US-12621910-B2 - Luminaire with multicolor neural network control

US12621910B2US 12621910 B2US12621910 B2US 12621910B2US-12621910-B2

Abstract

The chromaticity and luminous flux output of a multicolor luminaire with multiple independent wavelength outputs is controlled by a controller that has fewer input channels than there are output channels. A trained neural network is used such that the multi-color wavelength settings of the luminaire produce the specified chromaticity and luminous flux output while optimizing luminous efficacy or color rendering capabilities, or satisfying other constraints applied to the luminaire.

Inventors

Ian Edward Ashdown

Assignees

SUNTRACKER TECHNOLOGIES LTD.

Dates

Publication Date: 20260505
Application Date: 20230623

Claims (13)

1 . An illumination system comprising: n light-emitting elements (LEEs) each having a different peak wavelength; a dimming control with 3 input channels, wherein: n>3, where n is a whole number; each input channel provides a control output; and the 3 control outputs together represent a chromaticity and a luminous flux; and a neural network having n output channels that each correspond to a different one of the peak wavelengths, the neural network connected to the dimming control to receive the 3 control outputs and trained to provide a driver signal on each output channel such that the n driver signals together: control the LEEs to provide said chromaticity and said luminous flux while maximizing a color rendering index of light emitted by the LEEs.
2 . The illumination system of claim 1 , wherein the 3 input channels are: red, green and blue; or International Commission on Illumination (CIE) tristimulus values.
3 . The illumination system of claim 1 , wherein: an efficacy of the illumination system; a melanopic flux to luminous flux ratio in the light; a luminous flux to melanopic flux ratio in the light; or a ratio of a first wavelength flux to a second wavelength flux in the light; is maximized with lower priority than maximizing the color rendering index.
4 . The illumination system of claim 1 , wherein the neural network accepts the 3 control outputs as floating point or discretized integer inputs and generates the n driver signals as floating point or discretized integer outputs.
5 . The illumination system of claim 4 , wherein the neural network is a radial basis function neural network.
6 . The illumination system of claim 1 , wherein the neural network converts the 3 control outputs to the n driver signals in real time.
7 . The illumination system of claim 1 , wherein the peak wavelengths are between: 400 nm and 700 nm; or 200 nm and 3000 nm.
8 . The illumination system of claim 1 , further comprising one or more sensors connected to the neural network and positioned to monitor: (a) the light; (b) one or more package temperatures of the LEEs; (c) one or more heat sink temperatures of the LEEs; (d) ambient air temperature; or (e) any combination selected from (a) to (d); wherein outputs from the one or more sensors are used by the neural network to convert the 3 control outputs to the n driver signals.
9 . A method for producing light comprising: receiving, by a neural network, 3 control outputs from a dimming control in an illumination system, the 3 control outputs together representing a chromaticity and a luminous flux; receiving, by the neural network, a constraint to maximize a color rendering index of the illumination system; converting, by the neural network, the 3 control outputs to n driver signals, wherein: n>3, where n is a whole number; each driver signal is for a different one of n light-emitting elements (LEEs) in the illumination system, each LEE having a different peak wavelength; and the n driver signals together correspond to said chromaticity and said luminous flux and satisfy said constraint; and sending the n driver signals to drivers for the n LEEs; wherein, when the n LEEs are driven by the drivers: the n LEEs produce the light with said chromaticity and said luminous flux while maximizing the color rendering index of the light.
10 . The method of claim 9 , wherein the neural network maximizes: an efficacy of the illumination system; a melanopic flux to luminous flux ratio in the light; a luminous flux to melanopic flux ratio in the light; or a ratio of a first wavelength flux to a second wavelength flux in the light; in lower priority to maximizing the color rendering index.
11 . The method of claim 9 , wherein said converting is in real-time.
12 . The method of claim 9 , further comprising: monitoring, with one or more sensors: (a) the light; (b) one or more package temperatures of the LEEs; (c) one or more heat sink temperatures of the LEEs; (d) ambient air temperature; or (e) any combination selected from (a) to (d); and using, by the neural network, outputs from the one or more sensors to convert the 3 control outputs to the n driver signals.
13 . A method for training a neural network comprising: (a) specifying, with 3 parameters, a target chromaticity and a target luminous flux for an illumination system; (b) using a genetic algorithm to determine multiple solutions for the target chromaticity and the target luminous flux, each solution comprising a set of at least n driver signals, wherein n>3, n is a whole number and each driver signal is for a different one of at least n different peak wavelengths; (c) calculating, for each solution, a color rendering index value, used to maximize a color rendering index of the illumination system; (d) selecting, from the multiple solutions, the solutions that are within a predetermined tolerance of the target chromaticity and the target luminous flux; (e) repeating (a) to (d) multiple times, each time for a different target chromaticity and a different target luminous flux; (f) using some of the selected solutions to train the neural network; and (g) using others of the selected solutions to validate the neural network.

