CC BY
36
y^K 51-3
DOI: 10.17277/vestnik.2015.03.pp.393-398
ADAPTIVE OPERATION ALGORITHM OF AN INTELLIGENT HUMIDITY SENSOR
S. V. Artemova, A. A. Artemov, P. A. Podkhvatilin, I. Yu. Koshelev
Department "Design of Electronic and Microprocessor-Based Systems", TSTU;
sartemova@yandex. ru
Keywords: adaptive algorithm; intelligent humidity sensor; neural network; sensor failure.
Abstract: The article describes an adaptive operation algorithm of an intelligent humidity sensor that allows determining a moisture content of a paste material with a reasonable accuracy in case of failure of device components.
Intelligent sensors are traditionally improved by developing their designs and elements, as well as by using methods for the description and subsequent analysis of measuring procedures [1]. An analytical approach is commonly used in the synthesis of intelligent sensors that contributes to the development of optimal data processing algorithms based on increasingly complex stochastic models of an impact on a sensor. An alternative to this strategy is a method of smart sensors implementation using neural networks. Functioning of intelligent sensors is provided by using actual data, which characterize a physical object of research.
It is necessary to perform the following steps for creation of algorithms:
- determination of operation modes of intelligent measuring system in a variety of situations, including an extreme ones;
- training of a smart sensor considering peculiarities of a process of research;
- formation of logic of diagnostics system functioning in case of failure of some sensor.
The basis of intelligent humidity sensor (IHS) hardware is a MSP430 microcontroller from Texas Instruments. This device uses SH15 humidity and temperature sensors from Sensirion, external ones to measure the initial temperature and humidity of drying agent and internal ones to measure the current temperature and humidity in the sections. There is also a conveyor speed sensor, a general exhaust fan switching-on sensor, common to the entire dryer, as well as individual sensors of controlled cameras: degree of opening of release gates and air intake windows. The required information is stored in non-volatile DataFlash memory device.
One of important properties of the monitoring system is a self-diagnostics feature. The ability to self-control characterizes a property of sensor to retain the characteristics within acceptable limits for a consumer in the event of sensor or information channel failure.
Evaluation of pasty material humidity is carried out using a neural network with multilayer perceptron architecture, trained by back propagation algorithm.
Network inputs are normalized values: x1 is initial humidity of material, x2, x3 are ambient temperature and humidity, x4 is speed of a plate conveyor, x5 is switching on or off of agitator, x6 is percent of time when exhaust fan works, x7, x8 are average weighted temperature and humidity of drying agent in the i-th camera; x9, x10 are percentage of opening of air intake window and release gate. Training of a neural network is organized by exemplary measures of technological parameters in the adaptive range of the i-th camera with fixed precision.
Data normalization is performed using the following formula:
xn _ x K n + K n
q = 1,®, (1)
where q is serial number of network input; x^, xq are normalized and non-normalized
values of q-th network input; Kmp, K^Jf are amplitude and offset scaling factors;
® is number of network inputs used to calculate humidity <pk in i-th camera [2].
Analytical model of multilayer perceptron for evaluating moisture content in i-th camera, trained by algorithm that mentioned above, is represented as:
9 =
1 + exp
Pout
Z wj out) (l + exp (-ß(i)
fßf
w|i). x1 + w^j x2 +w3,j x3 + w^j x4 +
, (i) — , (i) — , (i) — , (i) — , (i) — , (i) -'
+w5, j x5 + w6fj x6 + w7, j x7 + w8, j x8 + w9, j x9 + w10,y x10
^(i) 3 j
))-1
^(i) out
-1
(2)
where for sigmoidal activation functions of neurons are denoted by B(/\ B(y"+1)
angles for hidden layer and B!
( n) out
ß^ - for output layer of neural network, öj^,
0(,"+1) - shifts for hidden layer and eon), 0onu++1) - for output layer of network; w(j
wg+1> and wjn out),
w
(n+1 out)
j
- weighting relation factors for hidden and output
network layers, Ni - number of neurons in hidden network layer when calculating moisture content in i-th camera of dryer; normalized values of material humidity data
, initial humidity of material x1, ambient temperature and humidity x2, x3; speed
of a plate conveyor x4, switching on or off of agitator x5, percent of time when
exhaust fan works x6, temperature and humidity of drying agent x7
percentage of
opening of air intake window and release gate x9, x10 [2].
