A self-evolutionary model for automated innovation of construction technologies

Automation in Construction 27 (2012) 78–88

Contents lists available at SciVerse ScienceDirect

Automation in Construction journal homepage: www.elsevier.com/locate/autcon

A self-evolutionary model for automated innovation of construction technologies Wen-der Yu a,⁎, Shao-tsai Cheng a, Chih-ming Wu b, Hou-rong Lou a a b

Department of Construction Management, Chung Hua University, Hsinchu 300, Taiwan, ROC Ph.D. Program of Technology Management, Chung Hua University, Hsinchu 300, Taiwan, ROC

a r t i c l e

i n f o

Article history: Accepted 18 April 2012 Available online 9 June 2012 Keywords: Computer‐aided innovation TRIZ Genetic operation tree Construction technology innovation

a b s t r a c t Previous approaches for innovation of construction technologies are constrained by the existing processes or engineer's experience and knowledge, thus are essentially incremental. This paper presents a selfevolutionary approach to assist automated innovation of construction technologies. The proposed approach integrates a text mining technique, patent analysis, and a Genetic Algorithm (GA) to form a prototype automated radical technology innovation model that has not been developed before. Previous technology information stored in the public technological repositories (e.g., published specifications, public patent databases, etc.) is adopted as the design knowledge for building the function model of a target technology. It is then translated into a genetic operation tree (GOT) for the self-guided evolution with a GA. Finally, the innovative solution is recovered as a function model and realized in a 3D model. A traditional road manhole construction technology is selected as demonstration case study to show the feasibility and potentials of the proposed method for automated innovation of construction technologies. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Creative economy has been identified as the major driving force for the next wave of world's economic growth [5,8]. It is especially meaningful to the current challenges of global change to the construction industry [29]. No matter the challenges are due to the extreme weather effects or over-urbanization, they have never been confronted and dealt with by the human beings before. Robert Harris pointed out that the Construction Engineering and Management (CEM) academia has overlooked the construction technology research in the past [11]. Shen identified construction technology research as a key approach to meet the future challenges of construction sustainability [29]. Due to the adopted research techniques, mostly related to process-based simulations [9] or classical constructability improvement methods [35], the technology innovation of the previous research was relatively limited and incremental [43]. Just recently, patent analysis and the theory of inventive problemsolving (TRIZ) were adopted in the innovation of construction technologies [21,22,27]. Such methods escape from the existing processes-based approach and generate innovative solutions based on global human intelligence (mainly from the public patent databases), thus result in automated and radical innovations that may bring in significant benefits for human beings [43].

⁎ Corresponding author. Tel.: + 886 3 5186748; fax: + 886 3 5370517. E-mail address: [email protected] (W-d. Yu). 0926-5805/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.autcon.2012.04.018

The paper presents preliminary results of a research project on developing a Model for Automated Generation of Innovative Alternatives (MAGIA) that integrates the most advanced computer‐aided innovation (CAI) techniques and a specialized genetic algorithm (namely genetic operation tree, GOT) to form a prototype automated radical innovation model for construction technologies. The proposed MAGIA intends to lower the uncertainties involved by human judgments (and thus resulted in local optimum) and reduce the manual efforts required for processing huge amount of technological information during innovation process. A specialized technology modeling technique and a self-guided model optimization algorithm are adopted to automate the alternative generation of technology innovation process. A prototype system is developed to implement and test the proposed MAGIA. A real world example of traditional road manhole construction technology is selected as case study to demonstrate the feasibility and potentials of the proposed method. Issues regarding to the implications to future construction challenges, the assumptions, and the limitations of the proposed model are addressed and discussed. 2. State of the art in construction innovation Innovation of construction technologies has resulted in dramatic revolutions in construction practice. For example, the introduction of Portland cement in 1824 has brought up thousands of new construction technologies and equipment that completely change the way of construction engineering. In the first quarter of the 20th century, the steel structural technology was invented and introduced to construction industry, which triggered a second wave of revolution

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

for construction technologies. During the late 1970s, the construction industry suffered in low productivity, hence inspired the next generation of construction innovation. Issues such as constructability [25], prefabrication, modularization [35], and automation [31] have drawn numerous researchers to devote in the innovation of construction and management processes. In spite of the tremendous efforts spent, innovation in construction industry has been relatively slow [39]. Lack of a common framework, as pointed out by Halpin [10], may be blamed for this lag. Previous researchers have exploited different approaches for organization process innovation [34], technology evaluation [42], and advanced technology repositories [13]. However, few of these efforts target were directly related to design of new technologies. Halpin proposed a CYCLONE model for analysis and improvement of construction processes [9]. Many efforts on construction process simulation followed him, e.g., COOPS [19], STROBOSCOPE [23], etc. Most process simulation techniques are still limited to the modeling of existing processes, rather than the invention of new processes or technologies. Just recently, a new area of construction innovation has been developing on patent analysis (PA) [27] and the Theory of Innovative Problem Solving (TRIZ) [21,22,36]. The former innovates the target technology based on existing technologies of the other areas, which are stored in public patent databases; the latter applies a systematic procedure to identify potentially improvable engineering contradictions with tools provided by TRIZ [2]. Unlike the simulation-based approach that innovates on the existing construction processes, the PA- and TRIZ-based technology innovation methods seek a different dimension of technology improvement by introducing technological information from outside of construction domain. The former belongs to “incremental innovation”, and the latter belongs to “system innovation” or “radical innovation” according to the classification of Sarah Slaughter [30]. Although the PA- and TRIZ-based methods open a new window for radical innovation of construction technologies, they still suffer in requirement of heavy human involvement during the alternative generation in the innovation process. Such requirements do not only hinder the adoption of those methods but also form a bottleneck of technology innovation. 3. Function modeling for construction technologies Before an automated technology innovation model is developed, it is required to define a “common language”, as appealed by Halpin [10], for describing the characteristics of a target technology. For this end, the technology characteristics should be translated into a model that is operational for computer‐aided innovation. This section describes a relevant model widely adopted in mechanical and product design, namely function model (FM) [26], from the viewpoint of construction technologies. 3.1. Vocabulary of function modeling Functional modeling is a critical method and the key step in a product design process [32]. It has been widely adopted in mechanical and many other areas and has demonstrated its benefits for assisting product designers [16,26,32]. There have been numerous functional modeling methodologies proposed by different researchers ([16,26,38,44-46]). All of which follow a similar procedure: they begin with an overall product function and then break that function down into sub-functions via a process similar to the Work Breakdown Structure (WBS) method that is widely adopted in construction project management. The early efforts on function modeling research traced back to the value analysis of a product [1,20]. In value analysis, the functions of a product are defined in terms of a verb plus a noun (V+ N). A list of verb-noun functions was suggested to represent the common functions associated with a product. The functional modeling method proposed

