ML on industrial and engineering data is used most often for visual quality control, geometry search, early anomaly detection, and decision support based on historical patterns.
How it works
A neural network is trained on labeled examples - good part versus defect - and can then classify new objects on the production line in fractions of a second.
Weaknesses
Training requires large volumes of labeled data. If a new defect type appears on the line that was not present in the training set, the model may miss it.
Sub-technologies (3)
▶
Classical ML
4 applications
Engineering-stage quality predictor
Quality & Testing · Engineering
Prototype screening and test data learning
Quality & Testing · Prototype
Production quality drift detection
Quality & Testing · Production
Design-stage failure mode prediction
Quality & Testing · Design
▶
Deep Learning on Geometry
4 applications
Reference-shape mining during concept phase
Design & R&D · Concept
Geometry similarity search during design reuse
Design & R&D · Design
3D defect detection on prototype scans
Quality & Testing · Prototype
Inline visual and geometric quality inspection
Quality & Testing · Production
▶
Physics-informed & Hybrid ML
3 applications
Physics-informed engineering models
Engineering & Simulation · Engineering
Hybrid service-condition models
Service & Maintenance · Service
Physics-informed process models for production engineering