Vous êtes sur la page 1sur 6

Towards a Factory-of-Things:

channel modeling and deployment assessment in PetroEcuador Esmeraldas oil refinery

S. Savazzi , B. Ramos , J. M. Winter , S. Kianoush , V. Rampa , E. del Rosario , T. Chavez and O. Cevallos §

CNR, IEIIT institute, Milano, Italy. E-mail: (stefano.savazzi, sanaz.kianoush, vittorio.rampa)@ieiit.cnr.it ESPOL, FIEC, Guayaquil, Ecuador. Email: (bramos, elderos, tchavez)@espol.edu.ec UFRGS, Porto Alegre, Brazil and E-Aware Technologies, Porto Alegre, Brazil. Email: jean.winter@ufrgs.br. § EP PetroEcuador, Quito, Ecuador. Email: oscar.cevallosn@eppetroecuador.ec.

Abstract—Industry 4.0 and industrial Internet of Things (iIoT) trends are pushing towards the transformation of factories to pro- vide more flexible production systems through the use of wireless networks. Technologies enabling the “Factory-of-Things” (FoT) paradigm allow the safe deployment of wireless field devices in industrial plants thanks to their low-battery usage that makes the maintenance cycle quite low, and highly reliable. The widespread adoption of these technologies should be paired with tools for pre- deployment network design and prediction of the wireless link quality to mimic the planning procedures applied to conventional industrial wired equipment. In factory sites, the strength of the radio signals is impaired by frequency, spatial and time-domain fading that influence the wireless link stability. In this paper, based on an extensive measurement campaign performed inside an active oil refinery, we propose and validate a novel channel model tailored for industrial wireless networks operating over 2.4 GHz and supporting a time-slotted channel hopping (TSCH) policy. Post-layout network performance verification has been finally carried out based on a WirelessHART industry standard system deployed in selected sites.


The adoption of wireless communication in industrial re- fineries is becoming of strategic interest for many manufac- turers and operators in oil & gas and petrochemical industries. Compared to wired system, wireless technology has the advan- tage of low-cost, mobility, energy-efficiency, compactness and flexibility in self-configuration or in overcoming any obstacle. Recently, the industrial Internet of Things (iIoT) paradigm [1][2] has emerged as an evolution from a large number of systems employing machine-to-machine communications for

This work has been performed in the framework of the project “A Wireless Sensor Network Deployment Study in Esmeraldas Refinery”, funded by PetroEcuador and under CNR-IEIIT and ESPOL Memorandum de Acuerdo Interinstitucional, Oct. 2015. Stefano Savazzi, Sanaz Kianoush, Vittorio Rampa are with Consiglio Nazionale delle Ricerche (CNR), Institute of Electronics, Computer and Telecommunication Engineering (IEIIT) Milano, Italy. Boris Ramos, Edison del Rosario and Tanny Chavez are with Escuela Superior Politcnica del Litoral (ESPOL), Facultad de Ingeniera en Electricidad y Computacin (FIEC), Guayaquil, Ecuador. Jean M. Winter is with Federal University of Rio Grande do Sul (UFRGS), Department of Electrical Systems, Automation and Energy, Brazil and with the spin-off E-Aware Technologies. Oscar Cevallos is with Empresa Publica (EP) PetroEcuador, maintenance supervisor of Esmeraldas refinery.

process automation. Wireless networks are the critical enabling technology for iIoT implementation, and are now considered as an instrumental tool for the Industry 4.0 trend towards the transformation of manufacturing technologies into a Factory of Things (FoT) [3][4].

Many FoT-related applications have often to satisfy stringent requirements in terms of reliability and timeliness. Therefore, the network must be accurately engineered during a “pre- deployment” stage in order to be robust against packet losses as well as meet stringent delay deadlines. Existing network design procedures are often not suitable to address complex and harsh environments as typical in refinery environments [3]. Industry-standard design procedures for wireless network optimization must be instead able to certify the reliability of radio links under harsh conditions and be applied before the deployment of the network. Most of the currently proposed in- dustrial systems are built on the IEEE 802.15.4 physical (PHY) layer [5] over the ISM 2.4 GHz band. Many standards have then redesigned the Medium Access Control layer (MAC) to address specific industrial needs. One of these MAC designs, namely the Time Slotted Channel Hopping (TSCH), gained popularity as standardized in the IEEE 802.15.4e and currently implemented in many products, such as WirelessHART [6] and ISA 100.11a [7]. The TSCH protocol implements a slow fre- quency hopping policy at the MAC sub-layer. The goal of the hopping scheme is to harness frequency diversity to improve communication reliability: the physical carrier frequency of each communication link thus changes according to a pre- defined hopping pattern [8] over a set of carrier frequencies defined by the physical layer of the standard IEEE 802.15.4.

