martes, diciembre 11, 2018

Big Data Congress 2018: The road to predictive Management of industrial value chains (by Gabriel Aranzadi - Eurecat)

La segunda de las conferencias en este bloque del Big Data Congress 2018 en Barcelona llamado Lessons Learnt from successful applications fue de Gabriel Aranzadi de Eurecat titulada “The road to predictive Management of  industrial value chains”


Algunos challenges que se plantean:

  •          Flexible production
  •           Linking Physical-Virtual
  •           Autonomous Production
  •           Predictive Management
  •           Knowledge Sharing
  •           Zero Waste
  •           Zero Defect


Digital Opportunity:



Inductry Transformation powered by:



Right Decission at the Right Level



Self-Diagnosis, Optimization, Organization


En el desarrollo de proyectos debe aplicarse un acercamiento que contemple tanto a los expertos de datos como también a los expertos del dominio, es decir del tema en el que se está trabajando.



Deploying industrial data driven projects


A.-Problem? Understand the concept and ptoblem objective from domain point of view
B.-Inputs - Outputs?
B.1.-Measurements (Really? How?)
B.2.-Relation with results
B.3.-Output parametres which match objectives
B.4.-Iterative and tireless process
C.-Data Exploration
C.1.-Graphing
C.2.-Strage things (always are there)
C.3.-Meetings to clarify doubts
C.4.-Propose changes / improvements in the structure of the data
C.5.-Homogenize the data
C.6.-Do we have enough data, samples?
D.-Data Processing - Data Cleaning, Feature engineering
D.1.-Time series? Batch?
D.2.-Noise
D.3.-Interest regions
D.4.-Data fusion?
D.5.-Metrics for new models
D.6.-PCA, PLS?
D.7.-Graphical representation
E.Data Modelling
E.1.-Classifier? Regressor? Optimizer? Anormally detector? ...
E.2.-Test framework
E.3.-Test data representations
E.4.-Initial algorithms selection
E.5.-Debugging
E.6.-Improvements, statistics, etc
E.7.-If they do not work
E.8.-Important: Logic of the results. Distrust from excellent outcomes (overfiting)
F.Results: Do not save efforts on tools that facilitate understanding and use of the prediction/prescription models


Un ejemplo práctico con: Predictive Quality: Plastic Injection



Success Stories: KPIs



Lessons Learnt:

  • -          Define the target, question to answer
  • -          Data availability – quality
  • -          Operation accuracy
  • -          Interoperability
  • -          ROI – OEE – TCO – TPM
  • -          Models usability
  • -          Cultural barriers

The Pathway:

  1. Map your Digital Strategy
  2. Create initial Pilot Projects
  3. Define the capabilities you need
  4. Become a virtuoso in data analytics
  5. Transform into a data driven company
  6. Actively plan an ecosystem approach





No hay comentarios: