revista robotica edição -100

7/24/2019 Revista Robotica Edio -100

1/148

ARTIGO CIENTFICO

Skill memory in biped locomotion (2. Parte)

A robotic platform for edutainment activities

in a pediatric hospital

AUTOMAO E CONTROLO

Automatizao de sistemas de bombagem

DOSSIER SOBRE INDSTRIA QUMICA

Qumica, a indstria com boa exportao

Reduzir o risco de transbordo

ESPECIAL SOBRE IMPRESSO 3D A primeira impressora 3D Made in Portugal

Imprima-Se em 3D

Utilizao das tecnologias de fabrico aditivo

no desenvolvimento de sapatos para pessoas

com paralisia cerebral

Contribuio para o desenvolvimento da Impresso 3D

CASE STUDY

Lubrigupo: Signum Oil Analysis: o poder de prever

WEGeuro: Gama de motores WEG W22 Super

Premium reduz perdas em 40%

Schaeffler: Guias lineares 4.0

Weidmller: Blocos de equalizao potencial JB 2550

e EBB 2550/16

RUTRONIK: A Indstria 4.0 tem de provar que vale

o investimento

ENTREVISTA

a nossa recente Cer tificao Ambiental

um fator impor tante para a nossa competitividade

no mercado, Snia Silva, WEGeuro

fornecedor lder em solues e produtos vocacionados

para a produtividade, Armando Mainsel, Europneumaq

Roadshowda Endress+Hauser em Portugal,

Paulo Loureiro

O mercado, devido crise, ficou muito mais exigente,

Jos Meireles, M&M Engenharia

100

ISSN 0874-9019

9 770874 901000

nmero 100 | 3. trimestre de 2015 | Portugal 9.50

PUB


2/148


3/148

FICHATCNICA.SUMRIO

1

robtica

FICHA TCNICA

robtica 100

3.oTrimestre de 2015

Diretor

J. Norberto Pires, Departamento de Engenharia Mecnica,

Universidade de Coimbra [email protected]

Diretor-Adjunto

Adriano A. Santos, Departamento de Engenharia Mecnica,

Instituto Politcnico do Porto [email protected]

Conselho Editorial

A. Loureiro, DEM UC; A. Traa de Almeida, DEE ISR UC;

C. Couto, DEI U. Minho; J. Dias, DEE ISR UC;

J.M. Rosrio, UNICAMP; J. S da Costa, DEM IST;

J. Tenreiro Machado, DEE ISEP; L. Baptista, E. Natica, Lisboa;

L. Camarinha Matos, CRI UNINOVA; M. Crisstomo, DEE ISR UC;

P. Lima, DEE ISR IST; V. Santos, DEM U. Aveiro

Corpo Editorial

Coordenador Editorial:Ricardo S e Silva

Tel.: +351 225 899 628 [email protected]

Diretor Comercial:Jlio Almeida


Chefe de Redao:Helena Paulino


Design

Luciano Carvalho [email protected]

Webdesign

Ana Pereira [email protected]

Assinaturas

Tel.: +351 220 104 872

[email protected] www.engebook.com

Colaborao Redatorial

J. Norberto Pires, Adriano A. Santos, J. Andre, C. Santos,

L. Costa, Joo Messias, Rodrigo Ventura, Pedro Lima,

Joo Sequeira, Paulo Alvito, Carlos Marques, Paulo Carrio,

Frederico Lucas, Paula Domingues, Miguel Malheiro,

Lus Arajo, Francisco Mendes, Amrico Costa,

Jorge Lino Alves, Lgia Lopes, Ana Dulce Meneses,

Carlos Alberto Costa, Rosrio Machado,

Ricardo S e Silva e Helena Paulino

Redao, Edio e Administrao

CIE - Comunicao e Imprensa Especializada, Lda.

Grupo Publindstria

Tel.: +351 225 899 626/8 Fax: +351 225 899 629

[email protected] www.cie-comunicacao.pt

Propriedade

Publindstria - Produo de Comunicao Lda.

Empresa Jornalstica Reg. n. 213 163

NIPC: 501777288

Praa da Corujeira, 38 Apartado 3825

4300-144 Porto

Tel.: +351 225 899 620 Fax: +351 225 899 629

[email protected] www.publindustria.pt

Publicao Peridica

Registo n. 113164

Depsito Legal n.o372907/14

ISSN: 0874-9019 ISSN: 1647-9831

Periodicidade: trimestral

Tiragem: 5000 exemplares

INPI: 365794

Impresso e Acabamento

Grficas Anduria

Avda. de San Xon, 32

36995 POIO (Pontevedra)

Os trabalhos assinados so da

exclusiva responsabilidade dos seus autores.

SUMRIO

da mesa do diretor

2 O mundo virado do avesso

artigo cientfico

4 Skill memory in biped locomotion (2. Parte)

10 A robotic platorm or edutainment activities in a pediatric hospital

empreender e inovar

16 Sentido da vida

automao e controlo

18 Automatizacao de sistemas de bombagem

eletrnica industrial

22 Fabrico de circuitos em PCI

instrumentao

28 Vlvulas de segurana e alvio

30 notcias da indstria

48 dossier sobre indstria qumica

49 Qumica, a indstria com boa exportao

52 Reduzir o risco de transbordo

56 especial sobre impresso 3D

57 A primeira impressora 3D Made in Portugal60 Imprima-Se em 3D

62 Utilizao das tecnologias de abrico aditivo no desenvolvimento de sapatos para pessoas com paralisia cerebral

66 Contribuio para o desenvolvimento da Impresso 3D

informao tcnico-comercial

68 igus: Transerncia de dados mais segura para aplicaes mveis na Indstria 4.0

70 Bucim Ex da Weidmller

72 Omron: Aranow Packaging Machinery

74 Zeben: DataloggersMSR: pequenos ormatos multiuncionais

76 Schaeffler Iberia: 7,6 milhes de euros de indemnizao pela distribuio de rolamentos FAG alsificados

78 Rittal apresenta nova gerao de ar-condicionados Blue e+

80 LusoMatrix: Unictron antenas CHIP

82 WEGeuro: Eficincia energtica em silos de armazenagem de gros

84 M&M Engenharia Industrial: Esquemas em metade do tempo

86 EGITRON/MECMESIN Controle a qualidade dos materiais da sua embalagem

88 JABA-TRANSLATIONS: criao e traduo de documentao tcnica

case study

90 Lubrigrupo: Signum Oil Analysis: o poder de prever

94 WEGeuro: Gama de motores WEG W22 Super Premium reduz perdas em 40%

96 Schaeffler Iberia: Guias lineares 4.0

98 Weidmller: Blocos de equalizao potencial JB 2550 e EBB 2550/16

100 RUTRONIK: A Indstria 4.0 tem de provar que vale o investimento

entrevista

102 a nossa recente Certificao Ambiental um fator importante para a nossa competitividade no mercado,Snia Silva, WEGeuro

104 fornecedor lder em solues e produtos vocacionados para a produtividade, Armando Mainsel, Europneumaq

106 Roadshowda Endress+Hauser em Portugal

110 o mercado, devido crise, ficou muito mais exigente, Jos Meireles, M&M Engenharia

reportagem

112 SEW-EURODRIVE PORTUGAL comemora 25 anos

114 bibliografia

116 produtos e tecnologias

138 calendrio de eventos

140 eventos e formao

144 links

www.robotica.pt

Aceda ao linkatravsdeste QR code.

/revistarobotica

Apoio capaMore Perormance

Simplified.u-remote.

Weidmller Sistemas de Interface, S.A.Tel.: +351 214 459 191 Fax: +351 214 455 [email protected] www.weidmuller.pt


4/148

obtca

AMESA

E

O mundo virado do avesso

J. Norberto Pires

Pro. da Universidade de Coimbra

Olho para a imagem daquele beb e

digo peremptrio que poderia ser meu

filho; alis, mesmo. Nele tudo me to

prximo. Os sapatos, penso que a minhafilha mais nova tem uns iguais, as calas,

o ormato da cabea... a maneira como

est deitado, como que a dormir mais

uma vez, a minha filha dorme assim, co-

loca os braos nesta posio e aunda a

cabea no travesseiro numa imagem

que tem tudo para ser serena. E , na ver-

dade, terrivelmente serena. Nunca vou

esquecer esta imagem.

Olho para os comboios de Budapes-

te cheios de gente desesperada, marca-da como se osse dierente, que nasceu

do lado "errado" do mundo, cheia de vida

e sem esperana, a correr reneticamente

procura de um osis de paz e de espe-

rana. Leio que so interrompidos na sua

viagem para a Alemanha e para a ustria

para serem levados a campos desenha-

dos para pessoas "diferentes" como eles.

S os iguais seguem a sua vida a cami-

nho de casa, e de um destino que para

os outros seria uma oportunidade de

sobrevivncia. Imagino outros comboios,noutros tempos que, em sentido inver-

so, traziam gente considerada "diferente"

para stios sem esperana, sem vida e

onde o destino era a morte.

Olho para os dirigentes europeus e

no vejo urgncia. Quando estava em

causa o dinheiro as reunies eram mar-

cadas em menos de 14 horas, mas neste

caso estando em causa vidas a urgncia

relativa e mede-se em semanas. Ouo

as suas palavras e pasmo de vergonha.Estamos perante uma "praga" de gen-

te dierente que no problema nosso,

mas antes da Alemanha. Ouo e leio isto

de pessoas que dirigem pases e no tm

a mnima vergonha, nem sobressalto hu-

mano e cvico, de o dizerem alto e em

pblico.No deixa de ser curioso ouvir a chan-

celer alem, com toda a razo, a apelar a

uma resposta coordenada da Europa,

unida em torno de valores superiores, de-

pois da campanha de desunio e desin-

teresse que promoveu para o problema

grego. Agora percebe as consequncias

de to mesquinha e irrefletida atuao. A

Europa escolheu a autodestruio e deu

voz aos nacionalismos mais primrios.

