Periodic Reporting for period 2 - AI-TWILIGHT (AI powered Digital twin for lighting infrastructure in the context of front-end Industry 4.0)
Reporting period: 2022-05-01 to 2023-04-30
The main goal of AI-TWILIGHT is to generate self-learning/adaptive virtual twins of LED-based product for design and operation using AI and analytical techniques. Those digital twins will serve as input for predicting performance and lifetime of product and infrastructure design and management in an autonomous world. Targeted applications for AI-TWILIGHT are automotive, horticulture, general and street lighting.
The key technical and exploitation objectives of the AI-TWILIGHT consortium are:
• To create and digital twins of LED light-sources and electronics (driver)
• To create self-learning models using AI and analytics techniques
• To facilitate the implementation of the digital twins in digitalized design flow (for SSL product design) and facilitate their applications upstream, up to digital twins of lighting systems of large infrastructures (e.g. for building design).
• To implement the AI-TWILIGHT methods, models and tools within consortium partners to harvest its benefits
When translated to business goals, objectives will result in the introduction of more customized and connected products by 20% while reducing the time to market by 30%, and reducing by 25% the total cost of ownership of a “AI-TWILIGHT powered system.
The key technical and exploitation objectives of the AI-TWILIGHT consortium are:
• To create and digital twins of LED light-sources and electronics (driver)
• To create self-learning models using AI and analytics techniques
• To facilitate the implementation of the digital twins in digitalized design flow (for SSL product design) and facilitate their applications upstream, up to digital twins of lighting systems of large infrastructures (e.g. for building design).
• To implement the AI-TWILIGHT methods, models and tools within consortium partners to harvest its benefits
When translated to business goals, objectives will result in the introduction of more customized and connected products by 20% while reducing the time to market by 30%, and reducing by 25% the total cost of ownership of a “AI-TWILIGHT powered system.
Year 1 concentrated on laying a solid foundation for the rest of the project in particular by defining the requirements and specifications, setting the required testing and test methods and establishing the use cases.
Year 2 put the emphasis on establishing a common understanding and defining concrete actions of modeling options for SPD, driver, ageing using AI and analytical techniques.
The focus moved to developing test, measurement and characterization methods and models required to implement a digital twin for lifetime prediction. Modeling activities are targeting the four selected use cases.
AI-TWILIGHT.EU or AI-TWILIGHT.org public website is disseminating 51 different activities. A LinkedIn page was setup and provides regular updates to the general public.
Work achieved so far
•Requirements and specification
a. Selection of 4 main use cases across the application areas (Automotive, street lighting, general lighting, horticulture), targeting specific purposes (eg : CLO / multichannel/ design and operation). Backup UCs pursued for specific applications.
b. End user requirements for LED system lifetime, survival rate, flux degradation and colour shift. The results were published in a public report.
•Data collection
a. Development and release to the consortium of a database platform for collecting field, measurement, characteristion data sets.
b. Definition of the testing requirements in correlation with the use cases (elaboration of mission profiles)
c. Establishing a dataset of pre-existing LED stress experiments in the form of LM-80 and product qualification data, and performing an initial load of the LIMS (laboratory information management system) with >60,000 experiment records.
•Measurements and characterisation
a. Detailed description and measurements of each use case in close relation with modelling requirements : LED chip lab measurements, driver description, thermal characterization of the luminaires, full description of the materials and optics, optical measurements.
b. Two different kind of test setups are built up for the measurement of fast optical (and electrical) transients of LED packages in order to support data corrections to be applied in the new high throughput test methods being developed through proper modelling.
c. Developing and setting up new accelerated aging tests for the selected LEDs types.
d. Launch aging tests of selected carrier devices in order to allow proper modelling of LED packages that are foreseen to be used in certain use-cases.
•Modeling and Digital twin generation
a. Identification and definition of the modeling extensions required to extend model-based design methodology to lifetime prediction.
b. Methodology approach for creating DIgital Twins in the context of SMEs
c. Review the steps towards “industrializing” the prior digitalized design and product development workflows of LED based luminaires as proposed by the prior Delphi4LED project through an improved, fully automated parameter extraction procedure that is capable of the identification of parameter sets of the Delphi4LED multi-domain LED models as well as parameters of SPD models and is open be part of an AI-based model identification approach aimed at the identification of the parameters of the dynamics of degradation of LEDs’ light-output properties.
d. A further achievement in this regard is completion of a web-based tool aimed at the fast and automated parameter extraction for the Delphi4LED Spice-like multi-domain LED model. Besides ease of use, robust and quick operation, the requirements against the test data used as input for the parameter identification got significantly released compared the past recommendations of Delphi4LED. Thus, it is sufficient to perform IVL characterization at about a dozen operating points (opposed to 30+ operating points recommended previously) and there is no need to set the junction temperature for the IVL characterization to a specific value; all together helping reduce testing efforts.
e. In terms of LED driver design an electro-thermal modelling and simulation workflow is proposed, assuring precise temperature values for components critical from a reliability perspective.
•Standardisation
a. Supporting the generation of improved AI-based aging models and test-related standardization activities with new aging measurements and LEDs degradation data.
Year 2 put the emphasis on establishing a common understanding and defining concrete actions of modeling options for SPD, driver, ageing using AI and analytical techniques.
The focus moved to developing test, measurement and characterization methods and models required to implement a digital twin for lifetime prediction. Modeling activities are targeting the four selected use cases.
AI-TWILIGHT.EU or AI-TWILIGHT.org public website is disseminating 51 different activities. A LinkedIn page was setup and provides regular updates to the general public.
