Sensor Fault Diagnosis in Autonomous Underwater Vehicle Based on Fuzzy Logic

A0055 Anton M. Pisarets Far Eastern State Technical University

A0056 Alexey N. Zhirabok Far Eastern State Technical University

The problem of ensuring safety and reliability operation of AUV is

very important. Since malfunctions and faults occurring in the AUV

systems can lead to erroneous mission fulfilment or losses of a

vehicle, it is necessary to give much attention to solution this


Navigation-piloting sensors such as the meters of trim and course, the

meters of angular velocities, the meters of velocity and depth are

important components of the AUV system from which its reliable

operation depends on. Thus the task of early fault detection and

isolation in this sensors are important.

Among different methods of dynamic systems diagnosis using analytical

redundancy model-based methods well established in practice. To

describe dynamic behaviour of the AUV, differential equations are used.

To provide diagnosis process, observer-based methods are used. The i-

th observer is driven by the AUV control signals and sensors output

signals but the i-th one. The difference between the i-th sensor

measurement and corresponding observer output signal is a residual

which is equal to almost null in the normal AUV operating and not null

when a fault occurs. This residual is sensitive to faults of those

sensors which measurements use for its obtaining. Diagnosis based on

analysis of all residuals values is performed.

In the simplest case, for fault detection and isolation, the residual

value is compared with certain threshold. If it is more than this

threshold, it is concluded that the fault occurs in some sensor. Use

of the threshold allows one to find sudden faults (i.e. step-like

changes in the sensors), but this method has small sensitivity to

incipient (slowly developing)faults which is a characteristic property

of the majority of sensors.

To increase the efficiency of the diagnosis process, it is offered to

use the fuzzy logic. In this case, the residual is evaluated as

"small ", "medium ", "big ", etc which reflects a human notion about

residual size. Mathematically, this evaluation is performed using so-

called membership functions. The correspondences of the residual

values to these functions are assigned by the membership grades which

are numbers from interval [0, 1]. These membership grades are used

for faulty sensor isolation. This diagnosis process allows one to

evaluate technical state of the AUV sensors more precise in comparison

with the threshold test.