On 20:th Sept 1999 two Gripens were involved in 1 vs 1 WVR low altitude combat. 10 min into the flight, during a steeply decending turn behind the target aircraft the student passed through the other aircraft's tip vortices which made it exceed its alpha limits. While the pilot had little control authority in a near vertical dive at 1250 m, he got an imminent ground collision warning indicating pull out was impossible and according to standard procedures ejected without taking time to analyze the situation. He landed safely in water.
The two Gripens started as a pair with the leader being the instructor who would also act as a target during the exercise. The goal was to train manouvring visual air to air combat at low and minimum altitude against an aircraft with inferior performance, in order to prepare the student for training with a larger number of aircraft.
The student had already flown all missions described in this phase of the training, but on the present day the number of available aircraft prevented a larger exercise, so one more mission than required in this phase of his training was flown. He had a total of 1063 flying hour, with 103 on JAS 39 Gripen, of which 13 the last 30 days. His status, including medical, was higher than required.
Flying safety requirements for this exercise are:
On the day in question visibility was 20 km with some high altitude haze and no clouds.
After arriving in the training area the pair split and the fighter was led by a fighter controller on the ground towards the target.
The target flew straight and level at 200 m with M 0.66 and was to start manoevring after gaining visual contact with the fighter. Manouvring limits were set at 6 G and 17 deg alpha.
Engagement started head on at 30 km distance with the fighter first getting radar contact with the target and at 8 km visual contact whereupon he acted to get into position to shoot a Sidewinder. The aircraft met several times and started a dogfight which lasted about 1 minute.
The dogfight mainly consisted of a series of barrel roll like manouvres with increasing altitudes and decreasing speeds. In the final part the fighter trailed the target by 400-500 m, the target then started a climbing left turn up to 1400 m altitude where he initiated a half roll like decending manouvre, with an alpha momentarily close to 20 deg. Speed was about 400 km/h, G loads 3-4 G and he pulled out at 400 m. The following fighter choose to open up and took a wider and higher flight path before following down. Estimated distance was about 1000 m.
While decending with the control stick in the softstop position with a nose down angle of 50-70 degrees he felt a brief powerful turbulence. He then pushed the stick forward to unload. When out of the turbulence, he again pulled the stick back, but didn't get the pitch response he expected, nor from the maximum roll and pitch inputs he commanded. While doing this, he perceived a ground collision warning level D with G-load larger than 10 G required to aviod the ground and ejected. Ejection was 5.6 s after the first turbulence.
The initial disturbance occured at 1250 m, pitch down angle -70 degrees, bank angle about 180 degrees, speed 305 km/h, alpha 20 degrees and G load 2.4 G.
It's not been possible to determine exactly the distance to the leading aircraft.
Beta angle didn't vary more than 4 degrees, so the disturbance was in pitch only. What happened was that alpha reduced from 20 to 5 degrees and G decreased to 1.5. The pilot kept commanding 20 degrees alpha which resulted in the aircraft responding with fastest possible canard and elevon movement.
Control surface movement got about half way to maximum deflection when the disturbance disappeared and they started to move the other way, but the available deflection speed wasn't enough to keep alpha below 20 degrees, so it increased to 45 degrees with a corresponding G load increase to 5 G and speed reduction, before the control system as designed with priority over pilot inputs brought alpha down to within limits.
Pitch varied between -75 and -28 degrees during the 3.2 s the aircraft was outside of MLL, when back inside the envelope again the nose pointed nearly vertically down.
The pilot pushed the stick fully forward when he noticed the disturbance. When at 45 degrees he felt the disturbance to be over and had no sense of it being stalled and brought the stick back to the mid range. This didn't give any response so he pulled it back to the softstop and gave a right roll command during 1.5 s. The roll signal was partially acted on by the FCS, but both the pilot's perception of what happened and the recorded memory doesn't indicate the aircraft actually rolled.
He then pulled the stick back to the hardstop. The aircraft was then still over 20 deg alpha and pitching down (max phase difference between control input and aircraft response). As he released the stick to grab the ejection handle the aircraft simultaneously re-entered its normal envelope.
