· Nimrod Rosenberg · Embedded UAV test pilot
Envelope Expansion, Step by Step: How I Open the Flight Envelope of a New UAV
How to expand the flight envelope of a new unmanned aircraft safely: build-up logic, one variable at a time, stop criteria, data to watch and how to decide the next point.
What "the envelope" actually means
The flight envelope is the set of conditions the aircraft can fly in safely: airspeed, altitude, weight, centre of gravity, wind, load factor, temperature, and for VTOL aircraft the transition corridor between hover and wing-borne flight. A new aircraft starts with an envelope that is mostly a prediction. Envelope expansion is how we replace the prediction with evidence.
On the site I draw it as a field of test points that fill in from the middle outwards. That picture is accurate: you start where you're confident, and you move outward in steps.
The rules I don't break
One variable per point. If I increase airspeed, weight stays the same. If I add weight, I go back to a speed we've already cleared. When something unexpected happens, there should only be one suspect.
Predict before you fly. Every point has an expected result written on the card: expected pitch attitude at that speed, expected motor output in hover at that weight, expected altitude loss in transition. If we can't predict it, we start further inside.
Stop criteria are written down. Not "if it looks bad". Things like: attitude error above an agreed value, motor output above an agreed percentage, airspeed below an agreed number in the turn, vibration above an agreed level. The pilot and the test conductor both know them.
Clear the data before the next point. For critical points, we land and look at the log before stepping. For lower-risk build-ups, a live telemetry check may be enough. We agree which is which in the plan.
Steps get smaller near the edge. Early steps can be generous. As we approach a predicted limit, the increments shrink.
A worked example: airspeed expansion on a quadplane
Say the predicted cruise is in the middle of the speed range, and the team wants to know the real minimum and maximum usable airspeed in fixed-wing mode.
- Start at a speed comfortably inside the prediction. Straight and level, stabilised mode, light weight, calm conditions. Record trim attitude and throttle.
- Step up. Increase the speed demand by a modest increment. Hold it, record, check attitude and control surface deflections.
- Watch the trend, not just the point. If pitch trim is changing faster than predicted, or a control surface is getting close to its travel limit, the next step gets smaller.
- Stop at the agreed criterion. Maybe that's a structural speed the engineers set, maybe it's a control margin. We don't go past it to "see what happens".
- Repeat downward for minimum speed, with smaller steps and the lift motors armed as a safety net if the aircraft has them.
Then repeat the whole sequence at the next weight.
The variables, in the order I usually take them
- Weight and CG. Start light with a forward CG (usually more stable), then move toward heavier and further aft. CG changes are one of the places I want the smallest steps.
- Airspeed. As above.
- Manoeuvre severity. Bank angle, pitch rate, turn rate. Doublets and step inputs at increasing amplitude.
- Altitude and distance. Climbs to higher operating altitudes, longer legs. This is also where link performance enters the picture.
- Wind and gusts. Started in calm, extended to the agreed limit. Wind is the variable you can't dial in exactly, so you choose test days for it.
- Transition corridor for VTOL: covered in quadplane transition tuning.
What I watch during expansion
- Attitude tracking: how closely the aircraft follows what the controller asked for. Growing error is an early warning.
- Control margin: servo deflection and motor output relative to their limits. An aircraft that's using most of its authority to fly straight has no authority left for a gust.
- Energy: airspeed and altitude together. In fixed-wing flight, the speed and height controller trading them badly shows up as wandering altitude.
- Vibration and sensor health: if the navigation filter is unhappy, nothing else in the log can be trusted. I go through this in reading ArduPilot logs.
- The pilot's own workload. If I'm working hard to hold a point, that is data too, and it goes in the report.
When a point doesn't pass
It happens on every campaign. The rule is simple: don't step past it.
- Land. Read the log while the details are fresh.
- Decide whether the problem is the aircraft, the configuration, the conditions or the test itself.
- Fix and re-fly the same point, or go back to the last good point and approach again with smaller steps.
A failed point is not a bad day. It's exactly what the campaign is for. The bad day is the one where nobody noticed.
What the team ends up with
At the end of expansion the engineering team should have a cleared envelope backed by flown points, a list of limits with the evidence for each, open items for the next design iteration, and the operating limits that go into the operator's procedures. That last part is what turns a test result into something an operator can use.
If you're planning expansion on a new aircraft, the first step is the plan itself. I wrote about how I structure the campaign around it and what goes on each test card. There's also an older post on envelope expansion for UAV developers.
Questions
Is envelope expansion the same as certification testing?
No. Expansion is how the development team learns where the aircraft's limits are. Certification or qualification is a separate process with its own authority. I can assess an aircraft against agreed criteria, but that is not a certification.
Do we need a simulator before expanding the envelope?
It helps a lot. A model gives you a prediction to compare against, and a large gap between prediction and flight is itself a finding. Without one, the steps need to be smaller.
What do you do when a point fails?
Stop, land, read the data. Then either fix the cause and re-fly the same point, or go back to the last good point. You never step past a point you didn't pass.