WO2023031122A1

WO2023031122A1 - Method for producing a set of light sources

Info

Publication number: WO2023031122A1
Application number: PCT/EP2022/073961
Authority: WO
Inventors: Chento Didden
Original assignee: Summa Ip B.V.
Priority date: 2021-08-30
Filing date: 2022-08-29
Publication date: 2023-03-09
Also published as: NL2029081B1

Abstract

The invention relates to a method for producing a set of light sources having a predefined resulting spectral emission, comprising: - providing a series of light sources; - providing a trained neural network, trained using a training dataset comprising at least one time series of reference operating parameters from a set of reference light sources which are statistically representative of the series of light sources, said reference operating parameters comprising for each reference light source a time series of a representative of a power drawn by the reference light source, and a time series of a representative of a spectral output; - measure operating parameters of each light source of said series of light sources, said operating parameters comprising a representative of said power drawn at at least one measurement time and an operating parameter representative of said spectral output at said measurement time; - categorize said series of light sources, using said trained neural network and based upon said measured operating parameters, into at least one set of at least two light sources that comply with said predefined resulting spectral emission.

Description

METHOD FOR PRODUCING A SET OF LIGHT SOURCES

FIELD OF THE INVENTION

The present invention relates to a method for producing a set of light sources, a production assembly for producing a set of light sources, and a computer program product for producing a set of light sources.

The invention further pertains to control unit for a lighting system, a computer program product for generating control settings for a lighting system, a lighting system, and a method for operating a lighting system.

BACKGROUND OF THE INVENTION

Already in 2000, US6441558 in its abstract states “An LED luminary system for providing power to LED light sources to generate a desired light color comprises a power supply stage configured to provide a DC current signal. A light mixing circuit is coupled to said power supply stage and includes a plurality of LED light sources with red, green and blue colors to produce various desired lights with desired color temperatures. A controller system is coupled to the power supply stage and is configured to provide control signals to the power supply stage so as to maintain the DC current signal at a desired level for maintaining the desired light output. The controller system is further configured to estimate lumen output fractions associated with the LED light sources based on junction temperature of the LED light sources and chromaticity coordinates of the desired light to be generated at the light mixing circuit. The light mixing circuit further comprises a temperature sensor for measuring the temperature associated with the LED light sources and a light detector for measuring lumen output level of light generated by the LED light sources. Based on the temperatures measured, the controller system determines the amount of output lumen that each of the LED light sources need to generate in order to achieve the desired mixed light output, and the light detector in conjunction with a feedback loop maintains the required lumen output for each of the LED light sources."

And 15 years later, WO2015/200615 in its abstract states “Illumination devices and methods are provided for calibrating and controlling individual LEDs in the illumination device, for obtaining a desired luminous flux and chromaticity of the device over changes in drive current, temperature, and over time as the LEDs age. In some embodiments, the illumination device may include a phosphor converted LED, configured for emitting illumination for the illumination device, wherein a spectrum of the illumination emitted from the phosphor converted LED comprises a first portion having a first peak emission wavelength and a second portion having a second peak emission wavelength, which differs from the first peak emission wavelength. In such embodiments, methods are provided for calibrating and controlling each portion of the phosphor converted LED spectrum, as if the phosphor converted LED were two separate LEDs. An illumination device is also provided herein comprising one or more emitter modules having improved thermal and electrical characteristics “

Thus in semiconductor-based lighting elements, such as LEDs, the color spectrum and the brightness (intensity) change. In addition, LEDs are also affected by a dispersion of their technical properties with regard to brightness and color during manufacture. This can be compensated for by the manufacturer using so-called “binning,” in which semiconductor elements are sorted according to a predetermined dispersion.

W02008/001259 according to its abstract relates to “a method of controlling a lighting system with multiple controllable light sources 3a, 3b and a system therefor. According to a first aspect, influence data of the lighting system are obtained, which data represent the effect of one or more of the light sources 3a, 3b on the illumination of one or more sections of an illuminated environment. In an optimization method, sets of control commands are continuously determined, a predicted light distribution for these control commands is determined from the influence data, and a colorimetric difference between the predicted light distribution and a target light distribution is determined. A plurality of adjustment steps are performed to minimize the colorimetric difference. According to a second aspect, a neural network is trained with the influence data and a set of control commands for controlling the lighting system is determined with the use of the neural network.”

There are thus several applications in which a desired light color is generated from three LED light sources with red, green and blue color spectra. The light emitted by the three LEDs can be detected by a three-section filter, and the measured RGB value can be converted to the so-called CIE XYZ color space (CIE=Commission Internationale de I'Eclairage [International Commission on Illumination]). The measured value vector is compared with an XYZ target value in a control unit which functions as a P controller and which, depending on the error, acts upon a drive unit such that the drive unit is made to supply electrical power to the light sources accordingly. By carrying out the actions as mentioned and using the control unit and other means, compensation for changes in the brightness and color of the light emitted by the LED light sources can be provided. However, a disadvantage in this respect is that the sensor has to be adjusted to the frequency spectra of the LEDs for the control unit to function sufficiently. Furthermore, with this system, a lighting device with more than three light sources having different color spectra can no longer be controlled, because the result of an algorithm of said control unit is no longer unequivocal in view of the fact that several lumen settings of at least four light sources can generate the same color impression in the XYZ color space.

There are also different applications that focus on processes for determining the light current components of individual LEDs via a v(lambda)-adapted sensor. The operationally conditioned color and brightness changes of the individual LEDs are determined by measuring the spectral component with the aid of measuring the operating temperature of the LED (board and junction temperature). These measured values are determined individually for the particular controlled LED. This has the disadvantage that only one individual light source can always be observed by the measuring method used. Even a detection of the color shift of an individual light source can be determined only indirectly with the information of the temperature. Non-temperature-dependent color changes of the light source cannot be differentiated.

SUMMARY OF THE INVENTION

It is an objective of the present invention to provide a lighting system comprising a lighting unit, an optical part and a drive unit, and also a method for operating such a lighting system, the system and method being suitable to provide optimized control of light sources with different color spectra and to enable differentiation of temperature changes and non-temperature-dependent color changes in real time. An objective is further or alternatively to provide a lighting system that is robust under production fluctuations.

It is an alternative or further object to handle lumen and color maintenance of at least one light source without a feedback loop, while taking into account factors that have an impact on light source depreciation. This also includes the desired user input. A user will have full control over the lights with the advantage that the light and color will be accurate throughout the lifetime of the light source.

