1. SUPREEMO: Smart Monitoring for Energy Efficiency and Predictive
Maintenance – Application to Electric Motors Retrofitting
Effie Marcoulaki & Sarantis Kotsilitis
2. Brief introduction of beneficiary and pilot
• Beneficiary: DEMOKRITOS is the largest multidisciplinary research center in Greece,
with critical mass in expertise and infrastructure on the fields of Nanotechnology,
Energy and Environment, Safety/Security, Biosciences, Particle and Nuclear Science,
Informatics and Telecommunications.
• The System Reliability and Industrial Safety Lab has long experience on equipment
reliability assessment, as well as on the development of decision support systems,
sensory devices for energy monitoring, and ML/DL techniques for the analysis of
electric load data.
• Pilot: ELSAP is a third-generation family SME established in 1983. The company is a
leader in pomace oil extraction and refinement in Greece, comprising three
extraction plants and the only edible oil refinery specializing in refining only olive
pomace oil and olive oil.
• The ELSAP processes involve a series of electric energy intensive steps for
centrifugation, drying and refinement.
4. AS-IS scenario, challenge, problem to be solved
• Monitoring of total electric power consumption of the plant, and specific zones.
Data collected once every 15 min, and transmitted to a cloud infrastructure
• Critical equipment monitored by SCADA systems [not in stage 3]
• No systematic tools to analyse the measurements of the energy monitoring
system and the deployed sensory equipment
• Equipment maintenance is planned periodically (preventive mode)
• Human inspection and control of the different units on 24/7 basis at each
process stage. Suspected devices are flagged to notify the next shifts
• Production losses due to unexpected failures (run-to failure mode)
• Lack of ICT tools to keep a digital maintenance log
5. TO-BE scenario and description of the
experiment workflow
Development, debugging, printing of custom PCBs for HF sampling
HF sensor deployment on energy-demanding equipment (stage 3 of ELSAP)
Development of custom ML/DL/AI cloud/edge algorithms for data
transmission & analysis
Model training using the HF sensor data collected, along with the equipment
maintenance / repair actions and the process operation logs during SUPREEMO
Development and user-evaluation of a DSS to improve energy efficiency and
support predictive maintenance
6. Architecture and technical details:
Data collection system
• Cloud:
• Data fusion, storage, and analysis using ML/DL
• Identification of electric signal anomalies
• Fog:
• Data compression and local storage
• Preprocessing and feature extraction
• Data transmission to cloud
• Edge:
• Design, printing, installation & evaluation of
PCB system to monitor the powerlines
• Sampling frequency = 64 kHz
9. Business objectives & impact
Reduction of
energy costs
Improvement of energy efficiency, environmental footprint, and process
sustainability, together with the reduction of operational costs *
Improvement of
process efficiency
Improvement of equipment availability and efficiency, better management of
spare parts, scheduling of small scale maintenance tasks when needed, lower
production losses due to unavailable equipment, shorter response time to
malfunctions, better scheduling of machine’s production and maintenance
activities *
Data-informed
decision making
Easy employee access to information regarding the health of the machinery, to
facilitate the identification and analysis of problems, improve their response to
malfunctions, and reduce impact of unexpected machine failures and unplanned
repairs (e.g. production losses).
*The developed tools have wider application to EMDSs inside and outside the pilot facility.
10. • KPI-1 Reduction of electricity consumption costs > 3-5%
• Set of energy efficiency actions that are feasible to be applied in the case of
the ELSAP refinery.
• Each one of the recommended actions meets the KPI-1 target
• KPI-2 Fault prediction accuracy > 90%
• Unsupervised autoencoder DNN, trained to understand a variety of normal
operating states for each device
• Identified over 100 deviations > only 5% of them were classified as warnings
• All warnings were later (1d-2w) associated with machine malfunctions and
vice versa
• KPI-3 Overall user acceptance > 80%
• User feedback regarding the DSS
• Overall user satisfaction indicator = 85.6%
Experiment KPIs