Description

TECHNICAL FIELD The subject matter of the present invention relates to the dynamic control of luminaires with multiple color channels, wherein a trained neural network controls the luminous or radiant flux emitted by each color of the luminaire in order to satisfy a set of physical constraints. BACKGROUND The term “light-emitting element” (LEE) is used to define any device that emits electromagnetic radiation within a wavelength regime of interest, for example, the visible, infrared or ultraviolet regime, when activated, by applying a potential difference across the device or passing a current through the device. Examples of light-emitting elements include solid-state, organic, polymer, phosphor coated or high-flux light-emitting diodes (LEDs) or other similar devices as would be readily understood by a person knowledgeable in the art. An LEE may have two or more constituent LEEs. The introduction of high-power light-emitting diodes (LEDs) some twenty years ago led to the introduction of color-changing luminaires with red-, green-, and blue-emitting (RGB) LEEs (e.g., Speier and Salsbury 2006). By connecting the LEEs to three independent electronic drivers (“color channels”) and varying the drive current to each color group, it is possible to generate a wide range of colors, including “white” light with different correlated color temperatures (CCTs). Mixing red, green, and blue light will produce colored light with any chromaticity within the color gamut defined by the chromaticities of the red, green, and blue light sources. However, the lack of spectral radiant flux between approximately 550 nm and 600 nm, which we generally perceive as yellow light, results in yellow objects illuminated by RGB LEEs appearing to be dull and lifeless. In general, such RGB light sources have poor color rendering properties. To resolve this issue, many color-changing luminaires designed for architectural applications include amber-emitting LEEs. These RGBA LEE luminaires require four independent color channels, but the improved color rendering capabilities often justifies their additional cost. There is however a disadvantage. With three color channels, the chromaticity of the mixed light is uniquely defined by the ratios of red, green, and blue light. This is not true for four or more color channels. Taking an RGBA luminaire as an example, there are an infinite number of different ratios of red, green, blue, and amber light that will, when mixed, produce light with a specific chromaticity. The disadvantage is that each different set of ratios results in different color rendering capabilities. Further, the different color LEEs will have different luminous efficacies, so that each different set of ratios will result in different luminous efficacies for the luminaire. With each LEE having a maximum allowable drive current, different ratios may result in different maximum luminous flux from the luminaire for the specific chromaticity. This problem is compounded by the electrical, optical, and thermal characteristics of high-power light-emitting diodes. The luminous efficacy and spectral power distribution of an inorganic LEE depends on its junction temperature, forward voltage, dynamic resistance, and package thermal resistance. The problem is further compounded by theatrical luminaires, which often have to swiftly and sometimes instantaneously change both the luminous flux and chromaticity of the emitted light. In doing so, the LEE junction temperatures will change, with consequent changes in luminous efficacies (particularly for amber LEEs) and spectral power distribution as the LEE heat sinks reach new (if temporary) thermal equilibria. Unfortunately, there is no deterministic method to determine the optimal set of ratios for LEE-based luminaires with more than three color channels. Mathematically, it is an overdetermined system with four or more input variables (i.e., the luminaire's color channel settings) and only three output variables (i.e., the CIE tristimulus values X, Y and Z defining the luminaire's target chromaticity and luminous flux output). Given, for example, a seven-color theatrical luminaire, an operator can only guess at the drive current settings for seven independent color channels to achieve a desired chromaticity and luminous flux. Worse, the LEE electrical, optical, and thermal characteristics are nonlinear. To understand the problem, first consider color-changing LEE modules with red, green, and blue (RGB) LEEs. Generating a specific color is a simple matter of choosing the appropriate ratios of the red, green, and blue LEE intensities. For example, if we want to generate 4150 K white light, we might choose the color channel intensity ratios graphically shown in FIG. 1. There are however two problems with RGB LEE-based luminaires. First, their color gamuts are fairly limited. In particular, they are typically unable to produce saturated blue, violet, and cyan colors (e.g., see gamut 20 provided by LEEs