Yet another or alternative object is to create a platform that is capable to incorporate other domains to enhance the user experience of the computer program product.

Aspects of the present invention are set out in the accompanying independent and dependent claims. Features from the dependent claims may be combined with features from the independent claim as appropriate and not merely as explicitly set out in the claims.

At least one of the above-mentioned objectives is achieved by a method for producing a set of light sources having a predefined resulting spectral emission, comprising:

- providing a series of light sources;

- providing a trained neural network, trained using a training dataset comprising at least one time series of reference operating parameters from a set of reference light sources which are statistically representative of the series of light sources, said reference operating parameters comprising for each reference light source a time series of a representative of a power drawn by the reference light source, and a time series of a representative of a spectral output;

- measure operating parameters of each light source of said series of light sources, said operating parameters comprising a representative of said power drawn at at least one measurement time and an operating parameter representative of said spectral output at said measurement time;

- categorize said series of light sources, using said trained neural network and based upon said measured operating parameters, into at least one set of at least two light sources that comply with said predefined resulting spectral emission.

In this respect, the “resulting spectral emission” relates to the spectral emission of the set of light sources, i.e., of the combined light sources in the set.

At least one of the above-mentioned objectives is further achieved by a production system for lighting systems, comprising:

- an automated reference measurement system for measuring said reference parameters of said reference light sources;

- an automated measurement system for measuring said reference parameters of said light sources of said series of light sources;

- a data processor system running a computer program product which when running on said data processing system provides a neural network, retrieve said reference parameters from said automated reference measurement system and adds these to a training dataset, train said neural network using said training dataset comprising at least one time series of reference operating parameters from a set of reference light sources which are statistically representative of the series of light sources, said reference operating parameters comprising for each reference light source a time series of a representative of a power drawn by the reference light source, and a time series of a representative of a spectral output, retrieve measured operating parameters of each light source of said series of light sources from said automated measurement system, said operating parameters comprising a representative of said power drawn at at least one measurement time and an operating parameter representative of said spectral output at said measurement time;

- categorize said series of light sources, using said trained neural network and based upon said measured operating parameters of each light source of said series of light sources, into at least one set of at least two light sources that comply with said predefined resulting spectral emission;

- an assembling system for selecting from said series of light sources the light sources having the same category, placing these light sources together for assembling said set of light sources for producing a lighting system.

At least one of the above-mentioned objectives is further achieved by a computer program product for defining a set of light sources having a predefined resulting spectral emission, said computer program product when running on a data processing device:

- defines a trained neural network, trained using a training dataset comprising at least one time series of reference operating parameters from a set of reference light sources which are statistically representative of the series of light sources, said reference operating parameters comprising for each reference light source a time series of a representative of a power drawn by the reference light source, and a time series of a representative of a spectral output;

- retrieves operating parameters of each light source of said series of light sources, said operating parameters comprising a representative of said power drawn at at least one measurement time and an operating parameter representative of said spectral output at said measurement time;

- categorize said series of light sources, using said trained neural network and based upon said measured operating parameters of each light source of said series of light sources, into at least one set of at least two light sources that comply with said predefined resulting spectral emission.

At least the above-mentioned objective is further achieved by a control unit for a lighting system, which lighting system comprises a drive unit and at least one light sources with said drive unit functionally coupled to said at least one light sources, wherein the control unit is functionally coupled to said lighting system and comprises:

- a communication device for retrieving data remotely from the lighting system; - a data processor and a computer program product which, when running on said data processor:

* retrieve local operating parameters from said lighting system, said local operating parameters comprises a total local operations time of the at least one light source, a representative of a temperature of the at least one light source at said local operations time, and a representative of a power drawn by the at least one light source at said local operations time;

* generates a machine learning network;

* retrieve reference operating parameters from a set of reference light sources which are statistically representative of the at least one light source and which are based upon a reference operations time;

* retrieve measured reference light output parameters at said reference operations time from the set of reference light sources;

* update a machine learning network training dataset using said retrieved reference operating parameters and reference light output parameters and said reference operations time;

* update said machine learning network using said updated training dataset;

* provide said local operating parameters as input to said updated machine learning network;

* generate said local control settings using said updated machine learning network, and provide said local control settings to said lighting system.

There is further provided computer program product generating control settings for a lighting system that comprises a drive unit and at least one light source functionally coupled to said at least one light source, which computer program product, when running on a data processor:

* generates a machine learning network;

* retrieve reference operating parameters from a set of reference light sources which are statistically representative of the at least one light source and which are based upon a reference operations time; * retrieve measured reference light output parameters at said reference operations time from the set of reference light sources;

* update said machine learning network using said updated training dataset;

There is further provided a lighting system, comprising: a lighting unit which comprises at least two of said light sources having different color spectra; a drive unit which is functionally coupled to the lighting unit and which is configured to control power to at least one of the light sources of the lighting unit; at least one sensor which is configured to detect a representative of a temperature each of the at least two light sources, and the control unit described above, functionally coupled to the drive unit.

Reference is made to light sources in general. Currently, most used light sources are light sources based upon solid state components. Most know of these are light emitting diodes, LEDs. These are well known in the art and require drive units that regulate power. Current developments for light sources relates to quantum dots.

In general, lumen and color maintenance were found to be two complementary problems since LED light sources are based on mixing different channels (colors) by controlling their intensity. Depreciation can be forecasted and taken into account so that the light output is compensated by adding more contribution on the depreciated channels. Limited amount of data will be recorded and model predictions will have to apply generalizations for unseen data. Note that in the context of this software, SPD prediction is done together with Yxy prediction to get a better representation of light, however their functions may be overlapping (for example, light control).

Other definitions that may be used in the current application:

SPD - Spectral power distribution of a light source in visible range MacAdam steps - A measure that determines which chromaticity points are indistinguishable by a naked eye from from a given target chromaticity.

Chromaticity, (..xy), (..uv) - A 2-dimensional representation of color, Luminous flux, Lumen, PHIv - “intensity” of light.

COB - Chip On Board. Had multiple smd’s on it that emit red, green and blue light. LM80 - An LED depreciation report standard.

Contribution values - are values in range of 0 to 1 that indicate the intensity of each channel of a COB. In this respect, 1 means that the channel is emitting the maximum amount of light possible. This is done by adjusting the PWM/PDM duty cycles.