79

by Pahl and Beitz [26] may be the most well-known. A schematic representation of the basic FM based on Pahl and Beitz's methodology is shown in Fig. 1. They model the overall function of a product and decompose it into sub-functions operating on three types of flows: energy, material, and signals. Their function modeling approach was a great advance in engineering design, but their methodology did not provide a comprehensive list of sub-functions to describe all possible engineering systems or produce repeatable functions [15]. Two engineers may generate two different function models for a set of same product/ technology and customer needs. Many researchers have tried to improve the weakness of Pahl and Beitz's methodology by developing a common vocabulary for functional modeling [15,16,32,38]. A summary of the common vocabulary for the functions and flows is shown in Tables 1 and 2, respectively. 3.2. Subject–action–object function models In addition to the Pahl and Beitz's FM and their families, another popular FM that is widely adopted by commercial computer‐aided innovation (CAI) software (e.g., Goldfire Innovator®, CREAX®, etc.) is the Subject–Action–Object (SAO) function models. The SAO FM in systems engineering and software engineering is a structured representation of functions, activities or processes within the modeled system or subject area [24]. A SAO FM defines the relationships between system elements in terms of the functions they perform [18]. In a SAO FM, a function is an “action” that directly changes or maintains a controllable or measurable parameter of a (material) object." The SAO FM is depicted in Fig. 2. Examples of actions are move, remove, burn, weld, count, deposit, inform, rotate, hold, conduct, carry…, etc. The SAO FMs represent a system (describing the functions of a product/technology) with two natural language templates: (1) Action–Object (AO), e.g., move (A) table (O); (2) Subject–Action–Object (SAO), e.g., conveyer (S) moves (A) table (O). A parameter is a directly measurable or controllable characteristic associated with a material object which is affected by a function. Examples of parameters affected by actions are length, area, volume, mass, density, volts, bits, joules, coulombs, temperature, roentgens…, etc. A common way to identify the SAO FM for a product/ technology is by asking the following questions: (1) What parameter of the object is controlled of changed by the action? (2) How do I measure action object? Comparing the Pahl and Beitz's FM with the SAO FM, it is identifiable that the black box function of Fig. 1 (defined in Table 2) comprises Subject and Action of Fig. 2; the flow of Fig. 1 (described in Table 1) is relevant to the Object and the associated controllable parameter of Fig. 2. As a result, a Pahl and Beitz's FM can be exactly transformed into an SAO FM. Besides, the SAO FM is more relevant to the V+ N function definition in value analysis [20]. It is also more intuitive to the construction engineers who are familiar with the value analysis. As a result, the SAO FM is adopted to model a construction technology hereafter in this paper. 3.3. Process of FM generation The SAO FM provides a useful tool to capture the functional characteristics of a construction technology. In this sub-section, a generic process for generating function models is introduced. Kurfman et al. [15] proposed a five-step generic procedure for deriving the FM of a product/technology: (1) identify flows that address customer needs—identify the flows (the physical phenomena) that the Input (Flow)

Black Box Model (Function)

Output (Flow)

Fig. 1. Schematic representation of Pahl and Beitz's FM.

80

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

Table 1 Flow classes and their basic categorizations [15]. Class

Material

Signal

Energy

Sub-classes

Human Gas Liquid Solid Plasma Mixture

Status Control

Actuate Regulate Change Stop Biological

Action

Subject Electrical Chemical Electromagnetic Hydraulic Magnetic

Radioactive Mechanical Pneumatic Thermal

product is to operate on; (2) generate a black box model—generate the graphical representation of a required function with input/output flows based on the customer needs; (3) create function chains for each input flow—develop a chain of sub-functions that operate on the flow; (4) aggregate function chains into a FM—aggregate all of the function chains generated in Step 2 into a single complete FM; and (5) verify the FM with customer needs—check and ensure that each customer need is addressed by at least one sub-function, if not, iterate through the tasks again starting at Step 2. Kurfman et al. [15] also show that with the five-step procedure described above, the repeatability of the generated function models is significantly improved. That is, different engineers would have a higher likelihood to generate similar FM for the same customer needs. Although Kurfman et al.'s procedure provides a good guideline for engineers to develop their function models, two requirements are necessary before a successful FM is built: (1) the professional training and practical experiences in product/technology design for the engineers [15]; (2) the design knowledge source that provides plentiful alternatives of proven functions [32]. For the latter, the knowledge source of a construction technology can be obtained from technical documents such as engineering text books, research theses, technical reports, historical construction plans, and some technology repositories (e.g., published specifications and public patent databases). No matter which source of the technology information is from, it is recorded in form of text and figures. Several previous researchers have adopted public patent databases as the knowledge sources for technology innovation [3,27,43]. Yu et al. [41] further the adoption of patent information by employing a text mining technique to analyze the technological information stored in a patent document. In their method, individual SAO structures are identified by separating the key components (Subjects and Objects) depicted in the illustrative figures from the text of the claim of a patent document. Then, the individual SAO structures are aggregated to form a complete function model of a technology. Integrating the method of Yu et al. [41] and the procedure of Kurfman et al. [15], a five-step procedure for generating function models from a technology document is proposed in Fig. 3: (1) search technology database (e.g., specifications and patent databases)—find out the target technology based on the innovation problem (or customer needs); (2) identify key components—identify the important components (and their descriptive terms) of the target technology with the elements depicted in the figures of the document; (3) identify SAO chains—identify individual SAO structures from the claims of the technology document with a text mining technique; (4) aggregate SAO chains—form the complete FM of the target technology by integrating the individual SAO chains; and (5) verify the FM—make