Agora com a crise dos reugiados pura esimplesmente no tem resposta, porque

no existe como UNIO.

No tenho nenhuma esperana que

o beb que podia ser meu filho, os com-

boios de Budapeste ou a vergonha das

declaraes de certos lderes europeus

alterem seja l o que or. O mundo est

numa encruzilhada terrvel. Desaparece-

ram os valores e tudo mais importante

do que as pessoas, os seus sonhos e as

suas vidas. A prova que na Europa, antes

uma esperana de um mundo melhor, selevantam muros, se acicatam antasmas

e se adiam respostas. No importante,

pois no tem a ver, aparentemente, com

dinheiro. Lamento que tenhamos, de

novo, chegado aqui.

Nota final: observo os muitos mi-

lhes de euros que estamos disponveis

para gastar em pesquisa espacial, por

exemplo, para observar e compreender

novos mundos, e fico a pensar como no

sabemos nada do que se passa no nossoplaneta, nomeadamente com as pessoas

que c vivem. E fico a pensar na razo de

tudo isso.

"O mundo est numaencruzilhada terrvel.

Desapareceram os valorese tudo mais importante

do que as pessoas, os seussonhos e as suas vidas. A

prova que na Europa,antes uma esperana de um

mundo melhor, se levantammuros, se acicatamfantasmas e se adiam

respostas (...) observo osmuitos milhes de euros que

estamos disponveis paragastar em pesquisa espacial,por exemplo, para observar

e compreender novosmundos, e fico a pensar

como no sabemos nadado que se passa no nossoplaneta, nomeadamente

com as pessoas que cvivem. E fico a pensar na

razo de tudo isso."


5/148


6/148

a

g

en

f

4

obtca

J.Andre1,C.Santos2,L.Costa3

1DepartmentofIndustrialElectronics,UniversityofMinho,[email protected]

2DepartmentofIndustrialElectronics,UniversityofMinho,[email protected]

3DepartmentofProductionandSystem,UniversityofMinho,[email protected]

Abstract Robots must be able to adapt their motor behavior to

unexpected situations in order to saely move among humans.

A necessary step is to be able to predict ailures, which result in

behavior abnormalities and may cause irrecoverable damage to

the robot and its surroundings, i.e. humans. In this paper we build

a predictive model o sensor traces that enables early ailure de-

tection by means o a skill memory. Specifically, we propose an

architecture based on a biped locomotion solution with impro-

ved robustness due to sensory eedback, and extend the concepto Associative Skill Memories (ASM) to periodic movements by

introducing several mechanisms into the training workflow, such

as linear interpolation and regression into a Dynamical Motion

Primitive (DMP) system such that representation becomes time

invariant and easily parameterizable. The ailure detection mecha-

nism applies statistical tests to determine the optimal operating

conditions. Both training and ailure testing were conducted on

a DARwIn-OP inside a simulation environment to assess and va-

lidate the ailure detection system proposed. Results show that

the system perormance in terms o the compromise between

sensitivity and specificity is similar with and without the proposed

mechanism, while achieving a significant data size reduction dueto the periodic approach taken.

Keywords Reinorcement learning Bio-inspired Skill Memory

6. FAILURE DETECTION

In order to properly take advantage o the inormation stored

into the ASM, we propose a system that monitors continuously

the execution o a motor skill (in this case biped locomotion)

and looks or deviations that could evolve into movement ailu-

res. The ailure detection protocol we introduce in this work was

inspired by Pastor et al. [20], but utilizes a more refined statisti-cal analysis in order to achieve the best results possible. Once

again we anchor the whole process on the phase values at any

given time. At each instant n o the simulation, and

, reconstructed rom the trained DMPs, provide the

ASM values or the correspondent phase (n), upon which a

statistical z-test is perormed. Thus, a tolerance interval is esta-

blished or each sensor according to:

(16)

where ytrial

(n) is the sensor data o the current simulation; (n)

is the phase at the current instant n in this trial,z =2.57 or a

confidence level o 99%, and and represent the

mean and standard deviation values stored into the ASM. I the

condition in equation (16) is satisfied, then the null hypothesis

o being a successul trial is not rejected and the sensor data is

assumed to be in conormity with the expected values - there

are no signs o ailure conditions. I, on the other hand, ytrial

(n)

is out o the confidence bounds established in (16), then the-

re is a high probability o ailure occurring, or that movement

objective is not achieved at the end o the trial. Whether or not

the current trial is flagged as ailure depends on the thresholds

or ailure detection: the minimum number o sensors M andthe minimum number o consecutive instants ailingN. Simply

put, i the system detects at leastMsensors ailing orNinstants

consecutively, then it is predicted that, based on the previous

experiences stored into the ASM, task execution will ail.

6.1. Detection Accuracy

Failure detection was interpreted as a simple two-class

classification problem, characterized by the sensitivity, conside-

red to be the probability o detecting a ailure on unsuccess-

ul trials and, similarly, by a specificityvalue, the probability o

rejecting ailures on successul trials. Conclusions about the per-

ormance o the ailure detection algorithm were based on adetection scorecomputed rom sensitivity and specificity values:

(17)

which can be interpreted as inversely proportional to the distan-

ce to the optimal operation point o maximal (100%) sensitivity

and specificity - a higher detection score implies better peror-

mance when detecting ailure conditions.

Figure 4.ROBOTIS DARwIn-OP humanoid robot in Webots.

6.2. Optimal Parameterization

MandN, thresholds or the number o sensors and number o

consecutive time steps, respectively, have a significant impact

on the perormance o the ailure detection system. With no a

prioriknowledge or practical know-how about how the sensor

readings vary, it is hard to estimate the optimal values or these

Skill memory in biped locomotionUsing perceptual informationto predict task outcome2. Parte

robtica100,3.oT

rimestred

e2015


7/148

artigocientfico

5

robtica

parameters. They have opposite impact on detection sensitivityand specificity, and as such optimal parametrization oMandN

values was achieved by plotting the ROC curve.

7. RESULTS AND DI SCUSSION

The robot used in the simulations is the DARwIn-OP by ROBOTIS

[14, 26], a small open-platorm humanoid with 20 DoF (6+6 in

both legs), using digital position controlled servos, measuring

45.5 cm and weighting 2.9 kg. The robots body is equipped with

a 3-axis gyro, a 3-axis accelerometer and each oot is equipped

with our orce sensing resistors (FSR) distributed through the

our corners. A simulated model o the robot is used in the 7.1

Webots simulation sotware, using the ODE physics simulator,

and can be seen in Figure 4.The system is integrated conside-

ring the Euler method with 8 ms fixed integration step, resulting

in a sampling rate o 125Hz. At each sensorial cycle, sensory in-

ormation is acquired. In order to replicate real world behavior in

the simulation sotware, noise magnitude was set at 5% in all ro-

bot sensors (oot orce sensors, gyroscope and accelerometer).

On the other hand, white Gaussian noise with an amplitude o

1.5% was generated and added to the joint positions computed

through the CPG dynamics

7.1. Feedback Mechanism

Locomotion is achieved in flat ground, with and without the

proposed eedback mechanism. However, an increase in loco-

motion quality is observed when compared to open-loop dy-

namics o CPG locomotion. In act, there is a 54.9% decrease in

the standard deviation in stride duration, as seen in Table 3. Inaddition, when comparing both types o locomotion, there is a

decrease in the average trial cost and in the number o ailure

trials. There is also a decrease in swing time when compared to

stance duration on a gait cycle, rom 42.85% to 34.44%.

Results o locomotion with and without eedback are sho-

wn in Figure 5, regarding right leg movement. Specifically, top

panel shows GRFright

, middle panel the phase right

, and bottom

panel the ankle pitch motion right,j

. Stance and swing phase

are highlighted in all panels. Oscillator phase is mainly modified

by the local sensory eedback during the stance phase, resulting

in steady walking. At the end o the stance phase the leg conti-

nues to bear a load, thus a phase delay is introduced which pro-vides enough time or the opposite leg to enter the stance pha-

se. As soon as the opposite leg begins to support the body, the

load on the right leg decreases accordingly. Meanwhile right leg

oscillator phase advances and the eedback effect decreases.

Thus, the right leg enters the swing phase. This illustrates the

behavior o the eedback loop implemented: the stance-swing

transition is delayed while the oot sensors are measuring non-

zero values. Besides, as discussed in [17], adaptation o the gait

depends on the body properties in a quantitative manner.

Figure 6 depicts the Centre o Mass (CoM) trajectory or

several noise magnitudes. It is notorious that open-loop CPGlocomotion is much less stable and more prone to deviation

rom the desired path. Besides, CPGs with sensory eedback

always travel more distance. Even when considering unrealistic

noiseless locomotion in a simulation environment, open-loop

CPGs have a tendency to deviate to the let. This deviation is

Figure 5.Simulation results o locomotion in the right leg movement with () and without () eedback, during the first 4 seconds o simulation. Top panel: gait dia-

gram (right and let leg), GRFright

and GRFleft

, respectively. Middle panel: phase right

. Bottom panel: ankle pitch motionright,AnklePitch

. The stance (grey areas) and swing

phase are highlighted in all panels.

Figure 6.Centre o Mass (CoM) trajectory with (-) and without (-) sensory eedback, or noise magnitude o 0, 0.5, 1.0 and 1.5% respectively. The noise seed used was the

same in all o these situations, in order to keep comparisons as realistic as possible.


8/148

a

g

en

f

obtca

significantly attenuated with the implemented sensory eedba-

ck. With noise levels o 1.5%, open-loop CPG dynamics are no

longer able to achieve stable orward movement, instead devia-

ting rom the path as soon as the simulation starts.

