Scholarly article on topic 'Sensor Systems for Prognostics and Health Management'

Sensor Systems for Prognostics and Health Management Academic research paper on "Electrical engineering, electronic engineering, information engineering"

Share paper

Academic research paper on topic "Sensor Systems for Prognostics and Health Management"

Sensors 2010, 10, 5774-5797; doi:10 3390/s100605774



ISSN 1424-8220 www m dpicom /pumal/sensors


Sensor System s for Prognostics and H ealth M anagem ent

Shunfeng Cheng 1,M ichaelH . Azarian 1 and M ichael G . Pecht 12*

1 Center forAdvanced Life Cycle Engineering (CALCE), University ofM aryland, College Park:,

M D 20742, USA; E-M ail:: chengf calceum dedu (S FC .); mazarian@ calceum dedu (M HA.)

2 , ...

Prognostcs and Health M anagem ent Center, City Uniersity of Hong Kong, Hong Kong;

E-M ail: mgpecht@ city^ed^hk

* Authortowhom aonespondence should be addressed; E-M ail: pecht@ calicoumdedu; Tel: +1-301-405-5323; Fax: +1-301-314-9269.

Received: 20 April 2010; in revised form : 27 M ay 2010 /Accepted: 28 M ay 2010 / Published: 8 June 2010

Abstract: Prognostics and health m anagem ent (PHM ) is an enabling discipline consisting of technologies andm ethods to assess the reliabilityof a product inits actual life cycle conditions to determ ine the advent of failure and m itigate system risk. Sensor system s are needed for PHM to monitor environmental operational, and perforaance-related characteristics. The gathered data can be analyzed to assess pioduct health and predict remaining life. In this paper, the considerations for sensor system selection for PHM applications, including the param eteers to be m easured, the perform ance needs, the electrical and physical attributes, reliability, and cost of the sensor system , are discussed. The state-of-the-art sensor system s for PHM and the em erging trends in technologies of sensor systems forPHM are presented.

K eywords: sensor system ; failure m odes; m echanism s and effects analysis FM M EA ) ; Prognostics and health m anagem ent (PH M )

1. Introduction

Prognostics and health management PHM ) genelaliyaombines sensing and interpretation of environm ental operational, and perform anoe-reHated param eters to assess the health of a product and predict rem aining useful life. A ssessing the health of a product provides inform ation that can be used

to m eet several critical goals!: 1) providing advance warning of failures; 2) m inim izing unscheduled maintenance, extending maintenance cycles, and maintaining effectiveness through tmelyrepair actions; (3) reducing the life-cycle cost of equipm entby decreasing in^spectbncostiB!, downtim e, and inventory; and (4) im proving qualification and assisting in the design and logistical support of fielded and future system s [1].

The importance of PHM has been explicitly stated in the U S. Departmentof Defense 50002 policy GR FXPHQW RQ GHI HQ VH DFTXLVLWL R Q ZKLFK VWDWHV WKD W readiness through affordable, integrated, em bedded diagnostics and prognostics, em bedded training and testing, serialized item m anagem ent, autom atic identification technology, and iterative technology UHIU HVKPHQW- > @ 7 K X V D SUR JQRVWL FV FD SDEL OL W\ KDV I Departm entof Defend.

Traditionally, prognostics have been implemented using either a data-driven approach or a m odel-based approach [1]. The data-driven approach uses statistical pattern rteoogn^itonand m achine laming to detect changes in param eter data., isolate faults, and estim ate the rem aining useful life (RUL) of a product [1-4] .D ata-driven m etthods do not requite product—pecific know ledge of such things as m aterial properties, constructions, and failure m echranism s. In data-driven approaches, in-situ monitoring of environm entaland operational param eters of the product is carril out and the complex; reatinships and trends available in the data can be captured by data-driven m ethods without the need for specific failure m odels. There are m any data-drivenapproaches, such as neural networks NNs), support vector machines (SVM s), decision tree classifiers, principle component analysis PCA), particle filtering (PF), and fuzzy logic [1].

M odel-based approaches are based on an understanding of the physical processes and iterielatinships am ong the different com ponents or subsystem s of a product [5], including system modeling and physics-of-failure (PoF) modeling approaches. I system modeling approaches, m atthem atcal functions or m appings, such as differential equations, are used to represent the product. Statisticalestim aton techniques based on residuals and parity relations are then used to detect, isolate, and predict degradation [5,6]. M odel-based prognostic methods are being developed for digital electronics com ponents and system s such as lithium ion batteries [7], m icroprocessors in avionics [8], globalpositining system s [9], and sw itched m ode pow ersupplies [10].

PoF-EDVHG SUR JQRVWLF P HWKRGV XWLOL] H N QR ZO H G JH RI D geom ety, m aterial properties, and failure m echanism s to estim ate its RUL [11 -14]. PoF m etthodology is based on the identification of potential failure m echanism s and failure sites of a product. A failure m echanism is describedby the relationship betw een the insitum onitored stresses and varabilityat potential failure sites. PoF-based prognostics perm it WKH D VVHVVPH QW DQG S UHG L F W reliability under its actual application conditions. It integrates in situ m onitored data from sensor system s w itthm odels that enable identification of the deviatinor degradatinof a product from an expected norm alconditin and the prediction of the future state of reliability.

Param eter m ontoting andtthe analysis of acquired data using prognostic m odels are fundam ental steps for these PHM methods, while sensor systems are the essential devices used to monitor param eters for PHM .PHM relies highly ontthe sensor system s to obtainlong-term accurate in situ inform aton to provide anom aly detection., fault isolation, and rapid failure prediction..

Firstly, PH M requies m onioring a large num ber of product param eters to evaluate the health of a product. D epending on the com plexity of the m onitred product, it is possible to m onitor thousands of param esters in the entie life cycle of the product to provide the inform aton requined by PH M . These param esters include operational and environm entai loads as wdl as the perfora ance conditions of the product, fer example, temperature, vibration, shock, pre^ure, acoustic levels, strain, stress, voltage, current, hum idity levels, contam inant concentration, usage ftequency, usage severity, usage tm e, pow er, and heatdisspation. in each case, a variety of m onitoring fea-tures such as m agnitude, variation, peak level and rate of change m ay be required in order to obtain characteristics of param eters. Figure 1 isan example of PHM foran automobile to ±low the complexity of PHM application [15].

Figure 1. An example of PHM application foran automobile [15].

No. M cnitoriig and sensing No. M ar-tor:hg and 93ishg

1 CcllisTon avoidance, night visTon, and front crash detection, forward obstacle ^^r 13 Side airbag deploym ent sensor

14 Angular acceleration (suspension)

2 Vehicle distance sensor 15 Em issions sensor

3 Road condition sensor 16 Rack -ip collision, learvisicn cam era, rear radar:, rear obstacle sensor

4 Side obstacle sensor

5 Oil/fuel pressure and flow m onior-ng 17 Tem peratture and hum idity sensor, and com fort control

6 Speed sensois

7 Tire pressure m onioring 18 Rollover detection

8 Fire detection sensor 19 Rain sensor and wiper control

9 Driverm cnitCIiirg sensor 20 Pow er trail control m odule

10 Steering angle sensor and stability control 21 Throttle position m onitDring and control

11 Yaw and acceleration sensois forairbag deploym ent 22 Battery m ch-itDIirg

12 Side crash detection 23 Ignition and engine control m ch-itDrhg

Secondly, param eterm onitoring may be needed during all stages of the product life cycle, icluding m anufacturing, shipm ent, storage, handling, and operation, since failures m ay occur due to the abnorm al operational or environm ental conditions of all these stages. Sensor system s provide the m eans by which the param eterrs can be m onitored and the data can be acquired and processed.

