Case studies

End-to-end worked analyses on real datasets. Each case study is a single script or notebook that runs against live data hosted at openmv.net, so you can re-execute and modify it locally with no extra setup.

The chapters mirror the structure of the companion textbook Process Improvement using Data, so a topic in the book lines up with a case study here.

• Latent variable modelling
  • PCA on NIR tablet spectra
  • PCA on food texture
  • PCA with missing data: Kamyr digester
  • MBPLS with missing data: LDPE tubular reactor
  • MBPCA with missing data: LDPE tubular reactor
• Least squares modelling
  • Regressing cheddar taste on chemical composition
• Design and analysis of experiments
  • Factorial DOE: oil-company experiment
  • Screen, then optimise: a reactor with a heat-release limit
• Process monitoring
  • Control charts for rubber colour
• Batch data analysis
  • Batch PCA outlier diagnosis: the DuPont polymerization reactor
  • Batch PLS fault diagnosis: the simulated SBR reactor
  • Multiblock batch PLS: the FMC batch dryer

Sources and attribution

The narrative for each case study is adapted from the PID book (Kevin Dunn, CC BY-SA 4.0). Datasets are downloaded from openmv.net at notebook execution time.