PCS #2 : Process Qualification

Objective

To develop a user interface that displays information about frames.
To predict for each frame if the quality is good enough to use a shortcut during the process.

Being able to connect to existing tools, machines and databases.
Being able to scale up (around 40 Mo of new data incoming everyday)
Being able to build a prediction model that could have good accuracy on 25 000 ++ references with direct impact on production performance.
Organize smart design of experiments to minimize the cost (to build our training data set).
Have to integrate the tool in the middle of information and production data chain.

Concept creation and uphill validation.
Automatic retrieving of data from tracing machines (live from production site). More than 4000 files per day.
Extraction of meaningful data from measurement files (oma files).
Extraction of data from production system.
Merging data from different sources to create a brand new database around frames and their features (measurement, characteristics, suppliers, volume, etc …).
Definition and calculation of new features around frames and their quality.
Data analysis + Data visualization (first conclusion around frames).
Design of experiment : launch of experiments to evaluate the responses of different frames regarding a special mounting process.
Machine learning : from the experiments, prediction model building on how the other references will respond to our new mounting process (deep neural network implementation).
User interface definition and creation for final users : production, management, quality, and business with live upgrade of data and prediction results.

Here is one screenshot of what it looks like at the end :

Deature extractor from oma file.

SQL Database management with R and SQL

User interface using shiny.

and more…