Thirdly, PH M should have m inim um adverse influences on the reliability of the m onitored product and should have relatively low cost:. This m eans that additional parts!, such as sensor system s, should be selected carefully to m inim ize the adverse effects on the m onitored hostproducts. The features of PH M

require m any high perform ance sensor system s to continuously m onitor record, analyze, and transfer a ODUJ H DPRX QW RI SDUDPHWHU G DW5 LQ WKH SURGXFW^V

In this paper, the considerations of a sensor ^stem for PHM applications are diaxis^d. Even though general considerations for sensor ^stem election can be applied to all branches of ^ience and engineering, the features discussedabove concerning PHM applications provide unique perspectives onthese considerations andassociated issues. The state-of-the-art of current com m ercialllly available sensor ^stem s for PH M is presented by a survey. The em erging trends of the sensor system s for PH M are alto predicted.

2. Considerations of Sensor System Selection for PH M

Considerations of sensor system election for PHM may include the parameters to be measured, the perform ance needs of the sensor system , the electrical and physical attributes of the sensor ^stem , reliability, and cost [16]. Sensor ^stem s w ith multiple sensing abilities, m iniature size and light weight, low pow er consum ption, long range and high rate data tran^ ission, large onboard m em ory, fast onboard data processing, low cost, and high reliability are specifically advantageous to PHM applications.

A generic sensor^stern will typically have ^nsing elements, onboard analog-to-digital converters, onboard m em ory, em bedded com putational capabilities, data transm ission, and a pow er source or supply, as shown in Figure 2. In this figure, the internal sensor elements, onboard memory, and onboard proces^rs are typical internal devices. The external m em ory, com puters, and external sensor m odules are typical external devices. The pow er sources can be internal, external, ora com bination of both, andthey provide pow er for the entire sensor ^stem . The w ired or w ireless data tranam ission interfaces connect the external andinternal devices. Not every PHM sensor system w ill necessarily contain all these elem ents, and not all sensorsystem sare suitable forthe implem entation of PHM . The user needs to understandtthe requiem ents of the PH M application to choose an appropriate sensor system .

Figure 2. Integrated sensorsystem forPHM [17].

Internal devices

Internal sensor m odules

O nboard m icroproaesa)r (with analog-to-digital converter)

O nboard m em ory (data storage, em bedded software)



W ired or wireless data transm ission interfaces

External devices

External PDAs, computers, cellphones,

External sensor modules

21. Param eters to Be M onitored

in general, unorder to assess the health of a product, the param eters to m onitor tor PH M include performance parameteis (eg., the speed of the fan in a laptop); physical characteristics (eg., the pressure change in an oil pipeline or the strain of a printed circuit board when it is bended) ; electrical characteristics (eg., the resistance of a resistor or the current through and voltage over a resistor) ; environm ental conditions (eg., tem perature, vibration, pressure, acoustic levels, andhum iditylevel) ; and operational conditions (eg., usage frequency, usage severity, usage time, power, and heat dissipation). The param eters can alto be classified based on differentdom ains, as show n in Table 1.

Table 1. Examples of param eters for PH M applications [17].

Dom ain Examples

M echanical Length, area, volum e, velocity or acceleration, m ass flow , force, torque, stress, shock, vibration, strain, density, stillness, strength, angular, drrecton, pressure, acoustic intensity orpow er, acoustic spectral distribution

Electrical Voltage, current, resistance, inductance, capacitance, dielectric constant, charge, polarization, electric field, frequency, pow er, noise level, im pedance

Therm al Temperature (ranges, cycles, gradients, ramp rates), heat flux, heat dissipation

Chem ical Chem ical, species concentration, gradient, reactivity, m ess, m olecularweight

Hum idity R elative hum idity, absolute hum idity

Biological pH , concentration of biologicalm olecules, m iroorganism s

Electrom agnetic radiation and ionizing radiation Intensity, phase, wavelength (frequency), polarization, reflectance, transm ittance, refractive index, distance, exposure dose, dose rate

M agnetic M agnetic field, flux density, permeability, direction, distance, position, flow

For a specific PH M application, the param eters tobe m onitored can be identified basedontheir relationship to functions that are crucial ibr the safetyof the product, are likelytobe im plicated in catastrophic failures, are eœential for m iion com pleteness, or result in long dow ntim es. Selection is alto based on know ledge of the critical param eters established by past experience and field failure data from similarproductsand by qualification testing.

System atic m ethods, aich as failure m odes, m echanir s, and effects analysis (FM M EA ), can alto be used to determ ine param eters that need tobe m onitored [18] .FM M EA is a m ethodology used to identify the critical failure m echanism s and m odels for all potential failure m odes of a product under expected operational and environm ental conditions. The output of the FM M EA process is a list of critical failure m odes and m echanism s that enable us to identifythe param eters to m onitor andthe relevantphysics-of-failure m odels to predict the rem aining life of the com ponent.

Figure 3 is a schematic flow chart of FMM EA . The product is divided into lower level subassem blies for investigation; these subassem blies are potential sites of failure. For each possible failure site, the functions and associated possible failure m odes are analyzed. A failure m ode is the way in which an item fails to perform its intended design function orperform s the function but fails to m eet its objectives [18-20]. Potential failure m odes of an electronic product can be identified by analysis of the subassembly functions, and an understanding of how their impairmentmay be manifested. For exam ple, for a cable in an electronic product, the failure m odes may be stretching, breaking, kinking, or fraying.

Figure 3. Flowchartof FM M EA process [18,19].

For each failure m ode, the potential failure causes should be identified a failure cause is the Specific process, design, and/or environm ental conditionthat initiated the failure and w hose rem oval will elim inate the failure. Possible failure causes are considered by investigation of the conditions of the life cycle of the product, including m anufacturing/assam bly, test, storage, transportation and handling, operation, and m aintenance [18]. For exam ple, the failure of m ulttlayer ceram ic capacitors M LCCs), which are widely used in various devices, m ay be observed as param eter shits, such as a decrease in insulation resistance or an increase in dissipation factor-. Exam ples of the potential causes of these failures include im proper am bient tem perature and hum idity conditions during storage or transportation. The im properbias voltage conditions in operation also contribute to failure.

Failure mechanisms are the processes by which specific combinations of physical, electrical, chem ical, and m echanical stresses induce failures. The potential failure m echanism s are identified for

each failure m ode andsie basedoncauses, loads, anddesign (geom etry andm ateral) [12]. Failure m odels are used as tools to assess failure propensity. 3hfailure m odels, the stresses and the various stress param eters and their relationships to m aterals, geom etry, and product life are considered. Each potential failure m echanism can be represented by one or m ore of failure m odels. FM M EA prioritizes the failure m echanism s based on their occurrence and severity in order to provide guidelines for determ ining the m ajDroperatbnal stresses and environm entaal and operational param eters.

The follow ing is an exam ple dem onstrating the application of FM M EA to param eter identification for PHM . In this case, the parameters that can be used to monitor the health state of flexioHe-termination M LCCs under teem perature-hum idity-bias (THB) conditions are identified by FM M EA . Figure 4 is the schem atic structure of the M LCC when it is soldered onto a printed circuit board [21]. The M LCC has precious m etaalelectrodes PM E) m ade of silver-palladium . The ceram ic dielectric m ateral is betw een electrodes.

Figure 4. Schematic figure of PM E-basedM LCC [21].

Electrode C eram ic dielectric

Printed circuitboard

C apacior term inaton Solder joint \-Solderpaid

Table 2. FM M EA of the M LCC underTHB condition.