Total on-time of a channel: Total on-time is calculated by integrating the used contribution values over the lifetime of the COB. a channel on half contribution has half on- time compared to another channel on a full contribution, on a same COB.

Operating temperature - temperature under which a COB operates when each channel contributes at a particular intensity. The heat is caused because of emission of light.

In the current context, a reference light source which is statistically representative of a (local) light source in an embodiment is a light source of the same type as the (local) light source. Usually it has the same specified emission spectrum. Often, it is one selected from the same production batch. In case light sources are produced in a continuous production process, the statistically representative of the light source can be light sources that are produced at/in predetermined time ranges. These time ranges are selected such that the light sources that are selected are statistically representative. This requires insight in the production process and the variations over time. Usually, selection is such that a light source is selected as representative for at least 100 light sources. In particular, a light source is selected as representative for at least 1000 light sources. More in particular, a light source is selected that is representative for at least 10.000 light sources. To be able to select a light source that is representative, at least one light source is selected out of less than 100.000. For example, suppose that 1000 light sources are produced in an hour and the production process provides light sources that vary significantly over a time interval of 5 hours, then in an example three statistically representative light sources are selected in an interval of 5 hours. Thus, the three selected light sources statistically represent 5000 light sources and possibly even the variation in this time range of 5 hours. These three light sources are measures very accurately during operation time. The number of selected light sources to statistically represent the local light sources also depends on the fluctuation in production, the required accuracy and requirement on the local light sources, and the measurement accuracy of local operating parameters, of reference operating parameters and of reference light output parameters. These statistical relations determine the ratio and number of reference light sources to local light sources. Using the statistically representative light sources, a dataset with various parameters against operating time can be measured, for instance selected from temperature, power supplied, voltage, current, flux, emission spectrum, luminescence, spectral irradiance, and a combination thereof.

Machine learning techniques (mainly deep learning) are to design and train a model given an input of the same type (RGB image, infrared, etc.) as the system perceives. The model is trained on a large amount of annotated data, a training dataset. In the current case, measured data from reference light sources. In the case of deep learning, a detection framework such as Faster-RCNN, SSD, R-FCN, Mask-RCNN, or one of their derivatives can be used. A base model structure can be VGG, AlexNet, ResNet, GoogLeNet, adapted from the previous, or a new one. A model can be initialized with weights and trained similar tasks to improve and speedup the training. Optimizing the weights of a model, in case of deep learning, can be done with the help of deep learning frameworks such as Tensorflow, Caffe, or MXNET. To train a model, optimization methods such as Adam or RMSProb can be used. Classification loss functions such Hinge Loss or Softmax Loss can be used. Other approaches which utilize handcrafted features (such as LBP, SIFT, or HOG) and conventional classification methods (such as SVM or Random Forest) can be used.

On the basis of a set of training data with measurements one or more machine learning algorithms and statistical classification algorithms can be applied. Example algorithms may include linear classifiers (e.g. Fisher's linear discriminant, logistic regression, naive Bayes, and perceptron), support vector machines (e.g. least squares support vector machines), clustering algorithms (e.g. k-means clustering), quadratic classifiers, multi-class classifiers, kernel estimation (e.g. k-nearest neighbor), boosting, decision trees (e.g. random forests), neural networks, Gene Expression Programming, Bayesian networks, hidden Markov models, binary classifiers, and learning vector quantization. Other example classification algorithms are also possible.

The process of categorization may involve the computing device determining, based on the output of the comparison of the one or more subsets with the one or more predetermined sets of scene types, a probability distribution (e.g. a Gaussian distribution) associated with the one or more subsets. Those skilled in the art will be aware that such a probability distribution may take the form of a discrete probability distribution, continuous probability distribution, and/or mixed continuous-discrete distributions. Other types of probability distributions are possible as well.

A current computer program product and/or method will use machine learning techniques (mainly deep learning) to design and train a model using a training dataset. The model is trained on a large amount of measured data.

The lighting unit may include at least four light sources, and the control unit is arranged for using an optimization algorithm which, as a main condition, optimizes a value of color consistency of each of the light sources, such as the color rendering index (CRI), which can be calculated from the predetermined primary data and the instantaneous secondary data.

The reference operating parameters may include previously measured data for each of the reference light sources and/or a reference optical part and/or a reference drive unit. Said measured data may be provided as a specification from the manufacturers of at least one of the reference light sources, the reference optical part and the reference drive unit. The measured data for the reference light sources may include color spectrum, peak wavelength, dominant wavelength, and beam angle in full width and half maximum for each one of the reference light sources.

The control unit is configured to use the optimization algorithm which may result in controlling values of the individual light source. The optimization algorithm may compensate any influences on the color and brightness change, in particular since a redundancy in determination regarding the color impression is generated by using the at least two light sources as compensation source. Additionally, color adaptation can also take place under reduction of the total brightness of the light in that the optimization is carried out in an XYZ color space affected by brightness.

The control unit is configured to carry out the setting of the lighting unit, by means of the drive unit, and by using the optimization algorithm that includes two or more optimization criteria. The optimization goal is to optimize the color consistency of the light sources and/or to maximize the life time of each of the light sources, wherein the optimization settings are calculated from local operating parameters and reference operating parameters. This data that include predetermined measured values for the individual light sources, the optical part and the drive unit, and local operating parameters that include the junction temperature of the light sources and/or the temperature of the optical part measured by the sensor. The sensor may be configured to detect the junction temperature of the light sources in the connection area. The sensor may be located near to each of the light sources as close as possible to the drive unit. The number of sensors can be chosen according to the number of the light sources that are used in the system. The temperature difference depends for each connection area on the thermal power to be dissipated from the respective connection area. Since brightness of each of the light sources defined with different wavelength depends on the junction temperature, the measured characteristic lines of the brightness as function of the junction temperature may show a powerdependent curve shape.

The control unit may provide temperature-dependent color correction onto the drive unit.

Spectral emission has the following background.

Wikipedia describes it as follows: In radiometry, photometry, and color science, a spectral power distribution (SPD) measurement describes the power per unit area per unit wavelength of an illumination (radiant exitance). More generally, the term spectral power distribution can refer to the concentration, as a function of wavelength, of any radiometric or photometric quantity (e.g. radiant energy, radiant flux, radiant intensity, radiance, irradiance, radiant exitance, radiosity, luminance, luminous flux, luminous intensity, illuminance, luminous emittance)

Physics

Mathematically, for the spectral power distribution of a radiant exitance or irradiance one may write:

M(A) = 5²0 Z(5A 5A) « I (A AA) where M(A) is the spectral irradiance (or exitance) of the light (SI units: W/m³ = kg nr' s’³); is the radiant flux of the source (SI unit: watt, W); A is the area over which the radiant flux is integrated (SI unit: square meter, m2); and A is the wavelength (SI unit: meter, m). (Note that it is more convenient to express the wavelength of light in terms of nanometers; spectral exitance would then be expressed in units of W m"² nm"¹.) The approximation is valid when the area and wavelength interval are small.