Table 3.Mean and standard deviation o Step duration, during T = 1000 trials with

1.5% random noise at joint-level.

In this situation, eedback locomotion eventually alls as well,

however is able to cover more ground distance than open-loop

CPGs. Under the assumption that more regular and stable loco-

motion leads to less variance in step duration, we conclude that

the inclusion o eedback has improved locomotion quality by

providing a more regular gait. Additionally, as described in this

work, this type o phase eedback is useul as an anchoring pointo the data normalization process, and can be used as reerence

when perorming ailure check.

7.2. Failure Detection Parameterization

The ASM prediction model proposed implements a strai-

ghtorward ailure detection system by treating detection as

a two-class classification problem, in terms o sensitivity and

specificity. M and N, thresholds or the number o sensors

and number o consecutive time steps, respectively, have a

significant impact on the overall perormance o this ailure

detection system. However, there is no a priori knowledge or

practical know-how about how the sensor readings vary, andthus it is hard to estimate the optimal values or these parame-

ters. In order to find out the best possible parameter values, the

optimal operating conditions are achieved by assessing system

perormance o the system using ROC analysis.

SeveralM/Ncombinations were tested in both cases (with

and without data normalization) ,which resulted in the ROC cur-

ves presented in Figure 7.

The minimal Euclidean distance to the optimal operating

point (0,1), associated with maximal (100%) sensitivity and

specificity values, was used as evaluation criterion - the para-

meter pairing closest to this point achieves best perormanceand was used throughout this paper. Best perormance is thus

accomplished with (M,N) = (12, 2) beore normalization and

(M,N) = (22, 7) with normalization. It is noteworthy to observe

that ater normalization, a larger number o sensors ailing or a

greater number o consecutive instants is necessary to achieve

detection reliability. This may be caused by the act that ater

normalization ASM reerence values become a more generic

representation o the movement, and local deviations rom the

nominal values occur more requently.

7.3. ASM training and testing

Figure 8 depicts a subset o the 32 recorded signals used in the

ASM training phase. Unsuccessul trials (a total o 436 in 500) are

depicted in red and success ul ones (64 in 500) in green. The

weighted mean (black dashed line) and the weighted stan-

dard deviation (black solid line), beore normalization, are com-

puted rom the successul trials. On the remaining training runs,

56;60 successul trials and 444;440 unsuccessul trials composed

the training set. As previously explained, in the testing phase the

other 500 trials containing 55;63;59 successul and 445;437;441

unsuccessul trials were used.During the testing phase, sensor data is being monitored

online. DMP reconstruction o the memory data occurs at the

beginning o the trial, or each sensor. At each simulation time

step, the oscillator phase is used as reerence in order to pin-

point the current stage o the movement (i), and the corres-

pondent and or phaseivalue is extracted

rom the reconstructed DMP. Then the ailure detection z-test is

perormed using and .

This process is illustrated in Figure 9, where the data rom 3

different sensors (gyroscope CoM angle xx

, Let Foot Back Let

orce sensor GRFLFBLand the CPG oscillator phase i) is plot-ted or a successul and unsuccessul trial, during a small part

o the simulation and including two complete gait cycles. The

Figure 7.ROC curves or ailure detection beore (let) and ater (right) data nor-

malization (plotted values:N2{1, 2, 3, 4, 5} andM2{6, 8, 10, 12, 14, 16} beore

normalization andN2{3, 4, 5, 6, 7} andM2{16, 18, 20, 22, 24} ater normalization);

the optimal operation point in each case is the closest to (0,1).

Figure 8. Sensor traces recorded during 1000 simulation trials, or sensors #1-3 (gyroscope readings o CoM angle ), #11 (Let Foot Front Let Force sensor),

#28 (right,KneePitch), during the first 1.6s o the simulation trials. Unsuccessul trials are depicted in red and successul ones in green, with the weighted mean drawn as a

black dashed line and the weighted standard deviation as a black solid line.


9/148

artigocientfico

7

robtica

boundaries between movement periods are identifiable by

the values o right

. The light green shaded areas represent the

confidence intervals provided by the ASM trained with the train

subset. Deviation rom these intervals are easily perceptible in

the case o the unsuccessul trial (which resulted in a cost o

-0.2434 - no all occurred) but only occur momentarily in the

successul case (with a cost o -0.4295). Failure was only detec-

ted in the unsuccessul trial, as expected.

Table 4. Statistical detection results averaged over 3 training and testing runs,

beore and ater data normalization.

7.4. Detection Accuracy

A detection scorecomputed rom sensitivity and specificity va-

lues as given by (17) was computed to draw conclusions aboutthe perormance o the ailure detection algorithm. Similarly to

ASM training, ailure detection was conducted three times on

different Gtest

trial subsets.

From the results presented in Table 4, it is possible to veriy

that ASM normalization had little to no impact on ailure detec-

tion perormance, which is in line with the delineated objectives

or this work. Theoretically, normalization should provide no in-

crease in algorithm perormance, as its advantages are mainly

related to data representation and flexibility. Even so, detection

results improved slightly ater data normalization (higher de-

tection score). Despite achieving high sensitivity and specificity

values in both cases, the presence o alse positive and alse ne-

gative detections might be explained by ailure trials that have

cost values close to the threshold - a greater distinction betwe-

en success and ailure might be needed. Alternative reasons

might also include a deficient selection o sensor data. A useulmetric in these situations is the alse negative and alse positive

rates, which are computed as 1 sensitivy and 1 specificity

respectively, and quantiy the probability o occurrence o a al-

se negative/positive.

7.5. Prediction Interval

One o the main eatures o a properly trained ASM is allow

ailure detection ahead o time during uture task executions.

Detection o ailure conditions while executing a task occurs

the moment several perceived sensor signals deviate rom the

predicted ones. The prediction advance in the specific case orobot all during a trial is here quantified.

Figure 10 presents the whisker diagram drawn rom the pre-

diction interval tadvance

= tfailure

tdetection

values. Failures were

detected on average 3.63 seconds in advance without norma-

lization, although a decrease to 1.90 seconds was observed a-

ter normalization. This can partly be explained by two actors:

firstly, the value o N increased significantly with normalization

- deviation rom nominal values must occur or a larger interval

beore a ailure is detected, which leads to small deviations that

cause ailures without normalization being ignored; secondly,

as previously mentioned, the normalized ASM also takes into

account inter-step/period variations and thus leads to broaderconfidence intervals, and consequently a less specific sensor

memory. An example o a sensor trace in a successul trial is pre-

sented in Figure 9. The one-tailed t-test shows that we can con-

clude that the average prediction interval ater normalization

is significantly inerior to average interval beore normalization

(T=7.58, d=658, p-value


10/148

a

g

en

f

obtca

The corresponding plots or the horizontal CoM angle xx

(top panel), Let Foot Back Let (LFBL) orce sensor (middle pa-

nel) and right leg phase right

(bottom panel) are depicted in

Figure 9 in orange. The robot reaches the start o the ramp at

t 30.6 s (C, blue dotted vertical marker), where the inclination

is still very sot. The inclination continues to get stronger and

more notorious, as can be seen by the large GRFLFBL

peak near

C. The robot eventually alls at t 31.2 s (indicated byD, solid

red vertical marker), with an increasing deviation in the valueo

xx.

On the other hand, ailure detection happens ahead o time.

When data normalization is not considered ailure prediction

occurs at t= 28.7 s (A, dotted-dashed grey vertical marker), and

at t= 29.04 s with data normalization (B, black dashed verti-

cal marker). This is not considered to be a relevant difference

- a window o 2.5 and 2.16 seconds beore alling, respectively,

is reasonable or corrective actions, although it will ultimately

always depend on the specific context. Even so, we consider the

benefits acquired with data normalization largely outweigh the

limitations encountered.

7.7. Size of Stored Data

With conventional ASMs, sensor data enclosing theP= 10 lo-

comotion periods corresponds to around 1600 data points

per sample or each sensor, and or both the means and

standard deviations have to be computed. This leads to

2 32 1600 = 102400 values to be stored (627 kB raw data).

Ater normalization, considering Kphase

= 200, each mean

and standard deviation are represented by 200 values (0.5%

intervals), which leads to 232200 = 12800, a decrease o

about 87.5% rom the initial data size.

Additionally, ater DMP regression, both and

are represented by the sets o D = 100 DMP parameters ound

through DMP regression. This leads to 2 32 100 = 6400 va-

lues, to which are added 3 DMP additional parameters o dam-

ping constantsx

, yand

yand the 64 starting values or each

DMP, which in all entails 6467 values (approx. 47 kB o raw data

- a 93.7% decrease to 6.3% o the original amount). It is notewor-

thy that this decrease ends up being dependent on the original

number o data periods recorded, and is even greater when

more than 10 periods are part o the original ASM. Besides, we

intend to make the ASM continuously or regularly updated by

the robot.

8. CONCLUSION AND FUTURE WORK

This work introduces systematic and automated guidelines

to build a skill memory rom sensor data, with periodic, time

consuming movements in mind, leading to a simplified and

phase-indexed representation. This allows to continuously

and periodically monitor the execution o a specific skill in

real time and predict nonconormity (such as robot alls),

allowing proper corrections to the robots behavior and mi-

nimization o unwanted consequences associated with mo-

vement ailure.

Conclusions are inerred rom simulations on the DARwIn-

OP robot. It is shown the potential in building a normalized ASM

and use it in ailure detection without significant losses in detec-

tion sensitivity and specificity - in act, i properly tuned, norma-

lized data can be as precise, albeit with a decreased advance in

detection. However, the authors consider the advantages brou-

ght by data normalization to largely outweigh the small decre-

ase in detection interval, as it adds flexibility and compactness

to the ASMs.

