Вы находитесь на странице: 1из 5

Sensor Reliability Evaluation Method

In Multi-Sensor Control Systems

Paweł Łęczycki, Artur Andrzejczak, Piotr Pietrzak, Andrzej Napieralski


Department of Microelectronics and Computer Science
Lodz University of Technology
Lodz, Poland
pleczycki@dmcs.pl

Abstract— This paper describes the reliability evaluation


method for reliability evaluation of a sensor in multi-sensor II. TULCOEMPA
control systems, where communication between modules is based
on a binary information. At the beginning, main objectives of the A. Project outline
TULCOEMPA is are briefly presented. One of the main tasks of In the context of the Swiss-Polish cooperation in research
this project is the Truck Recognition System that uses various and development of the innovative methods for monitoring in
types of sensors, which are working for the same purpose – the civil engineering infrastructure, the Lodz University of
recognition of vehicles with total mass over 3.5 ton without Technology (TUL) and Empa1 realize carry out
weighting system. In the following, authors present algorithms on
multidisciplinary project entitled TULCOEMPA. The basis of
which mentioned method is based on and results with description
the project is the pioneer worldwide usage of the innovative
of simulations performed to preliminary test the method.
strengthening system of the selected bridge (located in
Keywords—TULCOEMPA; truck recognition; sensor; trust Szczercowska Wies) using unanchored prestressed carbon
coefficient; structural health monitoring; measurement system; fiber reinforced polymer (CFRP) laminates. Reliability of this
new technique will be confirmed by long-term monitoring of
the facility and any additional environmental influences
I. Introduction (temperature and humidity) as well as static and dynamic
effects on the object. Monitoring will be realized with using
During a research, Iin many cases, during trials, the there utilization of the multi-point measuring measurement system
is need to estimate the state or behavior of an object composed of the networks of different sensors network,
influenced by depends on many factors. Investigation of the transducers, nodal points, a base station to with the GSM
relationships between them requires model development or if wireless communications modem and computer system for
possible making experiment. If some of those factors are collecting, processing and data coding. In addition, no
known, it is possible to tell On this basis it is possible to existence lack of proper infrastructure in mentioned
determine what is happening with the object at the moment. A localization enforces usage of batteries and solar and wind
wide range of methods and tools used to analyze information energy harvesting from solar cells harvesters. As a
from multiple sources during experiment lets(?) us also predict consequence, particular attention is paid to the energy
what will happen with the object in the future. There are more efficiency of the system [1].
and more solutions, which are using involve more than one
kind of sensor to analyze the same phenomenon. These types The Mmultitude of the measurement data, which will be
of systems are used inter alia among others to collect aquire collected constantly by the measurement system, requires a
most accurate data or to evaluate which of the available selection. To achieve that, cooperation with a specialized
sources of information are most trustworthy realiable under system for which recognize recognition of weight, speed and
given conditions and circumstances. For example, a gyroscope number of vehicles crossing the bridge is needed. In terms of
and an accelerometer can be used to determine in what project, the Truck Recognition System will cause the
position, horizontal or vertical, is an object. But to increase limitation of irrelevant information. It will be realized by
precision it is needed to use data from both of sensors. Similar triggering the measurement system only in the situation when
problem occurs in a TULCOEMPA project, where the investigated bridge is loaded by vehicle heavy enough to
independent modules of the Truck Recognition System will be affect the research object its structure will appear. This
identifying appearance of a specific event using different solution will also significantly reduce an energy consumption
methods. Based on the information generated from by system by switching particular elements of the measurement system
modules, equipped with various types of sensors, a wireless
bridge measurement monitoring system will be triggered. 1
Eidgenössische Materialprüfungs- und Forschungsanstalt
(The Swiss Federal Laboratories for Materials Science and
Technology)
to standby mode and by cutting back transmission power of vote for not starting the measurement by leaving for certain
putting into standby mode communication modules. period of time input pin state on at low logic level.
It must be taken into account that accuracy and reliability
B. Truck Recognition System (including stability) of Identification Modules can be unstable
One of the main tasks in the TULCOEMPA project is the during a year. Changes of climatic conditions and lighting
development of the Truck Recognition System (TRS). conditions can affect sensors too. Damaged sensor must be
Vehicles approaching towards the monitored object will be excluded from voting. As it was mentioned before, there will
monitored to recognize the ones with total mass over 3.5 ton. be no communication between Modules of Truck Recognition
After detection such vehicle, the TRS will trigger the System and there will not be any confirmation that sensors
structural health monitoring system installed on the bridge [2] work properly (for example the Identification Module with
[3]. sound sensor can identify an agricultural machine as a truck or
There are various ways enabling verification of the vehicle camera can have reduced field of view in foggy evening). To
type. One of the most credible is weighting. But this solution ensure triggering of the measurement system only when
is not cheap. Especially when it comes to installation of specified event occurs, influence reduction of the sensors
weighting system, which allows for weight measurement of mistakes and fortuitous events has to be reduced. is needed.
moving objects. In addition, it is not always acceptable or Solution for this issue appears to be a trust coefficient, which
possible to install this type of the system (for example in will be assigned to every vote as weight. Literature review did
short-term monitoring in center of a city). These are the not reveal existence or usage of any method for dynamic
reasons why it was decided to conduct studies to find the most appraisal of trust coefficient, which in addition do not require
relevant alternative solution. Amidst available to apply learning process, for particular system elements [4][5][6].
Among possible methods, studies focused on the analysis of: Based on performed analysis and research work a new,
universal method for trust evaluation of information from
 magnetic field, sensors was developed. .Its task is to estimate a level of
effectiveness and assign proper weight to it. I this case
 sound, described majority voting system takes form of weighted
 vibrations, voting system.