Potential Failure Sites Potential Failure M odes Potential Failure Causes Potential Failure M echanism s

Electrodes Short, decrease in resistance, decrease in capacitance, crack Thermal stre^, m oisture, bias voltage, bending of printed circuitboard Silver m igration, corrosion, fatigue

Ceram ic dielectric Decrease in insulation resistance, decrease in capacitance, increase in diœipation, crack Aging of ceram ic dielectric

Table 2 is the FM M EA of the M LCC under THB conditions. Two failure mechanism s were dom inant forPM E-M LCCs underTHB conditions. The first one was silver m igration. Sato etal. [20] showed a three-st^> process ibr M LCC silver m igration failure involving the form ation of a m icroscopic crack, fellow ed by penetrationof m oisture into the cracks, andfinallythe m ovem ent of m aterial (eg, silver) that caused failure. Silver m igration can cause a short, which would appear as a low insulation resistance value. The second failure mechanic was overall degradation of the dielectric of capacitors, which would also cause a lowering of insulation resistance and capacitance and an increase inthe diœipationfactor [22]. This degradationm ay be causedby m oisture penetrating into voids and oxygen vacancies in the dielectric of the capacitors [22]. Based on this analysis, perform ance parameters (including insulation resistance, capacitance, and diœipation factor), environmental param eters (including t^ perature and hum idity), and operational param eters (including bias voltage) should be m onitored.

PHM requires integration of m any different param eters to a^^ the health state, detect and isolate faults, and predict the rem aining life cf a product. IE an individual sensor ^stem can m cntcr m ultiple param eters, this would simplify PHM and reduce the ccstof PHM . The ^nsing cf multiple param eters refers to one sensor system that can m easure m ultiple types cf param eters arch as tern perature, hum idity, vibration, and presaue. Structures that can conduct m ultiple sensing include the follow ing: a senscr^stem that contains several different ^nsing elem ents internally; a sensor system with flexible, add-on external ports that support various sensor plug-in nodes; and com binations cf these structures. Fcr the^ structures, som e com m on com ponents can be shared such as the pow er supply, A /D converter, m em cry, and data transm i^ion.

Figure 5 shows the ePrognostic sensor system , which includes multiple sensor elements. The ePrognostic sensor ^stem can monitor multiple parameters used tor prognostics, including tem perature, hum idity, m otion, shocks, andvioration. Table 3 shcw s the perform ance cf this senior ^stem . In PH M applications, this sensor ^stem can be used for the m cntcring cf the environm ental and operational conditions cf a product [17].

Figure 5. ePrognostic senscrsystem [17].

Table 3. Perform ance cf the ePIDgnostic senscrtag [17].

M easured Param eters

Perform ance

Range: -0 °C to 60 °C (standard); 10 °C to 100 °C ficrspecial CIdercptDh; Accuracy: ±1 °C overthe full standard ranges; Programmable sample time intervals: frcm 10 s tc 24 h

M ctCh

3D m otcm sensing; Sensitivity is multiple step g-fcrce level from 15 g tc 10 g;

g^crce, m cton, and tim e stam p are recorded_

_10 G m axirt um m easurem ent; 3D sensing; Preprogram m ed sensitivity: up tc 10 g

Higher-level 200 G shock can be m easursd in single dim ensicn; Preprogram m ed seosiitivily: up

Shock tc 200 g_

Vibration M aximum frequency approaches 2kHz, with an accuracy cf ±5% at top range_

Relative Range: from 10% RH tc 90% RH Programmable sample time intervals: from 10 s Humidity tc 24 h; Accuracy: ±10% RH over the full temperature range ( -0 °C tc 100 °C)

Fault detection and isolation is a necessary process in PH M to detect the occurrences of the faults and then identify the types, sites, and causes of the fault. Som e faults and failure m edhanism s can be identified by some sensor system s directly. For example, the resistance m easurem ent can be used to isolate the open or short locations ina sim ple circuit, and corrosion sensors can use electochem ical impedance spectroscopy (EE) to monitor the corrosion of structures directly [23]. For complex electrical packages, som e techniques or sensor system s can be used to isolate the failure sites. For example, the Scanning Superconductive Quantum In^t^ff^^ce Device SQUID m iroscopy) can be used to detect shorts in the m i::Ibpdoaessors; 3D x-ray radiography/tom ography can im age various levels of iterconnlectionls; and scanning acoustic m icroscopy can detect the iterfacial delam inations anddefects inthe packages [24]. O ther sensor system s using electrom agnetc nondestructive testing technologies [25],ultasonlic guidedwave technologies [26],odopt±all technologies [27] can detect cracks inside a product.

Fault isolation can also be done by using m athem atcal m odels, such as principle com ponent analysis PCA) [1,28] and residuals estimation [29], WR DQDO\]H WKH GDWD IURP tJHQ However, if the cbnLsidedEd sensor system has the ability to detect and isolate the faults or failure m echanism s, it will im prove the efficiency of PH M and provide m ore direct inform aton. This type of sensor system m ay have a com plete structure, including sensing elem ents, m em ory, processors, and display parts. They can m onitor the corresponding param of the product, process the data using computers included in the system , display plots, and provide alarm s. The selection cbnLsideIationLS prEsentEEd in this paper can also be used for the selection of this type of sensor system s.

22. Sensor System Perform ance

W hen the param to m onitor are identified, the characteristics of these param e^terES, such as the possible range and frequency, should be understood. These characteristics can be obtained based on the historic records of the data or the specifications of the products. These features of the param should then be translated into the requirem ents for the perform ance at1dbu1ES of sensor system s. Several relevant com m on perform ance at1dbu1ES of sensor system s include the follow ing:

. M easurEem ent Range: the lowest and highestvalues of the measurands that the sensor can sense. The m easurem em range of the sensor system s should be w ider than the actual range of the measurand.

. Dynamic Range: the ratio of the largest measurable output variationl to the smallest distinguishable output variation], usual]yexpdessed indB [30]; this is an im portant aspect of a VHQVRUtV DELOLW\ WR UHVSRQG WR VLJQDOV KDYLQ J ERWK

. Accuracy: the closeness of agreement between the measureemem and the true value of the measureed quantity [31]. It can be presentEd as the error that is the difference between the m easureem entand the true value.

. Sensitivity: generally, the ratio between a small change in output to a small change in input, usually a unit change ininput. Sensitivity represents the slope of the cal-foIationLcuIve [32] .m general, it can be described by the derivation of the output to the input. It may be a constant for all the inputs, but also may be different fbrdiffeIent pants of the input.

z R epeatability : closeness of the agreem ent betw een the results of successive m easurem ents of the

sam em easurand carred outunderthe sam e conditions of m easurem ent [31]. z Resolution: the m inia al change of the input necessary to produce a detectable change in the output [32]. The resolution of the sensoris specified by the unit of the m easured param eterorthe percentage of the range of the m easurd param eter. z Frquency R esponse : output-tb-input ratio (the output pow er dividedby the input pow er)of a sensor as a funaiion of frequency and often given by dB [32]. It can be represented by gain-frquency response or phase-frquency response. The frquency response indicates the range of frequencies of the input for whiah the output is adequate (ie., does not decrease or increase the error due to the inability of the device to operate at a frquency or range of frquencies). The frquency range is betw een the low er and upper points where the am plittude of the signal has fallen off 3- dB , or 0 707 times of the input. The frquencies atthese two points are cutoff frequencies. The frquency response range of sensor ^stem s should be wider than the m easurd param eters.


approached from two diFnentdirections [34]. . Linearty : the m aximum deviation of the output funations from an idealstraight line [33]. . R esponse tra e : the tme a senscr takes to r^ctto a given input. This attribute reprsents how fast.

the sensor^stem can respond to the change of the m easurd param eter. . Stabilization trae: the trae a sensor takes to reach a steady output upon exposure to a stable input.

