AI for Predictive Maintenance in Commercial Refrigeration Unit Systems
About the Pilot
Klimamichaniki was founded in Thessaloniki, Greece in 1984. It is an engineering consultancy and construction company with expertise on all HVAC systems:
- A/C & cooling
- Solar thermal
- Commercial refrigeration
- PV systems
The company recently pivoted towards research and development via incorporating and applying AI methods within its modus operandi, to develop innovative solutions.
PreMaCOOL’s scope was the utilisation of the generated sensor data from a commercial refrigeration system installation, aiming towards developing AI-based services for this specific application. The development of a AI models for energy forecasting and predictive maintenance is innovative for the specific domain, since no such tools exist in the market. Monitoring tools exist for commercial coolers, however, there are no commercial solutions for forecasting or fault prognosis in industrial systems, not even in the academic literature. Advanced ML and DL algorithms such as Isolation Forest (IF) and Long Short-Term Memory (LSTM) networks, were utilized for training several AI models that could offer advanced insights to engineers and customers. Since there are no such systems at the moment, being able to implement and test such an idea, enabled Klimamichaniki to gain new knowledge during this pilot, which will eventually lead to the development of an innovative service, offered complementary to the installations.
Within PreMaCOOL, Klimamichaniki has accomplished a set of achievements:
- Developed a set of anomaly detection models based on point-wise (One-class SVM, Local Outlier Factor, Isolation Forest, Autoencoders) and patter-wise (Hidden Markov Models, LSTM autoencoders, CNN autoencoders) detection mechanism. Visual inspection on the data showed accurate detection of present anomalies.
- Developed a set of energy forecasting models with hourly granularity and forecasting depth of 24 hours, aiming to predictict next day’s consumption patterns. The selected algorithms (SARIMA, LSTM, Transformers, Neural Prophet) managed to achieve mean square errors below 0.01.
- Through the engagement with the pilot and EOSC DIH, two new ideas were conceptualized on innovative solutions based on data and AI methodologies. One of them has been submitted for funding via the funding calls proposed by EOSC DIH.
- The competence of HVAC engineers, towards identifying potential data-driven solutions to existing market problems, has been significantly increased. The company’s engineers are now is a continuous engagement and discussions on whether implementation of data-driven AI solutions could potentially address a number of specific problems in the domain.
HOW DID KLIMAMICHANIKI BECOME INVOLVED WITH EOSC DIH?
It was shared in a dissemination event organized by EUHubs4data, where Klimamichaniki was participating with ARCTUS, one of the funded experiments of the second open call.
WHAT SERVICES WERE USED?
- OpenAIRE EXPLORE: Academic research
- EGI Notebook: Develop code
- Deep training facility: Train models
- ARGOS: Create data management plan
- B2SHARE: Networking and dissemination of results
ABOUT THE EXPERIENCE OF WORKING WITH EOSC DIH AND THE VALUE OF EOSC FOR THE PILOT:
“Being part of the EOSC DIH was an honour and a privilege. All people were most helpful and supportive, something that allowed Klimamichaniki to focus on development. Additionally, EOSC DIH member provided informative sessions about new funding opportunities, something that the company has taken advantage of, and has submitted several proposals for funding.”
“EOSC DIH provided the services and infrastructure that was crucial for Klimamichaniki to proceed with the creating of a strategy and the training AI models at this stage of development. Since the company is not a software company and has only recently started being involved in innovation concepts, EOSC DIH provided with the tools and the human support to guide us through this endeavour.”
Klimamichaniki aims to continue developing innovative solutions based on data and AI methodologies. The company plans to create additional services that will bring her ahead of the curve and on top of the competition, potential building solutions that will have impact in a larger scale (i.e. European).