In the uture, it would be valuable to, instead o building

reerence performance confidence intervals, the procedure

would be inverted to arrive at reerence failure sensor oot-

prints, in order to properly identiy specific ailure conditions,that could be useul when identiying the cause o ailure.

These failure ASMs could be used either independently or in

conjunction with a library o skill ASMs. On a parallel line o

work, we have modified the CPG locomotion dynamics o our

locomotion system [14] to be easily modulated by a small set

o DMP systems, which are then parameterized by RL algori-

thms with the goal o adapting locomotion to different en-

vironments without changing the core dynamics. Ideally, we

expect to be able to use an ASM library to properly identiy the

current circumstances the robot is in and/or causes o ailure,

and appropriately load the correspondent DMP to cope withthe situation.

9. ACKNOWLEDGMENTS

This work is unded by FEDER Funding supported by the Ope-

rational Program Competitive Factors - COMPETE and National

Funding supported by the FCT - Portuguese Science Founda-

tion through project PTDC/EEACRO/100655/2008 and Project:

FCOMP-01-FEDER-0124-022674.

REFERENCES1. Andr, J., Santos, C., Costa, L.: Path Integral Learning o Multidimensional

Movement Trajectories. In: Proceedings o the International Conerence o

Numerical Analysis and Applied Mathematics 2013 (ICNAAM 2013). Rhodes,

Greece (2013);

2. Aoi, S., Ogihara, N., Funato, T., Sugimoto, Y., Tsuchiya, K.: Evaluating unctional

roles o phase resetting in generation o adaptive human bipedal walking

with a physiologically-based model o the spinal pattern generator. Biological

Cybernetics 102(5), 373387 (2010);

3. Chen, M., Zheng, A., Lloyd, J., Jordan, M., Brewer, E.: Failure diagnosis using

decision trees. In: Proceedings o the 1st International Conerence on Auto-

nomic Computing (ICAC 2004) (IEEE, 2004), pp. 3643 (2004);4. Christensen, A.L.: Fault Detection in Autonomous Robots. Phd, Universit Li-

bre de Bruxelles (2008);

5. Dev Anand, M., Selvaraj, T., Kumanan, S.: Fault detection and ault tolerance

methods or industrial robot manipulators based on hybrid intelligent appro-

ach. Advances in Production Engineering Management 7(4), 225236 (2012)

6. Hobbelen, D., Wisse, M.: A disturbance rejection measure or limit cycle walk-

ers: The gait sensitivity norm. IEEE Transactions on Robotics and Automation

23(6), 12131224 (2007);

7. Hohn, O., Gerth, W.: Probabilistic balance monitoring or bipedal robots. Inter-

national Journal o Robotics Research 28(2), 245256 (2009);

8. Ijspeert, A.J., Nakanishi, J., Hoffmann, H., Pastor, P., Schaal, S.: Dynamical Mo-

vement Primitives: Learning Attractor Models or Motor Behaviors. Neural

Computation 25(2), 32873 (2013);

9. Ijspeert, A.J., Nakanishi, J., Schaal, S.: Learning Attractor Landscapes or Lear-

ning Motor Primitives. Advances in Neural Inormation Processing Sys tems

(15), 15231530 (2003);


11/148

PUB

10. Kalyanakrishnan, S., Goswami, A.: Learning to predict humanoid all. Inter-

national Journal o Humanoid Robots 8(2), 245273 (2011);

11. Karssen, J.G.D., Wisse, M.: Fall detection o two-legged walking robots using

multi-way principal components analysis. International Journal o Huma-

noid Robots 7(1), 7393 (2010);

12. Kim, J., Kim, Y., Lee, J.: A Machine Learning Approach to Falling Detection

and Avoidance or Biped Robots. In: SICE Annual Conerence, pp. 562567

(2011);

13. Kunihiro, O., Koji, T., , Yasuo, K.: Falling motion control or humanoid robots

while walking. In: 7th IEEE-RAS International Conerence on Humanoid Ro-

botics (Humanoids 2007), pp. 306311 (2007);

14. Matos, V., Santos, C.P.: Central Pattern Generators with Phase Regulation or

the Control o Humanoid Locomotion. In: IEEE-RAS International Coneren-

ce on Humanoid Robots. Osaka, Japan (2012);

15. Ogata, K., Terada, K., , Kuniyoshi, Y.: Real-time selection and generation o

all damage reduction actions or humanoid robots. In: 7th IEEE-RAS Inter-

national Conerence on Humanoid Robotics (Humanoids 2008), pp. 233

238 (2008);

16. Olson, D., Delen, D.: Advanced Data Mining Techniques. In: Springer-Verlag

(2008);17. Owaki, D., Morikawa, L., Ishiguro, A.: Listen to bodys message: Quadruped

robot that ully exploits physical interaction between legs. In: 2012 IEEE/

RSJ International Conerence on Intelligent Robots and Systems, pp. 1950

1955. Ieee, Vilamoura, Portugal (2012);

18. P.-B.Wieber: Viability and predictive control or sae locomotion. In: 2008

IEEE/RSJ International Conerence on Intelligent Robots and Systems

(IROS), pp. 11031108 (2008);

19. Pastor, P., Hoffmann, H., Asour, T., Schaal, S.: Learning and generalization o

motor skills by learning rom demonstration. 2009 IEEE International Con-

erence on Robotics and Automation pp. 763768 (2009);

20. Pastor, P., Kalakrishnan, M., Chitta, S., Theodorou, E., Schaal, S.: Skill learn-

ing and task outcome prediction or manipulation. 2011 IEEE International

Conerence on Robotics and Automation pp. 38283834 (2011);

21. Pastor, P., Kalakrishnan, M., Meier, F., Stulp, F., Buchli, J., Theodorou, E., Schaal,

S.: From dynamic movement primitives to associative skill memories. Ro-

botics and Autonomous Systems 61(4), 351361 (2013);

22. Pastor, P., Kalakrishnan, M., Righetti, L., Schaal, S.: Towards Associative Skill

Memories. In: 12th IEEE-RAS International Conerence on Humanoid Robo-

tics (Humanoids), pp. 309315. Osaka, Japan (2012);

23. Pastor, P., Righetti, L., Kalakrishnan, M., Schaal, S.: Online movement

adaptation based on previous sensor experiences. In: 2011 IEEE/RSJ

International Conerence on Intelligent Robots and Systems (IROS)

(2011);24. Plagemann, C., Stachniss, C., Burgard, W.: Efficient ailure detection or mo-

bile robots using mixed-abstraction particle filters. In: European Robotics

Symposium 2006, pp. 93107 (2006);

25. Renner, R., Behnke, S.: Instability detection and all avoidance or a huma-

noid using attitude sensors and reflexes. In: 2006 IEEE/RSJ International

Conerence on Intelligent Robots and Systems (IROS), pp. 29672973

(2006);

26. ROBOTIS: ROBOTIS DARWIN-OP e-Manual v1.23.00. http://support.robotis.

com/en/product/darwin-op.htm (Aug. 2014);

27. Santos, C.P., Matos, V.: CPG modulation or navigation and omnidirectional

quadruped locomotion. Robotics and Autonomous Systems 60(6), 912

927 (2012);

28. Ruiz-del Solar, J., Moya, J., Parra-Tsunekawa, I.: Fall detection and man-

agement in biped humanoid robots. Proceedings o the IEEE Inter-

national Conerence on Robotics and Automation (ICRA 2007) 2,

3323 3328 (2010).


12/148

a

g

en

f

obtca

JooMessias1,RodrigoVentura1,Pedro

Lima1,JooSequeira1,PauloAlvito2,CarlosMarques2,Pa

uloCarrio2

1InstituteforSystemsandRobotics,Instituto

SuperiorTcnico,UniversidadedeLisboa,{

jmessias,rodrigo.ventura,pal,jseq}@isr.ist.utl.pt

2IdMindEngenhariadeSistemas,Lda.,

{palvito,cmarques,pcarrico}@idmind.pt

Abstract Social Robotics is a rapidly expanding field o resear-

ch, but long-term results in real-world environments have been

limited. The MOnarCH project has the goal o studying the long-

term social dynamics o networked robot systems in human

environments. In this paper, we present the MOnarCH robotic

platorm to the research community. We discuss the constraints

involved in the design and operation o our social robots, and

describe in detail the platorm that has been built to accomo-

date the project goals while satisying those restrictions. We alsopresent some preliminary results o the navigation methodolo-

gies that are used to control the MOnarCH robotic platorms.

I. INTRODUCTION

Designing robots or social purposes has been a trendy topic

or the last decades. The literature in this area is huge and has

yielded valuable lessons [1], [2]. However, experiments where

robots and people coexisted or long periods o time, outside

lab environments, meaning periods longer than the transient

in the dynamics o human expectations, have seldom been

reported.MOnarCH1 (Multi-Robot Cognitive Systems Operating in

Hospitals, [3]) is an ongoing FP7 project with the goal o intro-

ducing (social) robots in real human social environments with

people and studying the establishment o relationships betwe-

en them.

The environment that acts as a case-study or the project is

the pediatric ward o an oncological hospital (IPOL). We intend

to introduce a team o robots in that environment, that coope-

ratively engage in activities aiming at improving the quality o

lie o inpatient children.

Key scientific hypotheses underlying the MOnarCH projectresearch are that (i) current technologies enable the acceptance

o robots by humans as peers, and (ii) interesting relationships

between robots and humans may emerge rom their interac-

tion. These hypotheses are supported by extensive existing

work on (i) autonomous and networked robotics, enabling

sophisticated perception and autonomous navigation, and (ii)

interaces or human-robot interaction and expressive robots.

MOnarCH addresses the link between these two areas, having

robots playing specific social roles, interacting with humans un-

der tight constraints and coping with the uncertainty common

in social environments.