The purpose of this paper is to propose a novel approach to pre-deployment link quality prediction that is based on a statistical channel model to assess the site-specific wireless propagation parameters in industrial environments affected by dense blockage and multipath. The proposed model extends the one proposed in [3] and it is tailored for wireless networks that are built on a TSCH standardized transmission mode. It accounts for both frequency and time-domain fading and

978-1-5090-5137-3/16/$31.00 ©2016 IEEE

provides a statistical model for the received signal strength (RSS) that influences the wireless path stability [8]. In order to validate the proposed model, an extensive experimental campaign has been carried out inside an oil&gas refinery site (Esmeraldas refinery) owned by PetroEcuador, the national oil company of Ecuador. As for typical refinery environments, the site is characterized by an harsh environment for short-range (10-150m) propagation with metallic structures, changing en- vironmental conditions, non-line of sight (NLOS) and pos- sible co-located wireless applications running over the same unlicensed spectrum. It thus shows relevant similarities with a dense micro-cellular site. Specific wireless field equipment has been designed to perform a long-term and real-time acquisition of received signal strength (RSS) measurements over different links and considering all the channels defined by the PHY layer of the IEEE 802.15.4 standard over 2.4 GHz band. The study is therefore relevant as it highlights the impact of both spatial, frequency and time-domain fading on the performance of TSCH-based industrial networks and provides support to pre-deployment planning. In contrast to previous experimental campaigns (see [9], [10]), the refinery sites target of the tests were active (in production): this allowed to verify wireless propagation inside a fully representative environment possibly subject to electromagnetic (EM) interference caused by spurious emissions of plant instruments as well as other wireless equipment (e.g. WiFi). Post-layout network deploy- ment verification has been finally carried out by installing a WirelessHART-compliant network in selected sites.


As depicted in Fig. 1, the goal of the channel hopping scheme, as supported by TSCH mode communication, is to let each communication link hop pseudo-randomly over a set of physical radio-frequency channels. The wireless link is thus organized in time-slots, while frequency hopping might involve up to M radio channels (M = 15 as prescribed by IEEE 802.15.4) according to a predefined pattern F that defines a sequence of consecutive PHY channels. The focus of the section is on the development of an ad-hoc model that provides a statistical representation of the key environment- dependent parameters that characterize the channel response, namely the RSS, over different PHY channels and links. In particular, the RSS observed over different channels is affected not only by path-loss (function of the geometric properties of the link) but also by multi-path effects that are both frequency and spatially varying (frequency/spatial fading), depending on the type of the configuration of the surrounding fixed scattering/absorbing objects. In addition, time-domain fluctuations (temporal fading) might be also induced by people or objects moving in the environment [11]. The RSS s i, (t) at time t, observed over channel i ∈ F and link of length d is modelled (in [dBm]) as:

s i, (t) = s 0, ρ i,

+ δ i, (t),

s¯ i,


provides a statistical model for the received signal strength (RSS) that influences the wireless path stability

Fig. 1. Example of multi-channel RSS data collection (Esmeraldas refinery, PetroEcuador) and TSCH frequency-hopping pattern.

where s 0, = s 0,



0 , γ P


µ 0 Γ



0 , γ P is the path

loss static component, with µ 0 = µ 0 (P T , d 0 ) being the channel gain function for transmitted power P T , measured at a reference distance d 0 (d 0 = 2m typical), and distance dependent loss function Γ(d /d 0 , γ P ) = 20 log 10 (d /d 0 ) + 10(γ P 2) log 10 (d /d Fre ) that follows the model [12] for short d < d Fre and long range d > d Fre , respectively. Fresnel distance is d Fre = 4h tx h rx assumes flat terrain with h tx , and h rx being the height from ground of the transmitter and the receiver, respectively. The static RSS excess attenuation ρ i, randomly varies over the space/frequency domain as due to both diffraction and multipath effects. The static term s¯ i, = s 0, ρ i, thus combines the path-loss component with the fading effects due to the presence of fixed scattering/absorbing objects. Temporal fading is an additional component δ i, (t) that models the superimposed RSS time-domain fluctuations due to objects or people moving in the surrounding to the link, as well as intermittent/transitory EM interference.

