Unsupervised Learning for Product Use Activity Recognition: An Exploratory Study of a "Chatty Device"

Area of Science:

• Engineering
• Computer Science
• Human-Computer Interaction

Background:

• Manufacturers need methods to understand product experiences in real-world use at scale.
• Industry 4.0 emphasizes data-driven design and manufacturing, with "Chatty Factories" as a key concept.
• Existing methods lack real-time, large-scale insights into how consumers interact with products.

Purpose of the Study:

• To propose a model for agile engineering product development using "chatty" products.
• To enable products to relay their usage experiences back to designers and engineers.
• To identify product use activities through sensor data and machine learning.

Main Methods:

• Collected sensor data at 100 Hz from a "Chatty device" during six everyday activities.
• Pre-processed and manually labeled the collected sensor data for product use activities.
• Compared four Unsupervised Machine Learning models (including Fuzzy C-means) for activity recognition.

Main Results:

• Demonstrated the feasibility of using unsupervised machine learning for product use activity clustering.
• The Fuzzy C-means algorithm achieved the highest performance with an F-measure of 0.87 and MCC of 0.84.
• Unsupervised learning effectively clustered distinct product use activities based on sensor data.

Conclusions:

• The proposed model enables "chatty" products to provide valuable usage insights for product development.
• Unsupervised machine learning, particularly Fuzzy C-means, is effective for recognizing product use activities.
• This approach supports agile engineering by providing real-time, data-driven feedback on product experiences.