The constraints o the social environment partially translate

into physical constraints on the robot platorms, such as its ma-

ximum allowable dimensions and velocities, and also behavio-

1 Reerence: FP7-ICT-2011-9-601033. Website: http://monarch-p7.eu/

ral constraints that can reflect on the methods that are used to

control those platorms, such as its navigation algorithms.

In this paper we present the MOnarCH robot platorm to the

research community. The platorm is well-suited to a wide range

o applications that extend beyond the MOnarCH case-study:

combining different high-level actuators and sensors, the base

can be used in the ofice, domestic or industrial environments

that are considered in the RoboCup@Home or @Work compe-

titions, or example.This document is organized as ollows. We will first provide

an overview o the constraints that were taken into account in

the design o this platorm (Section II). We will then describe the

robot hardware (Section III); and also o the methods that were

used to carry out its navigation (Section IV).

II. CONSTRAINTS ON ROBOT DESIGN & CONTROL

The MOnarCH project has a significant component o human-

robot interaction (HRI) to be carried out in a very specialized

social environment, namely that o IPOL pediatr ic ward. The na-

ture o this environment implies concerns and constraints onthe type o robots to be used, namely,

The range o allowable linear and angular velocities;

The volumetry o the ull robot;

Aesthetics;

Maximum height o the platorm;

Payload;

Power supply autonomy;

Sel-saety eatures;

Human-oriented saety eatures.

Moving naturally is an essential capability or a robot to be ableto survive in a social environment. In a sense, i a robot mo-

ves naturally, with velocities in the same order as those used

by humans moving, then other HRI interaces can be ocused

A robotic platformfor edutainment activitiesin a pediatric hospital

robtica100,3.oT

rimestred

e2015

Figure 1. A comparative study o linear velocities or common off-the-shel

platorms, and also or the MOnarCH robot platorm.


13/148

artigocientfico

11

robtica

and their behaviour does not need to depend on the motion

o the robot. Motion in 2D (as is the case in MOnarCH) is com-

pletely described by linear and angular velocities and hence the

ability to combine these two velocities determines the baseline

expressivity o the movement. This is a key aspect when desig-

ning a mobile platorm or socially embedded HRI purposes, as

in MOnarCH.

For example, in what concerns linear velocity, i a social ro-

bot playing with a child needs to ask him/her to wait becauseit cannot move as ast as the child then there is a significant

risk that the child looses interest in playing with the robot.

Moreover, uture interactions may be compromised because

the child may eel that the robot can not do a simple thing

such as moving the way he/she does. As or angular velocity,

its combination with the linear velocity determines i the robot

ollows the child graceully, i.e., with appropriate expressivity.

Motion capabilities are thus a basic eature that potentiates

the effect o all the other HRI interaces, namely voice, vision,

grasping, etc.

Differential drive mobile platorms can be ound off-theshelin a wide variety o ormats. Figure 1 shows the maximum va-

lues or the linear velocity o common robots. An adult moving

normally in an indoor environment reaches requently velocities

in the range o 2 to 2.5 m/s. Under teens children move slower

than adults when walking but may reach similar speeds when

running.

The physical presence o the robot has a large influence in

the way bystanders perceive the robot and its intentions. The

physical dimensions o the robot must not be perceived by chil-

dren neither as a menace nor as a physically diminished social

entity.

The average height o an under teen (11 year) is around1450mm and hence this determines the maximum height o

a MOnarCH robot. The volumetry is selected in order to be so-

cially acceptable and dynamically stable (not tilting under high

accelerations/decelerations). The ability to carry a large number

o sensors and interaces i a key eature in a social robot, this

meaning that payload is an important eature. Moreover, such

payload has to comply with the volumetry/height/aesthetics

concerns above.

Power supply autonomy severely constraints HRI capabili-

ties i the robot requires too much time to recharge batteries or

recharging occurs at an inadequate time. An HRI aware batterymanagement system limits the situations in which children may

perceive the robot as a flawed social entity.

O extreme importance are the saety eatures in the plat-

orm. In addition to basic physical saety o the people handling

the robots, saety concerns are directly related to Ethics issues

and o paramount importance when in social environments

such as that at IPOL.

Saety measures are embedded at both hardware and sot-

ware levels. Unexpected collisions trigger can be detected at

hardware level and bypass all decisions levels to stop the robot.

Each o the sotware layers has their own saety measures.

III. ROBOT DESCRIPTIONThe kinematics o a robotic platorm can greatly impact the type

o social interactions that it can be expected to perorm. As the

user case scenarios or the MOnarCH were being defined and

the constraints posed by the environment o operation were

being discussed, it became evident that the mobility capability

o the robots could be a critical issue to the achievement o pro-

ject goals. Based on this evidence, we have opted to develop an

omnidirectional robot platorm based on our Mecanum whe-

els, actuated by our independent motors. The use o this kind

o kinematics substantially increases the maneuverability and

perormance o the platorm. The development and assemblyo MOnarCH robots has been divided in two phases. The first

phase includes the platorm base mechanics with the motors,

batteries and low-level electronics. The resulting platorm can

be adapted to serve different applications. A second phase, whi-

ch specifically targets the MOnarCH scenario at IPOL, includes

the installation o high-level devices mounted over an upper

structure and the design o an outer shell. For this purpose, two

types o robots are being developed. Perception Oriented (PO)

robots will have as primary goal to act as active sensors. Social

interaction Oriented (SO) robots will target social interactions. As

aorementioned, the SO and PO robots are built over the same

platorm base, differing in the onboard equipment and externalappearance. An assembled platorm is shown in Figure 2. At this

time, the first phase o robot development has been concluded.

A. MOnarCH Robot Platform Base Main Features

All the robot platorms include the same basic configuration

which can be described through the ollowing design eatures:

Body: Polyacetal -POM (PolyOxyMethylene) 10 mm thick

plates; rigid PVC 4 and 6 mm; and transparent polycarbona-

te 2mm;

Robot kinematics: Omnidirectional 4 Mecanum wheels;

Robot weight: 24 Kg (with batteries); Payload capacity: 30 Kg;

Maximum Linear Speed: 2.5 m/s;

Maximum Angular Speed: 600 o/s;

Acceleration: 1 m/s2 (low-level programmed);

Emergency Stop Acceleration: -3.3 m/s2 (low-level pro-

grammed);

Mini-ITX computer Board with CPU, RAM and SSD;

Batteries:

Supports up to 4 batteries at the same time,

Capacity: (12v) 17-20 Ah 5.5 kg each,

Chemistry: lead acid or LiFePO4 block 12V batteries with

PCM,

Autonomy: 4 to 6 hours,

Actuators: 4 DC motors or locomotion;

Sensors:

Battery level,Figure 2. Assembled MOnarCH robot platorm.


14/148

a

g

en

f

obtca

Motor encoders,

Omnidirectional bumper,

4 ground sensors,

12 sonars,

Laser Range Finder (5m range),

Temperature sensors to measure the motors and drivers

temperature,

Temperature and humidity sensor to measure the envi-

ronment conditions, Installed Electronics Boards:

Sensor & Management Board,

Motor Control Board,

Sonars Board,

Ground Sensor Board,

IMU Board.

B. MOnarCH Robot Upper Body

The upper body o the platorm will include different high-le-

vel devices. Some o these devices are still being defined and

need urther experiments to validate their use in the MOnarCHrobots.

Two depth cameras with microphone (Kinect type);

Three servo motors to actuate two robot arms and a head

(only SO robots);

One 10 touch-screen (only SO robots);

One pico-projector (only SO robots);

One RFID reader;

One Hagisonic StarGazer localization sensor;

Audio amplifier with speakers;

LEDs on the robot body;

Capacitive cells on the robot body;

C. Sensors

The robot is equipped with perception, navigation, interaction,

environment and low-level saety sensors. For locomotion the

robot uses encoders to control the velocity o the motors, and

or navigation it uses an inertial sensor to determine the angular

speed and a laser range finder to detect obstacles and the ge-

ometry o the environment. For perception and interaction, the

robot will use a depth camera or people tracking, ace analysis

and body gesture recognition, and also microphones. For envi-

ronmental sensing the robot will be equipped with temperatu-

re and humidity sensors. Finally, the bumpers and sonar sensorsprovide low-level saety sensing. To increase the robustness o

localization, some other sensors/solutions are also being evalu-

ated, e.g., RFID, IR and UWB.

We now list the sensors that are used onboard.

1) Navigation Sensors:The robot will navigate in the environ-

ment while making a usion o measures provided by different

sensors. The robot will be able to use a depth camera, a laser

range finder, encoders odometry and the IMU sensor to estima-

te its position and orientation. For obstacle avoidance, mapping

and localization it can use the laser and sonar sensors.

Inertial Sensor IMU: MPU6050;

Function: Orientation estimation;

Position: In the robots kinematic center;

Front 2D laser range-finder: Hokuyo URG-04LX-UG01;

Function: Mapping, localization and obstacle avoidance;

Position: Frontal and horizontal;

Sonar Sensors: Maxbotix EZ4;

Function: Obstacle detection (e.g.: glass wall or objects);

Position: Ring o 12 sonars around the robot;

Depth camera: Asus Xtion;

Function: Obstacle detection, space geometry analysis;

Position: Top o the robot pointing to the floor;

Sensors being evaluated: RFID, UWB, and ToF 3D cameras.

2) Perception and Interaction Sensors:The robot will make use o

a depth camera or people detection and sense visual user ee-dback or natural user interaction. It can also be used to detect

changes in the surrounding environment. The perception sen-

sors are the ollowing.

Depth camera: Asus Xtion;

Function: Interaction, people and gesture recognition;

Position: Top and looking ahead;

Microphone array: Asus Xtion;

Function: Sound eedback or natural user interaction;

Position: Turned to the users;

10 Touchscreen (or tablet);

Function: User eedback on specific contents;Position: Turned to the user;

Capacitive sensors;

Function: User eedback on specific points;

Position: Under the shell;

Other sensors still being evaluated: RFID and UWB.

