Predicting LED Lifetimes

Recently, a corporate client asked me to specify only LED fixtures with a lifetime of 100,000 hours, and preferred fixtures with a life of 200,000 hours.  I don’t know where they came up with these numbers, but my reply was that an L70 of 100,000 hours or more cannot be validated through standard testing procedures.  Here’s why.

To begin with, LEDs themselves don’t experience catastrophic failure the way incandescent and fluorescent lamps do. The don’t stop making light, but their output declines over time.  Today the generally accepted calculation of the life of an LED is called L70, which is the length of time before the light output has fallen to 70% of initial output.

The IES approved lifetime calculation method begins by collecting data using the procedure described in LM-80 (ANSI/IES LM-80 Measuring Maintenance of Light Output Characteristics of Solid-State Light Sources). Please note that LM-80 measures “LED packages, arrays, and modules” not fully fabricated fixtures, and there’s some dispute about whether or not testing bare modules is appropriate.  However, it does permit module manufacturers to test once and derive a lifetime, rather than every fixture manufacturer testing every fixture with every module they want to offer, which would be incredibly burdensome and expensive.

LM-80 requires a minimum collection time of 6,000 hours (250 days) but sets no upper limit.  If manufacturers want to use the data they’ve collected and project future performance they use the calculation procedure in TM-21 (ANSI/IES TM-21 Projecting Long-Term Luminous, Photon, and Radiant Flux Maintenance of LED Light Sources).  Importantly, TM-21 only permits data to be projected to six times the LM-80 data collection time period.  This is because of uncertainties involved with longer predictions (see PS-10-08 IES Position on LED Product Lifetime Prediction at https://www.ies.org/advocacy/position-statements/ps-10-18-ies-position-on-led-product-lifetime-prediction/).  So, an L70 of 50,000 hours is based on at least 8,333 hours of LM-80 testing.  That’s 347 days.

Thus, to say that an LED has an L70 100,000 hour life would require a data collection period of 16,667 hours (695 days), or 1,390 days (3.8 years of continuous testing) for a life of 200,000 hours.  Today, no LED manufacturer conducts LM-80 tests for that extended period of time because the lifetime of a given LED product is too short.  By the time you’ve finished a 4 year long test, the LED being tested is out of production and replaced by something new.  In the future, when the LED industry has matured and we’re no longer seeing continuous improvements in efficacy, color rendering, etc., they may test for that long, but not now.

Where do these 100,000 hour and longer lifetimes come from?  It seems that some manufacturers are using an internally generated prediction to get to these numbers.  The thing is, we don’t know what’s involved in that prediction, which means we can’t validate it or compare it to any other prediction.  We just have to take their word for it. With the LM-80/TM-21 procedure, on the other hand, we know that testing labs, regardless of who or where, are using the same procedure and their results should be consistent and repeatable.  That allows us to reliably, confidently compare fixtures by any number of manufacturers.

Tony and I Talk Color Rendering

Podcast Album Cover

In September at ArchLIGHT Summit, Tony Esposito and I gave a series of demonstrations on the spectral flexibility of LEDs and the possibilities they present with regard to color rendering.  While there we spoke to Sam Koerbel on his LytePod podcast about the basics of the new measures introduced in Annex E, and discuss why TM-30’s multi-dimensional approach to quantifying color preference is superior to the old-standby in the industry: CRI.  Our discussion is now available.  Give it a listen.

26 Consumer Groups Urge D.O.E. to Take Action on Lamps

You may recall that in 2019 the Trump administration blocked a rule intended to phase out incandescent lamps and encourage a conversation to more energy efficient models, namely LEDs.  If you don’t remember the New York Times and NPR both had articles, among many others.

Last week, the Consumer Federation of America and the National Consumer Law Center, along with 24 other groups across the country, urged the U.S. Department of Energy (DOE) Secretary Granholm to implement the efficiency standard for household lighting products mandated by Congress as soon as is practicable.  They claim that “Each month of delay costs American consumers nearly $300 million in lost utility savings and results in another 800,000 tons of climate changing CO2 emissions over the lifetimes of the incandescent bulbs sold in that month.”