Relative SPD

Characteristic spectral power distributions (SPDs) for an incandescent lamp (left) and a fluorescent lamp (right). The horizontal axes are in nanometers and the vertical axes show relative intensity in arbitrary units. The ratio of spectral concentration (irradiance or exitance) at a given wavelength to the concentration of a reference wavelength provides the relative SPD. This can be written as:

Mrel ( A ) = M(A) I M(A₀)

For instance, the luminance of lighting fixtures and other light sources are handled separately, a spectral power distribution may be normalized in some manner, often to unity at 555 or 560 nanometers, coinciding with the peak of the eye's luminosity function.

Responsivity

The SPD can be used to determine the response of a sensor at a specified wavelength. This compares the output power of the sensor to the input power as a function of wavelength. This can be generalized in the following formula:

R(A) = S(A) I M(A)

Knowing the responsivity is beneficial for determination of illumination, interactive material components, and optical components to optimize performance of a system's design.

Source SPD and matter

The spectral power distribution over the visible spectrum from a source can have varying concentrations of relative SPDs. The interactions between light and matter affect the absorption and reflectance properties of materials and subsequently produces a color that varies with source illumination.

For example, the relative spectral power distribution of the sun produces a white appearance if observed directly, but when the sunlight illuminates the Earth's atmosphere the sky appears blue under normal daylight conditions. This stems from the optical phenomenon called Rayleigh scattering which produces a concentration of shorter wavelengths and hence the blue color appearance.

Source SPD and color appearance

The human visual response relies on trichromacy to process color appearance. While the human visual response integrates over all wavelengths, the relative spectral power distribution will provide color appearance modeling information as the concentration of wavelength band(s) will become the primary contributors to the perceived color. When referring to visual light, we in general refer to a wavelength range between 400 nm and 700 nm.

This becomes useful in photometry and colorimetry as the perceived color changes with source illumination and spectral distribution and coincides with metamerisms where an object's color appearance changes. The spectral makeup of the source can also coincide with color temperature producing differences in color appearance due to the source's temperature.

In an embodiment, said at least one time series is relative to a time of first operation, in particular said time series includes said time of first operation, more in particular said time series time base is based on operation time. Thus, when using the power-on time of the light source as a basis for time, it was found that aging of other light sources than the reference light sources could also be described in a better way.

In an embodiment, said at least one time series includes said measurement time. In particular, said measurement time of said light sources of said series of light sources is fixed time after first powering the light source, and said time series includes that same moment for the reference light sources. For instance, each measurement and time series includes measuring at said first powering of a light source and reference light source. In addition or alternatively, a measurement is done after a predefined amount of time after said first powering.

In an embodiment, said spectral output includes spectral output in a visual wavelength range, in particular including a wavelength range of 400-700 nm.

In an embodiment, in said categorizing at least one color vision result of a human observer is calculated using said spectral output and said categorization is based upon said at least one color vision result, in particular said categorization take into account said color vision result at a predefined simulated moment in time. For instance, using the spectral emission and a standard set of color samples, like a CIE ceramic tile set, humanly perceived color of the color sampler under the spectral emission are calculated.

In an embodiment, the at least one set of at least two light sources comply with said predefined resulting spectral emission, in particular a color vision result of a human observer calculated using said spectral output complies with a predefined color difference criterium, when a resulting deviation is statistically within a predefined criterium.

In an embodiment, in the categorizing, a relative positioning of said light sources is taken into account. If for instance light sources are placed on a line, the relative arrangement of light sources in/on that line can influence the resulting spectral emission of that set.

In an embodiment, the neural network comprises a deep learning neural network. In an embodiment, the neural network can be part of a further neural network of machine learning network. In an embodiment, the categorizing comprises composing from said series of light sources a series of said sets of at least two light sources, in particular at least three light sources. Thus, in fact a series of light sources is binned or divided into a series of sets of light sources.

In an embodiment, the series of light sources are produced in at least one selected from batch production, continuous production, and a combination thereof, and during said production additional reference light sources are selected for expanding said training dataset. Thus, the training dataset will expand over time, making the predictive capabilities and accuracy better.

In an embodiment, the training dataset is further provided with at least one production parameter of each reference light source, in particular a production parameter representative of at least one characteristics of raw materials used in the production of said light source. In an embodiment, this may even allow the neural network to be incorporated into a further system or method for predicting the spectral emission of a light source based upon raw material of production settings. This may even allow for instance setting a raw material composition of for instance phosphors in such a way so as to result in desired spectral emission.

In an embodiment, said reference operating parameters comprising for each reference light source a time series of a representative of a temperature, and said training dataset further comprises time series of said representative of said temperature.

In an embodiment, the representative of said power comprise at least one selected from voltage, current, and a combination thereof.

In an embodiment, the training dataset further comprises production parameters including raw materials characteristics.

In an embodiment, the set of light sources comprises a substantially red light source, a substantially green light source and a substantially blue light source. Using these basic light sources, a relative white light emitting lighting system may be produced. The current method can add light sources together in an optimized way. In fact, it allows composing the set of light sources in such a way that for instance after a predefined operation time the spectral emission still complies with a set spectral emission. Suppose the set is composed to result in D65 daylight emission. The method allows producing a set that is withing a predefined tolerance from absolute D65 after a predefined operation time.

There is further provided a method for producing a lighting system comprising providing a set of light sources using the method described above. An (further or additional) optimization goal of the control unit and/or of the method may further be optimizing the color spectrum of the light sources. From the reference operating parameters in respect of the reference light sources and for instance the measured junction temperature and the temperature of the optical part, an associated spectrum may be calculated, which is added to the calculated spectra of the other light sources to form a jointly calculated “predetermined” total spectrum. From this calculated total spectrum, the CRI value Ra is calculated in the usual manner, as in the case of measured spectral values. It is preferred for this calculation to occur in the CIE system.