In what follows, we highlight that some site-specific features

of the channel response are shared by links conforming to the

same “obstruction type” (or link class). Those features reveal


Type I

Type II

Type III

Type IV

Type V

γ C j






σ F,C j






σ T





that the time δ i, and the frequency-domain ρ i, sensitivity of the link can be captured by a common stochastic model that provides a statistical representation for RSS s i, , namely for the probability functions Pr[ρ i, ], and Pr [δ i, ]. The model in turn allows to make predictions of the link quality (or path stability) under the assumption of a TSCH transmission mode, and before the actual deployment of the network.

A. Link classification

The classification of the propagation is based on a set of C

link types {C 1 , C 2 ,

C C }: each category describes a specific

... configuration of the building blockage that maps onto specific space/frequency-domain fading profiles. Classification of links is based on the number and the density of the obstructions: it might adopt a 2D/3D map the plant [10], or on-site inspec- tion. For any link ∈ C j classified as Type j, frequency-


domain RSS fluctuations map onto a type-specific stochastic

model for spatial/frequency-domain Pr [ρ i, | ∈ C j ] and time-

domain Pr [δ i, |

C j

] fading terms. RSS excess attenuation

  • 2 F,C j ,

is herein modelled as log-Gaussian ρ i, ∼ N γ C j , σ

with attenuation γ C j as due to diffraction [12] and frequency- domain fluctuations with deviation σ F,C j . Such deviations are caused by multipath effects characterizing the specific link type. Measurements reveal that temporal fading is less sensitive to link type, but still reasonably log-normal, therefore δ i, ∼ N 0, σ T now with time-domain deviation σ T . The use of C = 5 mutually exclusive link categories is reasonable for the considered setting: this is also in line with previous activities [3]. In what follows, ordering of link types is based on increasing attenuation γ C j due to diffraction effects, frequency and time-domain fading parameters are detailed in the Table I. Type I: “LOS” (C 1 ) link type is characterized by line- of-sight (LOS) propagation as well as absence of obstacles (with dimensions larger than the signal wavelength λ) within the 1st Fresnel volume [12]. Obstacles thus might occupy the 2nd Fresnel volume. Although reflections from terrain (flat) are responsible for some multipath effects, frequency-domain RSS fluctuations σ F,C 1 = 2.38 are smaller compared to other link types. Compared to Type I, links of Types II-IV are characterized by obstacles that are either lying inside the 1st Fresnel volume or obstructing the LOS path. Propagation is thus subject to strong multipath effects as due to different obstruction configurations and this results in higher frequency- domain RSS fluctuations, i.e. σ F,C i > σ F,C 1 , i = 2, 3, 4. In particular, Type II: “near-LOS” (C 2 ) link type is observed in environments where the obstacles are located inside the


first Fresnel outer region. Type III “OLOS” (C 3 ) obstructed-

LOS link type is observed in environments where the obstacles

are now partially obstructing the LOS path [10]. Finally, Type IV “NLOS” (C 4 ) link type is characterized by objects completely obstructing the direct path between transmitter and receiver, but leaving a clearance zone inside the first Fresnel area. Type V “severe-NLOS” (C 5 ) link type refers to a propagation environment where the LOS path is blocked by large-size (i.e. concrete) buildings [12] typically located in the surrounding of the transmitter or the receiver, while the first Fresnel region is completely obstructed. Propagation is mostly influenced by diffraction effects caused by large obstructions, while multipath effects are less significant. The observed frequency-domain RSS fluctuations σ F,C 5 are thus smaller compared to other link types.

B. Model validation and experimental activity

Measurements have been collected from 4 relevant sites inside the Esmeraldas refinery: these locations, depicted in Fig. 2, are considered critical for wireless deployment and have been identified by PetroEcuador crew. The developed model considers a database up to 120 wireless links. RSS data collection is based on specific field devices, labeled as MF and MF-G, whose MAC layer has been purposely designed for multi-channel acquisition. The devices comply with the physical standard of the IEEE 802.15.4e: RF emitting source device (MF) is tailored to transmit a configurable number of data frames (i.e., ranging from 1-10) over the

same physical channel and then hopping according to a pre- configured pattern (see the example of Fig. 3). Each MF device is programmed to transmit at standard-compliant power of 4 dBm, although the output power can range up to 10dBm. One (or more) receiver devices (MF-G) can be configured to synchronously collect RSS data and the corresponding channel

