SightView itself could be extended to be more intelligent about this. In this way, the three cognitive tasks performed by the joint system-—reasoning, communication, and perception— are distributed over SightPlan, the CI, and SightView (Figure 4.30). are distributed over SightPlan, the CI, and SightView (Figure 4.30).

Figure 6.4: SightPlan includes the Context of the site in Delta, Utah These boundaries further define limits on the location of objects that will be positioned in

The system recognizes that there are two major aggregations of objects to be included, the IPP-CONSTRUCTION-ENTRANCE (ENTRANCE) and the IPP- CONSTRUCTION-FACILITIES (FACILITIES), and chooses to include those. Because they do not have a given location on site, they will need to be positioned. An aggregatior of objects is represented by a so-called aggregate object on the PROBLEM-BB, that is an object that includes other objects. As of yet, these two aggregate objects are shapeles: and undimensioned. Figure 6.6 shows the “includes” link between the ENTRANCE aggregate and its parts. In this example, each part has a prespecified geometry representec by the shaded rectangles drawn to scale to the right of the part’s label.

permanent facilities look like.) SightPlan draws them at their known location on site. railroads. (So far, the person laying out the site has identified what the site and the

Figure 6.6: Example of the IPP-CONSTRUCTION-ENTRANCE Aggregate Object Showing the includes Links to its Parts, and the Geometry of the Parts

No specific context is known, but all the parts of the ENTRANCE need to be included (Cycles 60-72, Figure 6.9). (Note that some intermediate steps are for display actions.) SightPlan picks the largest one as anchor of the arrangement (Cycle 77, Figure 6.10), with respect to which the other parts will be positioned.

Figure 6.9: PA2 Includes the Parts of IPP-CONSTRUCTION-ENTRANCE

Figure 6,11: Largest Part Meets “non-overlap” Constraint with the Anchor “adjacent-to” constraint (Figure 6.13) with the anchor (Cycles 82-136, Figure 6.14). of” and “between-short-sides” constraints (Figure 6.12), “closer-than” constraint, and/o1 “adjacent-to” constraint (Figure 6.13) with the anchor (Cycles 82-136, Figure 6.14).

Figure 6.12: Largest Part Meets “North-of” and “between-short-sides” Constraints

Figure 6.13: Largest Part Meets “closer-than”Constraint with the Anchor

Figure 6.15: SightPlan Defined Boundaries on Context of Partially Laid Out PA2 With this estimate, the system focuses again on the first arrangement PAI, and

and where they do not overlap with any occupied space (Cycle 162, Figure 6.17).

Figure 6.17; IPP-CONSTRUCTION-FACILITIES has met all constraints except the preference constraint “as-close-as-possible to POWER-UNIT-1”

updates the occupied space to account for this positioning (Cycle 166). SightPlan selects the one as close as possible to unit 1 (Cycle 164, Figure 6.18) and

main access road to the site and below unit 1. Figure 6.19 depicts SightPlan’s AE solution

Figure 6.20: Completed ENTRANCE Aggregate Layout Now that overall positions for each of the aggregates have been found, SightPlan can further detail the parts of these aggregates. Attention focuses on the second partial arrangement again. Of the ENTRANCE, the only part that has constraints with objects outside of the aggregate is the WELDER-TESTING. Its constraint, to be as close as possible to unit 1, is met. The constraints on the other parts are also met, object by object, and this results in a single location for each. Figure 6.20 shows the resulting layout. In a similar way, SightPlan can then proceed to create a third partial arrangement and layout out the parts of the FACILITIES aggregate. This concludes the task of the AE. layout out the parts of the FACILITIES aggregate. This concludes the task of the AE. In a similar way, SightPlan can then proceed to create a third partial arrangement and

space (Cycle 29), and by including the sub-areas of the site (Cycle 33-36). All long-term

SightPlan positions the largest LAYDOWN in its area by meeting the LAYDOWN’s constraints with the fixed facilities on site in the following order (numbers in parentheses are numeric weights for the matching constraints): zoned-in (0.98), zoned-outside-of (0.95), non-overlap-set (0.93), non-overlap (0.9), at-long-side (0.85), betw-short-sides (0.83), adjacent-to (0.8), north-of (0.7), south-of (0.7), west-of (0.7), east-of (0.7). closer-than (0.6), further-than (0.5), parallel (0.4), perpendicular (0.4), discrete-sample (0.3), as-close-as-possible (0.2), pick-one (0.1). When all constraints are met, the las: one to be applied is the as-close-as-possible constraint, which picks one single position from those that remain possible. Then the occupied space is updated. constraints with the fixed facilities on site in the following order (numbers in parentheses

