On the Impact of Local Processing for Motor Monitoring Systems in Industrial Environments Using Wireless Sensor Networks

Abstract

This paper presents a theoretical study for verifying the impact of using smart nodes in motor monitoring systems in industrial environments employing Wireless Sensor Networks (WSNs). Structured cabling and sensor deployment are usually more expensive than the cost of the sensors themselves. Besides the high cost, the wired approach offers little flexibility, making the network deployment and maintenance a complex process. In this context, wireless networks present a number of advantages compared to wired networks as, for example, the ease and speed of deployment and maintenance and the associated low cost. However, WSNs have several limitations, such as the low bandwidth and unreliability, especially in harsh environments (e.g., industrial plants). This paper presents a theoretical study on the performance of WSNs for motor monitoring applications in industrial environments, taking into account WSNs' characteristics (i.e., unreliability and communication and processing latency). The results obtained through mathematical models were analyzed together with experimental results, and it was demonstrated that employing intelligent nodes with local processing capabilities is essential for the applications under consideration, because it reduces the amount of data transmitted over the network allowing monitoring even in scenarios with high interference rate, paying off the extra latency resulting from local processing.

1. Introduction

The optimization of power usage, by decreasing costs and reducing environmental impact, has been of great concern to various sectors in the society. In this context, intelligent and low cost industrial automation systems for motor monitoring can be a very useful tool, by reducing electricity consumption in the industrial sector. Electric motors are used in most industrial processes and account for more than two-thirds of electricity consumption in this sector [1].

Wireless networks present a number of advantages compared to wired networks as, for example, the ease and speed of deployment and maintenance, in addition to their low cost. Wireless Sensor Networks (WSNs) extend the advantages with their self-organization and local processing capabilities. Therefore, WSNs appear as a flexible and inexpensive solution to build industrial monitoring and control systems [2, 3].

Mainly due to high cost of monitoring systems, in general only motors over $500$ HP are monitored. However, motors that have less capacity account for $99.7 %$ of the motors in operation, accounting for about $71 %$ of power consumption [4].

Among the parameters that can be monitored, torque is crucial to prevent motor failures, avoiding losses in the production process [5]. The shaft torque can be estimated from the motor electric signals. Although this technique is less accurate when compared to the direct measurement methods, it is less invasive and allows shaft torque monitoring using low cost voltage and current sensors that can be easily integrated into a WSN [6]. The shaft torque can also be used for estimating motor efficiency, which is also the most important factor when targeting power consumption reduction [7, 8].

Other methods for fault analysis in motors may also be employed, like methods based on electrical signature analysis, which verify variations in voltage and current signals, in order to relate the signature characteristics with electrical and mechanical conditions [9–11] or methods based on vibration analysis using accelerometers, which are based on parameters such as displacement, velocity, and acceleration for detecting faults in motors [12, 13].

However, employing WSNs in automation systems in industrial environments presents a number of challenging aspects. Wireless networks have unreliable communication links [14], which can be worsened due to noise and interference in the communication spectrum range. The unreliability of the transmission medium in wireless networks makes it difficult to define quality of service guarantees.

Furthermore, sensor nodes must have low cost, which results in a number of constraints such as low bandwidth and low processing power. For instance, the IEEE 802.15.4 standard presents a nominal throughput of $250$ kbps. However, even considering this limitation, such standard can possibly comply to the requirements of many industrial monitoring systems.

Industrial monitoring systems need to measure signals that change rapidly, in a dynamic manner [15]. Applications such as efficiency monitoring and fault detection in induction motors fit this type of application. Due to the limitations of WSNs, mainly regarding the low bandwidth and the lack of reliability in the transmissions, the implementation of such systems becomes even more challenging.

In this context, this paper aims at analyzing the impact of using smart nodes for motor monitoring applications in industrial environments employing WSNs, through theoretical and experimental studies. By performing local processing, it is possible to mitigate the intrinsic limitations of WSNs. However, the processing latency should be considered, especially because the sensor nodes usually have low processing power and reduced memory.

Mathematical models for estimating the information delivery rate in several scenarios are properly described, and the results obtained from the models were analyzed together with experimental results. To perform this analysis, we take into account applications for torque and efficiency monitoring in induction motors. This type of motor corresponds to about $90 %$ among all motors employed in the industry [1]. From these studies, it can be observed that the use of embedded processing in WSNs' nodes allows the monitoring even in high interference scenarios, whereas it would be impractical without local processing.

The proposed mathematical model can be used to assess the impact of using local processing with respect to the information delivery rate of different applications. Some factors affect this metric as, for example, processing delay, the amount of data gathered from sensors, and the packet error rate. Thus, it may be advantageous or not to use local processing, depending on these parameters. Specifically, for the motor monitoring applications analyzed in this paper, it is shown that local processing effectively improves the WSN performance, with a direct impact on the application itself.

2. Industrial Wireless Sensor Networks

In an industrial WSN, sensor nodes are deployed in machinery for monitoring critical parameters such as vibration, temperature, pressure, and efficiency. Measurements are transmitted wirelessly to a sink node, which later provides the gathered information for analysis in a central station. Based on this information, it is possible to repair or replace devices before major damages take place [16].

Although WSNs have several advantages, the deployment of this technology presents some challenges. Wireless communication is inherently unreliable and is subject to a larger number of transmission errors when compared to wired networks, mainly due to channel failures and interference. Nodes can suffer interference from the coexistence with other nodes in the network, from the coexistence with other networks, and from other technologies operating in the same frequency range.

In industrial environments, there may be other sources of interference such as thermal noise, interference from motors and devices that cause electromagnetic interference in the band used for communication [16]. Besides, in industrial environments there is usually a large amount of metallic objects, which can impair the communication. In general, industrial wireless systems are prone to high error rates and often with high variance [17].

2.1. IEEE 802.15.4 Standard

The IEEE 802.15.4 standard was designed for WSN applications. This standard provides wireless communication with low power consumption and low cost for monitoring and control applications that do not require high bandwidth. Compared to other standards, such as IEEE 802.11 (WiFi) and IEEE 802.15.1 (Bluetooth), the IEEE 802.15.4 standard has advantages related to energy consumption, scalability, reduced time for node inclusion, and low cost [18].

The IEEE 802.15.4 standard defines the physical layer and the MAC layer. It has three frequency ranges (868 MHz, 915 MHz, and 2.4 GHz). As these bands are unlicensed, radios share the communication medium with devices that implement other technologies. For example, the IEEE 802.11 and the IEEE 802.15.4 standards both operate in the 2.4 GHz frequency range. However, as the spectrum is divided into channels, it is possible to multiple networks operating simultaneously, without interfering much with each other.

The network topology can be organized in three ways: star, mesh, and tree. However, the standard does not define the network layer. There are two types of nodes: full function device (FFD) and reduced function device (RFD). The FFD nodes can act both as network coordinator or end node. The coordinator is responsible, among other functions, for the initialization, address allocation, network maintenance, and the recognition of all other nodes. RFD nodes work only as end nodes, which are responsible for the functions of sensing or action. FFD nodes can also perform the function of intermediate routers between nodes, without the intervention of the coordinator [6].