specified by the hopping scheme. For each scenario, the MF device has been moved in order to assess the link quality in positions potentially critical for com- munication. In fact, the position of the wireless instruments can be influenced by the presence of obstacles (e.g., pipes, metal structures, walkways, valves, etc.). During the tests, RSS data are repeatedly collected (alternating long and short records of 15 min.) and processed for all M = 15 channels defined by the TSCH mode. The impact of people moving in the surrounding of the link, as well as small deviations of the nominal position of the equipment [3], have been also addressed. In Fig. 5 we highlighted the RSS measurements and model (1) calibration. Time-averaged RSS vs channels E t [s i, ] and log-distance log(d /d 0 ) are first compared with model E t,i [s i, ] = s 0, E[ρ i, ] for all link types (left column). RSS excess attenuation empirical probability functions ρ i, and log-Gaussian moment-based fitting are then compared (center column). Finally, time-domain deviations σ T (right column) are analyzed: notice that differently from frequency-domain fading, temporal fading is less sensitive to link class as mainly influenced by moving people/objects in the surrounding of the link, or transitory EM interference.

Fig. 4. Model calibration and testing based on the experimental data collected in the Esmeraldas refinery.

Fig. 4. Model calibration and testing based on the experimental data collected in the Esmeraldas refinery. From left to right: RSS vs channels and log-distance E t [s i, ] and corresponding model E t,I [s i, ] (blue line), RSS excess attenuation ρ i, empirical pdf and log-Gaussian fitting, time-domain deviations σ T . Data are collected for the 5 link types considered in Sect 2.2.

Fig. 2. Candidate sites for the experimental campaign inside the Esmeraldas refinery. Fig. 3. Example of

Fig. 2. Candidate sites for the experimental campaign inside the Esmeraldas refinery.

Fig. 2. Candidate sites for the experimental campaign inside the Esmeraldas refinery. Fig. 3. Example of

Fig. 3. Example of field devices used for multi-channel RSS data collection (top). Time-frequency RSS data-set sample (bottom).


In this section, we highlight two specific deployment case studies in the Esmeraldas refinery using WirelessHART com- pliant devices. In particular, the tests are meant to verify the (pre-deployment) analysis carried out in the same sites based on the RSS model highlighted in Sect. 2. The first test site chosen for the deployment is the Effluents area (see Fig. 2):

with size approx. 9000 sqm, it is has been currently renewed

to carry out the various types of treatment that are usually practiced by refineries for treating waste-water. The use of a wireless system could therefore augment the existing wired equipment, and provide a flexible and cost-effective solution for monitoring without the costly (and unfeasible for logistic reasons) re-wiring over the existing plant. The 15.000 sqm Setria site inside the Esmeraldas oil&gas refinery has been chosen as an additional test location over a larger size open- area field. It consists of a section of an oil depot characterized by 6 large-size concrete fuel tanks. Revamping activities are currently planned to augment the existing control systems. For both sites we used IEC 62591 compliant wireless equip- ment: 1) Gateway (G) model 1420 (developed by Emerson) with transmit power: 4 dBm, omnidirectional antenna: ½ wavelength 6 dB gain; 2) HCF sniffer – FieldComm used for data collection; 3) field devices (developed by E-Aware Technologies). Link types are evaluated at first by using the 2D map of the plant as well as by visual inspection to compute the link path stability P S, (or packet successful rate). This is percentage of transmitted packets that have successfully reached their destination over a given path. For a link implementing the channel hopping pattern F, successful rate can be modelled in terms of RSS (1) as:

P S, = Pr [min iF (s i, ) > β] (2) being β = 85 dBm the receiver sensitivity for IEEE 802.15.4 devices. Path stability is a lower-bound to network reliability as it does not account for automatic MAC-layer re-transmissions. Modeling of path stability (2) is thus a conservative choice as far as pre-deployment assessment is concerned. In addition to path stability, other network metrics are tracked: 1) burst rate, namely the transmission interval to the Gateway; 2) missed update since initially joining the net- work; 3) number of nearby devices with reliable connections. Figure 5 summarized the performance results from the tests in Effluents area. In particular 3 topologies are considered. In topology 1, field devices labeled as E01, E02 present a highly reliable connection with the Gateway as Type IV but in short range. Device E04 has 1 missed update and a low path stability (1.7%) with its neighbor device E02 (being a Type V link). Device E05 has 2 neighbors, a high path stability (99.5%) is observed with device E02 (Type III). In topology 2, devices E01 and E02 have a high reliability with 0 missed updates registered during network life. Device E04 has 13 missed updates for 660 data packets transmitted during this setup. Device E05 failed to establish a successful connection with the Gateway as the link is obstructed by a concrete building (Type V) that is surrounding the Gateway side. In the third topology, devices E01, E02 and E04 have a high reliability with 0 missed updates (in 36 minutes). Being located behind a concrete building, device E04 may present more failed transmissions than E01 and E02, since the path stability is about 70% for all neighbors. Device E05 has a lot of messages delivered out of time (1041 missed updates):

it is possible to conclude that device E05 is not deployed in a favorable situation, since it has only one neighbor with