 camera image.
were chosen to be analyzed and implemented.
III. Sensor Reliability Evaluation
Identification Modules of the Truck Recognition System Method
will be working independently. Each of them will be equipped
with a sensor of one of above-mentioned types and will A. Trust coefficient as a source of information
generate a signal for a management unit only when a suitable Trust is a belief that properties and functionality of an
vehicle will be is recognized. It means that there will be only object are in accordance to our assumptions and knowledge. In
one-way communication. statistics, level of trust is defined as a trust coefficient and is a
real number in the interval of 0 to 1 (0% - 100%). The trust
coefficient indicates probability that occurrence of an event is
truthful inevitable. The trust coefficient indicates probability
that occurrence of an event will be truthful. Accordingly,
lower trust coefficient means lower level of reliability. Trust
analysis methods review leads to the conclusion that objects
are evaluated in context of affiliation to specified classes. For
experts, the trust coefficient is only an indicator of possibility
to occurrence of an irregularity or some distortions [5][7].
In the Truck Recognition System there is no place nor time for
to analyzeing data by a specialist. Decision about triggering
the structural health monitoring system must be taken
Fig. 1. Block diagram of the Truck Recognition System with an energy automatically and almost instantly. According to this, a
harvesting system. decision algorithm implemented in the management unit is
based on weighted voting system. This is a type of majority
The management unit will be an Open Multimedia voting system where not all voters are equal. In the decision
Application Platform (OMAP). Prepared software will be algorithm, weight of a vote will be same as a trust coefficient.
scanning general input-output ports and waiting for signals
from Identification Modules. Decision about starting a the
measurement will be made based on a voting system. Modules
with sensors, after event detection, will inform about this fact
by setting its input pin state to high. It will be interpreted as a
vote for starting the measurement. Rest of Modules will take a
Fig. 2. A simplified block diagram illustrating the flow of information from
sensors to the measurement system