The IEEE 802.15.4 radios reach a nominal throughput of $250$ kbps when operating in $2.4$ GHz band. Nevertheless, there is an underutilization of the channel due to the CSMA/CA mechanism, for each packet awaits at least one backoff period before transmission. Experimental studies by Lee [19] showed that the real maximum throughput is approximately $153$ kbps. For many WSN applications, this throughput is enough, but for applications, where sensors acquire a large amount of data in short periods of time, the available throughput may become insufficient. This problem can be mitigated with the addition of local processing in sensor nodes, which reduces the amount of data transmitted over the network.

Some upper layer protocols have been proposed for IEEE 802.15.4 based WSNs. Following, we present a short description for some of such protocols, as well as implementation issues that impact on the communication reliability.

2.1.1. ZigBee

The most employed protocol in WSNs' applications is the ZigBee protocol [12, 18, 20–24]. This protocol has a number of desirable characteristics for WSN applications as, for example, low power consumption and low cost.

ZigBee's network protocol supports the three available topologies (i.e., star, tree, and mesh), allowing the implementation of an ad hoc WSN. When using a mesh topology, the routing process becomes more complex, but the robustness and fault tolerance of the network increase, due to the ability to find and maintain routes.

ZigBee does not implement mechanisms to mitigate the coexistence problem. For instance, it does not switch channels during periods of high contention and interference; instead, it adopts only a low duty-cycle and medium access control algorithms to minimize data losses from packet collisions [25].

2.1.2. MiWi

The MiWi protocol [26], developed by Microchip, is an alternative for small networks with at most $1000$ nodes. Theoretically, a ZigBee network can contain up to $65536$ nodes although, in practice, it is not recommended having more than $3000$ nodes in a single network [23]. Another factor that limits the size of MiWi networks is the maximum number of hops allowed (i.e., four hops).

An interesting feature that sets MiWi apart from ZigBee is MiWi's ability to perform dynamic channel switching. This mechanism, called Frequency Agility, is optional and allows moving the network to operate into a different channel once the current operating conditions are not favorable. To set the new channel, a node, called initiator, performs an energy scan in all channels, for finding the least busy one. After that, the initiator broadcasts a message to all nodes conveying information regarding the new channel. If a node does not receive the broadcast message from the initiator (probably due to a transmission failure), it performs a resynchronization after many recurring failed transmissions. Resynchronization consists of scanning all channels to find out the channel currently in use by the network.

Although this mechanism tends to improve communication quality in general, it incurs an overload on initiators. The network may spend much time without providing new data, if the initiators perform scans very often. Another important factor is the scanning period. In case it is too long, it is possible to obtain greater accuracy in estimating the best channel; however, the network will be idle for too long. On the other hand, if the scanning period is too short, the network spends little time idle, but then it might present lower accuracy in estimating the best available channel.

Although the Frequency Agility mechanism is provided by the MiWi suite, there is a strong dependence on the application layer, since the application determines when a scan and a possible channel switch must occur.

2.1.3. WirelessHART

The WirelessHART standard is considered the first open communication standard designed for wireless industrial monitoring and control applications [27]. The other standards, such as ZigBee and Bluetooth, do not completely meet the requirements of industrial applications.

The WirelessHART is based on the physical layer of IEEE 802.15.4 but implements its own link layer. It is based on the $2.4$ GHz ISM band but adopting only 15 channels, because channel 26 is not allowed in some countries [28]. Instead of using CSMA/CA as defined by the IEEE 802.15.4 standard, it implements an MAC layer with Time Division Multiple Access (TDMA). By using TDMA, it reduces collisions and power consumption [29].

To improve the coexistence with other networks and other technologies based on the $2.4$ GHz band, the standard implements a frequency hopping mechanism. Another mechanism, called Blacklisting, is also defined. Using this mechanism, the channels with high level of interference are avoided. However, the blacklist is not done automatically but by a network administrator.

In a WirelessHART network all nodes on the network must be able to perform routing. It is used a mesh topology with redundant routes. This feature allows increasing the reliability and fault tolerance, since redundant routes can replace obstructed paths. The routes are generated by a central entity (network manager). The network manager is also responsible for scheduling time among the nodes of the network, ensuring the correct operation of the TDMA mechanism.

WirelessHART networks are centralized, because the entire network operation is managed by a single entity. In MiWi or ZigBee networks, end nodes discover their route to the destination. Moreover, each node can decide when to initiate a transmission independently, using the CSMA/CA mechanism. In WirelessHART, the network manager defines the moment when each node should transmit or receive packets.

Petersen and Carlsen [28] performed studies on the performance of WirelessHART radios. The performance of these radios was also verified when subject to interference from three IEEE 802.11 g access points (operating in channels 1, 6, and 11). The results showed that, during interference periods, nodes experienced an average packet error rate of $27.2 %$ . However, in the experiment all channels were enabled. With a better management of blacklists, the performance could be improved. However, radios do not have the ability to enable and disable channels automatically.

A large latency of around two seconds for the network operating without interference and around $2.7$ seconds when operating in coexistence with the access points was observed. Thus, we can see that the overhead due to the mechanisms implemented to increase reliability implies a large latency.

WirelessHART is a recent standard, released in 2007. Until 2009 there was no complying component available on the market [29]. However, more experimental studies should be conducted to verify the performance of WSN that comply with this standard.

2.1.4. ISA100

The Instrumentation, Systems, and Automation Society (ISA) idealized the ISA100 standard [30], which is also designated for industry. As the WirelessHART, the ISA standard is based on the IEEE 802.15.4 physical layer but defines its own MAC layer. The MAC layer characteristics are very similar to the characteristics presented on WirelessHART. It also applies TDMA and frequency hopping to improve reliability. The network layer is a bit different, since it uses header formats based on the IP protocol [27].

2.1.5. Comparison among the Standards

Table 1 presents a brief comparison among the standards under consideration with respect to some aspects.

Table 1

Comparison among the standards.

Standard	ZigBee	MiWi	WirelessHART/ISA100
Supported frequency bands	868 MHz, 915 MHz, and 2.4 GHz	868 MHz, 915 MHz, and 2.4 GHz	2.4 GHz
Physical layer	IEEE 802.15.4	IEEE 802.15.4	IEEE 802.15.4
MAC layer	IEEE 802.15.4	IEEE 802.15.4	Custom
Medium access mechanism	CSMA/CA	CSMA/CA	TDMA
Coexistence mechanisms	—	Frequency Agility	Frequency hopping and Blacklisting
Definition of routes	Distributed	Distributed	Centralized
Redundant routes	No	No	Yes

ZigBee is the only protocol that presents no special mechanism for coexistence. The MiWi protocol provides a mechanism for switching channels, but there is still much dependence on the application layer.

