Gill 03 10 2017 - Clemson University, South Carolinarate(combinations:(Fe10,Fe40,Fe80Be10,Be40,Be80Ne10,Ne40,N80! 29(complexitymetricswere(calculated(for(each(of(these(function - [PDF Document] (2024)

CED RClemson Engineering Design Applications and Research

Presented By:

Amaninder Singh Gill

Advisor: Dr. Joshua D. Summers

IMPACT OF CHAINING METHOD AND LEVEL OF

COMPLETION ON ACCURACY OF FUNCTION STRUCTURE-­

BASED MARKET PRICE PREDICTION MODELS

2/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Design Process and Design Representation

1.) Pahl, Gerhard, and Wolfgang Beitz. Engineering design: a systematic approach. Springer Science & Business Media, 2013.

3/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Types of Representation

• Early phase of design (first 5%) is of importance to researchers because of the cost implication on deciding the total cost of product development (70%)

• Different methods to model the functions in the conceptual design phase• Different models suited for different areas of application have been developed over the years

4/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Motivation -­ 1

l Function structure modelling is a designer/human centric knowledge representation

l Different designers end up with different function models even though they are describing the same product as is revealed by protocol studies (Sen 2014;; Thiagarajan 2017;; Nagel 2015;; Patel 2016)

l Hence, researchers tried to standardize the elements of the function models

l Knowledge Collection & Standardization Efforts: design repository, functional basis;; functional basis hierarchy;; composition rules;; quantification of information content;; impact of level of detail;; impact of level of detail on information content

l Calls for benchmarking

5/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Motivation -­ 2l Understand how designers create function structures – how are the function blocks and the function flows created

l Videotaping – Sequence, Patterns, Pauses – Code

l Three modelling strategies identified – Forward Chaining, Backward Chaining, and Nucleation

6/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Motivation -­ 3

• Coding results from Thiagarajan

• Mixture of the three schemas used by each individual

• Bias in favor of forward Chaining

• Protocol study results from Patel et al

• Number of functions added by students when seed function structure is created using one of the three methods

• Forward chained seed results in the least number of functions added

7/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Research Questionsl What is the amount of information content in each of the three chaining methods at various stages of chaining completion?

l At what level of completion do the chaining methods yield a significant difference in their ability to accurately and precisely answer specific query like Market Price?

l Can guidelines be developed to select the chaining method based on amount of completion and computational accuracy?

8/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Experimental Procedurel The three different levels of completion in this study were chosen for each chaining method: 10%, 40% and 80%

l Some of the smaller function models may not be able to add even a single a block if the completion percentages are placed too close to each other

l Twenty different electro mechanical consumer products were chosen from the Oregon State University Design Repository

l These function structure were recreated using the following chaining methods and completion rate combinations: F-­10, F-­40, F-­80;; B-­10, B-­40, B-­80;; N-­10, N-­40, N80

l 29 complexity metrics were calculated for each of these function structures and ANNs were trained using these inputs

9/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Forward Chaining

10/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Backward Chaining

11/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Nucleation

12/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Computational Procedure

Collect Training

Models (Function

Structures and

Market Price)

Convert assembly

models to

connectivity graphs

Convert connectivity

graphs to complexity

vector

Create predictive

models (18,00 ANN

models)

Averages and

variance returned to

user with respect to

info predicted

𝑅𝑒𝑠𝑖𝑑𝑢𝑎𝑙 𝐸𝑟𝑟𝑜𝑟 = 𝑇𝑎𝑟𝑔𝑒𝑡 𝑉𝑎𝑙𝑢𝑒 − 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑉𝑎𝑙𝑢𝑒

𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐸𝑟𝑟𝑜𝑟 = 𝑇𝑎𝑟𝑔𝑒𝑡 𝑉𝑎𝑙𝑢𝑒 − 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑉𝑎𝑙𝑢𝑒