B. The method desctiption


Sensor reliability evaluation method consists of two parts,
directly correlated to each other:, parts:
 decision algorithm - where votes from Identification
Modules are analyzed;
 trust coefficient evaluation algorithm - where votes Fig. 3. Illustration of reliability evaluation process for a single sensor.
weights are estimated based on last votes and last
results of the decision module.
C. Preliminary tests
The trust coefficient evaluation algorithm depends on For this moment, the Identification Modules of the Truck
rating: positive (if the decision algorithm result is same as the Recognition System, the application platform and the
Identification Module vote) or negative (if the algorithm result structural health measurement monitoring system are in the
is opposite to the Identification Module vote). At the implementation phase. To perform tests of the sensor
beginning, by default, all sensors (in case of TULCEOMPA reliability evaluation method, a simulation was proposed. to It
project - Identification Modules) have the same value of the verify verifies effectiveness and assumptions of this described
trust coefficient, so the voting is majority. If all but one of method and its effectiveness. Basing on the results and
Identification Modules do react to approaching truck, the one experience, adjustable parameters and some properties
receive negative rating. It leads to drop its trust coefficient necessary for proper working of the method were adapted to
and, in as a consequence, to lower meaning of votes in future was determined for the TULCOEMPA project requirements.
votings. The same situation is if the Module will report false- Simulations led also for to preliminary assessment of the
positive vote, but it will be outvoted. Weight of a vote is a presented method.
percentage of positive ratings in all ratings. And soIn this way,
after some time, majority voting takes form of weighted In the preliminary tests of the sensor reliability evaluation
voting. method Identification Modules with 4 different sensor types
was were simulated. The simulation assumed a one year of a
work with about 10 arriving vehicles, with total mass over 3.5
tons, per day. Each of the sensors had assigned a percentage
chance for of proper identification of the event. This parameter
for first two sensors has been set to 95% and 75% as a
constant value. For other sensors it was changing over the
year. Third sensor was able to properly report an event in 95%
of cases, but after some time this value falls to 60% and then
backs to 95%. For the last sensor, the chance for proper
identification has been changing between 30% and 80%.This
value is represented on charts by red curve. Blue one
represents the instantaneous value of the trust coefficient
estimated by the method. The aim of this simulation was to
check that estimated value of trust coefficient is approximated
to assigned percentage chance for proper identification of the
event.
The method has few editable parameters that can effect on appropriate message for an expert or a conservator of a
influence the accuracy and the rate of changes of the trust system. Lengthened reaction time is caused by, described
coefficient. First of them is a number of historical ratings of a before, parameter that determines number of a sensor
sensor. In presented case (fig. 4.) value of this parameter was historical ratings.
set to 200. It means that the trust coefficient was estimated
based on last 200 rates of a sensor. Lower values (from range
50-150) caused rugged curves and peaks on charts. On the IV. Summary
other hand, values higher than 250 make charts flat and This paper describes the method for sensor reliability
lengthen reaction time for a sensor capabilities. Next editable evaluation method of a sensor working in multi-sensor control
parameter is a dominance factor and it can be a real number systems, where communication between modules is based on a
from range -1 to 1. The domination factor increases (when binary information. Presented results of simulations show high
value is positive) or decreases (when value is negative) the effectiveness of the method. Works on tThe presented method
chances for to obtain positive result in the decision algorithm are is still developed. in progress and their The aim of the
in the case when the voting is near of nearly draw. On conducted work is mainly minimization of the delay
presented simulations, domination factor was set to 0.1. Low, introduced to the system answer. Only after development of a
positive value was not interfering significantly on voting result final form of the Truck Recognition System and installation it
and an extra run of measurement system has no negative on the target object, it will be possible The only possibility to
consequences in uncertainty situation. Simulations show that, fully test the sensor reliability evaluation method is to
with help of the sensor reliability evaluation method, about implement it in the real system. It will be possible after
96.7% of events were detected correctly. installation of the TRS. Reason for that is that there are some
The presented method can be used for identification of situations that are hard to predict or simulate are to be
sensors with some malfunctions. When sensor reliability fall expected. On Near to the target object will be installed weight
to 0 or other declared value, alarm signal can be triggered. weighing scale which will help to evaluate effectiveness of the
Figure 5. shows simulation of exclusion of a sensor from method. Data collected from final version of the Truck
voting process after 18 days from occurrence of damage. Recognition System will be analyzed and the results will be
When it happens in real system, it is possible to generate presented in the further publications.

Fig. 4. Result of a simulation for testing the Sensor Reliability Evaluation Method. The simulation assumed a one year of a work with about 10 arriving vehicles,
with total mass over 3.5 ton, per day. Percentage chance for proper identification of the event is marked as red and trust coefficient estimated by the method - by
blue.
Fig. 5. Simulation of damaging one of sensors. After 18 days follows exclusion of a sensor from voting process.

[4] H. F. Martz, B. S. Duran, “A Comparison of Three Methods for


References Calculating Lower Confidence Limits on System Reliability Using
[1] TULCOEMPA – project supported by a grant from Switzerland through Binomial Component Data”, IEEE TRANSACTIONS ON
the Swiss Contribution to the enlarged European Union; RELIABILITY, vol. R-34, NO. 2; June 1985
www.tulcoempa.com [10.02.2014] [5] B. Janeczek, “Analiza wiarygodności danych z wykorzystaniem maszyn
[2] A. Andrzejczak, P. Łęczycki, M. Makowski, B. Pękosławski, P. Pietrzak, wektorów podpierających SVM”, Warsaw University of Technology,
A. Napieralski, “Wireless Communication Solution for Distributed The Faculty of Electronics and Information Technology, The Institute of
Structural Health Monitoring”, INTL JOURNAL OF ELECTRONICS Computer Science, 2008
AND TELECOMMUNICATIONS, 2013, VOL. 59, NO. 2, PP. 177–184 [6] Z. Pawlak, “Rough Sets. Theoretical Aspects of Reasoning about Data”,
[3] A. Andrzejczak, P. Pietrzak, M. Makowski, A. Napieralski, “Wireless Wydawnictwo Politechniki Warszawskiej, Warsaw 1990
Communication Solutions for Distributed Strain Measure Systems in [7] I. Białynicki-Birula, I. Białynicka-Birula, “Modelowanie
Mechanical Structures”, MIXDES 2012, 19th International Conference rzeczywistości”, Wydawnictwo Naukowo-Techniczne, Warsaw 2007
"Mixed Design of Integrated Circuits and Systems", May 24-26, Warsaw
2012

Вам также может понравиться