Quality Control

AI screens physical and digital outputs for defects, then routes exceptions to human review.

Business impact

• Quality control throughput — Netflix says the model reduces QC effort on assets that are not defective, and Renault reports nearly 1,000 controls per vehicle at Double AI.
• Inspection volume — Renault reports the Customer Satisfaction Tunnel captures up to 130 images per vehicle.
• Defect detection / failure prediction accuracy — Netflix built a model to predict whether an asset will fail QC, and the goal is to identify likely defects before manual inspection.

Data requirements

• Production-line images (image) — BMW and Renault inspect vehicles using camera-captured images.
• Video asset metadata (tabular) — Netflix used metadata related to content and fulfillment partners as model inputs.
• Past manual QC results (tabular) — Netflix trained the model on results from prior manual QC checks.

AI methods and techniques

• Computer vision — BMW and Renault use visual inspection to find assembly issues and defects from production-line images.
• Deep learning — Axis describes BMW's inspection as computer vision technology based on deep learning.
• Supervised machine learning — Netflix uses past manual QC outcomes to train a model that predicts pass or fail.
• Cost-sensitive training — Netflix uses this to address class imbalance because defective assets are a minority of the data.

AI models and model families

neural network, predictive model

Industries

Real-world evidence

5 documented case studies on record.

Companies using this: BMW, Renault, Tesla, Netflix, Assessor-County Clerk-Recorder (ACR) Office at a large Southern California County.