For at least two light sources, there are unlimited possibilities or possibilities only limited by the resolution of the control to adjust a desired chromaticity coordinate of color by mixing the used primary colors. Depending on the mixing ratio, it can be optimized towards different parameters like lumen efficiency or color consistency. The desired chromaticity coordinate color may also be optimized towards the color reproductions properties of the optical part. When the optimization is done, desired chromaticity coordinates x/y may be adjusted.

The control unit may define an ecosystem using an Artificial Intelligence method that gets feedback/input from the light sources, the optical part and the drive unit. Said ecosystem is configured to control the drive unit. Therefore, the system is not bound by a specific drive unit. The Artificial Intelligence method may be combined with and/or comprise machine learning and/or the use of artificial neural networks and training and/or trained artificial neural networks. In a specific embodiment, the control unit comprises the computer program product which in addition updates a trained artificial neural network using an update of reference operating parameters and corresponding updated reference light output parameters.

Said ecosystem is configured to use a communication protocol depending on the predetermined specifications of the drive unit. Said communication protocol may be the DMX protocol. The DMX protocol may allow a setting of the drive unit current for each light source with a precision of 8 bit (that is 256 different values). Instead of the DMX protocol, other protocols may also be used, for example, protocols with higher precision. It is preferable to provide a control reserve of, for example, one additional bit, in order to appropriately take into consideration the decrease in brightness occurring as a result of aging processes.

In an embodiment of the system according to the invention, the predetermined primary data include a preset target lumen value. In an embodiment of the system according to the invention, the predetermined primary data include a preset target correlated color temperature.

The system may be configured to be adjustable by the choice of the light sources and may be controlled with the optimization algorithm that includes, e.g., adjusting the junction temperature to a desired color temperature, brightness and the like. For achieving a solution in real time, when the said temperature is found to be above a limit value, the control unit compensates the temperature changes in that area.

The predetermined (target) correlated color temperature and the preset target lumen value of the lighting unit as well as the optical part may be compensated with the optimization algorithm of the control unit in dependency on the junction temperature of the light sources and/or the temperature of the optical part, depreciation in lumen of the light sources, color rendering index and mixed-light capability with the optical part. Accordingly, the control unit may be configured to set control values for the target parameters.

In an embodiment the control unit is configured to control a lumen value of each of the light sources during operation of the lighting system.

In an embodiment, at least one light source is a solid state light source, in particular a LED light source, and/or said at least one light source comprises at least two light sources each having a different emission spectrum, in particular at least three light sources with each light source having a different emission spectrum. In many practical implementations, at least three LED light sources are combined in creating an emission resembling for instance CIE standard daylight D65, or for instance llluminants A, and other spectral power distributions.

In an embodiment, the control unit and/or method are configured to optimize a value of color consistency of each of the at least two light sources, in particular a ratio of each of the at least two light sources in order to generate a set light emission. A set spectral power distribution, for instance D65, or any other, may be input into the control unit via the communication device. In an embodiment, a user input can be received. Using the updated machine learning network or updated trained artificial neural network, local control settings can be generated and provided to the drive unit. For instance, the updated trained neural network may be used to predict spectral emission and the computer program product uses the predicted spectral emission in calculating local control settings.

In an embodiment, the control unit and/or method are configured to maximize a life time of each of the at least two light sources. In fact, in an embodiment, the trained updated machine learning network or updated trained artificial neural network provides parameters for maximizing the life time.

In an embodiment, the control unit further comprises a second communication device for communicating data to the lighting system, in particular the communication device using a second communication protocol different from the communication protocol of the communication device, wherein the computer program product, when running on said data processor, controls the second communication device for transmitting the local control settings to said lighting system.

In an embodiment, the computer program product transmits data to at least one lighting system at least once a day, in particular functionally real time, in particular real time. It allows continuous adjustment of the drive unit for attaining a set emission spectrum.

In an embodiment, the computer program product is set to receiving reference operating parameters at least on a daily base, in particular at least once every hour. In this way, the training dataset can be updated, the artificial neural network can be re-trained. Usually, the reference light sources already have an operations time that is longer than the local light sources. Thus, in fact, the training dataset already is in time advance of ahead of the local light sources.

In an embodiment each of the light sources comprises a semiconductor-based light source.

The lighting system may comprise a calibration unit comprising reference light sources, one or more measurement devices for measuring reference operating parameters and reference light output parameters. The calibration unit may be coupled to a series of control units. Each control unit may be coupled to a series of drive units each coupled to one or more light sources. For instance, one central calibration unit may comprise hundreds of reference light sources that are continuously measured. For instance. Reference light sources are measured each minute, each hour, depending also on changes in the light emission measured. A building or a house may comprise one control unit. The control unit is coupled to a series of light elements each comprising several light sources coupled to one drive unit.

In an embodiment, at least a portion of the semiconductor-based light source includes a light emitting diode (LED).

The lighting system may be implemented to any desired light source, particularly any type of light emitting diode, including organic light emitting diodes (OLED). It is also possible to use light sources of different type together, in particular LEDs and incandescent light bulbs.

In an embodiment the optimization algorithm is implementable in a Cl E standardized X, Y, Z color space.

In an embodiment the optimization algorithm is configured to realize or result in a value of the color consistency lower than 10 Kelvin.

In an embodiment, the control unit is configured to dim each of the light sources during operation of the lighting system.

The predetermined (target) correlated color temperature and the preset target lumen value of the lighting unit as well as the optical part may be compensated with the optimization algorithm of the control unit in dependency on the junction temperature of the light sources, depreciation in lumen of the light sources, color rendering index and mixed- light capability with the optical part. Accordingly, the control unit may be configured to set lumen values for the target parameters.

It can be understood that the embodiments of the method according to the invention may include a lighting system having any of the features or combinations of features that are disclosed herein in connection with discussions of the lighting system according to the invention. Accordingly, the entireties of the earlier discussions of the lighting system are hereby incorporated into this discussion of the examples of the method.

The invention further applies to an apparatus or device comprising one or more of the characterizing features described in the description and/or shown in the attached drawings. The invention further pertains to a method or process comprising one or more of the characterizing features described in the description and/or shown in the attached drawings.

The various aspects discussed in this patent can be combined in order to provide additional advantages. Furthermore, some of the features can form the basis for one or more divisional applications.

BRIEF DESCRIPTION OF THE DRAWINGS

Further features and advantages of the invention will become apparent from the description of the invention by way of exemplary and non-limiting embodiments of a lighting system.