3) Environment Sensors:The environment sensors are used to

detect environment variations that can affect the normal ope-

ration o the robot. These sensors are: temperature sensor and

humidity sensors.

4) Low-level Safety Sensors:The undamental sensors or low-

level saety are the sonar sensors, internal temperature sensors,

motor current sensing and the bumper ring switches.

D. Actuators

For actuation, the robot is equipped with locomotion and inte-

raction devices.

1) Locomotion Actuators: For locomotion, this omnidirec-

tional platorm uses our motors to drive its Mecanum wheels.

Four Maxon RE 35 90W 15V motor with a Maxon GP 32

HP 14:1 Gearbox and encoder HEDS 5540 with 500 pulses;

Function: Provide a omnidirectional locomotion sys-

tem to the robot;

Position: In the platorm, connected to the drive system.2) Interaction Actuators: Here ollows the list o interaction

devices. The robot is able to display the contents on the interac-

tion monitor or project them over a surace.

10 Monitor with Touchscreen (or tablet);

Function: Interaction with displayed contents (e.g., AR

contents);

Position: Front o the robot;

Video Projector (pico type);

Function: Projection o contents;

Position: Projecting to the ront o the robot;

Arms and head servo motors;

Function: Human robot interaction;

Position: Mounted on the robot body;

Body LED lights;

Function: Show robot expressions;

Position: Mounted on the robot body;


15/148

artigocientfico

13

robtica

Stereo Speakers;

Function: Content playback; robot communication;

Position: Turned to the user.

E. Electronic Power Architecture

The robot can be powered by several 12V 17-20AH batteries. It

uses one 12V battery to deliver power to the motor drivers. Up

to 3 other batteries to provide energy to all the other compu-

ters and electronic components. An individual charging unit isused inside the robot to charge each battery. The batteries and

the power in the robot is managed by the Sensor &Managment

Board that measures the battery levels, battery charge, and also

controls the units (motors, sensors, actuators and inverters)

powered by the batteries. All onboard electronic systems can

be powered by the battery system. The ATX computer power

supply provides regulated voltages (rom 5V to 12V). Figure 3

depicts the onboard power architecture. Several DC-DC conver-

ters are also be used to provide the necessary regulated power

or other DC-DC powered devices.

F. Low-level Communication Architecture

The onboard robot navigation computer communicates with

the two boards (Sensor& Management Board and the Motor

Controller Board) using 2 USB ports. In each board there are

USB-to-RS232 converters that convert the USB data packages

to serial RS232 packages or the board controllers. Each board

controller communicates with the other allowing the exchan-

ge o inormation between them. This communication chan-

nel allows the execution o low-level behaviours, or example,

react against an imminent collision, enter into charging mode

with motors shut down, reduce the motors velocity when the

batteries are low, or react to changes that can affect the robotsoperation, which is undamental to the improvement o the

overall system dependability. The main controller rom the

Sensor&Management Board communicates with other micro-

controllers using Inter-Integrated Circuit (I2C) communication

ports. The main controller acts as the master and the other

microcontrollers behave like slaves. The Sensor&Management

Board controls the battery management and charge, sensor

acquisition, devices actuators and sonar acquisition boards. The

Motor Controller Board connects to the PI Motor controllers and

also to temperature sensors. Each controller has a low-level ault

diagnosis that will check the operation state o each microcon-troller and also monitor all the communication between the

devices. The low-level communication architecture is depicted

in Figure 4.

G. High-level Communication Architecture

The MOnarCH robot connects to a local network. A wireless

Ethernet router provides the IP address to the onboard comput-

ers and allows the exchange o messages between them. The

Navigation Computer is connected to the navigation sensors

and to the platorm board controllers using USB ports. The In-

teraction Computer connects to the Projector using a HDMI

output and to the Sound System using the audio line out, and

will use USB connections to connect to the Interaction Board

that will control the body LEDs, the capacitive sensors and the

upper moving parts o the shell (arms and head). The high-level

communication architecture is depicted in Figure 5.

IV. NAVIGATION

For navigation we use a standard occupancy grid map [4], obtai-

ned rom off-the-shel SLAM sotware2This map is used both or

motion planning, using Fast Marching Method (FMM) [5], and

localization, using off-the-shel sotware3

.Motion planning is based on a FMM approach [5]. Unlike

other methods based on explicit path planning, e.g., RRT [6],

ollowed by path tracking, we adopt here a potential field ap-

proach. Given a map constraining the workspace o the robot,

together with a easible goal point, a (scalar) potential field u(x),

orx2R2, is constructed such that, given a current robot location

x(t), the path towards the goal results rom solving the ordinary

differential equation x(t)= u(x). In other words, given an arbi-

trary current location o the robotx, the robot should ollow a

gradient descent o the field u(x). Using potential fields or mo-

tion planning was proposed in the 80s [7] but they were ound

to be prone to local minima [8]. This problem can be solved by

the use o harmonic potential fields [9], however it does not gua-

2 GMapping (http://wiki.ros.org/gmapping, retrieved 16-Oct-2013).

3 AMCL, (http://wiki.ros.org/amcl, retrieved 16-Oct-2013).

Figure 3. MOnarCH robot power architecture.

Figure 4. Low-level communication architecture.

Figure 5. High-level communication architecture.


16/148

a

g

en

f

4

obtca

rantee absence o local minima at the rontier. Thus, we decided

to employ a more recent approach [10]. The use o FMM provi-

des: (1) local minima ree path to goal across the gradient, (2)

allows the specification o a spatial cost unction, that introduces

a sot clearance to the environment obstacles, and (3) does not

require explicit path planning and trajectory tracking.

The FMM is based on the Level Set theory, that is, the re-

presentation o hypersuraces as the solution o an equationu(x)= C. The solution o the Eikonal equation

(1)

wherex2 is a domain, the initial hypersurace, andF (x)

is a cost unction, yields a field u(x) [5]. The level sets o this

field define hypersuraces u(x)=Co points that can be reached

with a minimal cost o C. The path that minimizes the integral

o the cost along the trajectory can be shown to correspond to

the solution o x(t)= u(x) with the initial condition ox(0)

set to the initial position and the initial condition u() = 0 set atthe goal4. Intuitively it corresponds to the propagation o a wave

ront, starting rom the initial hypersurace, and propagating

with speed 1/F(x). This path minimization is usually considered

a continuous space version o the Dijkstras algorithm. FMM is a

numerically eficient method to solve the Eikonal equation or

a domain discretized as a grid. Its computational complexity is

O(NlogN), whereNis the total amount o grid cells, which is

comparable to Dijkstras algorithm or sparse graphs.

Since FMM employs a grid discretization o space, it can be

directly applied to the occupancy grid map, where domain

corresponds to the ree space in the map. As cost unction we use

(2)

4 is set to the boundary o an arbitrarily small ball around the goal.

whereD(x) is the distance to the nearest occupied cell in the map

andDmax

is a threshold to clip the cost unction. This cost unction

induces a slower wave propagation near the obstacles, and thus

making the optimal path to display some clearance rom them.

The clipping atDmax

prevents the robot to navigate in the midd-

le o ree areas, regardless o their size. TheD(x) unction can be

directly obtained using an Euclidean Distance Transorm (EDT) al-

gorithm taking the occupied cells as boundary. Figure 6 illustratesthe results o this approach: the cost unction or the given map,

allowing a certain clearance rom mapped obstacles, is shown in

(a), rom which, given a goal location, a field u(x), shown in (b)

is obtained (the goal corresponds to the minimum value o the

field), and in (c) the real path taken by the robot is shown.

Using FMM on a previously constructed map does not ac-

count or unmapped or moving obstacles. Thus, the fieldv(x)

used to control the robot in real-time results rom combining

the field u(x) obtained rom FMM with a repulsive potential field

r(x) o obstacles sensed by the LRF. This repulsive field is obtained

rom running EDT on a small window around the robot, such thatthe value o r(x) corresponds to the minimum distance between

any obstacle and pointx. The fields are combined using

(3)

where is a parameters speciying the strength o the repulsive

field (higher values o tend to increase the clearance rom per-

ceived obstacles). Note that (3) destroys the property o a single

local minima o the field. We acknowledge the need to comple-

ment our navigation approach with a mechanism or detecting

and coping with stuck robot situations, such as replanning or

asking or help.

The method described above have proven to be very effec-

tive, even in cluttered environments ull o people crowded

around the robot. We have demoed this method on a public

event the European Researchers Night (September 27th,

(a) F(x) (b) u(x) (c) real path

Figure 6. Motion planning using FMM: (a) the cost unction F(x) (darker means a higher cost), (b) the solution field u(x) (level curves) together with the gradient descent

x(t)= u(x) solution (rom the right to the let), and (c) the real path traveled by the robot.

Figure 7.Trajectory o ISR-CoBot autonomously navigating along the IPOL premises. The task consisted in a sequence o waypoints.


17/148

PUB

2013, in the Pavilion o Knowledge science museum, Lisbon) where

people rom all ages crowded around the robot.

We have also tested this method at IPOL, where we run exten sive

autonomous navigation tasks during several hours (Figure 7). These tests

were perormed on a previous platorm [11]. Even though that previous

platorm is differential, minor modifications on the guidance method

were required to adapt it to the MOnarCH platorm. A video showcasing

the application o these methods to the autonomous navigation o the

MOnarCH platorm can be ound at: http://tinyurl.com/olounbn.

V. CONCLUSIONS AND FUTURE WORK

In this work, we have introduced the robotic platorm that was develo-

ped in the context o the MOnarCH project. This development explicitly

took into account a set o constraints that are induced by the social na-

ture o the projects case-study environment, namely, the pediatric ward

at IPOL. We described these constraints; detailed the hardware that was

(and is being) included in the robotic platorm; and presented prelimi-

nary results regarding the methods that were developed or reliable ro-

bot navigation.The qualities o the MOnarCH platorm make it a good choice or

other applications beyond the projects case-study, such as the Robo-

Cup@Home or @Work scenarios.

