Aircraft electrical network design using Model-Based Design

This publication provides a transcription of the webinar "Development of the aircraft electrical network using Model-Based Design". The webinar was hosted by Mikhail Peselnik, engineer CITM Exhibitor.)

Today we will learn that it is possible to tune models for an optimal balance between the reliability and accuracy of simulation results and the speed of the simulation process. This is the key to using simulation effectively and making sure that the level of detail in your model is appropriate for the task you are about to perform.

We will also learn:

• How can you speed up simulations by using optimization algorithms and parallel computing;
• How to distribute simulations across multiple computer cores, speeding up tasks such as parameter estimation and parameterization;
• How to speed up development by automating simulation and analysis tasks using MATLAB;
• How to use MATLAB scripts for harmonic analysis and document the results of any kind of tests using automatic report generation.

We will start by reviewing a model aircraft electrical network. Let's discuss what our simulation goals are and look at the development process that was used to create the model.

Then we will go through the stages of this process, including the initial draft - where we refine the requirements. Detailed design - where we will look at the individual components of the electrical network, and finally, we will use the results of the simulation of the detailed design in order to adjust the parameters of the abstract model. Finally, we will look at how you can document the results of all these steps in reports.

Here is a schematic representation of the system we are developing. This is a model of a half aircraft, which includes a generator, an AC bus, various AC loads, a transformer-rectifier unit, a DC bus with various loads, and a battery.

Switches are used to connect components to the electrical network. As components turn on and off during flight, conditions on the power grid can change. We want to analyze this half of the aircraft power grid under these changing conditions.

A complete aircraft electrical network model must include other components. We have not included them in this half-plane model as we only want to analyze the interactions between these components. This is a common practice in aircraft and shipbuilding.

Simulation Goals:

• Determine the power requirements for the various components as well as the power lines that connect them.
• Analyze system interactions between components from different engineering fields, including electrical, mechanical, hydraulic, and thermal effects.
• And on a more detailed level, to carry out a harmonic analysis.
• Analyze power quality under changing conditions and look at voltages and currents at different network nodes.

This set of simulation goals is best met using models of varying degrees of detail. We will see that as we move through the development process, we will have abstract and detailed models.

When we look at the simulation results of these different versions of the models, we can see that the results of the system-level model and the detailed model are the same.

If we take a closer look at the simulation results, we can see that even despite the dynamics due to power switching in the detailed version of our model, the simulation results are generally the same.

This allows us to quickly iterate at the system level, as well as perform detailed analysis of the electrical system at a detailed level. Thus, we can effectively achieve our goals.

Now let's talk about the model with which we work. We have created several options for each component in the electrical network. We will choose which component variant to use, depending on the task we are solving.

When we explore options for power generation in the mains, we can replace the integrated drive generator with a cycloconvector type constant frequency variable speed generator or a DC coupled FSHD generator. We can use abstract or detailed load components in an AC circuit.

Similarly, for a DC network, we can use an abstract, detailed or multidisciplinary option that takes into account the influence of other physical disciplines such as mechanics, hydraulics and temperature effects.

More about the model.

Here you see the generator, the distribution network and the components in the network. The model is currently set up to simulate with abstract component models. An actuator is modeled simply by indicating the active and reactive power that this component consumes.

If we set up this model to use detailed component options, then the actuator is already modeled as an electrical machine. We have a permanent magnet synchronous motor, converters and a DC bus, as well as a control system. If we look at the transformer/rectifier unit, we can see that it is modeled using transformers and universal bridges that are used in power electronics.

We can also choose a variant of the system (on TRU DC Loads -> Block Choices -> Multidomain) that takes into account the effects associated with other physical phenomena (in Fuel Pump). For the fuel pump, we see that we have a hydraulic pump, hydraulic loads. For the heater, we see temperature effects taken into account, which affect the behavior of this component as the temperature changes. Our generator is modeled using a synchronous machine and we have a control system to set the voltage field for this machine.