Work achieved so far
•Requirements and specification
a. Selection of 4 main use cases across the application areas (Automotive, street lighting, general lighting, horticulture), targeting specific purposes (eg : CLO / multichannel/ design and operation). Backup UCs pursued for specific applications.
b. End user requirements for LED system lifetime, survival rate, flux degradation and colour shift. The results were published in a public report.
•Data collection
a. Development and release to the consortium of a database platform for collecting field, measurement, characteristion data sets.
b. Definition of the testing requirements in correlation with the use cases (elaboration of mission profiles)
c. Establishing a dataset of pre-existing LED stress experiments in the form of LM-80 and product qualification data, and performing an initial load of the LIMS (laboratory information management system) with >60,000 experiment records.
•Measurements and characterisation
a. Detailed description and measurements of each use case in close relation with modelling requirements : LED chip lab measurements, driver description, thermal characterization of the luminaires, full description of the materials and optics, optical measurements.
b. Two different kind of test setups are built up for the measurement of fast optical (and electrical) transients of LED packages in order to support data corrections to be applied in the new high throughput test methods being developed through proper modelling.
c. Developing and setting up new accelerated aging tests for the selected LEDs types.
d. Launch aging tests of selected carrier devices in order to allow proper modelling of LED packages that are foreseen to be used in certain use-cases.
•Modeling and Digital twin generation
a. Identification and definition of the modeling extensions required to extend model-based design methodology to lifetime prediction.
b. Methodology approach for creating DIgital Twins in the context of SMEs
c. Review the steps towards “industrializing” the prior digitalized design and product development workflows of LED based luminaires as proposed by the prior Delphi4LED project through an improved, fully automated parameter extraction procedure that is capable of the identification of parameter sets of the Delphi4LED multi-domain LED models as well as parameters of SPD models and is open be part of an AI-based model identification approach aimed at the identification of the parameters of the dynamics of degradation of LEDs’ light-output properties.
d. A further achievement in this regard is completion of a web-based tool aimed at the fast and automated parameter extraction for the Delphi4LED Spice-like multi-domain LED model. Besides ease of use, robust and quick operation, the requirements against the test data used as input for the parameter identification got significantly released compared the past recommendations of Delphi4LED. Thus, it is sufficient to perform IVL characterization at about a dozen operating points (opposed to 30+ operating points recommended previously) and there is no need to set the junction temperature for the IVL characterization to a specific value; all together helping reduce testing efforts.
e. In terms of LED driver design an electro-thermal modelling and simulation workflow is proposed, assuring precise temperature values for components critical from a reliability perspective.
•Standardisation
a. Supporting the generation of improved AI-based aging models and test-related standardization activities with new aging measurements and LEDs degradation data.
To date, self-learning predictive digital twin methodology for lighting applications is a first of a kind.
The envisioned self-learning lighting products use their own digital twins that are „aware” of the history of their own operation through information feedback from luminaire field data (acquired in the field via CPS and IoT approaches in the future). Thus, such a system is learning from luminaire field data.
The process of adaptation of the lighting product takes place autonomously by means of edge computing, using the LED/driver digital twin extended with the capability of predicting the remaining product lifetime.
Project highlights of year 2
-Database uploads increase to >60000 stress test records
-Development of two improved High throughput characterisation test setup
-Fast optical testing technique for isothermal IVL measurements
-Development of online tool for LED Parameter extraction
-Development of an histogram based approach of past mission profile for ageing prediction
-Design of power cycling testing of LED systems with validated numerical and analytical models
-Physical realisation of Street lighting, Horti, indoor, automotive use cases
-Innovative LED Driver Chip Development: tape out of first silicon / first prototypes; preliminary engine design kit
-Delivery of the measuring box for outdoor application allowing measurement of environment conditions and characterization of the LEDs with transient tests
-Total of ~70+ technical publications cumulating year 1 and 2 (IEEE, Elsevier, MDPI, CIE, Lighting year books), Project presented in 21 activities of national societies
-Key roles in TC6- of CIE international standardization body, Several founding members and active members of TC2-91
-Updated versions of JEDEC’s LED thermal testing standards documents (JESD51-50A, 51-51A, 51-52A and 51-53A). (WP5)
The envisioned self-learning lighting products use their own digital twins that are „aware” of the history of their own operation through information feedback from luminaire field data (acquired in the field via CPS and IoT approaches in the future). Thus, such a system is learning from luminaire field data.
The process of adaptation of the lighting product takes place autonomously by means of edge computing, using the LED/driver digital twin extended with the capability of predicting the remaining product lifetime.
Project highlights of year 2
-Database uploads increase to >60000 stress test records
-Development of two improved High throughput characterisation test setup
-Fast optical testing technique for isothermal IVL measurements
-Development of online tool for LED Parameter extraction
-Development of an histogram based approach of past mission profile for ageing prediction
-Design of power cycling testing of LED systems with validated numerical and analytical models
-Physical realisation of Street lighting, Horti, indoor, automotive use cases
-Innovative LED Driver Chip Development: tape out of first silicon / first prototypes; preliminary engine design kit
-Delivery of the measuring box for outdoor application allowing measurement of environment conditions and characterization of the LEDs with transient tests
-Total of ~70+ technical publications cumulating year 1 and 2 (IEEE, Elsevier, MDPI, CIE, Lighting year books), Project presented in 21 activities of national societies
-Key roles in TC6- of CIE international standardization body, Several founding members and active members of TC2-91
-Updated versions of JEDEC’s LED thermal testing standards documents (JESD51-50A, 51-51A, 51-52A and 51-53A). (WP5)