The person skilled in the art will appreciate that the described embodiments of the system according to the present invention are exemplary in nature only and not to be construed as limiting the scope of protection in any way. The person skilled in the art will realize that alternatives and equivalent embodiments of the object can be conceived and reduced to practice without departing from the scope of protection of the present invention.

Reference will be made to the figures on the accompanying drawing sheets. The figures are schematic in nature and therefore not necessarily drawn to scale. Further, equal reference numerals denote equal or similar parts. On the attached drawing sheets, figure 1 illustrates a schematic block diagram of a lighting system in accordance with an embodiment of the invention figure 2 schematically illustrates a measurement assembly and system for providing a trained neural network; figure 3 schematically illustrates a measurement assembly for determining light source properties and for categorizing light sources into a set of light sources.

DETAILED DESCRIPTION OF EMBODIMENTS

Figure 1 is a block diagram illustrating an exemplary lighting system 1 . The current lighting system 1 comprises a lighting unit 2, a calibration system 3, and a control unit 4. Each of these parts will be discussed in more detail below. Several of the lighting unit 2, calibration system 3 and control unit 4 communicate with one another and exchange data. In the depicted embodiment, the calibration system 3 and the control unit 4 communicate with one another. In fact, in the depicted embodiment the calibration system 3 sends data to the control unit 4. The control unit 4 processes the received data. The control unit 4 can also sent requests for data or even instructions to the calibration system 3.

In the depicted embodiment of figure 1 , the control unit 4 and the lighting unit 2 communicate with one another. In fact, the control unit 4 receives measurement data from the lighting unit 2. The control unit send local control settings 16 to the lighting unit 2.

In the way illustrated, these elements control unit 4, lighting unit 2 and calibration system 3 are functionally coupled. The coupling can be hard wired. In most embodiments, however, the elements communicate wireless. Thus, the functional coupling in such an embodiment is a wireless coupling. A skilled person will recognize that such a coupling can be via one or more means and protocols like WiFi, Bluetooth, Zigbee, via optical coupling, and the like. The elements may all be coupled to the internet and communicate via internet protocols.

In the embodiment illustrated, the lighting unit 2, calibration unit 3, and control unit 4 are depicted as separate elements. In many embodiments, these elements will be physically separated from one another. These elements in many embodiments are going to be remote from one another. For instance, the calibration unit 3 can be located at or near a production facility, the control unit may be located at a corporate headquarter, and (many) lighting units 2 may be located at stores, houses, manufacturing plants, and/or office buildings all over the world. Thus, a control unit 4 may be functionally coupled to a multitude of similar lighting units 2.

In the depicted embodiment, the control unit 4 receives measurement data from one calibration system 3 measuring many light sources 9. Alternatively, a series of control units 4 can receive data from one calibration system 3. Thus, a larger calibration system 3 can be defined that is functionally coupled with a series of control units 4. In this way, complex and accurate measuring devices 10 can be used which can automatically measure many light sources 9.

The elements discussed so far (lighting unit 2, calibration system 3, control unit 4) have been discussed on a communications level, and in an abstract manner. Below, these elements are going to be discussed on a more detailed level.

First, the lighting unit 2 will be discussed in more detail, then the calibration system, and finally the control unit 4.

The lighting unit 2 can be a lighting system based upon “light emitting diodes”, LEDs. These are well known in the art by now. Other solid state light sources may also be used, for instance based on quantum dots, or other light emitting devices. As is commonly known, most of these advanced light sources require driving electronics and can be combined to obtain a required emission spectrum I (A). Many different configurations of lighting units are possible. In the embodiment depicted in figure 1 , three basic light sources are coupled to one drive unit 6. Two of these drive units 6 are combined into one lighting unit 2. The lighting unit 2 depicted here comprises an optical part 7 that can combine the output (light emission) h(A) of all the light sources 5 into one emission spectrum l(A).

In some embodiments, the lighting unit 2 comprises a lighting unit control 8 that controls the drive units 6 and for instance allows data communication. A lighting unit 2 can be a physically separate entity. Like for instance for replacing the known light bulb.

Next, the calibration system 3 is going to be discussed. In an embodiment, as discussed, a statistically representative of the at least one light source is selected. In an embodiment, the statistically representative can be a light source that is from a same production batch, in case an element is produced in a batch process. For light sources that are produced in a continuous process, such a statistically representative can be elements that are produces in a certain time window around a light source. In the batch example, suppose that light sources are produces in batches of 5000 light sources. In such a situation, one can select for instance 5 light sources random from the 5000. The other light sources will get a batch indication. In the example of line production, suppose 5000 light sources are produces in an hour. Then at 5 random production times, light sources are selected. The other 4995 light sources produced in that hours will receive an identical identifier.

Next, the statistically representatives will be placed in a measurement setup. The measuring setup is provided with one or more measurement devices, like a luminescence photometer, a temperature sensor. At predefined time intervals, the measurement devices will measure characteristics of each light source. Thus, a data set is created of several characteristics of each light source against operation time. In an embodiment as depicted in figure 1 , the measurement device of devices and the objects to be measured will be circulated with respect to one another. For instance, the light sources can have fixed positions on a rail. A transport device will move the measurement devices past each of the light sources.

In this way, a data set is created of statistically representatives of the light sources.

As indicated in Figure 1 , the control unit 4 is configured to act on the drive unit 6 as a function of predetermined primary data 14 relating to the light sources 5, the optical part 7 and the drive unit 6 as well as instantaneous secondary data l( ), l,( ) obtained real-time from the lighting unit 2, the optical part 7 and the drive unit 6 during operation of the lighting system 1.

The system 1 may include any suitable light sources 5 having different color spectrums, particularly any type of light emitting diode, including organic light emitting diodes (OLED). It is also possible to use light sources 5 of different types together, in particular LEDs and incandescent light bulbs.

Optionally, the light unit 2 may include multiple light sources 5 that may be monochromatic or polychromatic. In some embodiments, each of light sources 5 may produce a monochromatic light having a single wavelength or a narrow SPD with a single peak. In other embodiments, each of light sources 5 may produce a polychromatic light having multiple different peaks in its SPD. Furthermore, in some embodiments, each of light sources 5 may be any type of light source capable of emitting single wavelength light or light with a narrow SPD with a single peak, such as an LED, high pressure sodium lamp (HPS), fluorescent lamp (FL), or the like, or any combination thereof. Considering different kinds of light sources, it is noted that multi-package LEDs are flexible in spectral composition, and spectrum proportions of each LED are easy to control. For example, in some embodiments, by choosing different drive units 6 like in figure 1 , a variety of LEDs with different spectra could be obtained.

