AI Applications at Discom

Discom is developing AI features that complement existing, domain-specific algorithms for error detection. The AI applications are based on large volumes of historical measurement data stored centrally in the Discom Data Hub (DDH). The goal is to provide valuable AI capabilities to a broad customer base and, following successful evaluation, integrate them into real-time analysis.

Key AI Applications

Anomaly Detection
Unsupervised models learn the normal behavior of assemblies (e.g., transmissions) based on feature vectors derived from domain-specific NVH parameters. Deviations from normal behavior are flagged as anomalies—even if no explicit NOK examples are available. This helps identify rare or new types of faults, detect configuration errors, and reduce the manual testing effort.
Setup Optimizer
AI helps optimize inspection parameters and threshold values by analyzing which features best distinguish between OK and NOK. This allows for targeted improvements to inspection processes.
Feature Extraction & Labeling
Raw data is converted into insightful feature vectors. In addition, production data can be tagged with custom labels (e.g., type of damage) to enable supervised learning and targeted analyses.
Monitoring Services & Notifications
Intelligent background services monitor production data, identify trends, patterns, or outliers, and proactively notify users via push or email notifications.

Technical Features

Use of domain-specific knowledge for feature engineering.
Use of methods such as Isolation Forest, LOF, GMM, or autoencoders for anomaly detection.
Integration of AI models into the DDH infrastructure and WebPal for end-to-end data analysis and visualization.

Your Benefits

Early Error Detection: AI-powered anomaly detection identifies deviations from normal behavior.
Optimized Inspection Processes: The Setup Optimizer helps specifically improve thresholds and inspection parameters and reduce scrap.
Proactive monitoring: Intelligent monitoring services detect trends and outliers and automatically notify you.
Seamless integration & scalability: AI models and the DDH Python SDK enable end-to-end data analysis and easy integration into common ML stacks.
Simulations before implementation: “What-if” analyses show the effects of threshold changes before you deploy them in production.

Have we caught your interest?

Do you have any questions?
Would you like a consultation?