A Case Study of using MATLAB, Simulink and Stateflow for Model-Based Design of a Fuel Management System for the Airbus A380
The Airbus A380 is one of the largest commercial aircraft these days. Its flight range is more than 12,000 kilometers. Such long distances are enabled because A380 has 11 fuel tanks which all together contain 320, 000 liters of fuel.
The problem which engineers from Airbus had to solve was to create a sophisticated fuel management system which manages fuel flows to engines and between tanks during the flight. The system was required to be able to move fuel between tanks to optimize the aircraft’s center of gravity, reduce wing bending, and keep fuel within the acceptable temperature range.
Challenge
Two main problems that engineers of Airbus had to overcome are:- the A380’s fuel management system must be able to safely handle any failure in the system, but because of the complexity of the system: 21 pumps, 43 valves, and other mechanical components, it’s challenging for engineers at the requirements stage to predict problems that can result from combinations of relatively minor failures;
- the complexity of the system resulted in ambiguous and difficult to understand specification document which even had a few conflicts in its requirement list.
Background
Engineering problems that engineers must solve nowadays are very challenging and time-consuming, they require a lot of man-hours. Computer programming became an essential part of today’s engineering process: cars use computers to measure and manage fuel supply, airplanes use computers for proper navigation, spacecraft use computers for trajectory computation, houses use computer-controlled video cameras for security purposes, etc. Moreover, scientists use computer simulations in order to analyze physical phenomena which is very expensive or impossible to analyze by performing experiments, aerodynamics, for instance, is one of these phenomena. A sequence of actions that usually was performed for solving such problems can be seen in (Figure 1).Unfortunately, it is very unlikely to create a software using a low-level programming language (like C/C++/Java) from scratch to research a particular engineering problem and expect good results from the very beginning. In order for the software to give valuable information, it will take researchers a lot of time to debug it and create necessary algorithms for
Figure 1: Old Sequence of Actions
a particular problem, which is not trivial and extremely time-consuming. The new way of attacking such problems can be seen in (Figure 2). Using a high-level programming language for
Figure 2: New Sequence of Actions
the research phase drastically reduces the time consumption. Moreover, because high-level languages are much easier, scientists can use them without the help of professional programmers and, therefore, the use of high-level languages reduces the number of people that have to be hired to solve the problem.Alternatives
There are not that many software packages for Model-Based Design, and MathWorks package which consists of Matlab, Simulink and Stateflow is the most developed one in the world. So the main question, in this case, is either to create a private closed software or to use a 3rd party applications (Matlab, Simulink, Stateflow) which will reduce the time of the development, but the presence of 3rd party applications eliminates the ability to make the software closed.The decision-making must be based on how much profit each decision would possibly make. Hence, in order to understand which solution is more preferable, we need to evaluate all pros and cons of each solution.
The process of “creating software” which can be seen in (Figure 1), (Figure 2). Because of the complexity of low-level programming languages, the “old” process of creating software, which is illustrated in (Figure 1), takes usually from 2 to 4 times more man-hours to complete comparing to the “new” one in (Figure 2).
The main criteria that we take into consideration are salaries for programmers and researchers, and the price for the MathWorks package which was used. According to the “glassdoor.com”, the average salary of programmers who use low-level languages is lower than the salary of those who use high-level languages:
Solution
Airbus used Model-Based Design to model the A380’s fuel management system, validate requirements through simulation, and clearly communicate the functional specification.
Airbus engineers used Simulink and Stateflow to model the system’s control logic, which comprises 45 top-level charts, almost 6000 states, and more than 8700 transitions.
- Python, Matlab: about $90,000-$110,000;
- Java, C++, C#: about $80,000-$90,000.
- Matlab: $2,150;
- Simulink: $3,250;
- Simscape: $2,150;
- Additional tools: $2,150 each;
- Stateflow: $3,000.
- “Old scheme”: 9 months of using low-level languages to perform a research and create the final program (a group of 20 people: $1,275,000). The overall cost is about $1,275,000;
- “New scheme”: 4 months of working with high-level languages and 3rd parties programming tools (a group of 10 people: $333,333), 2 months of using low-level languages to create the final program if necessary (a group of 10 people: $141,666) and the cost of MathWorks tools ($12,700). The overall cost is about $488,000.
Solution
Airbus used Model-Based Design to model the A380’s fuel management system, validate requirements through simulation, and clearly communicate the functional specification.
Airbus engineers used Simulink and Stateflow to model the system’s control logic, which comprises 45 top-level charts, almost 6000 states, and more than 8700 transitions.
This model defines modes of operation on the ground (including refuel, defuel, and ground transfer) and in flight (including normal engine feed, center of gravity control, load alleviation, and fuel jettison).
Using Parallel Computing ToolboxTM and MATLAB Distributed Computing ServerTM, the team performed Monte Carlo simulations on a 50-worker computing cluster. Over a weekend, they could run 100,000 simulated flights under varied environmental conditions and aircraft operational scenarios.
The team created a desktop simulator by generating code from the plant and control logic models with Simulink CoderTM. A MATLAB® based user interface enables suppliers, airline customers, maintenance engineers, and other Airbus teams to visualize how the fuel management system works and interacts with other aircraft systems.
After successful flight tests of the A380, the team used System Identification ToolboxTM to tune their plant model using measured flight test data. They used Signal Processing ToolboxTM to remove noise from the test data, and Curve Fitting ToolboxTM to evaluate differences between the measured data and predicted results and to predict system performance beyond the usual flight envelopes. While refining the plant model, they used Simscape Power SystemsTM to incorporate relays and other elements of the electrical power system.
Results
- Months of development time eliminated. “On earlier projects, it took up to nine months to integrate our fuel system design with the simulated cockpit, or iron bird. Using Model-Based Design on the A380, it took less than a month,” says Slack, computational analysis expert in fuel systems at Airbus. “Similarly, by reusing the model to commission the HIL rig, we saved three months of development and shortened the time from initial concept to first flight.” “With Model-Based Design, the model we used to represent the functional specification enabled us to validate requirements months earlier than was previously possible,” says Christopher Slack.
- Models reused throughout development. “The Simulink and Stateflow models enabled us to validate requirements early and communicate the functional specification to our suppliers, complementing the written requirements in conformance with ARP 4754,” says Slack. “These models were reused to create desktop simulators, commission our HIL test rig, run on our virtual integration bench, and demonstrate system functionality to customers.”
- Additional complexity handled without staff increases. “The fuel system of the A380 is three to four times more complex than that of the A340,” notes Slack. “Model-Based Design enabled us to handle a substantially more complex project with the same size engineering team.”