In some embodiments, chromaticity of each light source 5 may correspond to a specific chromaticity coordinate on a chromaticity diagram, which in turn may correspond to a specific color presented on the chromaticity diagram.

For example, as shown in Figure 1 , the lighting unit 2 may comprise six component light sources 5. As described above, each component light source 5 may emit light having a specific color. For example, in some embodiments, the colors may be red, amber, green and blue. In various embodiments, any colors presented on the chromaticity diagram may be used. A polychromatic desired light having desired optical characteristics may be produced by mixing the component lights according to certain proportions. In some embodiments, proportions of the component lights may correlate with each other. Particularly, in some embodiments, proportion of one component light may assume a linear relationship with proportion of another component light. It shall be noted that the above description of the light emitting device is provided for illustration purposes, and is not intended to limit the scope of the present disclosure. For persons of ordinary skill in the art, various variations and modifications may be conducted under the teaching of the present disclosure. For example, the lighting unit 2 may have any number of component light sources 5, each light source 5 may produce a component light of any color, and a component light may be a monochromatic or polychromatic light.

The drive unit 6 may drive the light sources 5 by providing them with voltage or current at calculated levels. The drive unit 6 may receive a command from the control unit 4, and adjust driving voltage or current for individual light sources 5 accordingly. The control unit 4 may be configured to select and determine parameters for spectrum optimization based on the local operating parameters 14 and the reference operating parameters 15. For example, the control unit 4 may calculate respective proportions of multiple component lights to be combined to generate a desired light having a desirable synthesized chromaticity which is defined by a desired color consistency. In some embodiments, the local operating parameters 14 may provide the control unit 4 information regarding a working condition of the lighting system 1. As used herein, the term “working condition” broadly relates to any condition or circumstance under which a lighting solution operates, which includes but is not limited to the purpose or goal of the lighting, the target object or environment to be illuminated, the requirement or input by a system default or a user, etc. In some embodiments, information regarding the working condition relates to conditions of an ambient environment of a target object and may be acquired by a detector, transmitted from a local storage device or a remote server, or manually input by a user, or the like, or a combination thereof.

In some embodiments, the control unit 4 calculates respective proportions of component lights based on the component chromaticity and the desired chromaticity. As used herein, the term “component chromaticity” refers to the chromaticity of a component light, and the term “desired chromaticity” or “synthesized chromaticity” refers to the chromaticity of the desired light.

The control unit 4 uses an optimization algorithm which is designed to calculate control settings of the drive unit 6 on the basis of the local operating parameters 14 and reference operating parameters 15 for optimizing a value of color consistency of each of the light sources 5 and/or maximizing life time of each of the light sources 5. The local operating parameters 14 may include values of the junction temperature 131 of the light sources 5 detected by a sensor, the temperature of the optical part detected by a sensor, power supplied to the light sources in at least one of the lighting unit 2 and/or the drive unit 6, during operation of the lighting system 1.

For example, with four component light sources 5, there might be unlimited possibilities or possibilities only limited by the resolution of the control to adjust a desired chromaticity coordinate color by mixing the used primary colors. Depending on the mixing ratio, it can be optimized towards different parameters like lumen efficiency or color consistency. The color consistency may be optimized towards the color reproductions properties of the optical part 7. When the optimization is done, desired chromaticity coordinates x/y may be adjusted.

The local operating parameters 14 may include previously measured data for each one the light sources 5, the drive unit 6 and the optical part 7. Said measured data may be provided as a specification from the manufacturers of at least one of the drive unit 6, the light sources 5 and the optical part 7. The reference operating parameters for the statistically representative light sources 9 may include a predetermined color spectrum, peak wavelength, dominant wavelength, and beam angle in full width and half maximum for each one of the light sources.

The present invention can be summarized as relating to a lighting system 1 with a lighting unit 2 which comprises at least two light sources 5 having different color spectra, with an optical part 7 which is configured to mix the color spectrums of the light sources 5, with a drive unit 6 which is connected to the lighting unit 2, with a sensor which is configured to detect at least one of the junction temperature T(t) of the light sources 5 at a position of a connection area between the drive unit 6 and the lighting unit 2 and the temperature T(t) of the optical part, and with a control unit 4 which is configured to optimize a value of color consistency of each of the light sources 5 and to maximize life time of each of the light sources 5, and configured to act on the drive unit 6.

In figure 2 in combination with figure 3, an embodiment of the method for producing a set of light sources having a predefined light emission and assembly is illustrated.

In figure 2, a series 9 of statistically representative light sources 20 is depicted. In this embodiment, only a limited amount of light sources is shown. In fact, over time thousands and even millions of light sources 20 can be included. These light sources can be individual LED’s, or each light source may be composed of several LED’s, for instance.

These light sources 20 are all powered and transmit light. The dotted triangles represent transmitted light. Using a measuring system 10, a representative of the light emission is measured. This can be the emitted light as a function of wavelength. It can also be derived parameters representative of emitted light. In the illustrated example, the measuring system comprises a series of measurement devices, but alternatively, light sources travel past a measuring device or a measuring device travels past the light sources. In this way, for each individual reference light source, an identifier (represented with #), time t, a representative of power as a function of time P(t) and a representative of emitted light as a function of time M(t) are determined and added to a training dataset 13. The power can be measured for instance as voltage V, and current I. Other reference parameters can be added, like temperature, production process parameters, parameters identifying or representative for raw materials used can be added, like purity, and other parameters.

The training dataset 13 is provided to a machine learning system 12, for instance a neural network like a deep learning network. Such a neural network can for instance be used to predict emission at a point in time. The output of the neural network 12 can for instance be the spectral output at an operating time. Thus, the neural network is used to predict spectral output after an input operating time. The spectral output can for instance be provided for wavelengths between 400 and 700 nm. Usually, this is given for wavelength intervals of 20nm, 10 nm and even 5 nm. This allows easy calculation of colorimetric values. Using the spectral output (for instance temperature of a black body emitter, for instance D65, Daylight 6500K), for selected color samples a colorimetric theoretically perceived color can be calculated. This can be compared to predefined colorimetric values, allowing qualification of a light source or set of light sources in view of actual effect.

