Data-Driven Smart Manufacturing

• The factory of the future: Smart maintenance; Better product development; Quality improvement; Market adaptation
• Upskill the workforce: Data visualization; Model building, fitting, and sensitivity analysis; Advanced sensors; Manufacturing process control; Machine vision; Advanced data analysis

 

Smart enterprise at small- and mid-scale

• Smart technology strategies from modeling to manufacturing systems to advanced data analytics
• Highly-connected, knowledge-enabled industrial enterprise
• Enhanced productivity, sustainability, and economic performance

Other Applications

• Big data storage and processing infrastructure (manufacturing data format, protocol, networking & computing)
• Building APIs for data consumption ("data pipeline")
• Building predictive algorithms using machine learning
• Creating visualizations and dashboards to help interpret data

 

Purpose:

• Classify good/bad motors based on time-series sensor data

Data & Results

• Dataset: motor vibrational noise sensor (Ford)
• Data collection: sensor 1~500
• Data size: 4921 (30MB)
• Analysis algorithm: auto-sklearn and auto-keras (automated machine learning)
• Result: Characterize normal and abnormal data
  Accuracy - 75% (1-hour run); 88% (8-hour run)

Observation:

• Visual inspection of the time-series vibration data does not tell good/bad products (shown on the right; good at the top and bad at the bottom). But ML does.