Flight cycles are selected using a MATLAB variable named Flight_Cycle_Num. And here we see data from the MATLAB workspace that controls when certain components of the electrical network turn on and off. This plot (Plot_FC) shows for the first flight cycle when components are enabled or disabled.

If we tune the model to the abstract version (Tuned), we can use this script (Test_APN_Model_SHORT) to run the model and test it in three different flight cycles. The first flight cycle is working, and we are testing the system under various conditions. We then automatically set up the model to launch the second flight cycle and the third. Upon completion of these tests, we have a report that shows the results of these three tests compared to previous test runs. In the report, you can see screenshots of the model, screenshots of graphs showing the speed, voltage and generated power at the output of the generator, graphs of comparison with previous tests, as well as the results of an analysis of the quality of the electrical network.

Finding a trade-off between model fidelity and simulation speed is key to using simulation effectively. As you add additional details to your model, the time required to calculate and simulate the model increases. It is important to customize the model for the specific task you are solving.

When we are interested in details such as power quality, we add effects such as switching in power electronics and realistic loads. However, when we are interested in issues such as generation or consumption of energy by various components in the power grid, we will use a complex simulation method, abstract loads and average voltage models.

Using Mathworks products, you can choose the right level of detail for the problem you are solving.

To design effectively, we need both abstract and detailed component models. Here's how these options overlap with our development process:

• First, we refine the requirements using an abstract version of the model.
• We then use the refined requirements for the detailed design of the component.
• We can combine an abstract and a detailed version of a component in our model, which allows for verification and combination of that component with mechanical and control systems.
• Finally, we can use the simulation results of the detailed model to tune the parameters of the abstract model. This will give us a model that runs fast and gives accurate results.

You can see that these two options - system and detailed model - complement each other. The work we do with the abstract model to refine requirements reduces the number of iterations required for detailed design. This speeds up our development process. The simulation results of the detailed model give us an abstract model that is fast and accurate. This allows us to achieve a match between the level of detail of the model and the task that the simulation performs.

Many companies around the world use MOS to develop complex systems. Airbus is developing a MOS-based fuel management system for the A380. This system contains over 20 pumps and over 40 valves. You can imagine the number of different failure scenarios that could happen. With simulation, they can run over a hundred thousand tests every weekend. This gives them the confidence that, whatever the failure scenario, their management system can handle it.

Now that we've seen an overview of our model, and our simulation targets, we'll walk through the design process. We'll start by using an abstract model to refine system requirements. These refined requirements will be used for detailed design.

We will see how to integrate requirements documents into the development process. We have a large requirements document that lists all the requirements for our system. It is very difficult to compare the requirements with the project as a whole, and make sure that the project meets these requirements.

Using SLVNV, you can directly link documents to requirements and a model in Simulink. You can create links directly from the model directly to the requirements. This makes it easier to verify that a certain part of the model applies to a particular requirement and vice versa. This connection is two-way. So if we are looking at requirements, we can quickly jump to the model to see how that requirement is met.

Now that we have integrated the requirements document into the workflow, we will refine the requirements for the electrical network. Specifically, we will look at the operating, peak and design load requirements for generators and transmission lines. We will test them in a wide range of power conditions. Those. during different flight cycles, when different loads are switched on and off. Since we are only focusing on power, we will neglect switching in power electronics. Therefore, we will use abstract models and simplified simulation methods. This means that we will set up the model in such a way as to ignore details that we do not need. This will make the simulation run faster and allow us to test conditions for long flight cycles.

We have an AC source that runs through a chain of resistances, capacitances, and inductances. The circuit has a switch that opens after a period of time and then closes again. If you run the simulation, you can see the results with the continuous solver. (V1) You can see that the oscillations associated with opening and closing the switch are accurately displayed.

Now let's switch to discrete mode. Double click on the PowerGui block and in the Solver tab select the discrete solver. You can see that the discrete solver is now selected. Let's run the simulation. You will see that now the results are almost the same, but the accuracy depends on the selected sample rate.