On the other hand, the WirelessHART and ISA100 standards offer more complex mechanisms to improve the coexistence for industrial WSN. The main drawbacks are the heavy network centralization and the high communication latency, which results in a low information delivery rate [31]. Furthermore, from [28] we can see that, if there is no proper blacklist management, network performance can suffer a significant drop in the presence of interference.

WirelessHART and ISA100 also implement redundant routes, which can increase the reliability, since multiple paths may be defined for data transfer. However, as this mechanism is implemented at the network layer, it can also be implemented in radios that comply with the physical and MAC layers of IEEE 802.15.4.

Although WirelessHART and ISA100 are intended for industrial WSN applications, these are pretty new standards, and they do not have high availability of complying transceivers on the market. On the other hand, there is a wide availability of transceivers that implement the physical and MAC layers of IEEE 802.15.4 and are compatible with ZigBee and MiWi.

The results of this paper consider radios fully compatible with IEEE 802.15.4. However, it is important to notice that they are totally applicable to other protocols. The use of local processing still remains very important, due to the low throughput of the IEEE 802.15.4 physical layer. In addition to that, WirelessHART and ISA100 present reliability concerns [28].

3. Motor Monitoring Systems

This section describes some works [6–8, 15, 32–35] focusing on the application of WSN in industrial environments. There is a relatively small amount of work towards the development of monitoring and control systems in industry based on WSNs. This is due to the complex requirements of the system and severe work environment [15]. Some recent works address the performance evaluation of radios in an industrial environment [2, 25, 36–38], while some other works address the challenges of using WSN technology in industry [16, 17, 39–41].

Salvadori et al. [32] proposed a digital system for evaluation of power usage, diagnosis, control, and supervision of electrical systems employing WSNs. The system is based on two hardware topologies responsible for signal acquisition, processing, and transmission: intelligent sensor modules (ISMs) and remote data acquisition units (RDAUs). However, only wired communication RDAUs are used to perform acquisition of voltage and current of motors. ISMs were used only for temperature measurements. The work focuses mainly on the energy consumption of sensor nodes and does not provide detailed studies on transmission errors and communication channel quality.

Hsu and Scoggins [42] presented a method to estimate motor efficiency from the air-gap torque, which is obtained from the motor electrical signals (current and voltage). It is the noninvasive method for determining torque and efficiency that has less uncertainty [43]. Recent works have also used this technique to estimate the efficiency and torque of induction motors [6–8, 35]. These studies have also employed WSN for data transmission.

Lu et al. [7, 8] identify in their work the synergies between WSNs and analysis of motors based on electrical signals, also following a noninvasive pattern. They propose a scheme to apply WSNs for online and remote monitoring and fault diagnosis of industrial motors.

The main limitation of the work presented in [7, 8] is derived from the low throughput provided by the WSN based on the IEEE 802.15.4, since the proposed system does not employ local processing. Thus, it is necessary to transmit a large amount of data to estimate the desired parameters. This limits, among other things, the acquisition rate from the sensors, which consequently limits the accuracy of the estimation. In a WSN with a large number of nodes, the situation becomes even worse, since all nodes share the same physical medium. Moreover, the wireless networks are inherently unreliable, which can result in transmission errors, affecting the estimation process.

Hou and Bergmann [33] developed a motor monitoring system using WSN with local processing. A prototype was implemented and validated in a single-phase induction motor in laboratory. Motor current signature analysis (MCSA) is employed in this application, where motor stator current signal waveforms are given under different working conditions. Using local processing, a reduction of around $90 %$ was obtained in the amount of data transmitted for analyzing the motor.

Hou and Bergmann [15] also developed a system for fault detection in motors using accelerometers and WSN. In this system, a reduction of $99 %$ was obtained in the amount of data transmitted for performing the failure analysis task when using local processing. However, both [15, 33] do not performed a detailed analysis of the WSN performance. For example, the information delivery rate was not verified, and the experiments were not performed in realistic environments. When employing local processing in WSN systems, it is also important to consider processing latency, as well as the quality of the communication medium, to effectively verify the gain obtained from local processing.

Hu [34, 35] presents a DSP-based system for motor monitoring using the air-gap torque method and WSN for data transmission. The estimation of various parameters such as power factor, efficiency, speed, and torque was proposed. However, the tests were conducted in laboratory, which does not characterize a realistic experiment. As in [15, 33] there was no detailed study on the impact of using local processing on the WSN performance.

In a previous work [6], we developed an embedded system integrated into a WSN for online dynamic torque and efficiency monitoring in induction motors. The air-gap torque method was employed for estimating the shaft torque and motor efficiency. The computations for estimating the targeted metrics are performed locally and then transmitted to a monitoring base unit through an IEEE 802.15.4 WSN.

Experimental tests were performed to analyze the torque values obtained by the system and then compared with torque values based on the workbench dynamic model. The paper also showed an experimental study aiming at identifying the correlation between spectral occupancy and packet error rate (PER) for the proposed WSN. The experiments were conducted inside a shed, with typical characteristics of industrial environments.

The study demonstrated that the addition of new interference sources can significantly affect the spectral occupancy by also having a direct impact on the communication performance. Even for harsh condition scenarios, the system was able to provide useful monitoring information, since all processing is done locally (i.e., only the computed metric is transmitted over the network). Without local processing, it might be impossible to use the WSN technology for this particular application, considering an unreliable transmission medium.

In [6] a theoretical study was conducted to determine the number of packets transmitted over the network, comparing the approaches with and without local processing, as well as with and without packet retransmission. In this paper, we extend the mathematical model described in [6], to verify the information delivery rate in different scenarios, with varying PER and taking into account processing delay. In addition to that, we analyze the impact of retransmissions and aggregation. The results obtained through the analytical model were analyzed together with experimental results to show the impact of using local processing for motor monitoring applications based on WSN in industrial environments.

4. System Description

In this section, we describe the system designed for performing our studies. The system (Figure 1) consists in a WSN running an application for torque and efficiency measurements in induction motors. More details about this system can be found in [6].

Figure 1

Proposed WSN [6].

End nodes are composed of embedded systems located close to the electric motors. Sensors provide motor voltage and current measurements, and the embedded system performs the required processing for computing torque and efficiency. After local processing, the metrics are transmitted to the base station through the WSN.

Internally, end nodes (Figure 2) are composed of current and voltage sensors and an acquisition and data processing unit (ADPU), which is responsible for data acquisition and A/D conversion, besides data processing. Finally, an IEEE 802.15.4 transceiver is used for communication in the WSN.

Figure 2

: Embedded system [6].

Torque and efficiency are computed from the motor electrical signals, using the air-gap torque method, since this is the noninvasive method that presents less uncertainty [43]. To perform the estimation using this method in a three-phase induction motor, it is necessary to acquire two voltage signals and two current signals. More details about the air-gap torque method can be found in [4, 6–8].

Figure 3 shows a flowchart of the embedded system's operation. When the system starts, all embedded system's parameters are configured, such as the A/D converter and the network parameters. Then, the system performs a cycle that includes the acquisition of current and voltage values, sampling shaping (e.g., offset removal), data processing to estimate torque and efficiency, and transmission of the target values through the WSN.

