Data-driven design for industrial applications

Park, Seyoung

Permalink

https://hdl.handle.net/2142/120536 Copy

Description

Title

Data-driven design for industrial applications

Author(s)

Park, Seyoung

Issue Date

2023-04-23

Director of Research (if dissertation) or Advisor (if thesis)

Kim, Harrison M.

Doctoral Committee Chair(s)

Kim, Harrison M.

Committee Member(s)

• Sowers, Richard B.
• Wang, Pingfeng
• Kontou, Eleftheria

Department of Study

Industrial&Enterprise Sys Eng

Discipline

Industrial Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Data-driven design
• Data mining
• Machine learning
• User-generated data
• Online customers

Abstract

With the development of data analysis techniques, user-generated data has become a valuable resource for customer analysis. Compared to conventional methods such as surveys and interviews, user data analysis has advantages in that it can collect large datasets in a short period with little cost. In product design research, many studies have utilized user-generated data and presented various design implications. However, they have limitations when it comes to industrial applications. In other words, there exist gaps between research and industry. This dissertation aims to bridge these gaps based on state-of-the-art data mining techniques and the author's field experience in the smartphone industry. The dissertation consists of three topics in data-driven design: feature detection (Chapters 3 and 4), customer analysis (Chapters 5, 6, and 7), and design strategy (Chapters 8 and 9). The first topic is feature detection. Previous studies in this area analyzed user-generated data in unstructured text formats and extracted product features of customers' interests and their opinions about these features. For example, they extracted the keyword `memory' from smartphone reviews and suggested customers wanted larger memories for smartphones. However, it is not enough for industrial applications because a product is manufactured by combining multiple components. A company needs to make a decision on each of these components. Therefore, product designers and engineers request information about whether the customers mentioned RAM (Random Access Memory), ROM (Read Only Memory), or external memory supported by SD Card. To close this gap, the dissertation proposes a new methodology for sub-feature extraction based on phrase clustering in Chapter 3. Also, it suggests a new method for detecting customer opinions for target sub-features based on context-based review classification in Chapter 4. The second topic is customer analysis. Many studies have utilized user-generated data, such as online reviews and Twitter mentions. These studies regard the online customer base as one group with similar characteristics. However, people have divergent interests and characters. In terms of consumer products, customers have different preferences for product features. To reflect this reality, the dissertation suggested a methodology for online customer segmentation in Chapter 5. The method extracts customer attributes from online reviews and finds social networks among the reviewers. Then the customer network is partitioned into segments, which properly reflect the preferences of different customer groups. Another limitation of previous research is that they rarely considered the effects of social events. However, significant events can alter customers' attitudes toward products, so this factor should be reflected in the design of next-generation products. To capture changes in customer preference, this dissertation analyzes the impact of COVID-19 on customer attitudes, such as the sentiment for features and the importance of features. Short-term changes are explained in Chapter 6, and long-term changes are presented in Chapter 7. The research suggests proper design strategies for the changes in both terms. The final topic is design strategy. Design implications from the previous research focused on the customer's linear tendency for product features, e.g., preferences for a larger screen or lower price. However, companies require more information than that because they need to make decisions on product components, as mentioned earlier. Before mass production, companies must narrow down the spec ranges of each component and sign contracts for part sourcing. Therefore, product designers need answers to “What inch is large enough? What is the minimum/maximum size customers will accept?” To answer these questions, this dissertation proposes a new methodology for extracting spec guidance from online data in Chapter 8. The method constructs a neural network that predicts customer choices among given alternatives, then interprets the trained model using SHAP (SHapley Additive exPlanations). The analysis of the SHAP values provides spec guidance, i.e., spec ranges that positively affect customers’ purchase decisions. In Chapter 9, the proposed methodology is extended to new features. The newly suggested concept of NMF (newness merit function) measures the values of new features, and the method incorporates them into the choice prediction model. The SHAP analysis gives design guidance for new features and existing features. In summary, this dissertation proposes industry-applicable design methods and implications. These methods can help companies detect customers’ opinions on products at an engineering level, identify significant trends in the market and devise appropriate strategies, and evaluate new product designs from customers’ perspectives.

Graduation Semester

2023-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2023 Seyoung Park

Owning Collections