. Sam pling rate: the num ber of sam pies per second (or other unit) taken from a œntinuous signal to m ake a discrets signal.

PH M needs a high perforra ance sensor system to identify the healtthof the m onitord ^stem and keep the uncertainty at a certain level. Gu et al. [35] studied the difffernt sources of prognostic uncertainty and found thatmeasurementiriaacuraay by the sensor^stem was one of the main sources leading to uncerainty in PHM applications. Understanding the perfora ance attributes of sensor ^stem s m entioned above can help to determ ine the level of uncertainty caused by sensor ^stem s and aœist in œntrolling the overalluncertainty of the entre PH M application.

2 3. Physical C haracteristics of SensDr System s

The physical characteristics of a sensor system include itts size, w eight, âiape, packaging, and m ounting of sensors into the host. In som e PH M applications, the size of the sensor m ay becom e one of the m ost signifiant selection adera due to lim itations of available spaœ forataahing the sensDror due to the inacc^^bility of locations to be sensed. in electrnic products, due to the high-densitty componenison the circuit board, it is difFicult to mount large sensor system s to measure certain local param eters. in the m otherboards of c:шmentcbm puters, som e sensons are em bedded in the chips to save gpace and im prve perfora ance. For instance, the tem pâture of the CPUs and graphic proceœors are m easured by thera al sensors built info the proceœor chips.

The weight of the sensor should also be considered in certain PHM applications, aich as for vibration and shock m easurem ents using acaelerom eters, since the added m ass can change the system

response. f a fixtre is requied to m ount a sensor ona piece of equipm ent the addedm ass of the sensorand fixtuue itelf m ay change the system characteritics. W hen selectng a sensor system , users should determ ine the availble srfze and weight capacitty that can be handled by the host environm ent and then consider the srfze and weight of the entie sensor system , which includes the batery and other accessories such as ar^tennas and cables.

Users should also consider the shape (round, rectangular, or fat) of the sensor system ; the packaging m aterials, such as m etal or plastic; and the m ethod forataching or m ounting the sensor, for exam ple, glue, adhesive tape, m agnets, fixtuues, or sœw s (bols) to affix the sensor system to the host. The folbw ing is an exam plle explaining these requiem ente for sensor system s in a PH M application. In this exam ple, the PH M m etthod is used to identify the healtth status of a laptop by m onitorhg the internal environm ental param eters, such as tem peratuue, shocki, and hum iditty. The express card sOott, as show n in Figure 6, on the sade of the laptop is used to place the sensor system to m easure these param eteis. Based on this placem ent, the sensor system must be in the shape and srfze of the typical express card, should have a plastic or m etal outside packaging, should be able to transm it the data tthrough the card interface orwireless prottocol, and should be able to be m ounted in the sGot.

Figure 6. Exam ple of sensing location [36] (Sensorcard isput in the express card sOotof a laptop).

2.4. FunctonalAttributes of Sensor System s

The electrcal attributes of the sensor system s that should be considered include the follow ing : onboard pow er and pow er m anagem ent abilitty; onboard m em ory and m em ory m anagem ent abilitty; program m able sam ping rate; the rate, distance, and security of data transm isson of the sensor system ; and onboard data processing capabilitty. Each of these attributeswillbe discussedbelow .

2.41. Pow er and Pow erM anagem ent

Pow er consum ption is an essentiail characteristic of a sensor system as it determ ines how long the sensor system can functon without connection to an external source of pow er. The pow erconsum ption of sensor system s can be divided into three m ain dom ains: sensing, com m unication, and data processing [37] .mthese three dom ains, w ieless data com m unicationconsum es the m ost energy. m order to attainthe requied durationof operationinsuch applications, a sensor system m ust have a sufHcientpow er supply and the abiiity to m anage pow erconsum ption.

Sensor system s can be divided into two m ain categories w ith respect to ttheir power sources: non-battery-pow ered and batery-powered. Non-batery-powered sensor system s are typically either

wired tc an externalAC powersource crthey use pcwerfrom an integrated host system . Forexample, a temperature sensor is often integrated within the m ircproaes^r on a motherboard and utilizes the F R PSXW H U1 V S RZHU VXSSO\

Battery-pow ered sensor system s are equipped with an onboard battery. Replaceable or rechargeable batteries alow sensor ^stem s to operate continuously w ithout replacing the entire ^stem . Rechargeable lithium ion batteries are comm only used in battery-pow ered sensor system s. In som e situations, the battery is inaace^ible. The use cf batteries w itha larger capacityor standby batteries may be required in such applications.

Power m anagem ent is u^d tc optim ize the pcwer consumptioncf the sensor ^stem incrder tc extend its operating tim e. Pcw er ccnsum ption varies for different operational m odes cf the system (eg., active mode, idle mode, and sleep mode). The sensor is in active mode when it is being used tc m cnitcr, record, transm it, cr analyze data. The pcw er consum ed for ^nsing varies depending on the param eter ^nsing m ethods andsam pling rate. C cntinuous sensing w ill ccnsum e m ore pcw er, w hile periodic cr event-triggered sensing w ill consum e less pow er. A higher sam pling rate w ill consum e more power because it senses and records data more frequently. Additionally, wireless data transm ission and onboard signalprooessing will consum e m ore pcw er-.

In its idle state, a sensor system consumes much less power than during active mode. Sleep mode consum es the low estpow er-. The tasks cf pcw erm anagem ent are tc track and m odel inccm ing requests cr signals in order tc determ ine when tc ^ itch between the active state and idle state, hcw long the idle state will be m aintained, when tc switch tc the sleep state, and when tc wake up the ^stem . Fcr exam ple, in continuous ^nsing, the ^nsing elem ems and m em cry are active, but if data transm ission is not required, the sensor ^stem can be put into sleep mode. Power m anagem ent willwake up the data transm ission circuit when it receives a request

2.4 2. Onboard M em cry and M em cry M anagem ent

O nboard m em cry i the m em cry contained within a sensor ^stem . It can be used tc stcre collected data as well as inform ation peraining tc the sensor system (e g., sensor identity, battery status), which enables the sensor system tc be recognized and tc communicate with other ^stems. Firmware (em bedded algorithm s) in the m em cry provides operating instructions tc the m ircprDaes&cr and enables it tc proce^ the data in real tim e. O nboard m em cry allow s form uch higherdata sam pling and save rates.

M em cry requiem ents are affected by ^nsing m ode and sam pling rate. Sensor ^stem s should allow the user tc program the sampling rate and set the sensing mode (ie., continuous, triggered, and threshold). These ^ttings affect the am ountcf data stored in the m em cry.

M em cry m anagem ent allow s a user tc configure, allocate, m cnitcr, and optim ize the utilization cf m em cry. Fcr m ultiple-sensing sensor system s, the data fcrm atw ill often depend on the sensing variable. M em cry managem ent should be able tc distinguish various data form ats and save them into corresponding areas cf the m em ory. Fcr exam ple, the sam pling rate, tim e stam p, anddata range cf tem perature are different- from those cf vibration data. In the m em ory, these different- data m ay be stored separately basedonalgorithm s incrder tc m ake them easy tc identify. M em cry m anagem ent

also should have the ability to show the usage status of the m em ory, such as the availability percentage, and indicate when the memory isbecom ing full.

The sampling mode determ ines how the sensor monitors the parameters andat whattimes itwill actively sam p]e the m easurand. C om m only used sam pling m odes include periodic and eventtdggeIed sam pling. The sam pling rate defines the num ber of sam pies per second (or other unit of tim e) taken from a continuous signal to m ake a discrete signal. The com binaton of sam pling m ode and sam pling rate controls the sam pling of the signal.