Output of network is denormalized by the formula for obtaining data about material humidity in real-world units:
=
9i - Kodfn
k:
dn amp
(3)
where Kadmnp, Kodfn are amplitude and offset denormalizing factors [3].
The resulting analytical dependence (2) allows evaluating material humidity in i-th camera of dryer in real time during its drying. Similar models are made for other cameras.
Let us extend the set of humidity determination models to eliminate the significant impact on accuracy of failure of one or a group of sensors. Models of the form (2) are used in case of a faultless operation of sensors, where information from all sensors is
used. In the event of failure of one or more sensors it is proposed to carry out calculations on models, which take into account these situations.
Let us distinguish four groups of sensors in order of importance of information coming from them to create an extended set of models.
Zero group is class of situations G0 (trouble-free operation of all sensors).
First group is class of situations G1 (soft sensor failures) includes sensors, information from which is interference on measurement channel, for example, external sensor of initial temperature and humidity of drying agent or sensor of switching on or off of agitator, that responsible for uniformity of the material layer on the tape. Off-season average temperature and humidity in workshop and average thickness of the paste are taken into account in the case of failure of sensors from that group.
Second group is class of situations G2 (medium sensor failures) that includes windows and gates opening sensors, information from which represents dryer cameras setup.
If a failure occurs in this group of sensors, the models without taking into account the degree of opening of gates and windows in the sections are used.
Third group is class of situations G3 (hard sensor failures) that includes conveyor speed, agitator switching on or off, temperature and humidity of drying agent in sections sensors, failure of which is most dangerous because it leads to errors of humidity measurement and drying process control.
Table shows the correspondences of classes with situations, missing model parameters and identifiers of models that used in this case.
Extended set of models includes the following models for n and n +1 controlled cameras of following groups:
- G0 is models m1 and m2 similar to model (2), wherein model m1 output is model m2 input.
- G1 is models m3, m4, distinctive feature of which is obtaining information about initial temperature and humidity of drying agent not from sensors but from database. Model m5 similar to m1, but without taking into account an agitator;
Classes of situations and failures of sensors
Class of situation Situation Unknown model parameter Model identifier
Go Trouble-free operation of all sensors of IHS - mi, m2
Gi Failure of workshop temperature and humidity sensor ^ x3 m3
Failure of agitator switching on or off sensor x5 m4
Failure of gates opening sensor x10 m5, m6
G2 Failure of windows opening sensor X9 m7, m8
Failure of initial material humidity X1 m9
Failure of conveyor speed sensor X4 m10
G3 Failure of exhaust fan switching on or off sensor x6 mll
Failure of temperature and humidity of drying agent sensor in sections x7, x8 m13
^ Start ^
Setting of an IHS
/ Ready for use notice / i
/Digital // Analog sensors // sensors survey // survey /
Sensors performance analysis
Definition class G, models
(Request from the database A input values parameters XAJ
Calculation of input values A",
J Screen output J
Storing input value; A", in the database
D
Request normalization factors Kj° ' and K^p' I from the database
Rationing of input values Xj
Choosing a moisture evaluation model
Request from the KB j the models mj and structure
Request from the KB f the models m s and W6 structure
( Request from the , [ KB the models mi [ \ and param etersy
(Request from the KB the models j and me param eters\
Y/ Request from ( -J the DB the V value X\
/Request from I the DB the
\ valued
/Request from j X7,X8H the DB the \ value X\q
Request from the KB the model structure
/Request from the KB the models m-j and m% structure
(Request from the KB the models m\2 and m i3 structure
Request from the DB the model parameters
■é
ï«7
ï«5
ï«8
»¡6
»¡10
Request from the KB the models m7 and m$ parameters
'«11
Request from the KB the models m\2\ and M?i3 parameters
m
»¡13
Displaying the value of the error/
Request denormalization factors from the database
X
Calculation of the material moisture content
Saving the moisture values in the database
Screen output
No Is the \Yes power button pressed
^ Finish ^
The adaptive operation algorithm of IHS
- G2 is models m6 and m7 without taking into account changing of opening degree of release gates; models m8 and m9 without taking into account opening degree of air intake windows in n and n +1 sections respectively;
- G3 is model m10 without taking into account operation periodicity of exhaust fan, model m11 without taking into account an exhaust fan work, model m12 without taking into account drying agent temperature, model m13 without taking into account drying agent humidity.
Thus, if a failure occurrs in the process of sensors survey, then the model from the knowledge base of IHS is automatically selected, by which pasty material humidity is determined with an acceptable error. As a result of the adaptive algorithm work IHS is a fault-tolerant system. The adaptive operation algorithm of IHS is shown in Figure.