Object (Controllable parameter)

Fig. 2. Example of SAO FM.

sure that each innovation problem or customer need is addressed with at least one SAO chain, if not, iterate through the steps again starting at Step 3. 4. Self-evolutionary model for automated generation of innovative technology alternatives Once a FM of the target construction technology is constructed via the process suggested previously, the remained problem for an automated technology innovation model is a computational method to generate radical innovative alternatives (i.e., the alternatives that escape from the restrictions of the existing technology) that achieve the overall objectives of the technology. A Genetic Algorithm (GA) is adopted for this purpose; since mutations of the evolutionary process may introduce the radicalness of the innovation. In this section, a self-evolutionary model, namely Model for Automated Generation of Innovative Alternatives (MAGIA), for generating innovative technology alternatives is proposed and described in details. 4.1. Problem formulation In developing the proposed MAGIA, three critical problems should be tackled first: (1) a modeling language to represent a construction technology in a GA should be developed; (2) an appropriate objective function to guide the evolution of technology innovation should be defined; (3) an automated optimization algorithm should be developed so that the optimal solution can be reached. 4.1.1. Modeling language To solve the first problem, a genetic operation tree (GOT) model originally proposed by Yeh and Lien [40], for constructing the formulas to predict compressive strength of High-Performance Concrete, is adapted to represent the SAO FM in a GA. The GOT is an extension of GAs, which is a tree-like diagram, structured hierarchically to represent or model the domain problem. Fig. 4 shows an example GOT for representing the SAO chain of a frame structure. Where the frame structure is broken down into two individual SAO structures: (1) Column (Subject) supports (Action) Beam (Object); and (2) Beam (Subject) supports (Action) Slab (Object). Each individual SAO structure constitutes a branch of a GOT. As a result, all SAOs collectively form a complete GOT of the target technology. Then a GOT is encoded into a genotype as shown in Fig. 5. 4.1.2. Objective function The objective function for technology innovation should be defined according to the desired functions provided by the target technology to solve the innovation problems or to meet the customer needs. As a result, the objective function is highly domain-dependent. Moreover a quantitative objective function calculated directly from a FM is very desirable, so that no human involvement is required during the evolution

Table 2 Function classes and their basic categorizations [15]. Class

Branch

Channel

Connect

Control

Convert

Provision

Signal

Support

Basic

Separate Distribute

Import Export Transfer Guide

Couple Mix

Actuate Regulate Change Stop

Convert

Store Supply

Sense Indicate Process

Stabilize Secure Position

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

S11

Start

O13 ...

A 12

Chromosome 1

1. Search technology database

4. Aggregate function chains

NO

YES End Fig. 3. Procedure of deriving a FM.

AppEPk ðIP i Þ ¼

process. After trials-and-errors, an objective function is defined to maximize the usefulness of key components (Subjects and Objects) and the potential of TRIZ improvement for Action links. The usefulness of the key components is measured by the number of input and output flows and relevance to the main function of the technology. This is defined in Eq. (1). The underlying idea is that the more input/output flows connecting to the component and the more relevant to the main function, then the more useful it is. n X

  Fit Compj

ð1Þ

j¼1

where Fittotal(Compj) is the overall fitness value of all key components in the FM; Fit(Compj) is a measure of the fitness value of the j th key component (Subject or Object), which is further defined in Eq. (2); and n is the total number of key components in the FM. Fit ðCompi Þ ¼ Relevancei  Importancei NL ¼ Ri  n i X NLj

ð2Þ

j¼1

Beam

Chromosome i

...

S n1

A n2

On3

...

Fitness

Chromosome n

ð3Þ

j¼1

where APPEPk(IPi) is the value of the applicability weighting of IPi for EPk; Frq(IPi) is the occurrence frequency of IPi that is applicable to improve EPk in the TRIZ contradiction matrix; the denominator is the summation of the occurrence frequencies for all IPs applicable to improve EPk. As a result, the potential for TRIZ improvement of IPi to EPk can be defined with the following equation.   Fit Actionq ðEP k ; IP i Þ ¼ AppEPk ðIP i Þ

ð4Þ

where Fit(Actionq(EPk, IPi)) is the fitness value of the q th Action link while applying IPi to improve EPk (relevant to the desired function). Finally, the overall fitness value of a FM is calculated by summing up all fitness values of the key components (Subjects and Objects) and Actions: n m   X   X Fit FM p ¼ Fit ðCompi Þ þ Fit Actionq ðEP k ; IP i Þ

ð5Þ

q¼1

where Fit(FMp) is the overall fitness value of the p th alternative (i.e., FMp); m is the total number of Action links in the FM; the other symbols are defined as those in Eqs. (2) and (4).

Frame Structure

Column

Oi3 ...

FrqðIP i Þ   40 P Frq IP j

i¼1

Support

A i2

where Fit(Compi) is the fitness value of the i th key component; Ri is the relevance value of the i th key component assessed by domain expert according to his/her judgment on the relevance of the component with respect to the main function of the technology; NLi is the total input/output link flows connecting to the ith key component; the denominator is the summation of the link flow numbers of all components in the function model; and n is the total number of key components. In Eq. (2), the importance of a key component is determined by the number of connecting links; while the relevance of the component is assessed by the domain expert (or engineer) based on his/ her knowledge. This is done only once before the evolution process. The potential for TRIZ improvement is measured by the frequency of applicable inventive principles (IPs) to the Action link using the “Single Engineering Parameter” method proposed by Chen and Liu [4]. The “Single Engineering Parameter” method defines the most relevant Engineering Parameter (EP) for the Action link. Then, the TRIZ contradiction matrix [2] is consulted to calculate the occurrence frequency of a specific inventive principle (IP) over the summation of all IPs for the relevant EP (i.e., the controllable parameter of the SAO structure). The applicability weighting of the i th IP (denoted as IPi) for the k th EP (e.g., EPk) is then calculated by Eq. (3) as suggested by Liu [17].

3. Identify SAO chains

Fit total ðCompÞ ¼

S i1

Fig. 5. The GA encoding of a sample GOT.