As immediate uture work, we will integrate the high-level sensors

and devices discussed in Section III, as part o the second phase o robot

development. This will endow the robot platorm with HRI capabilities,

establishing a basis or the uture development o the socially-aware in-

teraction methods that are crucial to the outcome o the project.

ACKNOWLEDGMENTS

Work supported by FCT projects PEst-OE/EEI/LA0009/2013 and FP7-ICT-9-2011-601033 (MOnarCH).

REFERENCES

[1] C. Breazeal, Designing sociable robots. The MIT Press, 2002;

[2] G. Metta, G. Sandini, D. Vernon, L. Natale, and F. Nori, The iCub humanoid robot: an open

platorm or research in embodied cognition, in Procs. of the 8th Workshop on Performan-

ce Metrics for Intelligent Systems, PerMIS 08, 2008, pp. 5056;

[3] J. Sequeira, P. Lima, A. Safiotti, V. Gonzalez-Pacheco, and M. A. Salichs, MOnarCH: Multi-

robot cognitive systems operating in hospitals, in ICRA 2013 Workshop on Many Robot

Systems, 2013;

[4] A. Eles, Using occupancy grids or mobile robot perception and navigation, IEEE Com-

puter, vol. 22, no. 6, pp. 4657, 1989;

[5] J. A. Sethian, Fast marching methods,SIAM review, vol. 41, no. 2, pp. 199235, 1999;

[6] S. LaValle and J. Kuffner Jr, Randomized kinodynamic planning,The International Journal

of Robotics Research, vol. 20, no. 5, pp. 378400, 2001;

[7] J. Borenstein and Y. Koren, Real-time obstacle avoidance or ast mobile robots, IEEE

Transactions on Systems, Man and Cybernetics , vol. 19, no. 5, pp. 11791187, 1989;

[8] Y. Koren and J. Borenstein, Potential field methods and their inherent limitations ormobile robot navigation, in Proceedings of the IEEE International Conference on Robotics

and Automation (ICRA-91), 1991, pp. 13981404;

[9] J. Kim and P. Khosla, Real-time obstacle avoidance using harmonic potential unctions,

IEEE Transactions on Robotics and Automation, vol. 8, no. 3, pp. 338349, 1992;

[10] S. Garrido, L. Moreno, D. Blanco, and M. L. Munoz, Sensor-based global planning or

mobile robot navigation, Robotica, vol. 25, no. 2, pp. 189199, 2007M

[11] R. Ventura, New Trends on Medical and Service Robots: Challenges and Solutions, ser. MMS.

Springer, 2014, ch. Two Faces o Human-robot Interaction: Field and Service robots.


18/148

obtca

empeende

e

n

a

Frederico Lucas o empreendedor responsvel pelo projeto Novos

Povoadores, que presta apoio a empresrios que pretendammigrar com as suas amlias para territrio rural. Nascido

em Lisboa, em 1972, o autor dos conceitos territoriais desoftwareterritorial e economia DNS. Territorial Developer

em projetos de dinamizao territorial em www.inoex.pt e considerado como um visionrio de um uturo com grandes

mudanas e criao de novos paradigmas.

Sentido da vida

A robotizao das bricas tem sido ob-

servada com grande reserva por uma

parte significativa da populao mundial.

O filme Os tempos modernos de CharlieChaplin no compreendido por todos.

A histria do mundo no se az com

recordistas de linhas de montagem. So

tareas banais e repetitivas, que algum

equipamento ter a capacidade de

executar.

Complexas so as atividades que exi-

gem cultura, sensibilidade e conscincia.

Valores que uma mquina no saber in-

terpretar, e que dierenciam cada um de

ns, consequncia dos nossos percursose contextos.

A robotizao um aliado do Ser Hu-

mano, porque o liberta de tareas manu-

ais. Conduz a Humanidade para as tareas

inteletuais, a sua vocao.

Na Europa, este processo decorre h

vrias dcadas, e os elevados nveis de

desemprego espalham algum receio na

sociedade. Estamos a alhar no essencial:

legtimo e natural, que a gerao dos

nossos filhos venha a trabalhar quatro

horas por dia. E que essas horas sejamtempo de criao, em oposio a tareas

burocrticas ou repetitivas, mas sempre

mal pagas.

O poder de compra descer em con-

traciclo com o poder da mente.

As fbricas retiraram milhes

de chineses dos campos,os robots esto a entrarnas fbricas e os operrios

no querem voltar a seragricultores.

Joo Vale de Almeida,

Diplomata

Complexas so as atividadesque exigem cultura, sensibilidade

e conscincia. Valores que uma

mquina no saber interpretar,e que diferenciam cada um de ns,consequncia dos nossos percursos

e contextos. A robotizao umaliado do Ser Humano, porque

o liberta de tarefas manuais.Conduz a Humanidade para

as tarefas inteletuais, sua vocao."

As bricas que outrora poluram rios,

estaro no uturo a azer o processo in-verso: a recolher resduos nesses rios e a

produzir capas para iPhones com esses

detritos, num modelo conhecido por

upcycling.

Esses rios regressaro sua uno:

espao de lazer para o reino animal, in-

cluindo Seres Humanos, que deixaram

de os requentar por falta de tempo.

Pelo exposto, sou um deensor da

robotizao das indstrias. Estejam os

robotsao servio da pigmentao de car-

roarias ou a filtrar mensagens de spam

nas nossas caixas de correio.

Acredito que a Europa, antes da

China, saber trilhar este caminho, re-

centrando a Humanidade no Sentido da

Vida.


19/148


20/148

obtca

A

MA

E

N

AdrianoA.Santos

DepartamentodeEngenhariaMecnica

InstitutoPolitcnicodoPorto

Automatizao de sistemasde bombagem

INTRODUOA automatizao de uma rede de rega

ou distribuio de gua pode, em geral,

ser realizada em escalas distintas. Se nos

sistemas de rega a mxima automatiza-

o se consegue atravs da programa-

o das regas, da ertilizao e outros

aspetos relacionados com a recolha de

dados agrometeorolgicos e de cultivo,

nos sistemas de distribuio a mxima

automatizao consegue-se atravs do

controlo e monitorizao dos ativos re-motos como as estaes de bombagem,

de elevao e das estaes de tratamen-

to. A um nvel mais baixo, o controlo po-

der ser realizado localmente atravs de

um programador que, ao nvel da rega,

controla cada um dos hidratantes e o

conjunto de vlvulas e eletrovlvulas. Ou

seja, pode-se automatizar uma rede de

distribuio de gua, quer seja de rega

quer seja domstica, recorrendo a um

computador central que, com base, por

exemplo, num sistema SCADA (Supervi-sory Control and Data Acquisition), contro-

la os hidratantes, as vlvulas e os sistemas

remotos.

Por ltimo, e no menos importante,

as estaes de bombagem que tambm

podem e devem ser reguladas e contro-

tizao completa da estao de bomba-gem, mediante a variao de velocidade

das bombas, sensores de presso e cau-

dalmetros controlados por um autma-

to programvel, PLC (Programmable Logic

Controller).

AUTMATO PROGRAMVEL

Um controlador lgico programvel, nor-

malmente conhecido como PLC, um

equipamento idealizado para o controlode processo tendo como base instru-

es lgicas programadas atravs de um

software especfico. So equipamentos

de reduzidas dimenses que, instalados

no quadro de comando, permitem con-

trolar o processo e a comunicao com

o exterior e receber ordens atravs dos

canais de comunicao o que permite

a configurao de sistemas de teleco-

mando e de telemedida, dando lugar

ao comando centralizado. Estes, embora

possam possuir capacidades muito supe-riores s exigidas para este tipo de con-

trolo, apresentam uma elevada capaci-

dade de processamento, uncionamento

em modo local, inteligncia distribuda e

comunicao via rdio ou por cablagem

(concentrador ou hub).

Estes elementos, praticamente indis-

pensveis nos processos de automatiza-

o, so caraterizados por uma estrutura

modular que permite o crescimento de

acordo com as necessidades do proces-so, so robustos, possuem elevada capa-

cidade para a implementao de progra-

mas complexos, acilidade de correo

e modificao dos programas e baixo

custo (Figura 1). Os PLCs podem ser, per-

ladas, adaptando-se s variaes de cau-dal e de presso da rede. Estas so cons-

titudas, normalmente, por uma ou mais

bombas encarregadas de transormar a

energia mecnica recebida dos sistemas

de acionamento em energia hidrulica.

A energia hidrulica ser, ento, utiliza-

da para aspirar gua subterrnea de um

poo, elevar gua de uma cota inerior

para uma cota superior ou injetar uma

presso adicional na instalao. Assim,

e pelo que oi anteriormente exposto,pode-se dizer que a automatizao mais

bsica de um sistema de bombagem

consiste no arranque e paragem auto-

mtica segundo uma regulao horria.

A automatizao mais completa consis-

tir no controlo automtico do arranque

e paragem da bomba, monitorizao da

abertura e echo das vlvulas, tempos de

abertura e echo das mesmas, nvel de

gua, entre outros.

Por outro lado h que considerar

que a aspirao e a elevao de cotas,do ponto de vista uncional, trabalha

com parmetros fixos, altura e caudal, e

com pequenas oscilaes devido va-

riao do nvel de gua e, como tal, no

necessitam de um sistema de regulao

de caudal, trabalhando sempre perto da

seu rendimento mximo. Ao contrrio do

que se passa nos sistemas descritos ante-

riormente, as redes de distribuio sobre

presso esto sujeitas a variaes de cau-

dal ao longo da sua utilizao, pelo queestas necessitaro dum sistema de regu-

lao de caudal que adeque a presso e

o caudal, impulsionado pelas bombas,

procura instantnea da rede. Neste caso,

uma boa regulao requer uma automa-

Figura 1. Autmato programvel (PLC), mdulo de comunicao, CPU e mdulo digital de I/O (Siemens).

