More information
Description
Modern energy systems, such as wind turbines, motor drives in industry, and electric vehicles are prone to failures, resulting in loss of production, unavailability of services, or environmental disasters in a worst case. Electrical, mechanical, and thermal stresses are directly or indirectly responsible for these failures. To prevent these issues, energy systems must be regularly checked through routines and schedule specified by the manufactures. This schedule-based condition monitoring approach provides very little information on the remaining lifetime of the devices and does not allow for their prognostic and full exploitation. Furthermore, it is costly and presents problems related to the fact that devices might fail in between the routine check, which causes environmental risks and unsustainable use of resources.
In this proposal, we present solutions for these drawbacks by combining Virtual Sensors (VS) with powerful Artificial Intelligence (AI) tools. We will develop models of the underlying devices that can run in real time and thus serve as Virtual Sensor fed by real operation data from the actual devices. The VS will monitor thermal, mechanical, and electrical stresses. The data from the VS will be used in failure models to predict the remaining lifetime of the devices allowing for fault-tolerant and overload usage of the said devices, as well as condition-based maintenance. This is possible if the models are used in combination with AI or machine learning engines running in the clouds. The data for training the AI-engines will be generated from physical models of the devices, such as the finite element models of electrical machines, or in some cases from reduced models of these devices, to speed up the learning process. We expect the methodology to detect localized failure potentials in critical components, such as bearings, gearboxes, motors and generators. The possibility to apply the methodology to power electronic devices will be investigated.
Summary of project results
The project addressed the challenge of the call related to Technologies and innovation development under the sub-topics: 1. Renewable energy for local energy systems and 2. Sustainable management of water and aquatic resources.
The Baltic region, like the rest of the world, is rapidly witnessing the implementation of modern electric power conversion systems, such as electric power converters for wind and photovoltaic power plants, inverters for improving the electric power quality, and frequency converters for electric motor drives in industry and electric vehicles. These devices use power electronics equipment whose semiconductor elements operate at high voltages and conduct large currents. Additionally, the failure rate is increasing due to thermal and mechanical stresses that often affect power conversion systems. Therefore, their failures are more common compared to low-power electronic devices. Faults in power conversion systems cause economic losses or even can lead to disasters.
To prevent these issues, electric power conversion systems must undergo regular checks according to routines and schedules specified by manufacturers. However, this schedule-based condition monitoring approach provides minimal information on the remaining lifetime of the devices and does not allow for prognostics and full exploitation. Furthermore, it is costly and presents problems related to the fact that devices might fail between routine checks, causing environmental risks and unsustainable use of resources.
In the project, solutions to these problems were presented by a consortium of research institutions through the combination of Virtual Sensors (VS) with powerful Artificial Intelligence (AI) tools.
In the project, solutions to the issues identified above were presented by a consortium of research institutions through the combination of Virtual Sensors with powerful Artificial Intelligence (AI) tools. Models of the underlying devices were developed that could run in real time and serve as Virtual Sensors fed by real operational data from the actual devices. The Virtual Sensors are monitoring thermal, mechanical, and electrical stresses. The data from the Virtual Sensors is used in failure models to predict the remaining lifetime of the devices, allowing for fault-tolerant and overload usage of the devices, as well as condition-based maintenance. This is possible if the models are used in combination with AI or machine learning engines running in the cloud.
A prototype of the cloud based monitoring system with integrated developed models was implemented. The data for training the AI engines is generated from physical models of the devices, such as finite element models of electrical machines, or in some cases, from reduced models of these devices to speed up the learning process. The methodology detects localised failure potentials in critical components, such as bearings, gearboxes, motors, and generators. The possibility of applying the methodology to power electronic devices was also investigated.
During the project it was investigated different approaches to compare, which approach is suitable to be adjusted for analysis of different sensors (to sensors available on a particular system), that could be trained on data without desired output (e.g. failure). Finally, the project have developed an approach detecting anomaly in sensor signal (e.g., temperature sensor signal) by modeling the dynamic behavior of sensor value changes at different loads and environment conditions. The anomaly is detected, when sensor readings start to differ from predicted ones and this indicates the unusual situation, related to malfunction of upcoming failure. Investigations on a public wind farm dataset showed the possibility to predict gearbox failure 32-37 days in advance by applying this approach. Such approach is capable to work in real-time and needs only proper one-time training-based initialization.
13 PhDs (12M/1F) were involved in the project activities including 2 postdocs. 6 PhD candidates were also supported and had an excellent opportunity to collaborate within the international team. PhD student Karolina Kudelina (TalTech, Estonia) notably winning the 2022 L’Oréal-UNESCO For Women in Science Young Talents Program – Baltics Award. This recognition acknowledges her substantial contributions to energy conversion system condition monitoring and diagnostics, closely tied to her work in the project.
Throughout the project, a focus was maintained on project objectives, with successful completion of all planned work packages. The investigations involved analysing predictive maintenance algorithms, exploration of power converters for reactive power compensation in grids with renewable energy sources, and studying predictive maintenance, anomaly detection and power generation forecasting.
The project gave researchers the opportunity to develop and supplement their competencies in machine learning (ML) based solution development for energy sector, establish practical collaboration with public and private companies in Lithuania, as well as research institutions in Norway, Denmark, Latvia and Estonia. They took part in developing new innovative solutions for this sector, focusing on more efficient energy production forecasting, balancing and consumption.
Over the reporting period from January 2021 to the submission of the final report, 21 joint papers were published in Elsevier, MDP and IEEE journals in collaboration with the donor state. Additional papers, acknowledged by the project, were published through local efforts. The total number of published papers in three years surpasses the project''s expected output.
Information about the project has been shared through various channels, including conferences like eStream 2021, ICEMS 2021 (South Korea), eStream 2022, RTUCON2023, and others. An online project website emondi.vilniustech.lt with an integrated data management platform was created.
Summary of bilateral results
In this project, the project partners anticipated to strengthen their collaboration by researchers and PhD student exchanges, shared data and software. They also look forward to collaborating further with partners, establishing a robust research network between partners, and further connecting with industry organisations in the respective countries. The project enhanced the mobility of researchers and PhD students between partner countries, facilitating the transfer of technology and knowledge.It should be noted that each partner organisation and country had its own specific field of research. Nevertheless, there was sufficient overlap to facilitate cooperation and add value to their respective research outcomes.Several industry collaborators from each participating country agreed to cooperate within the project by providing and obtaining information. These organisations confirmed their willingness through submitted letters. The active participation and collaboration of these organisations enhanced their relations and consolidated cooperation with the partner organisations. The project consortium gained valuable inputs from the industrial partners to focus on research aspects necessary for the industry.Despite the epidemiologic situation preventing planned visits and physical meetings, the synergies between the partners were not jeopardised. Each partner specialised in a specific research field necessary for achieving the project goals, creating synergy within the entire project team. The primary description of this synergy is evident through the number of joint publications.Project Committee meetings occurred every two months, discussing project management, scientific outputs, platform implementation challenges, and communication possibilities.