Now I can select the complex simulation mode, set the frequency - since the solution is only obtained at a certain frequency - and run the simulation again. You will see that only signal amplitudes are displayed. By clicking on this block, I can run a MATLAB script that will run the model sequentially in all three simulation modes and plot the resulting plots on top of each other. If we look closer at the current and voltage, we see that the discrete results are close to continuous, but they coincide completely. If you look at the current, you can see that there is a peak that was not noted in the discrete mode of the simulation. And we see that the complex mode allows you to see only the amplitude. If you look at the sample time, you can see that the complex solver took only 56 steps, while the other solvers took many more steps to complete the simulation. This allowed the complex simulation mode to run much faster than other modes.

In addition to choosing the right simulation mode, we need models with the right granularity. To refine the power requirements of components in an electrical network, we will use abstract models of general application. The Dynamic Load block allows us to set the active and reactive power that a component in the network consumes or generates.

We will define an initial abstract model for reactive and active powers based on an initial set of requirements. As a source, we will use the Ideal source block - an ideal source. This will set the voltage in the network, and you can use this to determine the parameters of the generator, and understand how much power it should produce.

Next, you'll see how to use simulation to refine the power requirements for the generator and transmission lines.

We have an initial set of requirements including the power rating and power factor of the components in the network. We also have a range of conditions under which this network can operate. We want to refine these initial requirements by testing under a wide range of conditions. We will do this by tweaking the model to use abstract loads and sources, and testing the requirements over a wide range of operating conditions.

We will set up the model to use abstract load and generator models and see the power generated and consumed over a wide range of operating conditions.

Now we will move on to detailed design. We will use the refined requirements to refine the design, and we will match these detailed components to the system model to detect integration issues.

Today, there are several options available for generating electricity in an aircraft. Typically, the generator is driven by communication with the gas turbine. The turbine rotates at a variable frequency. If the network must have a fixed frequency, then a conversion from a variable speed of rotation of the turbine shaft to a constant frequency in the network is required. This can be done with an integral constant speed drive upstream of the generator, or by using power electronics to convert variable frequency AC to constant frequency AC. There are also floating frequency systems, where the network frequency can change and energy conversion takes place at the loads in the network.

Each of these options requires a generator, and power electronics to convert energy.

We have a gas turbine that rotates at a variable speed. This turbine is used to rotate the generator shaft, which produces variable frequency alternating current. Various power electronics options can be used to convert this variable frequency to a fixed frequency. We would like to evaluate these different options. It is possible to do this using SPS.

We can simulate each of these systems and build simulation results under different conditions to evaluate which option is best for our system. Let's switch to the model and see how it's done.

Here is the model we are working with. Variable speed from the gas turbine shaft is transmitted to the generator. And the cycloconverter is used to output fixed frequency alternating current. If you run the simulation, you will see how the model behaves. The top graph shows a variable speed gas turbine. You can see the frequency is changing. This yellow signal on the second graph is the voltage from one of the phases at the output of the generator. This fixed frequency alternating current is generated from variable speed using power electronics.

Let's take a look at how AC loads are described. A lamp, a hydraulic pump and an actuator are connected to ours. These components are modeled using blocks from the SPS.

Each of these blocks in SPS includes configuration settings to allow you to accommodate different component configurations and to fine-tune the level of detail in your model.

We set up the models to run a detailed version of each component. Thus, we have many opportunities to simulate loads in an AC network, and by simulating detailed components in discrete mode, we can see much more detail of what is happening in our electrical network.

One of the tasks that we will perform with the detailed version of the model is the analysis of the quality of electric power.

When a load is applied to the system, it can cause waveform distortion at the voltage source. This is an ideal sine wave, and such a signal will be at the output of the generator if the loads are constant. However, as the number of components that can be turned on and off increases, this waveform can become distorted and result in such small spikes.

