An Aircraft Navigation System Fault Diagnosis Method Based on Optimized Neural Network Algorithm

Jean-dedieu+Weyepe

【Abstract】 Air data and inertial reference system （ADIRS） is one of the complex sub-system in the aircraft navigation system and it plays an important role into the flight safety of the aircraft. This paper propose an optimize neural network algorithm which is a combination of neural network and ant colony algorithm to improve efficiency of maintenance engineer job task.

【Key words】 ADIRS Intelligent faults diagnosis neural network ant colony

1 Introduction

Airline Company nowadays needs more sophisticated aircraft maintenance system to provide their fleet safety and comfort for the passenger while increasing their profitability.

In recent years， artificial intelligence， especially in intelligent fault diagnosis system research， has made remarkable achievements. Artificial intelligence changed the traditional fault-based numerical fundamentally， deffect diagnosis method can only be applied to a specific device， the already eestablished diagnostic systems， slight modifications can be applied to other devices troubleshooting.

2 Case Study

With a complex system such as the above description of airspeed related analysis，it is still difficult to provide an accurate diagnosis result.Airbus A320 maintenance as example，uses the TSM（Troubleshooting Manual）to trouble shoot faults meaning，when an anomaly is detected on the airplane，technicians or maintenance engineer refer to this manual to take appropriate action.But in most cases，from a fault effect there are many possible causes given from the manual，therefore the engineer needs to blindly go through isolation method describe for this given fault information.Fig.6 illustrates the a case where the TSM might not figure out the main cause of fault.

3 Neural Network and ant Colony Optimization

3.1 Neural Network

A neural network is used for its learning non-linear problem solving ability.

This section describes several common learning rules.

A.Hebb learning rule：in 1949，U.S.scientists first proposed a theory of Hebb prominent correction hypothesis， it is believed that when the two are simultaneously in the inhibition of neurons are connected together，the connection strength between them should be greatly reduced.And proposed neural network learning and training signal is equal to the output neurons of the actual results：

（1）

With representing the transfer function，Oj been the neuron for the output ，Xi been the neuron for the input i， been the synaptic weights assigned i to the neurons to the neuron ， been the learning rate， is the goal of the ideal output value .endprint

Adjusting the formula to define the weights to assigned to：

（2）

Equation shows that there is a relationship proportional to the amount of weight assignments to adjust the input and output of the product.

3.2 Ant colony algorithm

Ant cllony algorithm is based on natural ant colony behavior.They have an ability to find the shortest path to reach their food and continuously search for new possibilities.

Implementing this with neural network will help solve the local minimum problem through the global search ability of the ant colony algorithm.This paper will not go in deep description of ant colony algorithm.Here below is the structure of the complete combination.

4 Results and Conclusion

Using the TSM information and some other information coming from theory analysis， the table below could be created.

An AirbusTSM task taken as test data is given：{Task Number 34 13 00 T 810 998： Airspeed Discrepancy between CAPT PFD and F/O PFD.Possible causes：static probe，AOA sensor，AOA sensor 3，pitot probe，ADM}

Possible causes are outputs and the result gives Pitot probe=0.623；AOA sensor=0.142；ADM=0.011.

The result been heuristic，that means pitot probe have the higher possibility to be faulty，this way an un-experienced engineer can have a chance to detect the problem at its first isolation procedure to get the job done.

Reference：

[1]X.S.Wen，Y.C.Xu，X.S.Yi，G.1.Liu and 1.L.Xu，"Research on the concept and connotation of intelligent built-in test，"Computer Engineering and Applications，vol.14，pp.29-32，2001.

[2]Dorigo， M.，Gambardella， L.M.：Ant colonies for the travelling salesman problem. Biosystems 43（2），73-81（July1997）.

[3]Jovanovic， R.，Tuba，M.：An ant colony optimization algorithm with improved pheromone correction strategy for the minimum weight vertex cover problem. Applied Soft Computing 11（8），5360–5366 （December 2011）.endprint

Adjusting the formula to define the weights to assigned to：

（2）

Equation shows that there is a relationship proportional to the amount of weight assignments to adjust the input and output of the product.

3.2 Ant colony algorithm

Ant cllony algorithm is based on natural ant colony behavior.They have an ability to find the shortest path to reach their food and continuously search for new possibilities.

Implementing this with neural network will help solve the local minimum problem through the global search ability of the ant colony algorithm.This paper will not go in deep description of ant colony algorithm.Here below is the structure of the complete combination.

4 Results and Conclusion

Using the TSM information and some other information coming from theory analysis， the table below could be created.

An AirbusTSM task taken as test data is given：{Task Number 34 13 00 T 810 998： Airspeed Discrepancy between CAPT PFD and F/O PFD.Possible causes：static probe，AOA sensor，AOA sensor 3，pitot probe，ADM}

Possible causes are outputs and the result gives Pitot probe=0.623；AOA sensor=0.142；ADM=0.011.

The result been heuristic，that means pitot probe have the higher possibility to be faulty，this way an un-experienced engineer can have a chance to detect the problem at its first isolation procedure to get the job done.

Reference：

[1]X.S.Wen，Y.C.Xu，X.S.Yi，G.1.Liu and 1.L.Xu，"Research on the concept and connotation of intelligent built-in test，"Computer Engineering and Applications，vol.14，pp.29-32，2001.

[2]Dorigo， M.，Gambardella， L.M.：Ant colonies for the travelling salesman problem. Biosystems 43（2），73-81（July1997）.

[3]Jovanovic， R.，Tuba，M.：An ant colony optimization algorithm with improved pheromone correction strategy for the minimum weight vertex cover problem. Applied Soft Computing 11（8），5360–5366 （December 2011）.endprint

Adjusting the formula to define the weights to assigned to：

（2）

Equation shows that there is a relationship proportional to the amount of weight assignments to adjust the input and output of the product.

3.2 Ant colony algorithm

Ant cllony algorithm is based on natural ant colony behavior.They have an ability to find the shortest path to reach their food and continuously search for new possibilities.

Implementing this with neural network will help solve the local minimum problem through the global search ability of the ant colony algorithm.This paper will not go in deep description of ant colony algorithm.Here below is the structure of the complete combination.

4 Results and Conclusion

Using the TSM information and some other information coming from theory analysis， the table below could be created.

An AirbusTSM task taken as test data is given：{Task Number 34 13 00 T 810 998： Airspeed Discrepancy between CAPT PFD and F/O PFD.Possible causes：static probe，AOA sensor，AOA sensor 3，pitot probe，ADM}

Possible causes are outputs and the result gives Pitot probe=0.623；AOA sensor=0.142；ADM=0.011.

The result been heuristic，that means pitot probe have the higher possibility to be faulty，this way an un-experienced engineer can have a chance to detect the problem at its first isolation procedure to get the job done.

Reference：

[1]X.S.Wen，Y.C.Xu，X.S.Yi，G.1.Liu and 1.L.Xu，"Research on the concept and connotation of intelligent built-in test，"Computer Engineering and Applications，vol.14，pp.29-32，2001.

[2]Dorigo， M.，Gambardella， L.M.：Ant colonies for the travelling salesman problem. Biosystems 43（2），73-81（July1997）.

[3]Jovanovic， R.，Tuba，M.：An ant colony optimization algorithm with improved pheromone correction strategy for the minimum weight vertex cover problem. Applied Soft Computing 11（8），5360–5366 （December 2011）.endprint