TTMD Uluslararası Sempozyumu – Yapay Zeka ve Veri Merkezleri oturumları (TTMD International Symposium report: AI and data-centre sessions)

Overview

This document is the report of the international symposium of the Turkish Society of HVAC and Sanitary Engineers (TTMD) focused on AI and data centres. It was published in issue 150 of TTMD Dergisi (April-June 2026, 'Bizden Haberler' section); its authors are Omer Yigit Turkarslan, Ismail Can, Sinan Alkas and Kivanc Aslantas. The report reflects the current priorities of practitioner engineers in Turkey.

The technical session addressed AI in data centre thermal management. One talk presented a physics-constrained neural network (PCNN) that reconstructs the full temperature field of a data hall from sparse sensors: mean absolute error fell from 4.35 °C to 0.75 °C against a classic neural network (about 83% better), and prediction time fell from about 60 minutes for a typical CFD run to milliseconds, enabling real-time monitoring and control, with the model's deviations being conservative. AI methods were mapped to thermal problems as surrogate models, PINNs, reinforcement learning and evolutionary optimisation.

The 'Future of Data Centres' panel covered free cooling, liquid and hybrid cooling, cooling strategy from an operator's view, the role of commissioning and TAB, and natural/low-GWP refrigerants with total equivalent warming impact (TEWI). The operator panellist stressed commissioning and TAB as essential parts of data centre cooling strategy for investment, operation and continuity. The source is an event report, not a peer-reviewed paper; results are given as presented in the talks.

Key points

1. A physics-constrained deep learning model (PCNN) reconstructed DC temperature fields from sparse sensors, cutting mean absolute error from 4.35 °C to 0.75 °C (about 83% better) versus a classic neural network.
2. Prediction time fell from about 60 minutes for a typical CFD run to milliseconds, enabling real-time monitoring and control; the model's deviations are conservative.
3. AI methods mapped to DC thermal problems: surrogate models, PINNs, reinforcement learning and evolutionary optimisation.
4. The operator panellist stressed commissioning and TAB as essential parts of cooling strategy for investment, operation and continuity.
5. Refrigerant selection was discussed via natural refrigerants and total equivalent warming impact (TEWI).
6. The panel also covered free cooling and liquid and hybrid cooling.
7. The report appeared in TTMD Dergisi issue 150 (April-June 2026).

Using it in design

• Offer sparse-sensor temperature-field reconstruction in monitoring and digital-twin features as a fast alternative to CFD, flagged as model-based.
• Add GWP to refrigerant fields in the cooling catalogue and support TEWI calculation.
• Define commissioning and TAB (testing, adjusting, balancing) as an explicit step in the cooling design workflow.

Who should read it

Engineers designing and operating data centre cooling in Turkey, and specialists following the role of AI in thermal management.

Publication details

Authors

Ömer Yiğit Türkarslan; İsmail Can; Sinan Alkaş; Kıvanç Aslantaş

Publisher

Türk Tesisat Mühendisleri Derneği (TTMD) - TTMD Dergisi

Year

2026

Document number

TTMD Dergisi Sayı 150 (Nisan-Haziran 2026), Bizden Haberler

Event

TTMD Uluslararası Sempozyumu 2026 - Yapay Zeka ve Veri Merkezleri oturumları

Document language

English

Source type

Industry body

Access

Free to read

Licence

Copyright retained; citation only — the publisher’s terms apply to the original document

Citation