Program m able sam pling m odes and ratEEs are preferred for PH M applicatonls, since these features affect diagnostic and prognostic pow er consum pton and m em ory requiem ents directly. U nder the sam e sam pling m ode, a low sam pling rate consum es less pow edahd m em ory than a high sam pling rate. Buta sam pling rate less than twice the frequency of operation leads to distortion in reconstructed signals and m ay reduce the likelihood of capturing interm ittEnt or transient events that are needed for fault detection. A dditionally, if the userwants to utilize a sensor, for exam ple, to m on]itodvi>IationL and temperature at the same time, the sensor system should allow the user to set the sam pling mode and rate fbdthese different types of param etas individually.

2.4 3. Onboard Data Processing

Signal processing consists of two parTs: one is embedded processing that is integrated into the onboard processor to enable im m ediate andlocal-izedpIDaessilg of the raw sensor data; the other is processing conducted in the host com puter. W hen selecting sensor system s, one should cbnLsided both of these functions.

O hboadd processing can significantly reduce the num berof data points and thus free up mem ory for m ore data storage. This in turn reduces the volum e of data thatm ustbe transm ited out to a base station orcom puter, and hence results in low erpow er consum pton by data transm issionL. In the case of a large number of sensor system s working ina network, this wouHdallow decentralizatonLof computatonal pow edand facilitate efficient parallel processing of data..

Onboard processing can also facilitate efficient data analysis for environmental monitoring applications. O hboadd processing can be set to provide real-tim e updates for taking im m ediate acton such as powering off the equipm ent to avoid accidents or catastrophic failure. Htcan also be set to provide prognostic horizons to conduct future :depaiI and m aiLtehance activities.

Currently, onboard signal processing includes feature extraction (eg, !ai]flow cycle counting algorithm ), data com presson, fault recognition, and faul1pr^i::tbh.. Ideally, the on]boaIdpIDaessor should display its calculation results and execute actons when a fault is detected, and it should be program m able.

However, the abilities of the onboard processor are lim ited by some physical constraints. One constraint is the available pow err. If processing requites extended calculation and high calculating Speeds, it will consum e m uch m ore pow err The othedCbnLStIailt is onboard m em ory capacity. Running complex software requires a lot of memory. These two constraints make it challenging to embed complex algorithm s into onboard processors. However, even using smple algorithm s and routines to process raw sensor data can achieve significant gains forin-siti analysis.

2.4.4. Data Transm ission

Once data is collected by the sensor ^stem , it is typically transm itted tc a base station cr com puter fcrpost-analysis. In general, the m ethods fcr data transm i^ion are eitherwirele^ orwired. W ired data transm ission can offer high speed transm itoicn, but this transm i^ion is Him ited by the need for transm ission w ires, and the cost is increased by w ires. W ireless transm ission has em erged as a promising technology that can impact PHM applications. W ireless transmissjonrefers tc the transm ission cf data over a distance without the use cf a hard-wired conniection. The distances involved may be short a few m eters, as in a television rem ote control) crvery long (thousands cr even m ill ions cf kilom eters for radio com m uncatcns). W itele^ sender ^stem s can be used tc rem otely m onitcr inhospitable and toxic envircnm ents and transm it m easured data tc a centrallylccated proce^ing station. Alo, since wirele^ sensor system s are not dependent on extensive lengths cf wires fcrthe transm ission cf m easurem em. data, they save instaIaltDh and m aintenance costs. The advantage cf w ireless ^nscr system s can be greatly enhanced by em bedding m icrD-cDhtDller5 that have data analysis capabilities within the wireless sensor.

A lot cf w irele^ technologies can be used for w irele^ data transm issionof sensor system s, for example, Radio Frequency Bentifcai.tion (RFID), Bluetooth, W i-Fi (IEEE 802.11), Ultra-W ideband UW B), Certified w ireless USB W USB), W iM ax W orldwide Interoperability forM icrcwave Access and IEEE 80216), and Zigbee (IEEE 80215 4). W hen selecting the wireless technology tc use for a particular application, the user shcuUdccn^sider the range andrate cf communication, power consum ption., ease cf im plem entatcn, and data security.

An RFID senscr^'stem combines the RFID tag with the sensing element. An RFID tag isan object that can be attached tc or incorporated into a product, anim al or person for the purpose of identification orm cntcring using radio waves [38]. This RFID tag uses sensing elem ems tc detectand record tem perature, hum idity, m ovem ent or even radiatcn data, and then utilizes RFID tc identify the sensor as well as to transfer the raw data cr processed data. RFID ^stem s use many different frequencies, for example, low-frequency (around 125 KHz), high-frequency (13 56 MHz), and ultra-high-frequency UHF, higher than 860 MHz). In order tc communicate tc readers, RFID tags have tc be tuned tc a frequency in com m on with the readers. [39].

The ePrognostic sensor ^stem showninFigure 5 is en RFB-basedsenscr system . The m ultiple arising elements, such as temperature, vibration and hum idity, are embedded in the RFID tag. M cnitcred data is transferred by RFID tc a reader connected tc a computer, The RF frequency cf an ePrognostic sensor tag is typically 915 MHz cr 2 4 GHz. The com m uncatcns range from a few feet (for^curity) tc 100 m .

Bluetooth is en RF-based wireless data transmi^ion technology. It operates in the 2 4 GHz tc 2 485 GHz ISM industrial, scientific, medical band utilizing low -transm itpower radios andtthe frequency-hopping spread spectrum technique [40]. Bluetooth devices hop though 1,600 frequency channels per second, of which 800 channels are transm it channels and the other 800 channels are receive channels. The channels span 79 MHz with 1 MHz spacing betw een the neighboring channels. Thus, Bluetooth is designed to be functional even in very noi^ RF envircnm ems [41]. The data rate

is 1 M bps fbdVersionL 12, up to 3 M bps supported fbdVэIsdbnL 20 + Enhanced Data Rate EDR), and up to 24 M bps supported fbdVersionL 3.0 + HiherSpeed (HS) [41].

W i-Fi CDEEE 80211) is anothedRF-based wireless technology that can be used in sensor system s. A typical W i-Fi setup includes one or m one access points APs) andone or m ore clients (com puIeis, video gam e consoles, mobiles, sensors, etc.). The pIim arry job of an access point is to broadcast a

ZLUHOHVV VLJQDO WKDW FRPSXWHUV -FiEnLetwQrks opeiatWirH thEE W4 GHDz Q G (80211b/g/n), 5 GHz (80211a/n), and 3.7 GHz (80211 y) radio bands, with 11 M bit/ (80211b), 54 M bit/ (80211a/g/y), orup to 600 M bit/ (80211n) data !atES, respectively. The outside transfer range can be up to 120 m (80211 a), 140 m (80211b/g), and 250 m (80211 n) [42].

Zigbee (IEEE 80215 4, Low -rate W ireless Personal Area Network (W PAN) standard) is used for industrial controls (for exam ple, process control and energy m anagem ent, em bedded sensing, m edial data collecton, sm oke and intulded w aming, building autom ation (for exam ple, access control and energy m onLioring), and hom e autom atbn (eg, sm ait lighting, tem perature control, andsafety and access control). Zigbee works at 2 4 GHz and 868/915 M Hz. The data rate is 250 Kbps at24 GHz, 40 Kbps at 915 MHz, and 20 Kbps at 868 MHz. The transm issdon range is betw .en 10 and 75 m and up to 1,500 m fbrZigbeE Pro, although it is heavily dependenton the partcularEnvioinm ent [43].