Fig. 5. Example topologies in the Effluents area and WirelessHART network performance analysis. a good link

Fig. 5. Example topologies in the Effluents area and WirelessHART network performance analysis.

a good link (device E04). Figure 6 summarized the results from the WirelessHART layout verification in the Setria area. For topology 1, as confirmed by the model, each device has at least one neighbor with a high path stability and thus high reliability. Device E05 is the only device with no direct connection with the Gateway: 1 missed update is observed (from 32 publications). The worst path stability is between device E02 and the Gateway (49%), however packets reaching device E02 can follow two alternative paths (through E02-E01 and E04). In the second topology, path stability is lower on average. However, excluding Type V links, all the equipment are reliably connected with the Gateway in multi-hop mode, with 0 missed updates (217 minute registration). For stress- testing purpose, device E05 was finally configured with a burst time of 16 seconds, while 13 missed updates have been observed during 144 minutes.


The paper presented the results of an extensive experimental measurement campaign in the oil refinery Esmeraldas. A novel channel model is proposed to characterize site-specific spatial, frequency and time-domain fading features of the wireless channel response that are shared by links conforming

Fig. 6. Example topologies in the Setria area and WirelessHART network performance analysis.
Fig. 6.
Example topologies in the Setria area and WirelessHART network
performance analysis.

to the same obstruction type. The measurements analysed highlight an accuracy of the proposed model with an error on predicted link path stability below 6dB for the considered sites. Results from the post-layout deployment verification based on a WirelessHART system confirm the effectiveness of the virtual planning approach.


REFERENCES M. R. Palattella, et al., “Internet of Things in the 5G era: enablers,

architecture, and business models,” IEEE Journal on Selected Areas in Communications, vol. 34, no. 3, Mar. 2016. [2] A. Bader, H. Ghazzai, A. Kadri, and M. S. Alouini, “Front-End Intelligence for Large-Scale Application-Oriented Internet-of-Things,” IEEE Access, vol. 4, 2016. [3] S. Savazzi, V. Rampa, U. Spagnolini “Wireless Cloud Networks for the Factory of Things: Connectivity Modeling and Layout Design,” IEEE Internet of Things Journal, vol. 1, no. 2, pp. 180-195, Apr. 2014. [4] L. D. Xu, W. He and S. Li, “Internet of Things in Industries: a survey,” IEEE Trans. on Ind. Inform., vol. 10, no. 4, pp. 2233-2243, Nov. 2014. [5] IEEE Standard for Local and metropolitan area networks - Part 15.4:

Low-rate wireless personal area networks (LR-WPANs) amendment 1:

MAC sublayer, IEEE Std 802.15.4e™-2012, Apr. 2012. [6] Standard IEC 62591: 2010, Industrial communication networks - Wireless communication network and communication profiles - Wire- lessHART™, Edition 2.0, 06 Nov. 2015. [7] Standard ISA100.11a-2009 “Wireless systems for industrial automation:

process control and related applications,” ISA, July 2009. [8] S. Petersen, S. Carlsen, “WirelessHART versus ISA100.11a: the format war hits the factory floor,” IEEE Ind. Electronics Mag., vol. 5, no. 4, pp. 23-34, Dec. 2011. [9] L. Tang, et al., “Channel Characterization and Link Quality Assessment of IEEE 802.15.4-Compliant Radio for Factory Environments,” IEEE Trans. on Ind. Inform., vol. 3, no. 2, May 2007. [10] S. Savazzi, S. Guardiano, U. Spagnolini, “Wireless sensor network modeling and deployment challenges in oil and gas refinery plants” International Journal of Distributed Sensor Networks, Hindawi vol. 2013, Article ID 383168, 17 pages, 2013. doi:10.1155/2013/383168. [11] S. Savazzi, V. Rampa, F. Vicentini, M. Giussani, “Device-Free Human Sensing and Localization in Collaborative HumanRobot Workspaces:

A Case Study,” IEEE Sensors Journal, vol. 16, no. 5, pp. 1253-1264, March, 2016. [12] D. J. Y. Lee et al., “Propagation prediction in and through buildings,” IEEE Trans. on Vehicular Technology, vol. 49, no. 5, Sept. 2000.