Figure 3

Flowchart of the embedded system.

5. Theoretical Analysis

5.1. Information Delivery Rate

In this section we present a model to analyze the latency for information delivery (torque and efficiency values) in the WSN. The propagation time was considered null in the model. Thus, the latency for information delivery by the WSN is computed only in terms of transmission time and processing time. We considered a WSN with star topology, which is the most prevalent topology today [29].

The latency $(L)$ for receiving one torque value and one efficiency value from the WSN, in the scenarios without retransmission, is given by the following equation:

\begin{matrix} L = Q_{t} (Q_{p} (\frac{P_{size}}{W} + e (p)) + L_{p}), \end{matrix}

(1)

where

Q_{t}

is the number of trials until information is delivered;

Q_{p}

is the number of packets transmitted in each trial;

P_{size}

is the number of bits of each packet; W is the network bit rate;

e (p)

is the error due to the assumptions of the model;

L_{p}

is the processing latency.

Note that the value of W represents the number of bits transmitted per second, regardless of transmission errors.

The latency $(L_{r})$ for receiving one torque value and one efficiency value from the WSN, in the scenarios with retransmission, is given by the following equation:

\begin{matrix} L_{r} = Q_{t} (Q_{p} (\frac{P_{size}}{W} + \frac{H_{ack}}{W} + e (p))) + L_{p}, \end{matrix}

(2)

where

H_{ack}

is the number of bits of an acknowledge packet (ACK).

5.1.1. Impact of Packet Loss

The transmission of a packet consists on a Bernoulli event with successful probability p, with a number of trials until the first success as defined by a geometric distribution. In a geometric distribution, the average number of events until the first success is $1 / p$ .

Thus, the probability of successfully transmitting the data necessary for estimating the torque and efficiency is $p^{Q_{p}}$ , and the number of attempts before the first success is $Q_{t} = 1 / p^{Q_{p}}$ , considering a scenario without retransmission of lost packets. As in each attempt $Q_{p}$ packets are transmitted, then we have an average of $Q_{p} / p^{Q_{p}}$ transmissions per efficiency and torque measures.

Considering that the probability of successfully transmitting an acknowledgment packet is also p, then the probability of successfully transmitting a packet in the scenario with retransmission is $p^{2}$ , and the number of trials until the first success is $1 / p^{2}$ . Thus, the average number of transmissions until the first hit is $Q_{t} = Q_{p} / p^{2}$ .

5.1.2. Scenario without Local Processing

Each node in the WSN obtains data from two current sensors and two voltage sensors. Current and voltage signals have a frequency of 60 Hz, and computing one value of air-gap torque requires a complete cycle of voltage and current. Therefore, the total number of bits for estimating the target values $(Q_{b})$ can be defined according to the following equation:

\begin{matrix} Q_{b} = \frac{4 b T_{a}}{60}, \end{matrix}

(3)

where

T_{a}

is the acquisition rate of the analog to digital converter (ADC) and b is the amount of bits per sample (which depends on the resolution of the ADC).

Then, without using local processing, the number of packets, $(Q_{p})$ , required for transmitting the data acquired by the sensors is given by the following equation:

\begin{matrix} Q_{p} = ⌈ (\frac{Q_{b}}{C_{m}}), \end{matrix}

(4)

where

C_{m}

is the maximum payload of a packet (defined by the standard).

The average size of the packet payload, $(C_{avg})$ , is given by the following equation:

\begin{matrix} C_{avg} = \frac{Q_{b}}{Q_{p}} . \end{matrix}

(5)

Thus, in the scenario without local processing $P_{size} = C_{avg} + H$ , where H is the number of bits of the packet header.

In the scenario without local processing, only the time to acquire the values, equal to $1 / 60$ seconds, was considered as the total processing latency; that is, $L_{p} = 1 / 60$ .

As demonstrated in Section 5.1.1, $Q_{t} = 1 / p^{Q_{p}}$ for the scenario without retransmission, and $Q_{t} = 1 / p^{2}$ for the scenario with retransmission.

Therefore, the latency for receiving one torque value and one efficiency value from the WSN, in the scenario without local processing and without retransmission $(L_{n lp})$ , is given by the following equation:

\begin{matrix} L_{n lp} = \frac{1}{p^{Q_{p}}} (Q_{p} (\frac{C_{avg} + H}{W} + e (p)) + \frac{1}{60}) . \end{matrix}

(6)

The latency for receiving one torque value and one efficiency value, in the scenario without local processing and with retransmission $(L_{r_{n lp}})$ , is given by the following equation:

\begin{matrix} L_{r_{n lp}} = \frac{1}{p^{2}} (Q_{p} (\frac{C_{avg} + H}{W} + \frac{H_{ack}}{W} + e (p))) + \frac{1}{60} . \end{matrix}

(7)

5.1.3. Scenario with Local Processing

With local processing, the nodes acquire $Q_{b}$ bits to estimate torque and efficiency, but it is necessary to transmit only a constant number of bits ( $V_{s}$ ), regardless of the values of b and $T_{A}$ . With local processing, only one packet is needed to transmit the information; that is, $Q_{p} = 1$ . Additionally, it is possible to aggregate various torques and efficiency values into a single packet.

Let $V_{a}$ be the amount of values aggregated into the same packet; we have that $P_{size} = V_{a} V_{s} + H$ , where $V_{a} V_{s} \leq C_{m}$ .

Let $P_{t}$ be the processing time to estimate the values of torque and efficiency; we have that $L_{p} = V_{a} * (P_{t} + (1 / 60))$ .

Since $Q_{p} = 1$ , we have that $Q_{t} = 1 / p$ for the scenario without retransmission, and $Q_{t} = 1 / p^{2}$ for the scenario with retransmission.

Therefore, the average latency for receiving one packet from the WSN in the scenario with local processing and without retransmission $(L_{lp})$ is given by the following equation:

\begin{matrix} L_{lp} = \frac{1}{p} (\frac{V_{a} V_{s} + H}{W} + e (p) + V_{a} (P_{t} + \frac{1}{60})) . \end{matrix}

(8)

In the scenario with retransmission and with local processing, the average latency $(L_{r_{lp}})$ for receiving one packet from the WSN is defined according to the following equation:

\begin{matrix} L_{r_{lp}} = \frac{1}{p^{2}} (\frac{V_{a} V_{s} + H}{W} + \frac{H_{ack}}{W} + e (p)) + V_{a} (P_{t} + \frac{1}{60}), \end{matrix}

(9)

where

e (p)

is the error on the estimation of the transmission time and processing time, due to the assumptions. The mathematical model does not take into account propagation delay and the processing overhead regarding the MAC protocol, among other factors.

5.1.4. Error Estimation

The mathematical model described here still presents some limitations, due to the assumptions made. In the scenarios with retransmission, the processing time from the acknowledge mechanism and the timeout for lost packets was also not considered.