2. Identify key components

5.Verify functional model

...

81

Support

Beam

Slab

Fig. 4. A sample GOT model of a frame structure.

4.1.3. Optimization algorithm To tackle the third problem, a simple GA [12], is adopted for the optimization of the objective function defined previously. Before GA is applicable, the SAO FM of the target technology is translated into a GOT first; then the GOT model is encoded into a genotype as shown in Fig. 5; finally, the optimization of the objective function is achieved by performing GA operations on the GOT.

82

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

4.2. Technology knowledge repository As pointed out by Stone and Wood [32], one critical factor that affects a successful technology design is the existence of design knowledge archival for the technology domain. The patent database containing 90–95% commercially valuable technology information [3] can be employed for this purpose. Utilization of technology knowledge recorded in other public technology documents, e.g., the published specifications, provides alternative sources for the technical knowledge required by MAGIA. However, the knowledge stored in the technology documents needs to be interpreted by human user before it is useful for technology innovation. To attack this problem, text mining techniques including keyword extraction, full-text information searching, document classification, and text summarization may provide a solution [7,33]. In this paper, a text mining technique proposed by Yu et al. [41] is adopted to extract the SAO function models from a technology document. 4.3. Operational procedure

individual SAO structures are integrated to form a complete FM. Technology repositories can be consulted in this step to obtain the technological information of the target technology. (3) GOT encoding The GOT model of the target technology is constructed based on the FM. In the GOT, the key components (Subjects and Objects) are placed at the bottom of the tree and linked with the Action nodes, see Fig. 4. The GOT is then translated into a genotype of GA, see Fig. 5. (4) Key component relevance assessment The relevance of the Subjects/Objects (S/O) elements with respect to the desired functions is assessed and evaluated by the domain experts (or engineers). The assessment is made by evaluating the relative contribution (scored by “1” as less important and “2” as more important in this paper) of the considered component, in the whole structure to the main function of the technology. The relevance values will be utilized later for fitness calculation. (5) Evolution operation

A ten-step procedure is proposed to implement the proposed MAGIA, see Fig. 6. Details of the proposed procedure are described as follows: (1) Problem and main function definition The first step of MAGIA starts with problem definition. Usually, the root cause analysis (RCA) of the problem is analyzed to identify the root problem causes (or customer needs). (2) Function modeling The second step is to build the SAO structures for the target technology. In this paper, a text mining method is adopted [41]. At first, all Subjects, Actions, and Objects of the target technology are identified. The links among Subjects and Objects are connected, and the

The evolution operation is performed on the GOT. In the first run, the initial fitness value is evaluated to give a datum of fitness improvement. Then, GA operations such as reproduction, crossover, mutation, and selection, are conducted on the GOT to generate new off-springs of the parent genotype. (6) Fitness calculation Fitness values of each generated genotypes are evaluated using Eqs. (1)–(5) to provide an indicator for evaluating the superiority of alternatives. (7) Stop criterion judgment The stop criteria such as maximum number of generations, target fitness value, difference between two generations, etc. are checked with the evolution results to determine the timing of evolution termination. If the termination criterion is not reached, go back to Step 5. (8) Decoding

1. Problem and main function definition

Target technology

The genotype with the best fitness is decoded to recover the associated GOT. (9) FM construction

2. Function modelling

The FM of the recovered GOT is constructed for the resulted innovative technology alternative. 4. S/O relevance assessment

3. GOT encoding

5. Evolutionary operation 6. Fitness calculation

No

In order to test the feasibility of the proposed MAGIA method, a prototype MAGIA system integrating a computer program developed by VB.net and the commercial GA software, GeneHunter® is developed. A real world case of traditional road manhole construction technology is selected for case study.

Yes

8. Decoding

9. FM construction

5.1. Case background No (revise FM)

Yes

10. Accept?

Fig. 6. Flowchart of MAGIA.

Finally, the resulted FM is verified with the customer needs/ problem requirements to determine the acceptance of the innovative alternative. If the alternative is acceptable, it is implemented in the next stage; otherwise, the resulted FM is revised by the engineer to produce another alternative. 5. Demonstration case study

7. Stop?

Implementation

(10) Check acceptance

The selected technology innovation case is a road manhole construction method. As Taiwan has become a developed country, many infrastructure systems are aging, including highways, utilities (e.g., electricity, water supply, and gas), sewage, communication

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

(e.g., TV, telephone) conduits, etc. As common conduits are rarely adopted in Taiwan, most of the pipeline conduits are constructed separately under roadways in the urban and suburban areas. Manholes are constructed in order to maintain the utility pipelines. It is found that the average lifetime of the traditional road manhole is less than two years. The damaged manholes have become one of the major causes of deterioration of road pavement. Developing a more durable manhole technology will extend the lifetime of the facility and thus reduce the wastes of construction materials and other resources used in road rehabilitation to improve construction sustainability.

Cover

5.2.3. GOT construction A GOT is constructed based on the extracted FM of the target technology; see Fig. 9 (partial presentation). 5.2.4. Genotype encoding The GOT is encoded into a genotype, where each SAO gene consists of three elements: (1) Subject—the subject elements are numbered as follows: Base-1, Form-2, RC-3, Neck-4, Bolt-5, Frame-6, and Cover-7; (2) Action—at first the most relevant engineering parameter

Cover

AC Road surface

Bolt RC

Cover Frame Neck

Base

Soil

connect

Frame

support surround

RC

Bolt connect

surround

Neck

support

Form

In the following, the MAGIA method is applied to the innovation of a traditional cast-on-site road manhole construction technology. The procedure depicted in Fig. 6 is followed with detailed explanations.

5.2.2. Key components extraction and function modeling The key components of the target technology are extracted from the construction specification using a text mining technique [41]. Totally, 11 SAO structures are extracted. The extracted SAO structures are integrated to form a complete FM as shown in Fig. 8.

hold

support

5.2. Application of MAGIA for innovation of road manhole technology

5.2.1. Target technology The target technology for innovation was selected from a published manhole construction specification [6], see Fig. 7. The target technology constructs road manhole with a precast manhole unit and an on-top cast-on-site RC neck structure to adjust the manhole cover to the level of road surface. The cast-on-site RC neck requires a lengthy curing period before service. In Fig. 7, the asphalt pavement (AC) tends to deteriorate due to the crack of the RC neck structure of the manhole.