These surges in the waveform at the voltage source can lead to problems. This can lead to overheating of the generator due to switching in the power electronics, this can create large neutral currents, and also cause unnecessary switching in the power electronics, as they don't expect this buzz in the signal.

The harmonic distortion factor offers a measure of the quality of AC electrical power. It is important to measure this ratio under varying network conditions because the quality will change depending on which component is turned on and off. This ratio is easy to measure with the MathWorks tools and can be automated for testing under a wide range of conditions.

Read more about the coefficient of non-linear distortion at Wikipedia.

Next, we will see how to power quality analysis using simulation.

We have a model of the electrical network of the aircraft. Due to various loads in the network, the voltage waveform at the output of the generator is distorted. This leads to a deterioration in the quality of nutrition. These loads are disconnected and connected to the network at various times during the flight cycle.

We want to evaluate the power quality in this network under various conditions. To do this, we will use SPS and MATLAB to automatically calculate the harmonic distortion factor. We can perform coefficient calculation interactively using a graphical interface or use a MATLAB script for automation.

Let's go back to the model to show you this with an example. Our aircraft electrical network model consists of a generator, an AC bus, AC loads, and a rectifier transformer and DC loads. We want to measure the power quality at different points in the network under different conditions. First, I'll show you how to do this interactively for the generator only. Then I'll show you how to automate this process using MATLAB. We will first run a simulation to collect the data required to calculate the harmonic distortion factor.

This graph (Gen1_Vab) shows the voltage between phases of the generator. As you can see, this is not a perfect sine wave. This means that the power quality of the network is affected by the components on the network. At the end of the simulation, we will use the Fast Fourier Transform to calculate the harmonic distortion factor. We will open the powergui block and open the FFT analysis tool. You can see that the tool is automatically loaded with the data that I recorded during the simulation. We will select the FFT window, specify the frequency and range, and display the results. You can see that the harmonic distortion factor is 2.8%. Here you see the contribution of various harmonics. You have seen how you can calculate the harmonic distortion factor interactively. But we would like to automate this process in order to calculate the coefficient under different conditions and at different points in the network.

Now we will look at the possibilities that are available for modeling loads in a DC circuit.

We can model purely electrical loads as well as multi-disciplinary loads that contain elements from different engineering fields, such as electrical and thermal effects, electrical, mechanical and hydraulic.

Our DC circuit includes rectifier transformer, lamps, heater, fuel pump and battery. Detailed models can take into account effects from other areas, for example, a heater model takes into account changes in the behavior of an electrical part with a change in temperature. The fuel pump takes into account effects from other areas to also see their effect on the behavior of the component. I will return to the model to show you what it looks like.

This is the model we are working with. As you can see, now the transformer-rectifier and the DC network are purely electrical, i.e. only effects from the electrical domain are taken into account. They have simplified electrical models of the components in this network. We can choose a variant of this system (TRU DC Loads -> Multidomain), which takes into account effects from other engineering areas. You can see that in the network we have the same components, but instead of the number of electrical models, we have added other effects - for example, for the tricky one, a temperature physical network that takes into account the influence of temperature on behavior. In the pump, we now account for hydraulic effects from pumps and other loads in the system.

The components you see in the model are assembled from Simscape library blocks. There are blocks for accounting for electrical, hydraulic, magnetic and other disciplines. Using these blocks, you can create models that we call multidisciplinary, i.e. taking into account spectacular from different physical and engineering disciplines.

Effects from other areas can be integrated into the electrical network model.

The block library in Simscape includes blocks for modeling effects from other areas, such as hydraulic or thermal. Using these components, you can create more realistic network loads and then more accurately define the conditions under which these components can operate.

By combining these elements, you can create more complex components, as well as create new custom disciplines or areas using the Simscape language.

More advanced components and parameterization settings are available in specialized Simscape extensions. More complex and detailed components are available in these libraries, taking into account effects such as efficiency losses and temperature effects. It is also possible to model XNUMXD and multibody systems using SimMechanics.