You can read the entire press release at Consumerfed.org

Standard 189.1 Now Includes TM-30 Requirements

Yesterday an addendum to ANSI/ASHRAE/ICC/USGBC/IES Standard 189.1-2017 Standard for the Design of High-Performance Green Buildings was published. The addendum makes changes to Section 8.3.5, which covers lighting. One of the biggest changes is to add TM-30 color rendition criteria to the section on Indoor Lighting Quality. Here’s the relevant text:

8.3.5.3 Color Rendition. At least 95% of lighting power of nominally white lighting within each enclosed space shall be provided by luminaires that meet the following criteria at full light output in accordance with IES-TM-30, Annex E, P2 and F3:
1. Rf of at least 85
2. Rf,h1 of at least 85
3. Rg of at least 92
4. Rcs,h1 of at least -7% but no greater than +19%

Nominally white lighting is lighting that has chromaticity within the basic or extended nominal color correlated temperature (CCT) specifications of ANSI C78.377.

Where a lighting system is capable of changing its spectrum, it shall be capable of meeting the color rendition requirements within each nominal CCT of 2700 K, 3500 K, 4000 K, and 5000 K, as defined in ANSI C78.377, that the system is capable of delivering.

I hope that this is going to put more pressure on manufacturers to improve the color rendering of their luminaires as measured by TM-30, not CRI, and to provide TM-30 information on their cut sheets. If not, they’ll risk not being considered on projects that have TM-30 requirements.

A Look at Bridgelux’s Average Spectral Difference Metric

Today’s post was going to be a reminder to take manufacturer provided education with a grain of salt.  Last week I sat through a manufacturer’s presentation on color.  There were some big errors and some that’s-not-quite-right errors that angered me. The information presented wasn’t hard to confirm, but whoever created the presentation didn’t so some of it was wrong.  However, before I could start writing I received an email about a new color quality metric that was developed by Bridgelux.  Here’s the scoop.

Last Thursday, May 14th, Bridgelux announced a new metric, Average Spectral Difference (ASD), which they claim quantifies the naturalness of a light source.  The announcement is based on this white paper by Bridgelux.  The white paper asserts that since we evolved under fire light and day light, human-centric lighting should use spectra that mimic these “natural” sources.  Bridgelux says that, “ASD provides an objective measurement of how closely a light source matches natural light over the visible spectrum, averaging the differences of the spectral peaks and valleys between a light source and a standardized natural light source of the same CCT.”  

Basically, ASD is a measurement of the difference between a “natural” spectrum and that of an electric light source.  It is expressed as a percentage, with lower percentages equaling a closer match to the reference source and higher percentages equaling a larger difference between the two.

My first thought was, “Oh, it’s CRI – Natural Edition” but in some ways it’s even worse.  For starters, while Bridgelux presents a definition of “natural” light that is based on the illuminants we use as references for color fidelity calculations, there is no accepted definition of “naturalness” in the lighting industry, or most other industries for that matter. Obviously, a metric for something that has no industry-wide definition is of questionable value.  The white paper says, “The reference source used by Bridgelux is the blackbody curve (BBC) for light sources of 4000K and below, and the daylight spectrum (i.e. standard illuminants such as D50, D57, and D65) for light sources of 5000K and above.”  (Yes, there’s an obvious typo there because they’ve left a gap between 4000 K and 5000 K.)  Second, like CRI it presents a single number with no additional information about where in the spectrum the differences occur, or if they are increases or decreases relative to the reference light source.  Third, as a measurement of spectral difference alone, it disregards the fundamentals of human vision, including the principle of univariance and how perception changes with intensity, among other things.  

I emailed a few colleagues on the IES Color Committee and found that they were already examining ASD.  Some of the comments that came back were, “This is just a refresh of a spectral bands method. It says little about color rendering” and “This is very similar to the Film industry’s SSI developed by the Academy. It also suffers from the same problem. If the result isn’t 0% (or 100%) then it tells you nothing about where the differences are. Thus, it tells you nothing about whether two light sources will work together.”

Michael Royer at PNNL went further by looking at ASD with the sets of data in TM-30 Annex F that were used to develop the TM-30 Annex E recommendations.  Here’s what he had to say. (You may have to right click and open the graphs in a new tab to see them clearly.)

First, spectral similarity metrics are not new at all—they predated CRI (e.g., Bouma spectral bands method from 1940s). For some reason they gained popularity again in the last decade or so. Here are some other examples:

B. H. Crawford. 1959. Measurement of Color Rendering Tolerances J. Opt. Soc. Am. 49, 1147-1156

Crawford, B. H. 1963. Colour-Rendering Tolerances and the Colour-Rendering Properties of Light Sources. Transactions of the Illuminating Engineering Society, 28: 50–65.