83

support

encase

support

Base Fig. 8. The FM of the traditional road manhole technology.

(EP) is identified, eg., “EP-10: Force” is most relevant to Action “support” and thus numbered as “10”; (3) Object—the object elements are numbered as: Base-1, RC-2, Neck-3, Frame-4, Cover-5. The translation of the GOT of Fig. 9 is shown in Fig. 10, where Chromosome 1: “Base (Subject) supports (Action) RC (Object)” is translated into “1 (numeric symbol of Base)-10 (EP-10: force to characterize the Action support)-2 (numeric symbol of RC)”. The other chromosomes are translated similarly. 5.2.5. Fitness evaluation A fitness function consists of two parts: (1) fitness value of S/O key components; and (2) fitness value of Action links. The fitness of S/O key components is defined with Eq. (1). Table 3 shows the importance and relevance assessments of the key components in the original FM. From Table 3, Base, Form, RC, and Neck are considered as more important components for a manhole, thus are scored with relevance weighting value “2”; the other components are scored with relevance weighting value “1”. Two most relevant EPs, EP-25 (Waste of time) and EP-14 (Strength) are identified by the domain experts for the important SAO structures for the key components, since the curing time should be shortened and the strength needs to be improved for the traditional manhole. As a result, the fitness of Action links defined with Eq. (4) is assessed on the applicable IPs to improve EP-25 and EP-14. Using the “Single Engineering Parameter Method” proposed by Chen and Liu [4], the potentials of applicable IPs for EP-25 and EP-14 are shown in Table 4. It is noted that in the beginning of the evolution, the IPs are not applied to the EPs, thus the fitness function is zero. Finally, the overall fitness value of the complete FM

Road manhole

Soil

Deterioration Part support

Base

Fig. 7. Schematic diagram of traditional road manhole technology.

RC

encase

Neck Base

surround

RC

Frame

connect

Bolt

Frame

hold

Cover Frame

Fig. 9. The GOT model of target technology (partial).

84

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

Base supp. 1

10

RC

Neck

2

4

Chromosome 1

enc. Base 14

1

...

Cover

...

7

...

Chromosome 2

hold Frame Fitness 14

4

5.59

Chromosome 11

Fig. 10. The genotype for the GOT (partial).

is calculated by Eq. (5). Using the first SAO in Fig. 9 (Base support RC) as example, the fitness is “0.28 (Base) + 0 (in the beginning of evolution) + 0.36 (RC) = 0.64”. Applying Eq. (5) to all 11 SAO structures in the original genotype, the results are shown Table 5. The summed fitness is 5.60 (=0.64 + 0.64 + 0.59 + 0.32 + 0.45 + 0.73 + 0.59 + 0.45 + 0.55 + 0.32 + 0.32). 5.2.6. GA evolution With the fitness value defined previously, the genotype is readily adaptable with GA. Some important parameters for GA evolution are set including: (1) the maximum number of evolution generations is 60; (2) the population size is 200; (3) the crossover rate is 0.98; (4) the mutation rate is 0.1; (5) the Elitist strategy is adopted, i.e., only the best 200 genotypes are preserved at the end of each generation. A monitor of fitness value for GA evolution process is shown in Fig. 11. The optimum fitness obtained is 8.58. 5.2.7. Model realization The realization of the resulted innovative model is relatively involved compared to the other steps in the prototype MAGIA. It consists of the following steps: (1) applying the applicable IP's for the SAO structures in the genotype by selecting “appropriate Actions” (i.e., the application of a specific IP to improve the relevant EP of the Object); (2) recovering the SAO FM manually and refining the FM until all individual SAO structures are included; (3) replacing SAO structures with real world examples (the SAO examples in the original target technology is considered first in this step) until all SAO structures are included; (4) building a 3D model for the finalized SAO FM. Since the resulted FM suggests a “pre-action (IP-10)” applicable to improve the “Strength (EP-14)” of RC structure. The resulted innovative alternative suggests a “precast road RC neck structure for manhole” to replace the cast-on-site RC neck structure. The resulted FM of the innovative alternative is shown in Fig. 12. The schematic diagram of the innovative manhole technology is shown in Fig. 13. 5.3. Implementation and evaluation of innovative technology 5.3.1. Innovation implementation In order to implement the innovative road manhole construction method (i.e., the precast neck structure manhole technology), the detailed decomposition of the innovative technology is depicted in Fig. 14, where the ‘adjustment neck’ can be used to adjust the level of manhole; the ‘cover frame’ provides a block to the drainage

Table 3 Importance and relevance assessment for key components. Component

Out-link

In-link

Importance

Relevance

Fit(Compi)

Base Form RC Neck Bolt Frame Cover Total

1 2 1 3 2 1 1 11

2 0 3 1 0 4 1 11

0.14 0.09 0.18 0.18 0.09 0.23 0.09 1.00

2 2 2 2 1 1 1 11

0.28 0.18 0.36 0.36 0.09 0.23 0.09 1.58

Table 4 Potential of applicable IPs for EP-25 and EP-14. IP

10

Frequency for improving 15 EP-25 Frequency for improving 11 EP-14 Summation of 26 frequencies % 14% Weighting 0.14

35

28

18

3

15

29

26

40

14

10

12

2

1

6

5

0

12

6

3

12

9

4

4

9

26

16

15

14

10

10

9

9

14% 9% 8% 8% 5% 5% 5% 5% 0.14 0.09 0.08 0.08 0.05 0.05 0.05 0.05

channel so that soils of surrounding base layer would not leak into manhole. Since the precast neck structure is cured in the plant, the strength of the structure is improved. As a result, the asphalt pavement surrounding the manhole cover does not deteriorate. A nine-step construction process is planned for construction of the innovative manhole technology including: (1) site preparation— clear obstacles and set up fences and warning tapers around the construction; (2) excavation—dig the road to the required depth; (3) surveying—precisely position to the required elevation for the base of manhole; (4) set up the manhole—install the manhole structure; (5) set up adjustment neck—the adjustment rings are installed to the required elevation; (6) set up frame base—install the base of frame for the neck; (7) set up neck ring—install the ring between frame and cover as a cushion; (8) set up cover—put on the cover of manhole; (9) refill—refill the gap between manhole structure and the road structure, and pave the road to a smooth surface. The installation diagram of the innovative manhole technology is shown in Fig. 15. In order to implement the innovative technology, the prototype products are produced and assembled in the plant, see Fig. 16a. Experiments are conducted to test the compression strength of the prototype products to fulfill the requirement of regulations for public works by the government agencies, see Fig. 16b.

