Enabling Industrial AI: How Siemens and AIT Leverage TDengine and Ontop to Help TCG UNITECH Boost Productivity and Efficiency

In modern manufacturing, producing high-quality components at scale isn’t just a matter of precision — it’s a matter of resilience and foresight. A single faulty engine part can trigger massive recalls, customer dissatisfaction, and regulatory penalties. For TCG UNITECH, a leading Tier 1 automotive supplier in Austria, the challenge was clear: how do you ensure consistent, defect-free production across 33 die casting machines, while reducing downtime and empowering a new generation of operators?

To address this, Siemens AG Österreich and the Austrian Institute of Technology (AIT) collaborated on a research initiative — DG Assist — to build an integrated Industrial AI platform. By combining their domain expertise with TDengine and Ontop, the team delivered a unified solution for data- and knowledge-driven process monitoring and decision-making.

Motivation: Manual Inspections, Knowledge Silos, and Traceability Gaps

TCG UNITECH’s operations rely on aluminum high-pressure die-casting — a process where molten metal is injected into molds at high speed and high pressure, completing in less than five seconds. Each part may go on to become a housing for an electric gearbox in a BMW or other vehicle.

Before this project:

• Workers inspected parts visually and logged defects on paper.

• Expert knowledge was concentrated in a few individuals, unavailable during off-hours.

• If a defect reached the customer, engineers had to search across a variety of different systems to reconstruct what happened.

• The potential cost of failure was significant: undetected issues could lead to returned production batches, warranty claims, or even recall actions — posing risks to quality reputation and customer trust.

The Solution: A Semantically Integrated Industrial AI Platform

To overcome these challenges, Siemens and AIT designed a semantically integrated platform — a system that connects machine data, structured records, and domain knowledge into a unified model that enables advanced querying, monitoring, and AI-driven recommendations.

Introducing the Technology Stack

At the heart of the DG Assist solution is a layered architecture that brings together data collection, semantic modeling, and expert-guided analytics:

• TDengine serves as the time-series backbone, capturing over 1,100 signals per machine at speeds up to 2.5 kHz. It stores both high-frequency batch data and continuous streams with minimal storage overhead and high compression efficiency.

• PostgreSQL holds structured production data, including part metadata, inspection results, material IDs, shift assignments, and tool configurations—critical for contextualizing sensor readings.

• Ontop enables virtualization of TDengine and PostgreSQL databases as part of a dynamic knowledge graph. This allows users to interact with distributed data via SPARQL queries as if it were a single source—without duplicating data.

• In parallel, a static knowledge graph captures expert knowledge using AIT- and Siemens-developed ontologies, stored in an RDF triple store and accessed through standard semantic web technologies.

• Custom middleware, developed by AIT, standardizes timestamps, transforms source data into a unified format, and ensures consistency between the databases. It acts as the interface layer for OPC UA, REST-APIs, and binary data logs, and handles all the ETL logic needed for seamless ingestion.

This integrated stack enables use cases like:

• Real-time anomaly detection based on data directly from the machine, e.g., pressure and piston velocity profiles

• Automated traceability from part ID to full process history

• Operator assistance systems powered by hybrid AI and semantic rules

• Long-term knowledge retention via expert-encoded graph structures

What Is Semantic Integration — and Why It Matters

Semantic integration goes beyond connecting databases — it connects meanings. It models the relationships between machines, sensors, parts, processes, and defects in a way that both humans and AI systems can understand.

This enables:

• Linking sensor data (e.g. pressure curve) to process steps (e.g. injection phase) and then to product outcomes (e.g. short shot defect)

• Embedding expert knowledge as queryable rules (e.g. “if velocity is low + pressure is high → check for blockage”)

• Providing cross-system traceability — without copying data or building complex ETL pipelines

• Ensuring interoperability across vendor systems by abstracting data semantics from proprietary formats

Knowledge graphs also lay the foundation for explainable AI in manufacturing. By explicitly modeling causality and expert logic, these systems help engineers understand how conclusions are reached — which is crucial for adoption and trust.

Real-World Impact: Concrete Examples from the Factory

1. Pressure Curve Anomaly and Leak Detection

   Operators used a time-series dashboard powered by TDengine to visualize cavity pressure during the injection process. An unexpected increase in pressure revealed a leak in the system, later confirmed to be caused by a worn valve.

   Stephan Strommer

   Project Manager &
   Senior Research Engineer, AIT

   

   We saw the difference in pressure between cycles and the operator said, “Our system is defective, we have a leak.”