An U lta-W idEband (UW B) transm iter works by sending billbnLs of pulses across a veiy w ide Spectrum of frequencies several GH z in bandwidtth. The corresponding :dece:ived then tIansla1es the pulses into data E\ OLVWHQLQJ IRU D I DPLOLDU SX OVH V HTXHQFH VHQ of a larger spectrum , low erpow er, and pulsed data im proves spesd and reduces i1edference with other wifeless spectra. In the United States, the Federal Communicationis Cbmmissdbn FCC) has mandated that UW B radio transm issons can legally operate in the range from 31 GHz up to 106 GHz at a lim ited transm itpow erof —1 dBm/MHz [44].

Certified W ieelessUSB is the specdficationL of a wirelessextensibnL of the USB standard intended to fuithedincrease the availability of general USB -based solutions. Itenables products such as personal com puter:!, consum er electronics, and m obile devices to connect using a com m on interface at up to 480 M bit/ at3 m and 110 M bit/ at 10 m . Atclose range, dtds the sam e rate asHi-Speed USB . W ielessUSB was designed to operate inthe 31 GHz ICjS GHz range. Certified w irelessUSB has powerm anagem ent strategies. Sleep, listen, wake, and conserve m odes ensure that devices use only the m inimum power necessary. All Certified W ireless USB products are required to encrypt their data transm issdons [45].

The securityof w irEeless data transm issdonis an dm pbrant factor that Shouldbe conLs:deded. The standard approach is to encrypt the data transm ited by the sensors. In addition, it m —H also be necessary to provide authentication m echanism s -in odded to guarantee that the data source is the claimed sender [46,47]. W alters et al. [48] classified different security risks of wireless sensor networks and described defensive methods to protect data transm isson. One Should evaluate the securitystrtEEgy of the w ireless sensor system or custom ize the securitylevel to protect data during transm isson.

Currently, hybrid data transm issdon combines wireed data transm isson and wireless data transm issdon. This arangem ent can represent a com prom ise that dm proves data transm isson, pow er requirem ents, and cost. W ireed transm isson offers high speed transm issdon and consum es low energy.

W ireless data transm i^aon can offer convenient data com m un^icatcn and elm inate the need for wire, but it consum es m ore pow er-. There are trade-offs to be m ade for any given application. For exam ple, m any sensor ^stem s transfer data from a sensor to a receiving device w irelessly, and then the receiving device transfers the data to a com puterby a wired USB port

2 5. Cost

The selection of the proper sensor ^stem for a given PH M application m ust include an evaluaton of the ccst The cost evaluaton should addre^ the total cost of ownership including the purchase, instaIaltDh, m aintenance, and replacem enit of sensor system s. In fact, the initial purchase ccst of a SURGXFW FDQ EH OH VV WKDQ RI WK9H SURGXFW^V WRWI

In additicn to cost of ow nerSdp, justifying the cost of sensors can be quite challenging for products produced in large vclum es. Prognostics and health m anagem ent (PH M ) m ethods for these products are m ore com plex because m ore parram eters need to be m onitcred and m ore sensor elem ems or ^stem s are required. Hthis case, the total cost coulddilfer ccnsiderablyfor different sensor system integration strategies. O ne strategy is that the user purchases individual sensor system s at a low ccst and then integrates them . A second strategy is tc purchase integrated sensor ^stem s w ith m ultiple sensing abilities, ohboarIdpow er,m em ory, processing capabilities, andw irele^ data transm -tecn. Typically, the ccst for integrated sensor system s will be low er than the total ccst of individual sensor ^stem s and the integraton of these individual sensor system s. The third strategy is to integrate or build the sensor ^stem s in the product as part of the product. This provides the low est ccst for sensor system s, but it. will increase the cc^fcrthe prcductdue to new hardw are, software, and qualif-caton procedures, etc.

2.6. Reliability

Failed sensor ^stem s provide incorrect or incomplete data and cause PHM to generate wrong detections, alarm s, and predictions. If the sensor ^stem is used to m easure a critical parram eter or is installed at a lim ited access locaton, the adverse im pacts to PHM will be m ore severe. The reliability of a sensor^stem requires the ability of the sensorsystem to perform a necessary fUlhctioh understated conditions fora stated period. However, reliability inform ation, arch as the mean time between failures (M TB F) and failure rate undercertain environm ental and operational conditions, is rarely specified by the sensor ^stem manufacturers.

One strategy to im prove the reliability of sensor ^stem s is to use m ultiple sensors (redundancy) to m cnitcr the sam e product or system . By using redundancies, the ri^ of losing data due to sensor ^stem failure is reduced, but. the cost.increases.

Som e technologies, arch as sensor validaton, can also im prove the reliability of sensor ^stem s. For exam ple, sehsorvaIidatiDn [50] is used to a^^ the integrity cf a sensor ^stem and adjustor correct it. as necessary. This fuhchorlaI-ty checks the sensor perform ance and ensures that the sensor ^stem is working correctly by detecting andelim inatng the influence cf system atc errors. W hen selecting a sensor system , the user should check tc see if the sensor ^stem has the validaton funct±ms.

W hie it ise^ential tc consider the reliability cf senaor system s, it is equally necessary tc consider the effects of the sensor system on the reliability cf the product it is intended to m cnitcr Sensor ^stem s that are heavy m ay reduce the reliability cf the circuit. boards they are attached tc over tim e. In

addition, the m ethod of attachm ent (soldering, glue, or screw s) can reduce the reliability of the product LI WKH DW WDFKPHQ W PDWHULDO LV LQ FRPSDW LEOH ZLWK WKH

3. State-of-the-Art Sensor System s for PH M Im plem entation

In 2008, a survey was conducted by the authors to determ ine the comm ercial availability of sensor system s that can be used in PHM for electronic products and system s. The survey only included com m ercialy available sensor ^stem s having features desirable for PH M .

The survey results showed the characteristics of 33 sensor system s from 23 m anufacturers. The sensor system characteristics included sensing param eteis, pow er supply and pow er m anagem ent capability, sam ple rate, onboard m em ory, data transm itebnm ethod, availabilityof em beddedsignal proce^ing software, size, weight, andcost. The data for each sensor ^stem was collected from the

P D Q X I DF W X UHU 1 V Z HE V LWH -mailS aUdlevSauatiinLsWof de®oDprodDctV. TheHdHaWisV H

listed in Appendix A of [1].

Key findings from the arrvey are that state-of-the-art sensor system s: 1) can autonom ously perform multiple functions using theirownpower management, data storage, signal processing, andwirrele^ data transm iteion; (2) have m ultple, flexible, or add-on sensor ports that support various sensor nodes to m ontor various param eters arch as tem perature, hum idity, vibration, and presarte; (3) have onboard power supplies, such as rechargeable or replaceable batteries; (4) have onboard pow er m anagem ent, allow ing control of operation m odes (active, idle, and sleep), and program m able sam pling m odes (continuous, triggered, or threshold) and sam pling rates; (5) have divert onboard data storage capacity (flaáh m em ory), from several KB to m ore than hundreds of MB; (6) have em bedded signal proce^ing algorithm s that enable data com predion or sim plication prior to data transfer; and 7) use various wirele^ technologies including RFID, Bluetooth, W i-Fi, Zigbee, Ultra-W ideband, and W irelessUSB .

The survey ala) found that current sensor ^stem s should be improved in ^veral aspects for PHM applications. The fürstareas are size and weight, PH M for an electronics ^stem with a high density of com ponents requires ot all-size and light-weight sensor system s that can be placed on the circuit board and that m inim ize the weight increase and potential adverse effects on the reliability of the m mitered electrical ^stem .

O nboard pow er is also one of the m ain lim itations of current com m ercially available sensor ^stem s, especially forwirrele^ data transm ission sensor system s. The m ain onboard pow er is a battery, which needs be replaced orr^harged when it is used up. Higher capacity batteries with small size and light weight or battery-free pow er are needed for sensor ^stem s to operate longer in PHM applications.