5.3.2. Innovation evaluation The evaluation of the innovative technology is performed in two stages: (1) Stage I—preliminary feasibility evaluation of the innovative technology is conducted by the domain experts in terms of functionality, constructability, and cost effectiveness; (2) Stage II—performance evaluation is conducted for real world applications of the innovative technology in terms of a set of durability performance indicators. The preliminary evaluation results are shown in Table 6. Since the traditional manhole technology employs workers to construct the RC structure on site, it requires more than eight hours for construction and curing. The longer construction time implies a higher risk for the workers to be injured by traffic. The lengthy construction and curing hours can be significantly reduced to less than one hour with the precast structure. Moreover, the structural strength of the manhole is enhanced since it is well cured in the plant. As a result, the functionality and constructability are improved. However, the precast method requires additional equipment and precast unit, which will increase the initial cost, and thus it is less cost effective compared with the traditional method. According to the data collected by the research team, the average construction cost (excluding the precast manhole unit) for a regular traditional cast‐on‐site manhole (236 cm× 126 cm× 208 cm) is around USD$275–420, while a precast manhole of the same size will cost around USD$590. The innovative technology has been adopted by a major communication firm, Chunghwa Telecom, Taiwan, in constructing the manholes of underground communication conduit. The performance evaluation is conducted on the real world implementation cases of Chunghwa Telecom. Totally 17 implementation cases of the innovative technology are collected as experimental group; the other 19

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

85

Table 5 Initial fitness of individual SAO structures. No.

1

2

3

4

5

6

7

8

9

10

11

S A O Fitness Sum

Base Sup. RC 0.64 5.60

Neck Eca. Base 0.64

RC Sur. Frame 0.59

Bolt Con. Frame 0.32

Bolt Con. Neck 0.45

Neck Sur. RC 0.72

Neck Sup. Frame 0.59

Form Sup. Base 0.44

Form Sup. RC 0.54

Frame Sup. Cover 0.32

Cover Hold Frame 0.32

traditional manhole cases are selected for control group. The control group cases are selected near the experimental group cases to control the effects of environmental variables. The survey is conducted two years after the construction of manholes. Five performance indicators are investigated as recommended by [37]: (1) pothole—any pothole of the diameter > 10 cm is found in the surrounding of the manhole; (2) pavement repair—any repair is found in the surrounding of the manhole; (3) pavement cracks—any pavement crack is found in the surrounding of the manhole; (4) levelness—any more than 6 mm difference in levelness is measured with a 3 m scale in the surrounding of the manhole; (5) structural cracks—any structural crack is found inside the manhole. The results are shown in Table 7. It is found that the innovative precast manhole technology has significant improvements in all five performance indicators compared with the traditional method.

6. Discussions The prototype MAGIA has been successfully applied to innovation of a traditional manhole construction technology. It shows several meaningful implications for the research on radical innovation of construction technologies. However, several assumptions are made during the development of the proposed MAGIA, and thus induce inevitable model limitations. The implications, assumptions, and limitation of the proposed MAGIA are addressed in this section.

6.1. Implications to construction technology innovation 6.1.1. Implications to radical innovation The proposed MAGIA method innovates technology alternatives based on the design knowledge sources from different technological domains (e.g., published specifications and patent databases). It provides a method to incorporate design knowledge of cross-disciplines and thus has a better chance to trigger the “Medici Effect” [14] that brings in radical ideas for technology innovation. Moreover, the selfevolutionary model of the proposed MAGIA introduces mutations in the process of genotype evolution. It may induce radicalness of the innovative alternatives and thus provides a prototype model of radical innovation. Such a model is different from the traditional process

simulation-based technology improvement methods and has not been found in the literature. 6.1.2. Implication to automated innovation The proposed MAGIA adopts a text mining technique to automatically construct the FM from a technology document (e.g., technical report, specification, patent document, etc.). Moreover, the innovation process of technology FM is performed by a GA on the GOT. Such a process minimizes the involvement of human engineers, and thus improves the previous computer-aided innovation methods to achieve a higher degree of automation. 6.2. Assumptions and limitations of MAGIA 6.2.1. Model assumptions The fitness function of the proposed MAGIA consists of two parts: (1) fitness of key components—measuring the relevance of the components to the main function and the number of input/output links; (2) potential of applicable TRIZ IPs—measuring the occurrence of the applicable IPs to the relevant EPs. The two fitness measures are assumed to be relevant to the objectives of technology innovation. Such an assumption is a compromise to fulfill the requirements of self-evolutionary computation algorithm. Expertise about the relevance of key components to the main function of the technology is involved during the relevance assessment by the domain experts. The “more linked” is “more important” assumption tends to preserve a highly linked component in the off-spring generations. It deserves more future researches for further verification. The measure of potential improvement of TRIZ IP to relevant EP is still based on the underlying assumptions of TRIZ method. Moreover, the identification of relevant EPs may also affect the effectiveness of the fitness function. Such an assumption also needs more practical applications for further verification. 6.2.2. Model limitations The proposed MAGIA has adopted function model (FM) as the technology modeling language. As a result, limitations of existing function modeling methods are also limitations for MAGIA, e.g., the dimensional, material, strength, and other physical characteristics are omitted by the existing FM methodologies. Incorporation of physical information modeling methodology, such as Building Information Modeling

Base encase

Bolt

connect

Neck

support tightly seal

RC

support connect

Frame

precast surround

support

Cover Fig. 11. Monitor of fitness for GA evolution process.

Fig. 12. FM of the resulted innovative technology for road manhole construction.

