The j2 Flight Plug-In is used to review flight test data and to identify the model corrections required.
One of the challenges related to flight test matching is that establishing a great match for one flight condition and using that to predict the next often leads to less than desirable results starting a manual and iterative solution loop. With j2 Flight, multiple test points can be analysed as a batch and the optimum solution for all cases found.
The detailed analytical approach provided by j2 Flight introduces a teachable, robust and repeatable process. This reduces the pressure and requirement on heavily experienced engineers who have learnt through experience which aspects of model corrections work. By having complete control of the model, we are able to perform a lot more activities mathematically.
The matching process developed by j2 Aircraft Dynamics follows a very similar approach to that which is used across multiple companies already. However, with some of the extra steps/details that are available, we are able to reduce the burden on experience, and the manual aspects of flight matching by up to 30%. All the process can be run as batches of test points to avoid iterative divergent solution loops.
The initial stage is the import of flight test data into the j2 Universal Database. Data can come in many formats with various signal names and in a range of units. The import stage allows for the conversion of signals from the FDR into SI units. At the same time, the signal names are matched to those on the a priori model. The set-up of the units and signal mapping is only required once and can be saved for re-use over and over again.
Flight Path Reconstruction
Once the FDR data is in the j2 Universal Tool-Kit, the next stage is to rebuild a complete set of aircraft states based on both kinematical and observation models. Whilst signals may exist in the data to provide equivalent values, the data from the FDR may contain noise and possibly bias. The process of flight path reconstruction is used to take a set of input parameters (Accelerations and Angular Rates) and use these to calculate/predict all linear/angular velocities. Other signals from the FDR can be used to “check” the validity of the states when considering the noise/accuracy of the sensors.
Re-prediction is the process of taking signals from the reconstructed data and tracking them. Tracking a signal means that the a priori model has the value forced to that from the reconstructed data. This could be an aircraft state, control surface deflection or any parameter that has been recorded and matched via the import process. The re-prediction process can be split into two activities:
In the QTG stages, we are interested to see if the a priori model is able to reproduce the behaviour of the aircraft within specified tolerances. As such we want to enable the model to “fly free”. As such we track drivers into the system such as control surfaces, engine characteristics, atmospherics etc. The remaining parameters are then a result of the model characteristics and kinematics. Here we can now identify whether the resulting behaviour of the model is a sufficient match to the real aircraft.
In the event that the model does not lie within tolerances it is necessary to identify corrections. Historically, it is down to experience to look at the relevant discrepancies from the QTG and to estimate corrections, often through iterative trial and error. However, because we have complete control of the aircraft model, we are able to drive all the aircraft states. This means that we can force every state as well as the drivers on the model to be identical to the flight data. Knowing all the aircraft states we can then use kinematics along with the aircraft mass and geometry and can calculate the total aerodynamic coefficients from the flight data.
From the re-prediction analysis we have the coefficients for the flight data. We also have the calculated total aerodynamic coefficients on the predictive model. In the Matching Re-Prediction as we tracked all aircraft states and drivers we can compare the coefficient from the flight data with that calculated on the model at the same flight condition. A regression model enables us to define the contributions to the coefficients, these can be combinations of derivatives and look up tables. The analysis then enables us to identify, from the variation in the states across multiple flights and the structure of the coefficients, the magnitude of the contributions for both predictive and actual aircraft. The software then calculates the correction necessary to align the predictive model with the real aircraft.
Once the corrections have been identified, these need to be applied to the model. As the corrections can be numerous and complex, manually updating the model can result in errors being introduced. As such the corrections can be exported, collectively or per coefficient, for review before being imported onto the model. The model that the corrections are imported onto does not have to be the original, but can be a delta model. The updated model can then be re-run through the QTG tests automatically. This therefore creates a sandbox environment to test end evaluate the corrections prior to updating the baseline.