2. Knowledge-Based Root Cause Tracing and Explainable QA

   Siemens and AIT’s knowledge graph mapped over 20 known product defects to root causes. For example, a cold shot could be linked to low piston velocity. The system uses SPARQL queries to trace such issues to their potential machine parameters and recommend countermeasures — especially when senior engineers aren’t available.

   Herbert Kerbl

   Quality Expert in Product &
   Process Management, TCG UNITECH

   

   Previously, connecting inspection results with real-time machine data required manual effort, and oftentimes gut feeling. Now, we can correlate sensor trends, production events, and expert-defined failure modes in one integrated view. It’s a major step forward in structured, explainable quality control.

3. Microsecond-Level Traceability

   TDengine records process data at sub-second resolution. By segmenting each part’s lifecycle (start/end timestamps per process step), the system can retrieve the exact time window when a part was cast. Combined with PostgreSQL’s part records, this allows engineers to demonstrate to customers that the part was produced under acceptable conditions — or precisely why it wasn’t.

4. Operator Guidance for Less Experienced Workers

    Expert knowledge is encoded into a static knowledge base. When anomalies are detected via simulation or statistical outliers, the system delivers targeted recommendations such as “inspect cooling system” or “increase mold temperature by 3%.” These are shown via on-site dashboards accessible in real time.

In the Grafana dashboard powered by TDengine, the operators at TCG UNITECH observed abnormal cavity pressure curves during the injection process. Specifically, they noticed a pressure difference between two vacuum systems in the mold, which was a deviation from the expected pattern. This visual cue led them to suspect a leak in the system. Their intuition was confirmed when further investigation revealed the root cause: a worn valve.

This real-time visibility into sensor behavior — enabled by high-frequency time-series data from TDengine — allowed them to detect and address the issue before it led to defective parts or production downtime.

Why TDengine was Chosen as the Time-Series Database

The AIT team initially experimented with a PostgreSQL-based time-series database but encountered performance limitations.

TDengine was selected as the time-series backbone after AIT benchmarked it against other systems. Its performance and efficiency were essential for TCG UNITECH’s environment, which combines high-volume, high-frequency data with long-term traceability requirements.

Key Benefits of TDengine:

• 3–5x Higher Compression

  Compared to PostgreSQL-based time-series databases, TDengine reduced data size from 50 GB/month per machine to under 10 GB/month. This figure is based on one machine capturing 1,100 signals with sampling rates up to 2.5 kHz. Across 40 machines, the total savings become even more impactful — reducing data from roughly 2 TB/month to 400 GB/month. These savings become even more critical when you factor in all redundant backups to keep the data available for up to 25 years. The result is extended on‑site retention of raw data — vital for traceability, root‑cause analysis, and regulatory compliance.

• >1 Million Rows/Sec/Thread Ingestion

  TDengine handled massive batch uploads from OPC UA and binary logs, ensuring no data was lost even during network lags or end-of-shift uploads.

• <100 ms Query Latency

  For dashboard queries and SPARQL federation, TDengine consistently responded in real time — even under concurrent queries across multiple sensors.

• Seamless SQL Integration

  TDengine’s SQL interface made it compatible with Ontopic’s SPARQL federation layer, enabling semantic queries across TDengine and PostgreSQL without ETL.

• Lightweight Edge Deployment

  A complete TDengine node — including database, ingestion logic, and dashboard support — ran on edge servers using only 400 MB RAM and minimal setup effort.

Value Delivered by Industrial AI

Looking Ahead

The Siemens and AIT team is continuing development:

• Deploying the recommendation system across all 33 machines

• Expanding the static knowledge graph through continuous learning

• Integrating new data types, including document metadata (PDFs)

• Productizing the solution for use in other factories and industries

• Designing a new AI-based quality control loop using smart sensor systems detecting defects on die-cast parts

By aligning with digital twins and standardized ontologies, the platform also opens the door for cross-site analytics and interoperable simulations — crucial elements for the future of smart manufacturing.

Conclusion: A Blueprint for Semantic-Driven Industrial AI

By combining TDengine’s time-series performance, Ontopic’s semantic federation, and the domain expertise of TCG UNITECH, Siemens, and AIT, the DG Assist project delivers a complete solution for traceability, defect prevention, and smart operator support. This architecture demonstrates that with the right tools, even legacy production lines can be transformed into intelligent, resilient, and AI-powered systems.