It is also important to consider some relations, which are not directly supported by the models. Although the value of W is not directly affected by the value of p, since W is the number of bits transmitted per second, regardless of transmission errors, there is some correlation between these parameters. W is affected by the CSMA/CA mechanism, and when the interference level in the environment is high or the network has many nodes, the packet error rate is likely high. At the same time, there is an increase in backoff periods, having a direct impact on W. Therefore, some scenarios are very unlikely to happen as, for example, the scenario with $p = 0.1$ and $W = 150$ kbps (maximum real throughput).

The relation between packet size and packet error rate was not taken into account in the analytical model. Thus, in practice the WSN performance will be lower than the one computed through the models. However, the model is useful for comparing the performance of several possible approaches. Particularly, it can be observed that the use of local processing can significantly increase the WSN performance. It should also be noted that when local processing is not used, the real performance suffers even more with the assumptions made, because there are more and larger packets being transmitted over the network.

To verify the precision of the models, some tests were performed using the embedded system developed (described in Section 4) and a coordinator node, to measure the information delivery rate and compare the results with the results obtained using the models. It is important to note that only some scenarios were analyzed. It was not possible to validate the scenarios with retransmission, since the radios used in our testbed do not provide information about transmission errors. Thus, it was not possible to identify the packet error rate. When retransmission is not used, to obtain the packet error rate, it is only necessary to verify the number of packets transmitted and the number of packets received at the upper layer.

During these experiments, the radios were configured to operate on channel $11$ . First, we obtained results for the scenario without packet loss, given that there was no interference source in the environment. To obtain values for the scenarios with packet loss, we activate a WiFi network in the environment, operating on channel $1$ , which coexists with channel 11 of IEEE 802.15.4 in the 2.4 GHz band. Although no significant traffic has been injected in the WiFi network, it was possible to observe a drop in the communication performance of the IEEE 802.15.4 radios. The embedded system was configured to transmit 1000 packets, and the coordinator was instrumented to calculate the information delivery rate. The time for receiving the packets was also measured by the coordinator.

With the results obtained from the experiments, it was possible to compute the values of $e (p)$ for the experiments under consideration. As discussed earlier, there is a correlation between the values of p and W. $e (p)$ corresponds to the additional time of transmission and processing overhead related to the errors in the models. Therefore, the value of $e (p)$ must compensate these assumptions. Thus, a model was defined to estimate these values for each scenario from the value of p, since the smaller the value of p, the greater the error.

To relate these two parameters a nonlinear regression (exponential) was performed. Through these regressions, we obtained (10), (11), (12). Since the error is different in each scenario, a model for each one was obtained

\begin{matrix} e_{n lp} (p) = 0.0073 {(0.202)}^{p}, \end{matrix}

(10)

\begin{matrix} e_{lp} {(p)}_{1} = 0.0061 {(0.363)}^{p}, \end{matrix}

(11)

\begin{matrix} e_{lp} {(p)}_{18} = 0.029 {(0.115)}^{p}, \end{matrix}

(12)

where

e_{n lp} (p)

is the error for the scenario without local processing.

e_{lp} (p)_{1}

is the error for the scenario with local processing, and

V_{a} = 1

e_{lp} (p)_{18}

is the error for the scenario with local processing, and

V_{a} = 18

In the next section, we will analyze the model accuracy, compared with values obtained by the experiments.

6. Results

Using the generic mathematical model described above, we will conduct an analysis considering the IEEE 802.15.4 characteristics. In this standard, $C_{m} = 118$ bytes. $4$ bytes are necessary to store one torque value and one efficiency value (i.e., $V_{s} = 32$ bits). To obtain good accuracy from a simple numerical integration method, such as trapezoidal (the one used to implement the algorithm), it requires a sample rate greater than 2 kHz [4]. Based on these observations, $T_{a}$ was set to $3$ KHz, and $b = 10$ bits.

6.1. Number of Transmitted Packets

Figure 4 shows the number of packets that must be transmitted, on average, for the four scenarios. In the chart, the values for $p = 0.2$ and $p = 0.1$ were omitted, in order to improve the visualization. For $p = 0.1$ , without local processing and without retransmission, we have $3000$ transmissions, while in the approach with local processing and without retransmission only $10$ transmissions are required on average. In the chart, we counted only the data transmissions; that is, the transmission of the ACK packets was not taken into account.

Figure 4

Average number of transmissions.

It is important to notice that as torque and efficiency values occupy only $4$ bytes, we can aggregate multiple values in a single packet. Besides that, due to an increase in the number of transmitted packets, considering an approach without local processing, the packet error rate tends to increase; that is, the value of p tends to be lower.

6.2. Latency

At first, we considered $e (p) = 0$ in all scenarios. Figures 5 and 6 show the delivery latency for one torque value and one efficiency value, for the scenarios with local processing and without local processing, respectively. In Figure 6 $V_{a} = 1$ and $L_{p} = 14$ ms. The other parameters were defined according to the IEEE 802.15.4 standard. We conducted a performance test with the embedded system implemented in this work (considering $T_{a} = 3000$ Hz) to observe the processing time for the metrics under consideration. The results pointed out a processing time of around 14 ms.

Figure 5

Latency to receive information in the WSN (without local processing).

Figure 6

Latency to receive information in the WSN (with local processing).

Latency values are shown for various values of p and for different bit rates (10 kbps, 50 kbps, and 80 kbps). The maximum effective bit rate for the IEEE 802.15.4 standard is around 150 kbps, but this rate may be much lower with an increasing number of nodes in the network, leading to longer backoff periods before nodes perform packet transmissions. For example, in [19] it was demonstrated that the data rate in a WSN with four nodes reaches a maximum of around 80 kbps.

From the charts, we can see that when local processing is used, latency is short, even in scenarios with high PER and low bit rate. The charts do not include values for $p < 0.25$ , to enhance visualization. Without local processing and without retransmission, latency can reach up to 253 seconds for $p = 0.1$ and $W = 10$ kbps. Also for $p = 0.1$ , latency can be as low as $2.93$ seconds, for $W = 80$ kbps and with retransmission.

With local processing, in the worst scenario (i.e., $p = 0.1$ and $W = 10$ kbps), latency is around $2.43$ seconds when using retransmission. We can notice, from Figures 5 and 6, that retransmissions improve the overall performance in scenarios without local processing for all considered bit rates. This is due to the large amount of transmissions and minimum processing time. When using local processing, retransmissions may be advantageous only in some cases, when both throughput and p are high.

6.3. Information Delivery Rate

If many values of torque and efficiency are aggregated into a single packet, latency increases due to an increase in the overall processing time. However, the information delivery rate may be larger, because each packet carries more information. Therefore, to verify the WSN performance when using aggregation, we will analyze the information delivery rate, that is, the amount of torque and efficiency information received per second for a specific node (the inverse of latency).

The chart in Figure 7 shows a comparison between the values obtained from the experiments (reference) and the values obtained from the model (8), for the scenario with local processing, and $V_{a} = 1$ . Three curves are plotted from the model, besides the reference curve.