At ejection the aircraft was at 750 m altitude, pitch down 75 degrees, bank angle about 0 degrees, speed 350 km/h, alpha -8 degrees and G load 1.5 G. At spashdown the aircraft had accelerated to 600 km/h as the afterburner was lit.
Unfortunately, the video recorder tape was destroyed by water. Also, some interesting parameters weren't recorded and others sampled at a perhaps a bit low frequency. There were no write protected information between take off and splash down (8.8 minutes).
At the turbulence disturbance there were two warnings lights lit. First PRIMDAT [primary data], which probably (the crash memory records the warnings but not the underlying error codes) was because of code 263 indicating a discrepancy between the nose ADC sensors and those on the fin, with the latter being discounted and information gotten from the INS instead. One second later FELINFO [error information] was lit, most likely because of a code indicating reduced redundancy in the FCS.
Some data points:
543.43 s : FCS error, ADDI flag set (air data discrepancy) 544.50 : Main warning + PRIMDATA warning light (probably secondary air data error) 545.57 : FELINFO warning light (probably reduced FCS redundancy) 546.63 : Ejection 547.97 : Altitude warning
The pilot didn't notice any warning indicating the aircraft was outside of its envelope.
The crash memory doesn't register ground collision warnings, only indirect information like the altitude warning light and radar altitude with together with the INS-derived vector is used to generate the warnings.
The target didn't get any warning and the fighter followed it, so it's very unlikely the fighter got a warning according before the turbulence.
Manual calculations using the sparse data indicates a warning level A or B could happen a few seconds before ejection and level C possibly at the time of the ejection. The altitude warning light was recorded as lit 1.4 after ejection, but it's recorded so sparsely it could have been lit up 3 s before, or 1.6 before ejection.
The altitude used for the ground collision warning is usually gotten from the filtered radar altimeter data. But outside certain pitch and bank angles it goes into standby and air data altitude is used and during the out of MLL condition the nose was only close to horizontal briefly (up to -28 degrees), but during this time four data points were recorded with good correspondance to air data.
One theory is that the other aircraft was measured by the radar altimeter or interference between radar altimeters. This has happened several times in the Swedish air force, with false ground collision and altitude warnings, usually because of radar altimeter interference. But as it was in standby most of the time and when it wasn't agreed with the air data this isn't seen as likely.
A second theory is that the extreme alpha angles have disrupted the air data and indicated a suddenly lower altitude. This will also add to the calculated vertical vector and further increase the ground collision warning level. Air data disruptions have happened during high alpha testing, with sudden air data altitude loss of 300-400 m, which isn't enough of a discontinuity to trigger a monitortrip. If it happened this time cannot be verified because the data simply isn't available, but there's also no data which contraindicate it.
The third is misinterpretion by the pilot. If the HUD was in navigation instead of sight mode the wrong numerals might due to the dynamics of the display be associate with the warning symbol, but this is unlikely since the goal was simulated weapons launch and it doesn't explain the large dynamic arrows of level D.
It can be noted that at the time of the accident Gripen's behaviour in tip vortices wasn't validated, and the pilot instruction simply said they were to be avoided, with no further information about what could happen in them.
Before the accident there had been five occurences of similar tip vortix disturbances in air force service, all during air combat. In three cases the FCS has reconfigured, in one the alpha increased from 15 to 35 degrees and returned to normal values in about 1 s, the left and right alpha sensors have given different readings and leading edge flap monitors have triped and locked the flap.
The Swedish Board of Accident Investigation says it's clear that the exercise in question wasn't designed by someone who was familiar with the now known properties when passing tip vortices and thinks it wasn't compatible with the restriction mentioned in the pilot handbook [from a 1999 perspective]. One month after the accident the air force raised the minimum altitude for this kind of exercise.
The current software release at the time of the accident was R11:9, the following was R12:4 which had a higher limit both for max G load and max alpha. Taken together this means that the aircraft is both flying closer to the limit when pulling to the hardstop and that vortices likely are more powerful, so it's probably easier to exceed the MLL, but on the other hand R12:4 has in addition to improved high alpha laws an autoreturn-to-MLL mode which brings it back to the envelope in 10 s from an alpha of 65 degrees, something which doesn't happen with R11:9
The full report, in Swedish [PDF], original at Swedish Board of Accident Investigation.