Lunera GX23 Recall – What we learned as a Company and as an Industry
How the GX23 recall has changed Lunera
This recall exposed a weakness in both Lunera’s and UL’s safety test process. We have sought to plug this in the following ways:
- Lunera has added four burn-in corner testing to our product release process (voltage and temperature)
- Lunera has added in abnormal testing with a low impedance source into our product release process
- Lunera has significantly improved our design review process – individuals can make mistakes, but with multiple individuals reviewing and questioning each component choice and whether it is suited to the conditions present in the lamp the potential for a mistake making it through to the field is geometrically reduced
- Lunera has submitted to UL to modify the UL1598 test process for LED lamps – to include low impedance abnormal testing as standard in the test flow
The GX23 lamp is being re-released in March 2016 with the following changes:
- The fuse has been resized from 1A to 100mA and fully characterized as noted above
- Surge testing was completed with a variety of components and it was determined that the TVS did not improve surge immunity – it is a Zener diode capable of only absorbing a small amount of energy before a crowbar; this it was removed from the system. Surge immunity remains a 7 strikes of 3kV 100kHz ring-wave
We have done our best to make our customers and channel partners whole on the costs of this recall and published this in the interest of improving safety across the LED industry. More details on what we learned follow.
How It All Began
Lunera launched the GX23 product in October 2015 a few months later, just a few days before Christmas we began to receive reports of latent field failures. Christmas vacations were canceled and we moved forward with a voluntary recall with the Consumer Product Safety Commission (CPSC) and Health Canada to remove the units from the field and develop a fix to ensure the product could be safely re-released into the field.
It was the single worst experience of my career; I’m thankful that we did not cause any injuries or property damage as a result. But we learned a good deal and wanted to share that with the industry.
Lunera GX23 Architecture
Fig 1 Lunera GX23 Schematic
The Lunera GX23 lamp uses a single stage non-isolated buck architecture to convert 120V/277V AC with or without a ballast to the 65mA / 66V DC needed to drive its LED array. A simplified block diagram of that circuit is shown in fig 1.
Lamp Testing & Characterization in June – August 2015
The lamp contains two components designed to arrest surges which may be caused by lightning strikes or heavy equipment switching on/off in a non-graceful manner – a metal oxide varistor (MOV) designed to absorb most of the energy from surges and a transient voltage suppressor (TVS) was designed to absorb the rest. The ENERGY STAR spec for surge immunity is for 7 strikes at 2.5kV, this lamp passed up to 3.0kV and failed only at 3.5kV, which we were happy with.
The input filter is designed to isolate the electrical noise generated by the buck switch to ensure high power factor, low THD and compliance with the FCC class A conducted and radiated emissions requirements. The lamp successfully achieved a commercial standard of a PF > 0.9 and a THD <20% at both 120V and 277V and passed FCC testing – we were satisfied with the results.
When the lamp is hit with a surge that exceeds its capability, the MOV and TVS are designed to fail in what is called a crowbar method – they intentionally fail to short circuits causing the input fuse on the lamp to blow. In our testing as well as testing at UL, this condition was created and verified at 120V and 277V using a 1000VA electronic power supply and a 10A variac. The fuse failed safely and we were satisfied with the results.
The MOV that we choose was a Panasonic 5mm MOV rated to start conduction at 431V and to clamp up to 5A of surge at a peak voltage of 745V.
The buck converter in our application the buck converter was rated to a maximum input voltage of 600V thus we added in a TVS rated at 447-494V breakdown to clamp any incoming surge below the 600V FET breakdown rating.
In the event of a crowbar failure, we included a 1A input fuse to open and allow the lamp to fail safely.
In our testing as well as the safety testing at UL, we tested the full voltage range of operation (108V – 305V and tested the full thermal range of operation but neither test protocol includes all four corner nor was the sample size very large (UL typically requires about 5-10 pcs to complete their test, we tested a similar amount at Lunera).
When the TVS that we use was rated to a maximum standoff voltage of 400V; during nominal 277V AC operation the peak voltage is 277V * sqrt(2) = 385V which is fine. However, 277V lines can vary +/- 10% and at the upper range of 305V the peak voltage can be as high as 425V, which exceeds the DC standoff of the TVS but does not exceed it rated breakdown voltage of 447V min. At room temperature there is no issue; at high temperature (100’C) a percentage of the TVS family will crowbar in this condition causing a fault.
The input fuse that we choose was oversized, this should have been caught during our design review process and was not. The maximum current draw at low line (108V) is 56mA, the fuse required 20% thermal derating for 100’C operation and fuses are normally used to only 75% of their rated current otherwise one risks falsely tripping them. In retrospect, a 100mA would have been the desired size for the fuse, in our case we were 10x over-rated.
The input filter that we chose was designed for the 5W load and has a relatively high DC resistance of 18 Ohms due to its high inductance and narrow wire gauge.
Our analysis concluded the following cascade occurred:
1). TVS crowbar failure causing the lamp to short circuit;
2). High current from line through the filter inductor to the TVS crowbar resulting in thermal failure of the filter inductor – a mechanical pop which may be sufficient to dislodge the lens and a smell of burnt electronicsEventual opening of the fuse causing the cascade to end;
3). Eventual opening of the fuse causing the cascade to end.
Fuse Ratings & How It Affects the System
Fuses are essentially small wires that heat up and melt when their current rating is exceeded. In this case, the 1A fuse that we used had a rating of 6.6 to 9.2 A2s to opening.
In our system, with 18 Ohms of inductor resistance and 6.8mH of inductance when driven from a 277V input this fuse takes 43.78ms to open with a peak current of 21A, during this time the inductor L1 is receiving 3.8kW of power for a total of 167J of energy transferred to the fuse. This is more than sufficient to cause the fuse to mechanically fail and pop off the board dislodging the lens.
When we changed to the 100mA fuse with a rating of 0.02 A2s, the fuse opens in 1.28ms with total energy transferred to the inductor during opening reduced by 425x.
Net-net – over-rating a fuse to 10x causes almost 500x more energy to be released into the system before opening. It’s really important to get fuse ratings right.
How this mistake crept through the process
In Lunera’s in-house testing as well as abnormal testing at UL, we used a power supply capable of supporting approx. 1000VA – that is 277V @ approx. 4A of current. When current is limited to 4A, 1A the fuse opens in about 500ms and power transfer to the lamp is only 288W. The inductor suffers no damage when exposed to 288W for 500ms.
The missing step in both Lunera’s and UL’s testing was to use a low impedance (e.g. utility input) power source to test the abnormal condition. Although 288W for 500ms and 3.8kW for 44ms both have a similar amount of energy they result in a dramatically different outcome.