Figure 7

Validation of the model for the scenario with local processing, without retransmission, and $V_{a} = 1$ .

Since it is not possible to observe with accuracy the value of W, the two first curves were obtained from the model for $W = 150$ kbps and $W = 80$ kbps, respectively, and $e_{lp} (p)_{1} = 0$ . Also, the absolute errors between the curves obtained from the model and the reference curve, for the three cases are shown.

The third curve obtained from the model considers the value of $e_{lp} (p)_{1}$ , obtained from (11). To obtain the model of $e_{lp} (p)_{1}$ , the value of W was set at $80$ kbps. Although the value of W can vary depending on the value of p, the error resulting from the constant value of W is also compensated by $e_{lp} (p)_{1}$ .

For $W = 80$ kbps, the errors vary from $6.2 %$ to $8.8 %$ , and for $W = 150$ kbps the errors vary from $8.7 %$ to $11.2 %$ . We can notice that the smaller the value of p, the greater the error between the reference and the model. This occurs due to the relation between p and W.

When $e_{lp} (p)_{1}$ is considered, the error stays less than $1 %$ for almost all values of p, reaching a maximum of $1.41 %$ , as observed in Figure 7.

The chart in Figure 8 shows a comparison between the values obtained from the experiments (reference) and the values obtained from the model, for the scenario with local processing, without retransmission, and $V_{a} = 18$ .

Figure 8

Validation of the model for the scenario with local processing, without retransmission, and $V_{a} = 18$ .

When $W = 80$ kbps, the errors vary from $0.6 %$ to $2.36 %$ . When $W = 150$ kbps, the errors vary from $1.32 %$ to $3.07 %$ . Again, we can observe that the error increases when p decreases. The overall error was smaller in this scenario, even considering $e_{lp} (p)_{18} = 0$ , because the system spends more time performing processing to compute the torque and efficiency values, spending less time transmitting, since each packet carries $18$ values of torque and $18$ values of efficiency. Since the main error source in the analytical model comes from transmission time, neglecting it does not have a larger impact on this particular scenario.

When assuming $e_{lp} (p)_{18}$ , the error reaches a maximum of $0.72 %$ , as observed in Figure 8.

The chart in Figure 9 shows a comparison between the values obtained from the experiments (reference) and the values obtained from the model, for the scenario without local processing and without retransmission.

Figure 9

Validation of the model for the scenario without local processing and without retransmission.

When $W = 80$ kbps, the error is up to $13.33 %$ , and when $W = 150$ kbps, the error is up to $39.17 %$ . In this scenario, the error was larger for $e_{n lp} (p) = 0$ , because the system spends most of the time transmitting, and the processing latency is minimal, making the transmission time the main error source. However, when $e_{n lp} (p)$ is considered, the model presents a small error of up to a $3.99 %$ , even with the great number of transmissions in this scenario.

Figure 10 shows the information delivery rate (obtained by the models) with p varying from $0.1$ to $1$ , considering the values of $e_{n lp} (p)$ , $e_{lp} (p)_{1}$ , and $e_{lp} (p)_{18}$ .

Figure 10

Information delivery rate of the WSN, considering the values of e.

As it was not possible to validate the scenarios with retransmission, for this analysis we considered that the errors in the scenarios with retransmission are equal to the errors when retransmission is not used. However, the error values for the scenarios with retransmission are larger, due to the assumptions made regarding the ACK mechanism.

We can conclude that when employing local processing and for $V_{a} = 18$ , the information delivery rate is better for all values of p. Without local processing the information delivery rate tends quickly to zero when p approaches zero. Thus, monitoring may be infeasible in environments with a lot of interference and without local processing.

For $V_{a} = 1$ we believe that the performance is worse when retransmission is used. Although for some values of p the information delivery rate is larger, the model is very simplistic in the modeling of the time associated with the ACK mechanism, and the real performance is worse than the one observed in the chart.

The use of retransmission indicates an improvement in the information delivery rate for $V_{a} = 18$ . Although the real performance is less for the scenario with retransmission than the observed in the chart, we believe that for a large range of p the use of retransmission can be advantageous. This is due to the processing time to calculate $18$ values of torque and $18$ values of efficiency.

To verify the WSN performance when using retransmission, more experimental studies will be necessary. From these studies, it is also possible to refine the model for the scenarios with retransmission.

7. Discussion

In scenarios with high level of interference, the use of local processing becomes very important. Many recent works address the performance evaluation of radios in an industrial environment [2, 25, 36–38], when subject to interference.

In our previous work [6] a study to verify the impact of interference sources (microwave oven and IEEE 802.11g network) in the communication quality of the WSN described in Section 4 in an industrial environment was performed. We will relate these results with the theoretical results obtained in this paper.

Based on the theoretical study in Section 5, Table 2 shows the information delivery rate for some values of p.

Table 2

Information delivery rate.

	LP ( $V_{a} = 1$ )	LPR ( $V_{a} = 1$ )	LP ( $V_{a} = 18$ )	LPR ( $V_{a} = 18$ )	NLP	NLPR
p = 0.1	2.63	1.13	3.08	4.69	0.015	0.18
p = 0.2	5.33	4.34	6.22	14.27	0.12	0.72
p = 0.3	8.09	8.77	9.38	21.76	0.4	1.6
p = 0.4	10.92	13.36	12.57	26.14	0.94	2.79
p = 0.5	13.8	17.43	15.78	28.57	1.84	4.25

LP: with local processing.

LPR: with local processing and with retransmission.

NLPR: without local processing and with retransmission.

NLP: without local processing.

We can observe once more that when p is small, the information delivery rate is very small when local processing is not used. On the other hand, when using local processing, it is possible to perform monitoring even with high packet error rate.

From the studies in [6], we observed that when the WSN (operating on channel 18) was exposed to the interference of an IEEE 802.11g network operating in channel 6, the packet error rate reached about $90 %$ . In this case, the latency to transmit an information (values of torque and efficiency) can reach 68 seconds, when both local processing and retransmission are not used, and at least $5.5$ seconds when retransmission is used. Thus, in the best case, nodes deliver $0.18$ values of torque and efficiency per second. Moreover, the real information delivery rate is less than that, due to the overhead of the ACK mechanism, which was not considered completely in the model.

For a packet error rate of $90 %$ ( $p = 0.1$ ), when local processing is used, and $V_{a} = 18$ , in the worst case (without retransmission) the latency to transmit an information is about $0.32$ seconds; that is, it is possible to obtain about $3$ values of torque and efficiency per second from the nodes, even for this high interference scenario.

From these studies we can notice that the deployment of a WSN in industrial environment still presents serious challenges related to the communication reliability. However, despite the high packet error rate in some cases, it is important to notice that, due to the local processing capability, all packets that are received in the destination carry useful information, even considering the processing latency by the embedded system. Without local processing, it is necessary to transmit many packets to provide the desired information, and it is practically impossible to obtain useful data from the WSN without using local processing, for scenarios with high interference.