Kirkpatrick, D. 2004. Is solid state the future of lighting?” Proc. SPIE 5187, Third International Conference on Solid State Lighting.

J. Holm et al., “A Cinematographic Spectral Similarity Index,” SMPTE 2016 Annual Technical Conference and Exhibition, Los Angeles, CA, 2016, pp. 1-36, doi: 10.5594/M001680. Also: https://www.oscars.org/science-technology/projects/spectral-similarity-index-ssi

Acosta I, Leon J, Bustamante P. 2018. Daylight spectrum index: a new metric to assess the affinity of light sources with daylighting. Energies 11 2545

Spectral similarity measures, like ASD, don’t relate to perceived naturalness or preference at all. They’re more closely correlated with color fidelity (e.g., Rf) but perform even worse in terms of correlation with perceived qualities because they don’t account for how the visual system works (they might have more use for understanding cameras, as used by SMTPE with SSI, linked above). I guess people just assume that a Plankian/Daylight spectrum is ideal.  While smooth SPDs have advantages, Planckian/Daylight SPDs aren’t perceived as more natural or more preferred in typical architectural lighting scenarios. This has been shown over and over in experiments, where it’s become quite evident that certain deviations from Planckian are preferred/viewed more natural than others.

Here’s the correlation between ASD and rated naturalness/normalness, preference, and Rf for the three datasets used to develop TM-30 Annex E:

 If you’re not up on your statistics, r2 is a measurement of how well data fits to a prediction or to the data average.  1.0 is a perfect fit.  Generally, 0.7 or above indicate a strong statistical correlation, and values less than 0.3 indicate no relationship.

PNNL (combination of three studies):

Zhejiang:

Penn State:

Overall, it’s clear that ASD isn’t a tool for characterizing perceived naturalness (or preference) over a wide range of SPDs, and it probably has limited other uses. While spectral smoothness (as exemplified by the reference illuminants in ASD) is sometimes a useful goal, there are other metrics more rooted in human vision to better asses this characteristic. It’s a shame that ASD and the accompanying message will likely lead to confusion, especially when there’s enough to learn about color rendition already.

This is a good example of why it’s important to rely on metrics that have been vetted through a standardization process and to always be skeptical of marketing material.

So there you are. Take manufacturer’s education with a grain of salt. The same is true of their internally developed metrics. I’m not saying that they are intentionally deceiving anyone. but their goal is sales, not education. As Mike points out, this is why metrics need to go through a vetting process before we can use on them with confidence.

By the way, although I’ve mentioned the IES Color Committee and quoted a few of its members, this post doesn’t represent the opinions of the committee or of the IES.

Do LEDs Make You Look Orange?

Last Thursday Donald Trump spoke to a group of Republicans in Baltimore.  One of the things he said caught my attention: “The lightbulb. People said what’s with the lightbulb? I said, here’s the story. And I looked at it, the bulb that we’re being forced to use, No. 1, to me, most importantly, the light’s no good. I always look orange. And so do you. The light is the worst.”

Now, I’m not aware of being made to look orange under LEDs, nor have I ever noticed LEDs making my friends, colleagues, or students appear orange.  You can’t imagine how embarrassed I’d be if it turned out that a real estate developer and entertainer had more astute color perception than me, a lighting designer and Co-Chair of the IES Color Committee.  If our only means of evaluating the color rendering of a light source, and evaluating the orange content specifically, was CRI we would have no objective way of testing his statement.   CRI, technically Ra, is a single value that gives us an average of the match between the light source in question and its reference source (either a blackbody radiator or a CIE definition of daylight, depending on CCT) using only eight color samples. 

Colors used to calculate CRI Ra

Since Ra is an average value there’s no way to understand the rendering of any particular hue. I’ve talked about this here. However, one of the wonderful things about ANSI/IES TM-30 IES Method for Evaluating Light Source Color Rendition is that we can use it to test that claim.  TM-30 uses 99 color samples that are distributed across the color space and the visible spectrum. 

TM-30 99 color evaluation samples (CES)
TM-30 CES spectral reflectance functions

It also breaks the color space up onto 16 Hue Bins, each one covering a specific range of the color space.  In the case of orange, we want to look at Hue Bin 3.  Specially, we want to look at Rcs,h3 (the subscript CS stands for Chroma Shift) which quantifies the increase or decrease in the saturation or vividness of orange compared to the reference light source.  