86

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

AC

Cover

Bolt

Cover

Cover Frame Cover Frame

Neck Neck Adjustment Neck

Adjustment Neck Soil Base

Base

Fig. 15. The installation diagram of the innovative road manhole technology.

Fig. 13. Schematic diagram of the innovative road manhole technology.

6.3. Issues for extension to complex technology systems (BIM), for technology modeling of MAGIA may improve the above problems. However, the adoption of BIM will dramatically increase the amount of information in the technology model, which will cause heavy computation burden for the evolution process of MAGIA and result in technical problems of the computation engine. The other limitation of the MAGIA model is the required human involvement for relevance assessment of key components (Step 4 of Fig. 6) and model realization of FM to practical implementation of the innovative technology. In the demonstration case study, such processes are performed by engineers with domain knowledge only once in the innovation process depicted in Fig. 6. Future version of MAGIA may employ proven industrial BIM models that are associated with the relevant SAO structures, so that the implementation of FM can be further automated.

The proposed MAGIA method has been applied to simple technology innovation of the manhole structure. Such a method may also be extended to more complex technology systems, provided that the FM of a complex technology system can be constructed and translated into a GOT model. An issue of computation efficiency will rise when the GOT model becomes too large. Schemes for improving the computation efficiency of MAGIA need to be developed.

a) Production of prototype product

Cover Cover Frame Neck Adjustment Neck

b) Testing of compression strength

Base

Fig. 14. Decomposition diagram of the innovative road manhole technology.

Fig. 16. Production of prototype products in plant.

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

Acknowledgment

Table 6 Results of preliminary evaluation. Criterion

Functionality Constructability Cost effectiveness

87

Technology Traditional

Innovative

Poor Medium Good

Good Good Medium (high initial cost)

This research project was funded by the National Science Council, Taiwan, under project no. NSC 99-2221-E-216-041-MY2. Sincere appreciations are given to the sponsor by the authors.

References Table 7 Results of performance evaluation. Performance indicator

Frequency of deterioration (%) Traditional

Innovative

1. 2. 3. 4. 5.

17 (89.5%) 19 (100%) 18 (94.7%) 19 (100%) 19 (100%)

1 (5.9%) 4 (23.5%) 4 (23.5%) 1 (5.9%) 1 (5.9%)

Pothole Pave. repair Pave. crack Levelness Structr. crack

Improvement (%) 83.60 76.50 71.20 94.10 94.10

A very useful scheme to deal with complex technology system will be partitioning of FM to form relative small isolated FMs. By processing small isolated FMs will generate radical alternatives for technology innovation. The other scheme may be hierarchical modeling of FM by modular FM models similar to that of hierarchical and modular simulations [28]. All these issues are worth of efforts for future researches.

7. Conclusions and recommendations 7.1. Conclusions Lack of radical technology innovation method has been a bottle neck of technology improvement in previous technology improvement researches. As suggested by several previous researchers, radical innovation in construction technology provides a promising solution to meet future challenges of construction industry such as the sustainability issues. This paper presents a self-evolutionary model, namely Model for Automated Generation of Innovative Alternatives (MAGIA), for automated generation of innovative technology solutions. A prototype MAGIA system is developed and a traditional road manhole construction technology is selected as demonstration case study to test the feasibility and potentials of the proposed method. The testing results show several desirable features of the proposed MAGIA method for a radical and automated technology innovation method including the automated function modeling capability with a text mining technique, the cross-discipline design knowledge repository from public technological databases (e.g., published specifications, public databases, etc.), and the self-guided optimization algorithm with a GA. It is concluded that the proposed MAGIA method has demonstrated a promising direction for future research of construction technology innovation.

7.2. Recommendations Although the preliminary case study shows promising results, challenges are also identified for future research including: (1) a more appropriate fitness function is needed to define the customer's needs or domain problem, e.g., the more sustainable or greener technology solutions; (2) a more powerful algorithm is desired to improve the computation efficiency; (3) a more systematic approach for model realization to replace the current manual approach, e.g., the proven BIM models that are associated with standard SAO structures; and (4) extension application to complex technology systems. The abovementioned topics are recommended to interested researchers.

