**4.2 Experimental results and discussion**

The proposed wavelet-based fault detection system tested using the ServoHeli-20 RUAV system in manual mode.

Fig. 9. ServoHeli-20 fault detection experiment

Application of Wavelets Transform in Rotorcraft UAV's Integrated Navigation System 625

As is shown in the figure 11, discontinuity point of signal is displayed obviously, it is allocated very accurately in time-domain, and fault point of bias signal is contained in signal abruption. Using the <sup>1</sup> | ( , )| *W f s t* criterion, the fault detection system can detect locations of the bias fault at 7000 that we can see the local maximum value of module indicates the

<sup>0</sup> <sup>2000</sup> <sup>4000</sup> <sup>6000</sup> <sup>8000</sup> <sup>10000</sup> <sup>12000</sup> <sup>14000</sup> -10

<sup>0</sup> <sup>2000</sup> <sup>4000</sup> <sup>6000</sup> <sup>8000</sup> <sup>10000</sup> <sup>12000</sup> <sup>14000</sup> -10

<sup>0</sup> <sup>2000</sup> <sup>4000</sup> <sup>6000</sup> <sup>8000</sup> <sup>10000</sup> <sup>12000</sup> <sup>14000</sup> -5

<sup>0</sup> <sup>2000</sup> <sup>4000</sup> <sup>6000</sup> <sup>8000</sup> <sup>10000</sup> <sup>12000</sup> <sup>14000</sup> -5

<sup>0</sup> <sup>2000</sup> <sup>4000</sup> <sup>6000</sup> <sup>8000</sup> <sup>10000</sup> <sup>12000</sup> <sup>14000</sup> -5

t/0.02s

We also made a spike failure injection to RUAV system in manual mode to test the performance of the wavelet-based fault detection system. At the point of 7000, the compass

Similar to the bias failure experiment, the location of fault agree with the maximum values

From the results, it can be conclude that the proposed method is effective for detection the abrupt faults of the RUAV sensor system. Fault point could be also being described accurately at some certain resolution. Local characteristics of wavelet are represented well in

With the development of microelectronics technology, low cost MEMS gyroscopes begin to be used widely. It makes the great development of integrated navigation system, especially in UAV system. Compared with high costs gyroscope, the MEMS gyroscope devices have

original signal

signal singularity accurately.

0 10

0 10 scale-S3

> 0 5 scale-D3

> 0 5 scale-D2

> 0 5 scale-D1

**4.2.2 Failure of spike** 

time and frequency-domains.

Fig. 11. Spike failure and its wavelet transform(t=0.02s)

roll channel gets spike which the signal return to zero.

of the wavelet transform on different scales.

**5. Wavelet for gyroscope de-noising 5.1 MEMS gyroscope signal analysis** 

Amplitude (deg)

The use of the mathematical model makes it easier to test the wavelet-based fault detection system, but the characteristic of the datasets may not reflect the real flight environment and the actual actuator failures. On the other hand, real autonomous flight experiments with an injected sensor failure can be potentially dangerous for the helicopter because it can take the RUAV out of control and RUAV may crash. Thus, we planned to inject the sensor failure while the absence of the security problems of the RUAV with its manual mode. As is shown in the figure 9, the pilot controls the helicopter using radio controller. The onboard computer online detects the fault with wavelet-based algorithm (Qi & Han, 2007).

To demonstrate the effectiveness of the fault detection scheme, the failure scenario of abrupt bias and spike in compass roll channel is assumed.

A "db2" ("db" is define in Matlab) wavelet with a vanishing moment 2 is applied to these abrupt faults of sensor. Figure 10 and 11 show their wavelet transforms in scale-D1 to scale-S3 including the original data signals. In figure 10, scale-D1 to scale-D3 denote the details of the wavelet transform of the signals on scales 1 to 3, respectively, while the scale-S3 represents the approximation of them on scale 3.

Fig. 10. Bias failure and its wavelet transform(t=0.02s)

#### **4.2.1 Failure of bias**

In figure 10, an example sensor failure experiment is presented. At the point of 7000, the compass roll channel gets bias of 5 degree.

The local maxima of the first derivative are sharp variation points of [ ( )] ( ) *<sup>s</sup> f xt t* . For abrupt failure detection, we are only interested in the local maxima of <sup>1</sup> | ( , )| *W f s t* . When detecting the local maxima of <sup>1</sup> | ( , )| *W f s t* , we call also keep the value of the wavelet transform at the corresponding location.

As is shown in the figure 11, discontinuity point of signal is displayed obviously, it is allocated very accurately in time-domain, and fault point of bias signal is contained in signal abruption. Using the <sup>1</sup> | ( , )| *W f s t* criterion, the fault detection system can detect locations of the bias fault at 7000 that we can see the local maximum value of module indicates the signal singularity accurately.

Fig. 11. Spike failure and its wavelet transform(t=0.02s)