TM-30 hue bins
Example of TM-30 chroma shift bar graph by hue bin

So, let’s put the science of TM-30 to work and see if we really do know that LEDs make us look orange!

The TM-30 calculator contains a library of 300 SPDs (spectral power distributions), of which 137 are commercially available white LEDs.  The CCTs range from 2776 K to 6123 K.  If white light LEDs really did make us look orange we’d expect to see a large majority of them have a positive Rcs,h3, probably with an average chroma shift in excess of 10%.  In fact, the 137 SPDs have Rcs,h3 that range from -8% to 1% with an average of -3.6%, a decrease (not an increase) in the saturation of orange.  It’s not me, it’s him.  TM-30, which uses the most modern models of human vision and a set of colors that cover the color space and visible light spectrum, proves it.  What a relief!  

Don’t believe me?  Download TM-30 and the calculator for free from the IES web site and see for yourself.

Of course, I’m not saying LEDs are perfect light sources. Like any other product there are good ones and bad ones. However, TM-30’s measurements of fidelity and gamut (as averages) and measurements of fidelity, chroma shift, and hue shift (by hue bin) permit us to make a thorough evaluation of a light source to understand its color rendering characteristics. Using this knowledge, we can determine if a particular light source distorts colors and is appropriate for a project, or not.

I should take a moment to note another error he made when he said, “And very importantly—I don’t know if you know this—they have warnings. If it breaks, it’s considered a hazardous waste site. It’s gases inside.”   Perhaps you’ve heard the acronym SSL or the phrase solid state lighting.  LEDs are a version of SSL, which means that they are…well, solid. Unlike previous light producing technologies LEDs are a solid combination of materials.  As such, if one were to physically break (which is unlikely since LEDs are small, are mounted to a heat sink and often covered with a lens, so you’d have to break a lot of materials simultaneously) no gas, hazardous or benign, is emitted.  He’s thinking of fluorescent lamps and the small amount of mercury they contain.  Even then, a broken fluorescent lamp doesn’t turn the area into a” hazardous waste site.” Here are the EPA’s instructions for cleaning up a broken fluorescent lamp.

White House to Relax Energy Efficiency Rules for Light Bulbs – The New York Times

In 2007 Congress passed the Energy Independence and Security Act (EISA) with the goal of increasing energy efficiency across the economy.  Part of EISA has affected the lighting industry in the form of mandated efficacy of light sources.  The initial efficacy rules targeted A-Lamps (standard household light bulbs) and set the efficacy level above that of incandescent but below that of halogen lamps.  The result was a slow shift to the more energy efficient technology.  Over the years the energy efficiency requirements have been expanded to more lamp shapes, always in keeping with technological ability so that we never faced a lamp shortage or loss of a lamp shape.  Today, more than 50% of lamps sold are LED that exceed even the most stringent requirements.

On September 4th the administration announced that it was going to cancel a new set of requirements that would have taken effect in January 2020 that would have applied to products such as decorative medium base lamps and MR type lamps.  In my opinion, this is another example of the administration cutting off its nose to spite its face.  As with the threat to “investigate” automakers who agree with the State of California’s proposed energy efficiency requirements, this effort to undo energy efficiency despite the monumental consensus that we need to reign in our energy consumption isn’t going to go have any effect.  No lamp manufacturer is going to reopen or build new factories to make incandescent lamps when it’s obvious that A) the next administration is going to reinstate the efficacy requirements B) the public has embraced the energy savings of LED lamps, and C) the companies know that it would be bad for their image to turn their backs on mitigating climate change.

Source: White House to Relax Energy Efficiency Rules for Light Bulbs – The New York Times








Samsung Introduces Chip-on-Board LED Packages Optimized for Commercial Lighting – Samsung Global Newsroom

Source: Samsung Introduces Chip-on-Board LED Packages Optimized for Commercial Lighting – Samsung Global Newsroom

An interesting bit of news from Samsung this week.  They’ve developed an LED package especially designed to achieve an Rg value over 110, “a level that ensures lighting with outstanding color and whiteness.”

It’s important to note that increased saturation means decreased fidelity to the reference light source.  This is a lighting solution that will be desirable in some applications, such as retail,and undesirable in others, such as medical facilities.