𝑇𝑎𝑟𝑔𝑒𝑡 𝑉𝑎𝑙𝑢𝑒 𝑋 100

𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝐸𝑟𝑟𝑜𝑟 = |𝑇𝑎𝑟𝑔𝑒𝑡 𝑉𝑎𝑙𝑢𝑒 − 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑉𝑎𝑙𝑢𝑒|>

|𝑇𝑎𝑟𝑔𝑒𝑡 𝑉𝑎𝑙𝑢𝑒 𝑋 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑉𝑎𝑙𝑢𝑒| 𝑋 100

13/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Results & Interpretation

• Clockwise: (Bottom Left to Top Right): Residual Error Values for a) Nucleation b) Forward Chained c) Backward Chained

• Pattern Emerges

14/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Interpretation and Conclusionl The study also runs counter to the previous understanding that the more function blocks and flows are added, the more amount of information a function structure has and consequently should end up with better prediction capability as compared to a function structure that has a lesser amount of information

l Possible Explanation: for 10% completion, there is a lot more hom*ogeneity in the way that the chaining methods select the function blocks. There is a loss of this hom*ogeneity as the completion percentage reaches 40%. Some of this hom*ogeneity is again visible as the function structure is almost complete at 80%

l There is also a lot more room for how the user chooses to grow a function structure while using nucleation

l Inherent bias in the construction of the function models used

l Future work: Quantify and understand the effort involved in construction of the function structure;; develop tradeoffs

15/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

Referencesl Pahl G., Beitz W., Blessing L., Feldhusen J., Grote K.-­H. H., and Wallace K., 2013, Engineering Design: A

Systematic Approach, Springer-­Verlag London Limited, London.l Sen C., Summers J. D., and Mocko G. M., 2010, “Topological information content and expressiveness of

function models in mechanical design,” J. Comput. Inf. Sci. Eng., 10(3), p. 31003.l Study on How Designers Construct Function Structures in Novel Design,” Design Computing and

Cognition’12, J. Gero, ed., Springer-­Verlag, College Station, TX, pp. 247–264.l [21] Thiagarajan A., Patel A., O’Shields S., and Summers J. D., 2017, “Functional Thinking: A Protocol

Study to Map Modeling Behavior of Designers,” Design Computing and Cognition’16, J.S. Gero, ed., Springer, Evanston, IL, pp. 339–357.

l [22] Nagel R. L., Bohm M. R., Linsey J. S., and Riggs M. K., 2015, “Improving Students’ Functional Modeling Skills: A Modeling Approach and a Scoring Rubric,” J. Mech. Des., 137(5), p. 51102.

l [23] Patel A., Kramer W., Summers J., and Shuffler M., 2016, “Function Modeling: A Study of Model Sequential Completion Based on Count and Chaining of Functions,” International Design Engineering Conferences and Computers in Engineering Conference, ASME, Charlotte, NC, p. DETC2016-­59860.

l Mohinder C. V. S., Gill A., and Summers J. D., 2016, “Using Graph Complexity Connectivity Method to Predict Information from Design Representations,” Design Computing and Cognition’16, J.S. Gero, ed., Chicago, IL, p. no. 73.

l [33] Sridhar S., Fazelpour M., Gill A. S., and Summers J. D., 2016, “Accuracy and Precision Analysis of the Graph Complexity Connectivity Method,” Procedia CIRP, 44, pp. 163–168

l Caldwell B. W., Sen C., Mocko G. M., and Summers J. D., 2010, “An empirical study of the expressiveness of the functional basis,” Artif. Intell. Eng. Des. Anal. Manuf., 1(1), pp. 1–15.

l Mathieson J. L., and Summers J. D., 2010, “Complexity Metrics for Directional Node-­Link System Representations: Theory and Applications,” ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, ASME, Montreal, Canada, p. DETC2010-­28561.

16/162017.3.10

[emailprotected] RClemson Engineering Design Applications and Research

QUESTIONS … ?