Applying AI/ML to manufacturing process in order to find anomalies and predict failuresSITUATION
- A biomaterials company used a semi-automated production process for a complex proprietary polymer. Their quality was lower than expected and caused yield issues. It was hard to determine what may be causing the quality to fluctuate since many variables could influence the outcome during their highly sensitive process. They were recording some of the critical variables, but there were many “unknown unknowns”.
- The goal was for us to “digitize” their entire process and find patterns in the data to find root cause(s), predict failures, and optimize their process.
- Did the timing between process steps contribute to final quality?
- Was there consistency in how staff were performing various steps in different shifts?
- Which environmental variable were affecting quality?
- Etc.
- We conducted a deep review of their manufacturing process along with a top vendor that we had selected that had deep domain expertise in digitizing manufacturing operations
- Along with the vendor, we developed user requirements and created a software representation of their process (“digital twin”), imported their data sources, added sensors to their equipment to capture real-time data
- Setup AI/ML system to train the model in order to detect relationships between data-sets
- Client ran mfg process for 12 weeks and our system for trained using QC results for each production batch.
- In the end, we were able to find hidden relationships, identify risk condition in real time to generate notification of process failures and identify in-process variation that were linked to final quality and were not known before