O nboard data proce^ing ability should ala) be im proved. In the arrvey, only a few sensor ^stem s had onboard data proce^ing ability, and these only had very sim ple functions, for exam ple, data reduction. Onboard data proce^ing provides tim ely inform ation about the health of the system and can reduce the cost of the entire PHM application. Onboard data processing requires highspeed procerus, large-capacity m em ory, m ore pow er, and proce^ing algorithm s. The processing algorithm s can be developed based on specific PHM applications.

4. Em erging Trends in Sensor Technology for PHM

In general, PHM application requires that sensor technology should be headed toward extreme m iniaturization, battery-free pow er orulta-low pow er consum ption, and -intelligent wireless networks. Since electronic components and systems continue to decrease in size, sensors to monitor their environm ents and operation w ill also becom e sm aller and weigh less in order to be integrated. A s M icro Electro M echanical Systems M EM S) orNano Electro M echanical Systems (NEM S) and smart m aterial technologies m ature, M EM S sensors or nanosensors w ill integrate the sensing elem ent, am p^lificatTon, analog-to-digital converter, and m em ory cells into one m icrochip. The fabrication of MEMS and N EM S w ill offer significant advantages in term s of integration w ith electronics, fabrication of arrays of sensors, sm all size of individual devices, low -pow er consum ption, and low ercosts [51].

W iththe developm ent of new m aterials and energy technologies, battery-free sensor system s are being considered, especially for use in embedded, remote, and other inaccessible monitoring conditions. B attery-free sensor system s w illbe developedbasedonultra-low pow er electronics and energy-harvesting technologies.

Ultra-low pow er electronics will enable future sensor system s to consum e m uch less pow er. A lot of new ultra-low power consumption technologies are emerging. For example, in June 2009, STM icrbelec:trbnics announced a new ultra-low pow er technology platform for building a range of 8-bit and 32-bit m icroœntrollers, which w ill enable future generations of electronic products to consum e less pow er, m eet evolving energy-efficiency standards, and operate for longer tim es from their batteries. This new platform isbuilton a 130nm process, which ST has further optim ized with ultra-low leakage transistors for logic functions, low voltage transistors for analog functions, innovative low pow er em beddedm em ory, new low -voltage low -pow er standardperipherals, and an LQQRYDWLYH S RZHU PDQDJHPHQW DUFKLWHFWXUH t 7KHVH H static power consum ption, enabling forthcoming families of m irrbcbntr^Vlerq delivering better perfform ance per W att than the m ost frugal low - S RZHU GHYLFHV RQ WK] TiThe P D U ] em erging ultra-low pow er technologies enable the productionof ultra-low pow er consum ptionchips for all kinds of electronic products.

Energy harvesting is a process to extractenergy from the environm ent or from a surrounding system and convert it to usable electrical energy. Current energy harvesting sources include sunlight, therm al gradient, human motion, body heat, wind, vibration, radio power, and magnetic coupling. Several excellent. articles reviewing possible energy sources for energy harvesting can be found in the literature [53-57]. Som e large-scale energy harvesting schem es such as w indturbines andsolar cells have m ade the transition froom research to com m ercial products. The interest in sm all-scale energy harvesting for em bedded sensor system si, such as im planted m edical sensors and sensors on aerospace structures, is increasing.

Basic effects used in energy harvesting include electromagnetic, piezoelectric, electrostatic, and therm oelectric effects. For exam ple, the m echanical vibrationinside a device or am bient m echanical vibration can be converted into electric energy by piezoelectric m aterials or electrom agnetic induction.

Piezoelectric materials form transducers that are able to interchange electrical energy and m echanial viratbnLor tore. The electom agnetc inductbn system s are com posed of a coil anda perm anentm agnet attache to a spuing. The m echanial m ovem ent of the m agnet which is caused by a device or am brent viratbn, induces a voltage at the coil term inal and this energy can be dellvPEed to an electrical lad. The thermal energy is often converted into electric energy by thermoelectric generators TEG s) .W :thdecpn1 advances m ade innamotechnolog:ies, the fabIiatiohof M EM S -scale TEG devices has been actively studied. The com bined use of several energy harvesting sources in the sam e device can incrEEasE the harvesting capabilities indifferent situations amdappTiC:at±mls andcan m inim ize the gap betw epn the requirEd and harvests energy [58].

Distributed sensor networks (DSNs) consist of mult-ple sensor nodes that are capable of commun-cating with each other and collaborating on the same sensing goal [59]. The advantage of a DSN is that it allow s data from m ult-ple sensors to be com bined or fused to obtain inferences that m ay not be possible from a single sensor. The sensor nodes in a D SN are organized into a cooperative system . The nodes can com m unicate with each othedamd have the ability to self-organize.

W ieless transm iss-ons, such as W rrpeless USB, are being transferred to sensor products. The developm ent of wiEeless transm issnon technology will produce iitellient w ireless sensor networks with characteristics of long distance, high transm isson rate,ultra-low powercon]sumpton,more secure data com m umcationL, andhigh-speeddata fUlsionLpdOGessiinLg. Furtherm ore, future sm art sensor nodes

ZLO O EH KLJKO\ LQ WHO OLJHQ W ZLWK P R6U]. HThey Iw Xl QavFP Wuit-i Q V diagnostic and pdogmostic capabilities, which will m ake the entire spnsodnletwork m ore functional. The integrationof all of the above technology accelerates the developm ent of w iEeless -rntpTliigent sensor networks [61].

5. Discusston

This paper mainly focuses on com m PKcial]y available sensor system s that can be used for PHM applications w ithm inor or no design m bdificationLs. There are several lim itations w iththesp add-on sensor system s. For exam ple, they cannot be pffPctiply integrated w ith the m on—ored pioduct and hehcp cammo1 sense crii^aT variables thatarp integral to product opedationLamdpedfbrm ancp, such as voltage rais, pow prbuisps, etc. One way for PH M application to ovprcom e this lim —ation is to identify the c:orelatoms betw epn the param eters that can be m on]itoded by add-on sensor system s and the param eters integral to product operation and peifbrm ance based on historic data or training data.. Based on the correlation, PH M can ifprchanges in critical param ptersbased on the param eters m on]itoded by add-on sensor system s. For exam ple, Kwon pt al. [62] dem onstated that m on]itoded RF impedance increased as a physical crack propagated across the soldpr joint under stress conditions, and charactE^iz^ the extent of physical solder joint degradation associated with the changes in RF im pedancp. The other m ethodis tor PHM to integrate the infoum ationoffpred by the add-onspnsor ^tpms and the information obtained from the system buses of the product. For example, when applying PH M for a laptop, a sensor system that can sense teem perature, hum idity, Shock:, and position can be integrated into a card that can be inserted into the card slot of the laptop to m on—or these param eters. On the other hand, PHM also obtains the inform ation acquired by BIOS from the system

bus, for exam ple, the internal tem perature of the CPU and GPU, fan speed, m em ory usage condition, CPU usage condition, etc.

Otherlim itations, such as aaoessibility to the best sensing locations due to lim ited space for sensing or inhospitable or toxic environm ents, can be overcom e with the developm ent of sensor technologies that can m ake sm aller and m ore pow erful wireless sensor system s.

The selection of a sensor system for PHM applications requires analysis of the application, ientifration of the param eters to be m onitored and all the requirem ents for the sensor system s, and prioritization of these requirem ents based on the specific application. Then, sensor system candidates should be identified and evaluated based on these requirem ents. Finally, som e trade-offs m ust be m ade to selectproper sensor system s.

