We were contacted by a multi-technology tier 1 aerospace supplier who experienced recurring spindle failures in their milling machines, leading to unscheduled maintenance and lost revenue. Amygda's proprietary ML-stack works without any reliance on the OEM, and so we help the supplier without involving the OEM.

We set out to identify the exact dates for the spindle changes using a data-driven approach, as we were advised the datetimes could be weeks or months out. With more reliable dates in place, the next step was to develop a technique to detect and diagnose a developing spindle failure ahead of these change dates.

Our solution involved using our existing ML-stack, and building features without any reliance on the machine manufacturer (OEM). We built new health indicators that could detect failures due to high vibration levels, runout above tolerance, bearing failure, and more.

Insights were generated on production data on a subset of machines and events, which could form part of elaborate experiments across other machines and programs.

We detected 73% of the spindle failures, before they failed in production. This reduced the material scrap due to quality issues if the machine failed mid-production, and increased machine uptime by scheduling maintenance during operational intervals.