[1] K. Akiyama, Function Analysis: Systematic Improvement of Quality Performance, Productivity Press, Cambridge, MA, 1991. [2] G. Altshuller, 40 Principles: TRIZ Keys to Technical Innovation, Technical Innovation Center, Boston, MA, USA, 2002. [3] Evaluation of Dynamic Technological Developments by Means of Patent Data, in: K. Brockhoff, A.K. Chakrabarti, J. Hauschildt (Eds.), The Dynamics of Innovation: Strategic and Managerial Implications, 1999, pp. S107–S132. [4] J.L. Chen, C.C. Liu, An eco-innovative design approach incorporating the TRIZ method without contradiction analysis, Journal of Sustainable Product Design 1 (4) (2001) 263–272. [5] Creative Industries Task Force, Creative industries mapping documents, Final Report to the Department for Culture, Media and Sports, UK, 1998. [6] Chunghwa Telecom, The Smooth Road Manhole Construction Method, 10 pp. Manhole Construction Specification, Chunghwa Telecom Inc, Taipei, Taiwan, 2008, (in Chinese). [7] R. Feldman, I. Dagan, Knowledge Discovery in Textual Databases (KDT), Proceedings of the First International Conference on Knowledge Discovery and Data Mining (KDD 1995), 1995, pp. 112–117. [8] R. Florida, The rise of the creative class, Washington Monthly, May 1st, 2002. [9] D.W. Halpin, CYCLONE: method for modelling of job site processes, Journal of Construction Division, ASCE 103 (3) (1977) 489–499. [10] D.W. Halpin, Process-based research to meet the international challenge, Journal of Construction Engineering and Management, ASCE 119 (3) (1993) 417–425. [11] R.B. Harris, A challenge for research, Journal of Construction Engineering and Management, ASCE 118 (3) (1992) 422–434. [12] J. Holland, Adaptation in Natural and Artificial Systems, University of Michigan Press, Ann Arbour, Michigan, USA, 1975. [13] P.G. Ioannou, L.Y. Liu, Advanced construction technology system—ACTS, Journal of Construction Engineering and Management, ASCE 119 (2) (1992) 288–306. [14] F. Johansson, The Medici Effect: Breakthrough Insights at the Intersection of Ideas, Concepts, and Cultures, Harvard Business School Press, Cambridge, MA, USA, 2002. [15] M.A. Kurfman, M.E. Stock, R.B. Stone, J. Rajan, K.L. Wood, Experimental studies assessing the repeatability of a functional modeling derivation method, Journal of Mechanical Design, ASME 125 (4) (2003) 682–693. [16] C. Kirschman, G. Fadel, Classifying functions for mechanical design, Journal of Mechanical Design, ASME 120 (3) (1998) 475–482. [17] Liu, C.C."A Study of TRIZ Method Improvements and Eco-Innovative Design Methods," Ph.D. Thesis, Department of Mechanical Engineering, National Cheng Kung University, Tainan, Taiwan, 2003. (in Chinese) [18] Liu, H. C., "Design and Study of Automated Text Summarization for Extracting SAO Structures from Chinese Patent Documents," Master Thesis, Department of Information and Computer Science, National Chiao Tung University, Hsinchu, Taiwan, 2009. (in Chinese) [19] L.Y. Liu, Simulating construction operations of precast-concrete parking structures, Proceedings of the 1995 Winter Simulation Conference, 1995, pp. 1004–1008. [20] L. Miles, Techniques of Value Analysis Engineering, McGraw-Hill, New York, NY, 1972. [21] Y. Mohamed, S. AbouRizk, Technical knowledge consolidation using theory of inventive problem solving, Journal of Construction Engineering and Management, ASCE Vol. 131 (No. 9) (2005) 993–1001. [22] Y. Mohamed, S. AbouRizk, Application of the theory of inventive problem solving in tunnel construction, Journal of Construction Engineering and Management, ASCE Vol. 131 (No. 10) (2005) 1099–1108. [23] J.C. Martinez, P.G. Ioannou, General purpose simulation with stroboscope, Proceedings of the 1995 Winter Simulation Conference, 1994, pp. 1307–1313. [24] NIST, FIPS Publication 183, Released of IDEFØ by the Computer Systems Laboratory of the National Institute of Standards and Technology, USA, 1993. [25] J.T. O'Connor, S.J. Miller, Constructability programs: method for assessment and benchmarking, Journal of Performance of Constructed Facilities, ASCE 8 (1) (1994) 46–64. [26] G. Pahl, W. Beitz, Engineering Design: A Systematic Approach, 2nd ed. SpringerVerlag, London, UK, 1996. [27] C.A. Rothe, Using patents to advance the civil engineering profession, Civil Engineering, ASCE 76 (6) (2006) 66–73. [28] A. Sawhney, S.M. AbouRizk, Application of hierarchical and modular simulation to a bridge planning project, Journal of Construction Engineering and Management, ASCE Vol. 121 (No. 3) (2005) 297–303. [29] L. Shen, Key research agenda on sustainable construction, Proceedings of the 14th International Symposium on Advancement of Construction Management and Real Estate (CRIOCM 2009), Oct. 29~31, 2009, Keynote Speech, PRC, Nanjing, 2009. [30] E.S. Slaughter, Implementation of construction innovations, Building Research and Information 28 (1) (2000) 2–17. [31] E.S. Slaughter, Characteristics of existing construction automation and robotics technologies, Automation in Construction 6 (2) (1997) 109–120. [32] R.B. Stone, K.L. Wood, Development of a functional basis for design, Journal of Mechanical Design, ASME 122 (4) (2000) 359–370.

88

W-d. Yu et al. / Automation in Construction 27 (2012) 78–88

[33] D. Sullivan, Document Warehousing and Text Mining, Wiley Computer Publishing, New York, USA, 2001. [34] C.B. Tatum, Process of innovation in construction firm, Journal of Construction Engineering and Management, ASCE 113 (4) (1987) 648–663. [35] C.B. Tatum, J.A. Vanegas, J.M. Williams, Constructability improvement using prefabrication, preassembly, and modularization, Technical Report No. 297, Department of Civil Engineering, Stanford University, 1986. [36] A. Teplitskiy, Application of 40 inventive principles in construction, TRIZ Journal (2005), http://www.triz-journal.com,Mar 2005, visited 2011/08. [37] The World Bank, Performance-based contracting resource guide, Oct. 2011, http:// www.worldbank.org/transport/roads/resource-guide/index.html, visited. [38] Y. Umeda, T. Tomiyama, Functional Reasoning in Design, IEEE Expert Intelligent Systems and Their Applications 12 (2) (1997) 42–48. [39] L.R. Yang, J.T. O'Connor, J.H. Chen, Assessment of automation and integration technology's impacts on project stakeholder success, Automation in Construction 16 (6) (2007) 725–733. [40] I.C. Yeh, L.C. Lien, Knowledge discovery of concrete material using genetic operation trees, Expert Systems with Applications 36 (2009) 5807–5812.

[41] W.D. Yu, S.T. Cheng, C.M. Wu, T.F. Chiang, An Automated Construction Innovation Model to Meet Future Construction Challenges, PRC, Chongqing, 2011, 6 pp. [42] W.D. Yu, M.J. Skibniewski, Quantitative constructability analysis with a neurofuzzy knowledge-based multi-criterion decision support system, Automation in Construction 8 (5) (1999) 553–565. [43] W.D. Yu, C.M. Wu, Integration of research & development project management with computer aided tools for fast innovation of construction technologies, Journal of Project Management, TPMA, Vol. 2, No. 4, 2009, pp. 1–25. [44] K. Lai, W.R.D. Wilson, FDL - a language for function description and rationalization in mechanical design, Journal of Mechanics, Transmissions, and Automation in Design 111 (1) (1989) 117–123. [45] M.S. Hundal, A systematic method for developing function structures, solutions and concept variants, Mechanism and Machine Theory 25 (3) (1990) 243–256. [46] Y. Iwasaki, M. Vescovi, R. Fikes, B. Chandrasekaran, Causal functional representation language with behavior-based semantics, Applied Artificial Intelligence 9 (1) (1995) 5–31.