Besides the application for torque and efficiency monitoring, other applications may suffer even more with the unreliability and low bandwidth of WSNs. For example, in [7] the motor current signature analysis method for fault detection was employed. The ADC was configured to operate with an acquisition rate of 4 kHz and $12$ bits of resolution. To perform the analysis, it was necessary to acquire values during $10$ cycles of current, which results in about $8000$ bits of data (i.e., $(Q_{b} = 8000)$ ). Thus, nine IEEE 802.15.4 packets are required to store all the data. For example, for $p = 0.2$ , without local processing and without retransmission, we have $17578125$ transmissions in average, and with retransmission $225$ transmissions in average (data transmission plus acknowledgment) are necessary. When using local processing and without retransmission, on average only $5$ transmissions are needed. However, the processing latency of the fault detection algorithm must be analyzed, to verify the real gain obtained with local processing.

8. Conclusions and Future Work

In this paper, we have presented a theoretical study for verifying the performance of motor monitoring systems in industry employing WSN. First, a discussion about the standards and protocols already proposed for WSN and about some implementation aspects which can impact the quality of service in WSN based applications in industrial environments was performed. Finally, mathematical models were developed for verifying the performance of an IEEE 802.15.4 based WSN for applications of torque and efficiency monitoring in induction motors, which are widely used in industries. Methods for estimating torque and efficiency in a noninvasive manner have been studied for years, and some works [6–8, 35] have already proposed the integration of these methods with the WSN technology.

From the models, it was possible to verify the performance of the WSN in several scenarios and the benefits from performing local processing, taking into account the lack of reliability of the transmission medium, the characteristics of the IEEE 802.15.4 standard, and processing latency. The analytical results were validated and analyzed together with experimental results obtained in a previous work. It was shown that the use of local processing is essential for the application under consideration, mainly when the WSN is subject to interference sources. In scenarios with high level of interference, it can be almost impossible to perform monitoring without using local processing.

In other applications, such as the fault detection from motor current signature analysis, the use of local processing can be even more essential, since the amount of data that must be processed to perform the analysis is even higher in comparison with the estimation of torque and efficiency. Besides, by performing local processing, the end nodes can be configured to transmit data only if an important event (i.e., a failure) is detected. However, the processing latency to perform the fault detection is also high. Thus, it is important to analyze the performance of the WSN taking into account the node's processing capacity too. The models developed in this paper can be instantiated to evaluate the performance of other applications, such as the fault detection application, verifying the impact of local processing. As one of the future works, we intend to develop fault detection applications using WSN.

Using the results obtained from the model, it is possible to compare several configurations, which can guide de development of WSN applications. This kind of study is especially useful for industrial applications, due to the lack of reliability of wireless networks in this particular environment, and for any application that needs to acquire a large amount of data from sensors.

As future work, we intend to perform detailed performance studies using a WSN with a large number of nodes inside an industrial environment. From these studies we will verify the scalability of the system and characterize the main interference sources in the environment. We also intend to develop spectrum-aware protocols, in which radios can choose the operating channel dynamically, allowing embedded systems to self-adapt to the environment and improving the quality of service of the network. Through more detailed experimental studies, it will be possible to refine the analytical models, increasing their accuracy. Some topics of interest include (i)

experimental evaluation of the WSN with an increasing number of nodes in an industrial environment;

(ii)

development and evaluation of protocols for dynamic channel allocation;

(iii)

exploration frequency redundancy, employing multiple transceivers;

(iv)

experimental evaluation of the WSN using retransmission;

(v)

development of techniques for data summarization, reducing even more the amount of packets transmitted in the network.

Footnotes

Acknowledgments

This work was supported by the National Council of Scientific and Technological Development (CNPq, Brazil), by the Coordination of Improvement of Higher Education Personnel (CAPES, Brazil) and by Eletrobras—CEAL.

References

Hanitsch

Energy Efficient Electric Motors 2002

RIO 02-World Climate & Energy Event

Gomes

R. D.

Spohn

M. A.

Lima-Filho

A. C.

Anjos

E. G.

Belo

F. A.

Correlation between spectral occupancy and packet error rate inIEEE 802.15.4-based industrial wireless sensor networks

IEEE Latin America Transactions 2012 10 1 1312 1318

Baronti

Pillai

Chook

V. W. C.

Chessa

Gotta

Y. F.

Wireless sensor networks: a survey on the state of the art and the802.15.4 and ZigBee Standards

Computer Communications 2007 30 7 1655 1695

Habetler

T. G.

Harley

R. G. A.

Survey of efficiency-estimationmethods for in-service induction Motors

IEEE Transactions on IndustryApplications 2006 42 4 924 933

Lima-Filho

A. C.

Belo

F. A.

Gomes

R. D.

Tests prove out self-powered, wireless, pump torquemeter

Oil and Gas Journal 2008 106 46 43 48

Lima-Filho

A. C.

Gomes

R. D.

Adissi

M. O.

Silva

T. A. B.

Belo

F. A.

Spohn

M. A.

Embedded system integrated into a wirelesssensor network for online dynamic torque and efficiency monitoringin induction motors

IEEE/ASME Transactions on Mechatronics 2012 17 404 414

Gungor

V. C.

Online and remote motor energy monitoring and fault diagnostics using wireless sensor networks

IEEE Transactionson Industrial Electronics 2009 56 11 4651 4659

Benbouzid

M. H.

A review of induction motors signature analysis as a medium for faults detection

IEEE Transactions on Industrial Electronics 2000 47 984 993

Gandhi

Corrigan

Parsa

Recent advances in modeling and online detection of stator interturn faults in electrical motors

IEEE Transactions on Industrial Electronics 2011 58 1564 1575

10.

Antonino-Daviu

Aviyente

Strangas

Scale invariant feature extraction algorithm for the automaticdiagnosis of rotor asymmetries in induction motors

IEEE Transactions on Industrial Informatics 2013 9 100 108

11.

Bouzida

Touhami

Ibtioen

Belouchrani

Fadel

Rezzoug

Fault diagnosis in industrial induction machines through discretewavelet transform

IEEE Transactions on Industrial Electronics 2011 58 4385 4395

12.

Suratsayadee

Himanshu

Lee

Kwan

Wireless healthmonitoring system for vibration detection of induction Motors

Proceedings of the IEEE Industrial and Commercial Power Systems Technical Conference

2010

1 6

13.

Zhang

Zanchetta

Bradley

K. J.

Gerada

A low-intrusion load and efficiency evaluation method forin-service motors using vibration tests with an accelerometer

IEEE Transactions on IndustryApplications 2010 46 1341 1349

14.

Anastasi

Conti

Di Francesco

M. A.

Comprehensive analysisof the MAC unreliability problem in IEEE 802.15.4 wireless sensor networks

IEEE Transactions on Industrial Informatics 2011 7 52 65

15.