In figure 3, schematically a production line of light sources 20 is depicted. A quantity of these light sources is a series of light sources 22. The spectral output (dotted triangles) of each light source 20 is measured using the measuring system 10. Again, various options are possible, for instance a stream of light sources automatically passing a spectrometer, or a spectrometer passing a line of light sources 20. The light sources can for instance be measured after or during production. The measurement can be done the first time a light source is powered, at operation time 0 (‘zero’), of after a preset standard operating time ti. It may also be possible to measure a time series [t₀....ti] at time intervals dt, for instance the first 10 seconds end each second ([0... 10] with dt = 1 s), first minute, or the like after powering the light source for the first time.

The measured data is inserted into database 14 with an identification of the light source ('#’) and operating time. Using the trained neural network 12, the light sources of the series of light sources are categorized. The trained neural network can for instance be used to predict the spectral output over time of each light source. Using that prediction, statistical matching can be done to produce sets of light sources 23 that comply with a preset requirement. The preset requirement can for instance be the spectral output after a set operating time t. The categorization can be based upon a criterium like a predefined deviation from a predefined black body temperature, a predefined average humanly perceived color difference that is calculated for irradiating a set of color samples with a set of light sources compared to the perceived color difference that would result from irradiating the same set of color samples with a spectral emission of a spectral emission from a desired/preselected set of light sources.

Please note, as described in Wikipedia, In the field of color vision, a MacAdam ellipse is a region on a chromaticity diagram which contains all colors which are indistinguishable, to the average human eye, from the color at the center of the ellipse. The contour of the ellipse therefore represents the just-noticeable differences of chromaticity. Standard Deviation Color Matching in LED lighting uses deviations relative to MacAdam ellipses to describe color precision of a light source. As a measure of perceived color difference perception, MacAdam's results confirmed earlier suspicions that color difference could be measured using a metric in a chromaticity space. A number of attempts have been made to define a color space which is not as distorted as the CIE XYZ space. The most notable of these are the Cl ELU V and Cl ELAB color spaces. Although both of these spaces are less distorted than the CIE XYZ space, they are not completely free of distortion. This means that the MacAdam ellipses become nearly (but not exactly) circular in these spaces. Often, a standard set of ceramic tiles is defined for measurement and calibration of color measurement. These standards can be used in the simulation of light source color perception.

In fact, in general the spectral output M for many light sources is a function of operating time t, wavelength A, temperature T, power P.

For categorizing, the trained neural network can be combined with an additionally trained neural network that in fact groups light sources into optimized sets of light sources that provide the best (closest) combination providing a desired combined spectral emission of the set of light sources.

It will be clear to a person skilled in the art that the scope of the present invention is not limited to the examples discussed in the foregoing but that several amendments and modifications thereof are possible without deviating from the scope of the present invention as defined by the attached claims. In particular, combinations of specific features of various aspects of the invention may be made. An aspect of the invention may be further advantageously enhanced by adding a feature that was described in relation to another aspect of the invention. While the present invention has been illustrated and described in detail in the figures and the description, such illustration and description are to be considered illustrative or exemplary only, and not restrictive.

The present invention is not limited to the disclosed embodiments. Variations to the disclosed embodiments can be understood and effected by a person skilled in the art in practicing the claimed invention, from a study of the figures, the description and the attached claims. In the claims, the word “comprising” does not exclude other steps or elements, and the indefinite article “a” or “an” does not exclude a plurality. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference numerals in the claims should not be construed as limiting the scope of the present invention. REFERENCE LIST

1 lighting system

2 lighting unit

3 calibration system

4 control unit

5 local light sources

6 drive unit

7 optical part

8 lighting unit control

9 statistically representative light sources

10 measuring system

11 control unit communication device

12 Machine learning module

13 training data set

14 local operating parameters

15 reference operating parameters

16 local control settings

20 light source

22 series of light sources

23 set of light sources t time

T(t) temperature as a function of time

P(t) power as a function of time

M(t) light source output as a function of time, often also a function of wavelength, M(A, t)

L illumination instructions

Claims

-28- CLAIMS

1 . A method for producing, from a series of light sources, a set of light sources having a predefined resulting spectral emission, comprising:

- providing said series of light sources;

2. The method of claim 1 , wherein said at least one time series is relative to a time of first operation, in particular said time series includes said time of first operation, more in particular said time series time base is based on operation time.

3. The method of claim 1 or 2, wherein said at least one time series includes said measurement time.

4. The method of any one of the preceding claims, wherein said spectral output includes spectral output in a visual wavelength range, in particular including a wavelength range of 400-700 nm.

5. The method of any one of the preceding claims, wherein in said categorizing at least one color vision result of a human observer is calculated using said spectral output and said categorization is based upon said at least one color vision result, in particular said categorization take into account said color vision result at a predefined simulated moment in time.

6. The method of any one of the preceding claims, wherein said at least one set of at least two light sources comply with said predefined resulting spectral emission, in particular a color vision result of a human observer calculated using said spectral output complies with a predefined color difference criterium, when a resulting deviation is statistically within a predefined criterium.

7. The method of any one of the preceding claims, wherein in said categorizing, a relative positioning of said light sources is taken into account.

8. The method of any one of the preceding claims, wherein said neural network comprises a deep learning neural network.

9. The method of any one of the preceding claims, wherein said categorizing comprises composing from said series of light sources a series of said sets of at least two light sources, in particular at least three light sources.

10. The method of any one of the preceding claims, wherein said series of light sources are produced in at least one selected from batch production, continuous production, and a combination thereof, and during said production additional reference light sources are selected for expanding said training dataset.

11 . The method of any one of the preceding claims, wherein said training dataset is further provided with at least one production parameter of each reference light source, in particular a production parameter representative of at least one characteristics of raw materials used in the production of said light source.

12. The method of any one of the preceding claims, wherein said reference operating parameters comprising for each reference light source a time series of a representative of a temperature, and said training dataset further comprises time series of said representative of said temperature. -SO-

13. The method of any one of the preceding claims, wherein said representative of said power comprise at least one selected from voltage, current, and a combination thereof.

14. The method of any one of the preceding claims, wherein said training dataset further comprises production parameters including raw materials characteristics.

15. The method of any one of the preceding claims, wherein said set of light sources comprises a substantially red light source, a substantially green light source and a substantially blue light source.

16. A method for producing a lighting system comprising providing a set of light sources using the method of any one of the preceding claims.

17. A production system for lighting systems, comprising: