Here's the challenge for airbag engineers:
Design a computer capable of measuring and analyzing millions of possible combinations of crash forces acting on it, yet small enough to fit under a car seat.
Give it a hair-trigger response time as fast as 12 thousandths of a second - but make it discerning enough to overlook jarring potholes, speed bumps and other sudden jolts.
Build it to stand ready for instant action at any time over a period of a decade or more with no maintenance and constant exposure to light, water, dust, searing heat or bitter cold.
Have it ready for mass production in 18 months or less.
In fact, engineers can do all of the above most of the time; they just can't do it all the time. That's why airbags have been going off when they're not supposed to.
'Your testing program would have to be trillions of miles to be sure you've got 100 percent (reliability). That's Apollo-moon-shot levels of reliability,' says Phoenix-based engineering consultant Don Struble.
Algorithms are key
Better sealing of the airbag system against the elements will reduce one kind of inadvertent deployment, those caused by rogue electrical signals. But writing a foolproof algorithm - the controlling program the system uses to decide when it's time to fire the airbag - is more of a problem, say industry sources.
The algorithm is at the very core of an airbag system. It is a mathematical code, usually hundreds of lines long, that decides whether the stresses suddenly acting on the vehicle merit deployment of the airbags.
In larger cars, the algorithm has about 25 thousandths of a second to decide 'yes' or 'no.' In cars with less crush zone, it can be half of that.
To write the algorithm, automakers try to simulate real-world conditions by ramming cars into barriers and running them over corrugated test tracks while instruments measure dynamic forces. The most important of these measurements is the change in vehicle velocity during impact, or 'delta-V.'
For example, a delta-V of 40kph over 5 seconds is hardly dangerous; over 50 thousandths of a second, it could be deadly. Somewhere in the middle - such as 17kph over 25 thousandths of a second - a threshold is set for airbag deployment.
The industry has a rough rule, called '5 inches, 30 milliseconds.' This means that the airbag should be at full deployment within 0.03 seconds of a firing decision in a crash severe enough to move the occupant forward five inches (127mm).
But the actual firing threshold also depends on the vehicle's structural response to a collision. Testing can take up to 18 months.
One possible reason for airbag misfires is that most algorithms have boundaries for deployment and nondeployment that overlap, creating a 'window' of values in which the need for airbag deployment is unclear.
This gray area is caused by the need to accommodate unpredictable variables such as: the number of occupants in the car at the time of the crash and their collective weight; the weight of cargo; and the weight, speed, trajectory and composition of the struck object.
Put simply, a car packed with people and luggage striking an object may need an airbag deployment more than a single driver in the same, empty car.
Overlap tends to be larger in smaller cars because passengers and luggage have a greater proportional effect on total weight, says Said Nakhla, director of airbag system design for Breed Technologies Inc. in Lakeland, Florida.
Small cars also stop in less space during a crash because of their smaller crush zones, meaning the acceleration levels in the cockpit are much higher.
Preventing jolts from tricking the sensors requires 'creative thinking' to avoid misfires, Nakhla says. In practice, that depends on the amount of testing experience a company has amassed, he says.
Another complication for algorithm writers has been the industrywide move to single-point sensors, the eyes and ears of the airbag system.
Older airbag-equipped vehicles have multiple sensors placed in forward locations. But since the early 1990s, technology improvements have collected the sensors and control module into a single box, reducing weight, wiring and exposure to the elements.
With single-point sensors, 'all the intelligence is focused there,' said an airbag supplier executive who did not want to be identified. Accordingly, they must be that much more accurate.
Because a single-point sensor, by definition, can measure changes in velocity at only one point in the vehicle, 'The vehicle's structure must be stiffer,' the executive explained.
Engineers are considering returning at least one sensor to the bumper area to provide another set of eyes in a crash.
These so-called 'G-satellite' sensors could be simple, inexpensive devices to measure crash severity and also help distinguish localized bumps from serious impacts.
Nakhla says such sensors would provide airbag systems with more reliability, but add cost.