Aviation (quality control)
The production of aviation turbine blades requires data science models to be built in order to identify and optimize the decisions to scrap parts instead of shipping with some impacting defects. This requires applying advanced analytics techniques to help realize the ROI through this improved quality control. Typical blade operations consists of the following steps:
- Create router and drill EDM holes
- Grind dovetail and CMM inspect
- Heat treat (transfer data has run number in it FYI)
- FPI inspect
- Visual inspect
- Farm out transaction to OV
- Coat at OV
- Visual inspect
- Farm in transaction from OV
- Shotpeen
- Waterflow
- Airflow inspect
- X-ray inspect
- Final visual inspect and package for shipment
- Ship transaction
- Closes router
In a case study, the data engineering team created flattened data by serial number for these steps, which was the framework for all data analysis. This included data such as percent of pieces above and below control limits, by Coordinated Measuring Machine (CMM) dimension. It was identified that the top defects were due to EDM and grind. It was also identified how grind defects correlated to CMM dimensions. This was accomplished by grouping all serial numbers into two groups—those that had grind defects (red) and those that don't have any grind defects (green). The frequency distribution (d) of four key dimensions was plotted to see if there was a perceptible difference in the two groups. See the first figure of the chapter.
The data science effort involved predicting each KPI based on diverse data sources. Analytics models were developed for two KPIs—Part Defect Rate and Machine Uptime. It was identified that mostly EDM and casting issues lead to scrap and some grind defects can be recovered. Plots were generated for the count of serial # with at least one defect, the count of scrapped and shipped, as well as the defect count by type (EDM, cast, and grind). CMM data shows statistical differences in individual measurements for parts shipped healthy with no defects, shipped with grind defects and scrapped with grind defects. See the Common cloud protocols table.