"A automatizao de uma

rede de rega ou distribuio degua pode, em geral, ser realizadaem escalas distintas.

Se nos sistemas de regaa mxima automatizao

se consegue atravsda programao das regas,

da fertilizao e outros aspetosrelacionados com a recolha

de dados agrometeorolgicose de cultivo, nos sistemas de

distribuio a mxima automatizaconsegue-se atravs do controlo

e monitorizao dos ativosremotos como sejam as estaes

de bombagem, de elevao e dasestaes de tratamento."


21/148

AUTOMAOECONTROLO

19

robtica

eitamente, utilizados nas estaes de

bombagem no controlo do arranque e

paragem das bombas, na abertura e e-

cho das eletrovlvulas, no nvel de aspi-

rao e nvel dos depsitos de armazena-

mento, na regulao do caudal, presso,

entre outros.No que se reere ao uncionamento

dos autmatos programveis este rela-

tivamente simples. As entradas e sadas

do sistema so fisicamente ligadas aos

dispositivos de campo das mquinas,

aquisio de sinais, e aos processos a

controlar, envio de sinais.

As entradas transerem para a uni-

dade central do autmato inormaes

sobre o uncionamento do processo que

podem ser sinais discretos ou analgicos

(pressstatos, sondas de nvel, transduto-res de presso, e outros). O CPU, por sua

vez, em uno do programa lgico car-

regado, envia para as sadas inormaes,

tambm elas discretas ou analgicas,

para eetuar o controlo dos dispositivos

(rels das bombas, solenides, aciona-

mento de vlvulas, variao de veloci-

dade das bombas, entre outros). Assim,

e durante o tempo de uncionamento

deste, o ciclo de trabalho do autmato

executar trs aes cclicas, preponde-rantes para o controlo do processo como

a leitura das variveis de entrada, a exe-

cuo do programa lgico de controlo e

a escrita das variveis de sada, como se

depreende do esquema representado na

Figura 2.

Esta leitura sequencial das entradas,

execuo do programa e atualizao das

sadas conhecida como ciclo de varri-

mento (Scan) onde as entradas digitais

do tipo On/Off, permitem ler o estado

dos pressstatos, contactores, botes

de presso, comutadores, entre outros,

ou seja, o estado dos equipamentos li-

gados (On) ou desligados (Off). Por sua

vez, as entradas analgicas permitem

conhecer, em modo contnuo, quer em

mA (miliAmperes) quer em V (Voltes), o

valor dos sinais emitidos pelo controlo

de presso, caudal, volume, temperatura,

entre outros, ou seja, receber sinais do

tipo analgico.As sadas digitais so utilizadas em

manobras de abertura e echo de eletro-

vlvulas, arranque e paragem de bom-

bas, vlvulas motorizadas bem como de

sistemas de sinalizao e de piloto. As

sadas analgicas so utlizadas na atua-

o de variadores de requncia, variao

da velocidade das bombas, e posio de

abertura de vlvulas motorizadas, servo-

-vlvulas, entre outros.

VARIADORES DE FREQUNCIA

Os variadores de requncia so utiliza-

dos para variar a velocidade de uncio-

namento dos motores assncronos de

Corrente Alternada (CA) utilizados no

acionamento das bombas hidrulicas.

O princpio de uncionamento des-

tes equipamentos baseia-se no controlo

da velocidade e binrio do motor, por

intermdio de um sistema de comando

eletrnico. Quando um motor de indu-

o colocado em marcha, o seu ponto

de operao ajusta-se ao fim de algumtempo de uncionamento. Este ajuste

depende da voltagem (V) e da requn-

cia () aplicada ao motor. O ajuste pode

ser obtido atravs da curva caraterstica

do motor, ou seja, da relao V/ (V/Hz)

(Figura 3).

Note-se que a relao V/ linear

entre =0 e a requncia de base (50 Hz).

A tenso aplicada ao motor aumenta

linearmente at aos 50 Hz, situao em

que o motor atinge a sua tenso nominal(tenso da rede). At aos 50 Hz o bin-

rio desenvolvido pelo motor constante

com uma potncia crescente. Acima de

50 Hz, a tenso mantm-se constante e

igual ao valor nominal com binrio de-

crescente (zona a tracejado). O aumento

da requncia implica o aumento da ve-

locidade, o que se traduz na diminuio

do binrio.

O retificador gera uma Corrente Con-

tnua (DC) que posteriormente filtrada

e introduzida no inversor. A unidade decontrolo , normalmente, constituda por

um modulador PWM, limitador de cor-

rente e um controlador de velocidade do

tipo (PI) (Figura 4). O modo de operao

pode ser manual ou automtico segun-

do as necessidades do processo pelo

que estes podem ser operados por com-

putadores, PLC, atravs de sinais digitais e

analgicos ou de orma manual.

A utilizao de variadores de re-

quncia nos mais diversos processos ,hoje em dia, uma prtica corrente devi-

do, essencialmente, ao seu baixo custo,

melhoria energtica resultante da sua

utilizao e acilidade de se poderem

instalar com qualquer motor assncrono

Figura 2. Esquema geral de uncionamento de um autmato.

PROCESSO(Dispositivos de atuao)

ENTRADA

(Sensores, sinaisdigitais e analgicos)

AUTMATO

(Lgica programada)

SADA

(Atuadores, sinaisdigitais e analgicos)

Figura 4.Diagrama de blocos de um variador de requncia do tipo V/.

Barramento DC

FiltroRetificador

Redede

alimentao

Tensode

frequnciafixas

Alimen

taomotor

Tensoefrequnciavarivel

Inversor

Unidade de controlo microprocessada

0

V

M

10 20 30 40 50 60 70 Hz

Figura 3.Curva V/ e binrio para um motor

de induo.


22/148

A

MA

E

N

obtca

no interior dos quadros de controlo dos

processos (Figura 5).

No entanto, a variao de velocidade

de rotao do motor produz uma modi-

ficao da curva caraterstica da bomba

dando lugar a curvas paralelas curva davelocidade nominal, como se pode ver

na Figura 6, diminuindo o rendimento

da bomba que , claramente, mais baixo

do que o rendimento terico. A perda

de rendimento deriva principalmente

do rendimento do prprio variador de

requncia e, esta perda ser tanto maior

quanto mais baixa or a velocidade de ro-

tao da bomba.

AUTOMATIZAO DE SISTEMAS

DE BOMBAGEM

Como j oi reerido a automatizao dos

sistemas de bombagem pode ser realiza-

da com dierentes graus de automatiza-

o. Em geral, o grau de automatizaoser tanto maior, quanto maior or a pro-

undidade do poo e o caudal da bomba,

a cota de elevao e a presso de injeo

na rede de distribuio.

Por outro lado, e em qualquer um

destes processos, o arranque e a para-

gem, totalmente automatizada, podero

ser controlados por um autmato tendo

em considerao que a abertura e o e-

cho das vlvulas devem ser temporiza-

dos de modo a proteger o binmio mo-

tor-bomba, a tubagem e evitar golpes deariete (martelo ou choque hidrulico) na

ase de paragem dos processos.

No obstante, este tipo de automati-

zao com abertura e echo temporiza-

do de vlvulas, tem sido substitudo por

sistemas de variao de requncia. Os

variadores permitem o arranque e a para-

gem suave dos motores protegendo-os

de sobreintensidades e dos golpes de

ariete. Esta proteo conseguida com

o arranque e a paragem progressiva do

motor da bomba acionado pelo variador

segundo o programa de arranque e de

paragem estabelecido.

Outros pontos devem ser considera-

dos no processo de automatizao dos

sistemas de bombagem como, e no

menos importantes, os reservatrios,as vlvulas, os sensores (nvel, presso,

caudal, e outros) bem como as aes

de segurana e emergncia. Assim, e

no que se reere s aes de segurana

e de emergncia programadas no PLC,

estas passaro pela paragem de emer-

gncia por alta de comprovao da

abertura e echo das vlvulas, sinal de

retorno da posio, por valor de presso

do pressstato superior ao mximo ou

mnimo predefinido incluindo os nveisde gua nos poos e nos reservatrios,

isto , alha dos sinais das sondas de n-

vel mnimo e mximo de gua respon-

sveis pela paragem da bomba, perante

o nvel mnimo de gua e, no caso de

enchimento, evitar que a gua transbor-

de no reservatrio.

No que se reere s unes de con-

trolo e regulao estas devem incidir

no s sobre a regulao do caudal que

deve ser realizada em situaes em que

o ornecimento da gua inerior s ne-cessidades dos consumidores, nas horas

de ponta, aumentando o ornecimento e

diminuindo-o nas horas de menor con-

sumo, preservao dos nveis de gua,

mas tambm sobre o ator energtico e

na sua respetiva regulao que passar

pelo uncionamento dos sistemas, sem-

pre que possvel, durante as horas de

energia mais barata, ou seja, nos pero-

Figura 7.Esquema geral de um sistema de bombagem de injeo direta na rede.

Figura 5.Variadores de requncia (Siemens).

Figura 6.Curvas caratersticas de uma bomba

a dierentes velocidades de acionamento.

H (m)

P (kW)

Q (l/min)

1100 rpm

1150 rpm

1400 rpm

1400rpm

1450 rpm

1450rpm

0'70'75 0'8

0'80'75

0'81

500 1000

100

100200300400

75

50

1100rpm

Sonda de nvel mnimoSonda de

nvel de guaSonda de nvel

revista robotica edição -100

Documents