An IHS during a pasty material drying in a drum-tape dryer implemented at JSC "Pigment" (Tambov); its use has allowed increasing the production of quality products.
Thus, a technical task of rapid assessment of a moving pasty material humidity during its drying in real time is reached in order to ensure the quality of the resulting material.
The work was partially supported by the RFBR grant № 14-08-00198^.
References
1. Selivanova Z.M., Samohvalov V.A. Measurement Techniques, 2012, vol. 55, no. 9, pp. 1049-1056. doi: 10.1007/s11018-012-0073-1.
2. Artemova S.V., Gribkov A. N., Bryankin K.V., Nazarov A.S., Tambov State Technical University, Sposob otsenki vlazhnosti pastoobraznogo materiala v protsesse ego sushki v val'tse-lentochnoi sushil'noi ustanovke (A method of evaluating of a pasty material humidity in a process of drying in a drum-tape dryer), Russian Federation, 2012, Pat. 2444725,
3. Artemova S.V., Nazarov A.S., Podhvatilin P.A. Voprosy sovremennoi nauki i praktiki. Universitet imeni V.I. Vernadskogo, 2011, spec. issue (36), pp. 7-12.
Адаптивный алгоритм функционирования интеллектуального датчика влажности
С. В. Артемова, А. А. Артемов, П. А. Подхватилин, И. Ю. Кошелев
Кафедра «Конструирование радиоэлектронных и микропроцессорных систем», ФБГОУ ВПО «ТГТУ»; [email protected]
Ключевые слова: адаптивный алгоритм; интеллектуальный датчик; нейронная сеть; отказ датчика.
Аннотация: Приведен адаптивный алгоритм функционирования интеллектуального датчика влажности, позволяющий в случае отказа компонентов устройства с приемлемой погрешностью определять влагосодержание пастообразного материала.
Список литературы
1. Селиванова, 3. М. Интеллектуальная информационно-измерительная система для определения теплофизических свойств материалов и изделий / 3. М. Селиванова, В. А. Самохвалов // Измер. техника. - 2012. - № 9. - С. 38 - 42 .
2. Пат. 2444725 Российская Федерация, МПК в0Ш2500. Способ оценки влажности пастообразного материала в процессе его сушки в вальце-ленточной сушильной установке / Артемова С. В., Грибков А. Н., Брянкин К. В., Назаров А. С. ; заявитель и патентообладатель Тамб. гос. техн. ун-т. -№ 2010109120/28(012771) ; заявл. 10.03.2010 ; опубл. 10.03.12, Бюл. № 7. - 16 с.
3. Артемова, С. В. Микропроцессорная система мониторинга влажности пастообразных материалов / С. В. Артемова, А. С. Назаров, П. А. Подхватилин // Вопр. соврем. науки и практики. Университет им. В.И. Вернадского. - 2011. -Спец. вып. (36). - С. 7 - 12.
Adaptiver Algorithmus des Funktionierens des intellektuellen Feuchtigkeitsensors
Zusammenfassung: Es ist den adaptiven Algorithmus des Funktionierens des intellektuellen Feuchtigkeitsensors, der im Fall des Versagens von den Anlagekomponenten mit dem annehmbaren Fehler den Feuchtegehalt zu bestimmen erlaubt, angeführt.
L'algorithme adaptatif du fonctionnement de la sonde intellectuelle d'humidité
Résumé: Est donné un algorithme adaptatif du fonctionnement de la sonde intellectuelle d'humidité permettant en cas de défaillance des composants de l'appareil de déterminer la teneur en eau de la matière pâteuse avec une précision acceptable.
Авторы: Артемова Светлана Валерьевна — доктор технических наук, профессор кафедры «Конструирование радиоэлектронных и микропроцессорных систем»; Артемов Анатолий Анатольевич - доктор технических наук, профессор ФГБОУ ВПО «Тамбовский государственный университет им. Г. Р. Державина»; Подхватилин Павел Александрович - магистрант кафедры «Конструирование радиоэлектронных и микропроцессорных систем»; Кошелев Игорь Юрьевич -студент, ФГБОУ ВПО «ТГТУ».
Рецензент: Муромцев Дмитрий Юрьевич - доктор технических наук, профессор, проректор по научно-инновационной деятельности, ФГБОУ ВПО «ТГТУ».
