Quality assurance · Production

MakinaRocks Runway™ MLOps platform enables anomaly detection for semiconductor laser drills with one month advance warning

The problem

Manufacturing AI projects have a success rate of only nine percent due to the wide diversity of manufacturing environments and data, forcing each manufacturer to have a customized model that then fails when deployed into unpredictably different production conditions.

First attempt

Previous AI models for the laser drill client could not produce inference results when some sensors failed, causing frequent monitoring interruptions and client dissatisfaction.

Workflow diagram · grounded in source

Laser drill sensor data collected

trigger

“The client company was operating dozens of laser drills with identical specifications. Nonetheless, different AI models needed to be developed and deployed separately as each drill was operated with different core parts and parameters an…”

Autoencoder trains on normal data

ai_action

“MakinaRocks' solution thus featured an autoencoder model capable of compressing, restoring, and training with normal data”

Semi-supervised anomaly detection

ai_action

“designed to detect and identify signs of forthcoming interruptions with a high level of precision based on semi-supervised and continual learning”

Model switching on sensor availability

routing

“The solution alternated between these two models depending on the availability of data. The streaming-serving and model-shadowing features of Runway™, in other words, enabled the solution to continue to monitor the laser drills and detec…”

Maintenance alert one month ahead

output

“informed facility maintenance staff of these signs one month in advance so that they could maintain timely upkeep”

User-driven model retraining

feedback_loop

“general users can train and operate the AI model themselves, even during major events, without the help of data scientists or ML engineers”

Reported outcome

For a tech-sector client, Runway™ cut deployment time from six months to four weeks (reduction of about 80%) and reduced MLOps development manpower by about 50%.
For the semiconductor laser drill client, the solution detected anomaly signs one month in advance and maintained monitoring through sensor failures. General users can now retrain models without data scientists or ML engineers.

Reported metrics

Deployment period reduction (tech sector client)reduction of about 80%

MLOps development manpower reduction (tech sector client)about 50%

Advance warning time before laser drill interruptionone month

AI project success rate in manufacturing (Deloitte survey)nine percent

Show all 6 reported metrics

deployment period reduction (tech sector client)reduction of about 80%

MLOps development manpower reduction (tech sector client)about 50%

advance warning time before laser drill interruptionone month

AI project success rate in manufacturing (Deloitte survey)nine percent

ML models deployed (tech sector client)50+

solar power plants scaled (energy sector client)600+

Reported stack

Runway™Link™Jupyter

Source

https://mlops.community/blog/the-success-of-ai-depends-on-the-speed-of-iteration-an-mlops-strategy-for-ai-models-in-manufacturing

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

For a tech-sector client, Runway™ cut deployment time from six months to four weeks (reduction of about 80%) and reduced MLOps development manpower by about 50%.

What tools did this team use?

Runway™, Link™, Jupyter.

What results were reported?

Deployment period reduction (tech sector client): reduction of about 80%; MLOps development manpower reduction (tech sector client): about 50%; Advance warning time before laser drill interruption: one month; AI project success rate in manufacturing (Deloitte survey): nine percent (source-reported, not independently verified).

What failed first in this deployment?