Machine Learning drives Automated Fault Detection and Diagnostics and predictive maintenance

Digital twins based on Machine Learning (ML) will make the air conditioning, refrigeration and Heat Pump systems more reliable and efficient.
This presentation shares experiences of using ML for Automated Fault Detection and Diagnosis (AFDD) that reduce total cost of ownership and down-time.

Description

south_east

Speaker

south_east

When & Where

calendar_today

Thu, 10.10.2024, 11:20 – 11:40

place

Hall 8 / 8-516

Details

Format: Lecture

Language: English

Description

Background

Digital twins based on Machine Learning (ML) will change maintenance practices in the air conditioning, refrigeration and Heat Pump industry. Our industry uses 20 % of the global electricity and pressure to reduce the carbon footprint and total cost of ownership is increasing. Experience shows that an average saving potential of 25 % is realistic without replacing equipment.

Digital Twins are powerful tools for AFDD

The presentation highlights the potential and experience of using ML to increase accuracy and reduce engineering time for Automated Fault Detection and Diagnosis (AFDD). ML will also be used in BMS systems to reduce loads and optimise controls, but focus is on AFDD.

ML will drive the paradigm shift to Predictive Maintenance (PdM) as it reduces time required and reduce false warnings. Competent staff to optimise hundreds of millions of systems, without advanced automation, will not be available. To collect and analyse performance data over varying operation conditions is cost-effective but rarely a part of contracts as savings cannot be guaranteed before baseline is established. The solution will not be to train hundreds of thousands of technicians to become analysts. Analyst competence will be focused to centers that generate workorders for large numbers of sites when AFDD raise “Early Warnings” for performance drift.

Performance monitoring with AFDD using ML models increase accuracy and decrease cost. Most sensors are standard in installations since many years. It is a question of specifying what data should be collected, at what interval, and ensure that data is made into actionable information. International Energy Agency (IEA), Annex 52, compiled a comprehensive guideline for data collection for Ground Source systems that with slight adaptations is applicable for all systems. Emerging standards such as e.g. “Real Estate Core” establish good practice which streamline and decrease cost for secure data management.

Digital twins presented are developed on component level – compressor, condenser and evaporator as well as for all KPIs important for maintenance and in-direct leak detection. A database containing detailed field performance data from thousands of systems in all sectors has been used to train and verify ML models. Experience show that ML is extremely powerful to detect any “performance-drift” with higher accuracy and less engineering time than with traditional rule-based limits. ... read more