6.1.3 Discussion of the Expert Strategy on IPP

Figure 6.24: An Example of the Instantiation Problem Due to Early Commitment positioned on site, except that FACILITIES are just marginally closer to power unit 1 . 7 + 7 * ae rt ea oy: AI researchers have identified these types of problems and have proposed several ways for avoiding them. For example, one may assess the quality of the outcome of an action before taking the action, one may undo the results of actions that were taken earlier, or one may postpone commitment. I will return to postponed commitment and investigate how that strategy affects problem-solving in Section 6.3. Other methods for improving the least-commitment strategy have not been investigated.

TRAILER, and PARKING) will be referred to as long-term facilities in the remainder of this section 6.2. The system ends its definition of this partial arrangement by orienting it (Cycle 44, Figure 6.26). SightPlan zones each of the facilities as needed inside or outside the sub-areas of the site (Cycles 50-56). After all these constraints are met, the system pursues its early- commitment strategy in picking one facility and meeting all constraints between that facility and objects with fixed location on site (Cycle 63). Finally, SightPlan selects a single position by satisfying that facility’s preference constraint (pick the position as close as possible to another fixed facility or to a fixed object on site, or pick one position out of the facility’s set of possible positions at random) (Cycle 64). The occupied space is updated (Cycle 65) and the process repeats for the following facility, until all facilities have a single position in the layout (Cycles 63-84, Figure 6.27).

Figure 6.27: AM1 with Permanent and Long-term Temporary Facilities

KS). In the subsequent BB1 cycle, the object can be included in the PA. Figure 6.29: Four Objects on the AMERICANI-BB with Attributes Describing their Construction Period (from start-construction to end- construction) or their Time Period on Site (from mobilize to demobilize) the PA. The resulting new state makes the KSAR that proposes to include the object in

‘igure 6.37: PA1O with Time Frame 880630-88070 No short-term laydown areas are on site.

Figure 6.39: PA7 with Time Frame 881231-89053: No short-term laydown areas are on site.

because hard constraints are not always the most efficient ones to compute. When all constraints on all objects are met, and objects remain with sets of possible locations (Figure 6.43), SightPlan introduces additional constraints into the problem to heuristically reduce those sets. Each set of possible locations will be restricted by sampling the corner points of the rectangles in their essential area by examples of the SAMPLE-FOUR-CORNERS constraint (Cycle 182, Figure 6.44).

Figure 6.43: All Objects met their Non-Overlap Constraint with the Permanent Facilities Subsequently, SightPlan uses these limited locations for generating coherer

it considers them to be alternative solution arrangements (Cycle 239, Figures 6.45, 6.46, and & AT,

Figure 6.45: One Coherent Instance of a Solution Layou' Generated by the Computational Model Applied to IPP Figure 6.46: A Second Coherent Instance of a Solution Layou Generated by the Computational Model Applied to IPP Figure 6.46: A Second Coherent Instance of a Solution Layout Generated by the Computational Model Applied to IPP

Figure 6.47: A Third Coherent Instance of a Solution Layout Generated by the Computational Model Applied to IPP The system user, or SightPlan, can then evaluate those solutions and select the best one. For example, to evaluate the alternatives, SightPlan computes the sum of the distances between the centerpoints of each LAYDOWN and POWER-UNIT-1. COHI-1-2 rates 72,055, COHI-1-3 72,055, and COHI-1-4 74,687. Although these values provide a comparative measure, the reader will agree that this evaluation value does not assess the value of each layout for construction. I only programmed this feature in SightPlan to illustrate the concept. Significantly better measures for site layout evaluation do not exist to my knowledge but should be developed. The system user, or SightPlan, can then evaluate those solutions and select the best

Connect with 287M+ leading minds in your field

Discover breakthrough research and expand your academic network

Explore
Papers
Topics

Features
Mentions
Analytics
PDF Packages
Advanced Search
Search Alerts

Journals
Academia.edu Journals
My submissions
Reviewer Hub
Why publish with us
Testimonials

Company
About
Careers
Press
Help Center
Terms
Privacy
Copyright
Content Policy

580 California St., Suite 400

San Francisco, CA, 94104

© 2025 Academia. All rights reserved