On-Premise Data Preparation for Industrial AI Applications
Ertas Data Suite gives manufacturing companies a secure, air-gapped pipeline to prepare production data, quality records, and maintenance logs for AI training — keeping proprietary process knowledge on your factory floor.
The Challenges You Face
Production Data Contains Trade Secrets
Manufacturing process parameters, quality specifications, yield data, and equipment configurations represent years of engineering investment. Uploading this data to cloud AI services risks exposing the proprietary knowledge that gives your products their competitive edge.
Industrial Data Is Messy and Multi-Modal
Sensor readings, SCADA logs, maintenance tickets, quality inspection reports, and operator notes arrive in different formats, time scales, and systems. Preparing this data for AI training requires domain-specific normalization that generic ETL tools cannot provide.
Factory Floor Expertise Is Hard to Scale
Veteran operators and quality engineers carry decades of pattern recognition experience — they know what a bearing failure sounds like, what a surface defect looks like, and what process drift precedes a quality excursion. This knowledge lives in their heads, not in your systems.
OT Networks Must Stay Isolated
Operational technology networks are air-gapped from IT networks for safety and security reasons. AI tools that require internet connectivity cannot operate in the environments where the most valuable manufacturing data resides.
How Ertas Solves This
Ertas Data Suite runs as a native desktop application with zero network requirements, making it the first data preparation tool many manufacturers can actually deploy in their OT-adjacent environments. Install it on a workstation with access to production data exports and prepare AI training datasets without any data leaving the facility.
The Ingest module handles the diverse formats common in manufacturing — CSV sensor exports, XML quality records, PDF inspection reports, and structured database extracts. The Clean module normalizes time series, handles missing sensor values, and standardizes unit conversions. The Label module lets quality engineers and operators annotate data with their domain expertise — tagging defect types, failure modes, and process anomalies.
The Augment module generates controlled variations to address the class imbalance problem inherent in manufacturing data (failures are rare but critical). The Export module produces versioned datasets with complete provenance, ready for model training on a separate system.
Key Features for Manufacturing & Industrial
OT-Safe Air-Gapped Operation
Data Suite runs without any network connectivity, making it safe to deploy on workstations connected to OT network data exports. No risk of inadvertent IT/OT bridging, no firewall rules to configure, no security exceptions to request.
Industrial Data Format Support
The Ingest module handles CSV time-series exports, XML quality records, PDF inspection sheets, structured database extracts, and delimited sensor logs — normalizing everything into a consistent format for downstream processing.
Operator-Driven Labeling
Quality engineers and experienced operators use the Label module to annotate production data — tagging defect types, failure precursors, process anomalies, and quality classifications using terminology and categories they already understand.
Rare-Event Augmentation
Manufacturing AI models need to detect failures that occur once in 10,000 cycles. The Augment module generates controlled variations of rare events to ensure your training data adequately represents the failure modes your model needs to recognize.
Why It Works
- Manufacturing companies have used Data Suite to prepare predictive maintenance training datasets from 5+ years of maintenance logs, sensor data, and failure records — all processed on a single workstation within the facility.
- The air-gapped architecture satisfies IEC 62443 requirements for OT data handling, enabling AI projects that were previously blocked by IT/OT segmentation policies.
- Operator-labeled quality data has produced defect classification models that match the accuracy of experienced inspectors while processing parts 100x faster.
- Rare-event augmentation has enabled failure prediction models to achieve meaningful recall rates despite the extreme class imbalance typical of manufacturing failure data.
- Complete provenance tracking from raw sensor data through to exported training dataset satisfies the documentation requirements of ISO 9001 and IATF 16949 for AI-assisted quality processes.
Example Workflow
An automotive parts manufacturer wants to build a visual defect detection model. A quality engineer opens Ertas Data Suite on a workstation in the quality lab, ingests 50,000 inspection images and their associated sensor readings at time of production through the Ingest module.
The Clean module normalizes image metadata, aligns sensor timestamps, and filters out duplicates. Experienced quality inspectors use the Label module to tag defect types — surface scratches, dimensional deviations, material inclusions — using the same classification system they use for manual inspection. The Augment module generates controlled variations of rare defect types to balance the training set.
The Export module produces a versioned dataset with complete traceability from each training example back to its source image, sensor context, and labeler identity. The dataset is transferred via approved media to the engineering team's training infrastructure, where it produces a defect classification model deployed on the inspection line.
Ship AI that runs on your users' devices.
Early bird pricing starts at $14.50/mo — locked in for life. Plans for builders and agencies.