Now that we have completed the detailed design, we will use the results of the detailed simulations to tune the parameters of the abstract model. This will give us a model that runs fast while still producing results that match those of a detailed simulation.

We started the development process with abstract component models. Now that we have detailed models, we would like to make sure that these abstract models produce similar results.

The green color shows the initial requirements we received. We would like the results of the abstract model, shown here in blue, to be close to the simulation results of the detailed model, shown in red.

To do this, we will define active and reactive powers for an abstract model using an input signal. Instead of using separate values ​​for active and reactive power, we will create a parameterized model, and will tune these parameters so that the active and reactive power plots from the simulation results of the abstract model match the detailed model.

Next, we will see how the abstract model can be tuned to match the results of the detailed model.

Here is our task. We have an abstract model of a component in an electrical network. When we apply such a control signal to it, the output is such a result for active and reactive power.

When we apply the same signal to the input of the detailed model, we get the following results.

We need the simulation results of the abstract and detailed model to match so that we can use the abstract model to quickly iterate on the system model. To do this, we will automatically tune the parameters of the abstract model until the results match.

To do this, we will use SDO, which can automatically change parameters until the results of the abstract and detailed models match.

To configure these parameters, we will take the following steps.

• First, we import the simulation outputs of the detailed model and select these data for parameter estimation.
• We then specify which parameters to configure and set the parameter ranges.
• Next, we will evaluate the parameters, with SDO adjusting the parameters until the results match.
• Finally, we can use other inputs to validate the results of the parameter estimation.

You can greatly speed up the development process by distributing simulations using parallel computing.

You can run separate simulations on different cores of a multi-core processor or on compute clusters. If you are faced with a task that requires running many simulations - for example, Monte Carlo analysis, fitting parameters, or running multiple flight cycles, then you can distribute these simulations by running them on a local multi-core machine or computer cluster.

In many cases this will be no more difficult than replacing the for loop in the script with a parallel for loop, parfor. This can result in a significant speedup of the simulation suite.

We have a model of the electrical network of the aircraft. We would like to test this network under a wide range of operating conditions - including flight cycles, crashes and weather. We will use PCT to speed up these tests, MATLAB to tune the model for each test we want to run. We will then distribute the simulations across the different cores of my computer. We will see that parallel tests complete much faster than sequential ones.

Here are the steps we need to follow.

• First, we will create a pool of worker processes, or so-called MATLAB workers, using the parpool command.
• Next, we will generate sets of parameters for each test we want to run.
• We will run the simulations first in sequence, one after the other.
• And then compare that to running simulations in parallel.

According to the results, the total testing time in parallel mode is about 4 times less than in serial mode. On the graphs, we saw that, in general, the power consumption is at the expected level. Visible peaks refer to different network conditions when consumers are turned on and off.

The simulations included many tests, which we were able to run quickly by distributing the simulations across different cores of the computer. This allowed us to evaluate a really wide range of flight conditions.

Now that we have completed this part of the development process, we will see how we can automate the creation of documentation for each step, how we can automatically run tests and document the results.

System design is always an iterative process. We make a change to the design, test the change, evaluate the results, then make a new change. The process of documenting results and rationale for change takes a long time. It is possible to automate this process using SLRG.

Using SLRG, you can automate the execution of tests, and then collect the results of these tests in the form of a report. The report may include evaluation of test results, screenshots of models and graphs, C and MATLAB code.

In conclusion, I will recall the key points of this presentation.

• We saw many opportunities for model tuning to strike a balance between model fidelity and simulation speed - including simulation modes and levels of abstraction of models.
• We have seen how simulations can be accelerated using optimization algorithms and parallel computing.
• Finally, we saw how you can speed up the development process by automating simulation and analysis tasks in MATLAB.

Material author — Mikhail Peselnik, engineer CITM Exhibitor.

Link to this webinar https://exponenta.ru/events/razrabotka-ehlektroseti-samoleta-s-ispolzovaniem-mop

Source: habr.com