Hou

Bergmann

N. W.

Novel industrial wireless sensor networks for machine conditionmonitoring and fault diagnosis

IEEE Transactionson Instrumentation and Measurement 2012 61 2787 2798

16.

Gungor

V. C.

Hancke

G. P.

Industrial wireless sensor networks: challenges, design principles, and technical approaches

IEEE Transactionson Industrial Electronics 2009 56 4258 4265

17.

Willig

Matheus

Wolisz

Wireless technology in industrial networks

Proceedings of the IEEE 2005 93 1130 1151

18.

Santos

J. L. A.

Araujo

R. C. C.

Lima Filho

A. C.

Belo

F. A.

Lima

J. A. G.

Telemetric system for monitoring and automation of railroad networks

Transportation Planning and Technology 2011 34 593 603

19.

Lee

Performance evaluation of IEEE 802.15.4 for low-rate wireless personal area networks

IEEE Transactions on Consumer Electronics 2006 52 742 749

20.

Cheong

Chang

Lai

Sou

Tam

A zigbee-based wireless sensor network node for ultraviolet detection of flame

IEEETransactions on Industrial Electronics 2011 58 5271 5277

21.

Hanzálek

Jurčík

Energy efficient scheduling for cluster-tree wireless sensor networks with time-bounded data flows: application to IEEE 802.15.4/ZigBee

IEEE Transactions on Industrial Informatics 2010 6 3 438 450

2-s2.0-77955712042

10.1109/TII.2010.2050144

22.

Caione

Brunelli

Benini

Distributed compressive sampling for lifetime optimization in dense wireless sensor networks

IEEE Transactions on Industrial Informatics 2012 8 1 30 40

2-s2.0-84856350415

10.1109/TII.2011.2173500

23.

Ascariz

J. M. R.

Boquete

System for measuring power supply parameters with ZigBee connectivity

Proceedings of the IEEE Instrumentation and Measurement Technology (IMTC '07)

May 2007

1 5

2-s2.0-34648816351

24.

Filho

A. C. L.

Belo

F. A.

Alves Dos Santos

J. L.

Gomes Dos Anjos

Self-powered telemetric torque meter

Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 2011 133 4

2-s2.0-79954462508

10.1115/1.4003264

045001

25.

Guo

Healy

W. M.

Zhou

An experimental study of interference impacts on ZigBee-based wireless communication inside buildings

Proceedings of the IEEE International Conference on Mechatronics and Automation (ICMA '10)

August 2010

1982 1987

2-s2.0-78649253619

10.1109/ICMA.2010.5589256

26.

Microchip—MiWi Protocol, 2013, http://www.microchip.com/miwi/

27.

Song

Han

Mok

A. K.

Chen

Lucas

Nixon

Pratt

WirelessHART: applying wireless technology in real-time industrial process control

Proceedings of the 14th IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS '08)

April 2008

377 386

2-s2.0-51249117686

10.1109/RTAS.2008.15

28.

Petersen

Carlsen

Performance evaluation of WirelessHART for factory automation

Proceedings of the IEEE Conference on Emerging Technologies and Factory Automation (ETFA '09)

September 2009

1 9

2-s2.0-77949879785

10.1109/ETFA.2009.5346996

29.

Frey

Lennvall

Zurawski

Wireless sensor networks for automation

Networked Embedded Systems 2009

CRC Press

27.1 27.43

30.

ISA100, Wireless Systems for Automationÿ, 2013, http://www.isa.org/isa100

31.

Flammini

Marioli

Sisinni

Taroni

Design and implementation of a wireless fieldbus for plastic machineries

IEEE Transactions on Industrial Electronics 2009 56 3 747 755

2-s2.0-63149192223

10.1109/TIE.2008.2011602

32.

Salvadori

de Campos

Sausen

P. S.

de Camargo

R. F.

Gehrke

Rech

Spohn

M. A.

Oliveira

A. C.

Monitoring in industrial systems using wireless sensor network with dynamic power management

IEEE Transactions on Instrumentation and Measurement 2009 58 9 3104 3111

2-s2.0-69249221379

10.1109/TIM.2009.2016882

33.

Hou

Bergmann

N. W.

Induction motor condition monitoring using industrial wireless sensor networks

Proceedings of the 6th International Conference on Intelligent Sensors, Sensor Networks and Information Processing (ISSNIP '10)

December 2010

49 54

2-s2.0-79952322038

10.1109/ISSNIP.2010.5706739

34.

The application of wireless sensor networks to in-service motor monitoring and energy management

Proceedings of the 1st International Conference on Intelligent Networks and Intelligent Systems (ICINIS '08)

November 2008

165 169

2-s2.0-58049153172

10.1109/ICINIS.2008.30

35.

Jingtao

In-service motor monitoring and energy management system based on wireless sensor networks

Proceedings of the 11th International Conference on Electrical Machines and Systems (ICEMS '08)

October 2008

823 826

2-s2.0-67649544736

36.

Boano

C. A.

Tsiftes

Voigt

Brown

Roedig

The impact of temperature on outdoor industrial sensornet applications

IEEE Transactions on Industrial Informatics 2010 6 3 451 459

2-s2.0-77955710723

10.1109/TII.2009.2035111

37.

Lo Bello

Toscano

Coexistence issues of multiple co-located IEEE 802.15.4/ZigBee networks running on adjacent radio channels in industrial environments

IEEE Transactions on Industrial Informatics 2009 5 157 167

2-s2.0-65449152568

10.1109/TII.2009.2018541

38.

Tang

Wang

K.-C.

Huang

Channel characterization and link quality assessment of IEEE 802.15.4-compliant radio for factory environments

IEEE Transactions on Industrial Informatics 2007 3 2 99 110

2-s2.0-34249862901

10.1109/TII.2007.898414

39.

Willig

Recent and emerging topics in wireless industrial communications: a selection

IEEE Transactions on Industrial Informatics 2008 4 2 102 122

2-s2.0-44249108577

10.1109/TII.2008.923194

40.

Gungor

V. C.

Hancke

G. P.

Opportunities and challenges of wireless sensor networks in smart grid

IEEE Transactions on Industrial Electronics 2010 57 10 3557 3564

2-s2.0-77956607253

10.1109/TIE.2009.2039455

41.

Åkerberg

Gidlund

Bjorkman

Future research challenges in wireless sensor and actuator networks targeting industrial automation

Proceedings of the 2011 9th IEEE International Conference on Industrial Informatics (INDIN '11)

July 2011

410 415

2-s2.0-80054978662

10.1109/INDIN.2011.6034912

42.

Hsu

J. S.

Scoggins

B. P.

Field test of motor efficiency andload changes through air-gap torque

IEEE Transactions on EnergyConversion 1995 10 471 477

43.

Hsu

J. S.

Kueck

J. D.

Olszewski

Casada

D. A.

Otaduy

P. J.

Tolbert

L. M.

Comparison of induction motor field efficiency evaluation methods

IEEE Transactions on Industry Applications 1998 34 1 117 125

2-s2.0-0031674861

10.1109/28.658732