6. Conclusions

PHM isan enabling discipline that assesses the reliability of a product in itsactual life cycle. PHM for a product is a long-term effort that requires m onitoring the product by continuous or periodic m easuring, sensing, recording, and analyzing of different param eters in its life cycle to assess the health status of the product, identify abnorm al conditions, and predict the rem aining life of the product. In PHM applications, sensor system s are essential devices for conducting in situ m onitoring of the actual life cycle of a product. PHM requires that sensor system s be highlyintegrated w ith m ultiple sensing abilities, low power consum ption, low cost, long-range and high-rate data transm ission (wireless or wired), large onboard m em ory capacity, fast onboard data processing abilities;, m iniature weight and size, and high reliability.

A survey of the current com m ercially available sensor system s identified the state of the art of sensor system s. M any sensor system s have m ultiple functions, such as m ultiple sensing capabilities, onboard pow er and pow er m anagem ent ability, onboard m em ory and m anagem ent ability, w ireless data transm ission, and onboard data processing ability. However, the survey also identified several unm et needs of sensorsystem s for PHM applications, including size and weight of the sensorsystem s andthe lim itations of onboardpow er supply, m em ory capacity, onboarddata processing ability, etc. Several technologies such as M EM S or NEM S, energy harvest techniques, ultra-low power consum ption circuits, and intelligent w ireless sensor network techniques, have been emerging to provide m ore pow erful sensor system s inthe future. The trendinsensor system s is tow ardextrem e m iniaturization, w ireless transm ission w ith long range and high rate, low cost, ultra-low pow er consum ption, battery-free pow er supply, stronger onboarddata processing ability, m ore intelligence, and networking.

Ina specific PH M application, the user can identfythe param eters tobem onitored by a failure m echanism -based m ethod called failure m odes, m echanism s, and effects analysis FM M EA ). The requirem ents of PHM applications for sensor system s regarding sensor perform ance, electrical and physical attributes, reliability, cost, and availability of sensor system s m ust also be understood. Som e trade-offs m ustbe m ade to selectproper sensor system s as well.


1. Pecht M . Prognostics andHpalthM anagemen^t of Electronics; W iley-IntEEnscrenicP: New York, NY, USA, 2008.

2. DoD 50002 Policy Dccum ent Pprfbrm ancp Based Logistics. In Dpfensp Aoquisrtionl Guidebook; Defense AoquisitonLUniversrty: ForrtBelvoir, VA , USA , December2004; Chapter3 .

3. V -chare, N ; Pechlt, M . Prognostics and Health M anagem ent of Electronics. IEEE Trans. Cbmpon. Packag. Technol. 2006, 29, 222 -29.

4. Schwabacher, M ; GbPbel, K . A Survey of Artificial Intell—ence for Prognostics. In Prrocepdings of2007 AAAIFall Symposium : AI for PIogmostCs, ArEingtonL, VA,USA, November9 -1, 2007.

5. Liu, J; D jurdjanovic, D ; M arko, K ; Ni, JA. Divide and Conquer Approach to Anomaly Detectbn,Localization Diagn. M pch. Syst Sighal Peocpss. 2009, 23, 2488 -499.

6. Vachtspvanos, G ; Lew is, F ; Rcem pt, M ; H ess,, A ; W u, B . Intelligent Fault D —gnosis and Prognosis for Engineering Systems, 1sted; John W iey & Sons: Hoboken, NJ, USA, 2006.

7. Saha, B ; GcPbel, K ; Poll, S ; Christophersen, J. Prognostics M ethods fbd BatteryHealth M onlitorjnig Using a Bayesan Framework. Instum entaton and Measurement. IEEE Tians. Instrum . Meas. 2009, 58, 291 -96.

8. KaTgdEn, P; BayOutt, M ; Ginart, A ; M innella, C ; Rbempd, M ; Dabney, T. Application of Prognostic Health M anagem ent in Digital Electronic Systems. In ProceEdings of the IEEE AerospaceConLfPrpncp, Big Sky, M T, USA ,M arch 2007; pp. 1 -.

9. Brown, D ; KaTgrEen, P; Byingtonl, C ;Rbempr,J. Electronic Prognostics tA Case Study Using Global PositioningSystem (GPS). M itroelecttonL. ReT. 2007, 47, 1874 1881.

10. KuTkami, C ; Biswas, G ; Koutsoukos, X . A Prognosis Case Study foi Electrolytic Capacitor Degradation in DC-DC C onlvpiteis. In P rrocepdinigs of Annual C onLferenicp of the P rtognostics and Health M anagem ent Society, San Diego, CA, USA, Spptem ber2009.

11. Gu, J;- Barked, D ; Pecht M . Prognostics Implementaton of Electronics uhdPd Vibration Loading. M icropTectrQn.ReT. 2007, 47, 1849 -856.

12. Pecht, M ; D asgupta, A . Physics-of-Failrne: An Approach to Reliable Product D eveTopm entr J. Inst Environ. Sct. 1995, 38, 30 -4.

13. Andrew , H . Real Prognostics t & KD O OHQ JHV , V V X H V DQG /HVVRQV / H In P rrocepdinigs of the FiISt]m1EInatibn]alForrum onL!rLtegrra1ed System H^lthEmgineeI:-ng and

M anagem ent in Aerospace, Napa Valley, CA , USA , Novembei7 10, 2005.

14. Jata, K .V ; Parthasarathy, T A . Physics of Fa-Trie. Ih P rtocepdinigs of the 1st IntErnational Foium on !rLtegrra1ed System HeaTth Engineerting and M anagem ent in Aerospace, Napa Valley, CA , USA , November7 -0, 2005.

15. Kia M otois. Kia Safety: Future ofSafety. Available ohline: http://tfww kiaconz/lPlanietKia/Cias-Cbmm jme1-To-Sustaajrabi-:/C:a-Sftyaspx accessed on 1 June 2010).

16. Tuchbanid, B ; Cheng, S ; Pecht, M . Techmology A ssessm ent of Senscr System s Lot Prognostics and Health M onliortng. In Prrocepdings of IM APS on M iliary, Aerospace, Space and Homeland Security: Packaging Issues and Applications (M ASH ), Baltim ore, MD, USA, M ay 2007.

17. Cheng, S ; Tcm , K ; Thcmas, L ; Pecht M . A W irEelessSenscr System fbdPrognostcs and Health Management. IEEE Sens. J.2010, 10, 856 -62.

18. Ganesan, S.; Eveloy, V .; Das, D .; Pecht M . Ientf-caton and Utlizatori cf Failure M echani^s tc Enhance FM EA and FM ECA . In Proceedings cf the IEEE W crkshcp cnAccelerated Stress Testing and ReHab-Hiy (ASTR), Austin, TX , USA , Octcber3 5-, 2005.

19. IEEE Standard 14131-2002 tIEEE Guide fcr Selecting and Using Reliability Predictions Based on IEEE 1413; IEEE : New York, NY, USA, 19 February 2003.

20. Sato, K .; Ogata, Y .; Ohno, K .; Ikeo, H . M echani^ cf Ceramic Capacitor Leakage Failures Due tc Lcw DC Stress. In Proceedings cf the 18th Ir.tematiohal Reliability Physics Symposium , Lcs Angeles, CA ,USA ,April 1980; pp. 205 -12.

21. Keimasi, M .; Azararn, M .; Pecht M . Isothermal Aging Effects cnFlex Crackling cf M ulttiayer Ceramic Capacitors with Standard and Flexible Terminations. M icicelectiDn. Rel. 2007, 47, 2215 -225.

22. Donahoe, D .; Pecht, M .; Lloyd, I.; G anesan, S. M oistuire Induced D egradaton cf M ulttiaiyer Ceramic CapacitcrE-. M icroelectron. Rel. 2006, 46, 400 408.

23. Davis, G D .; Daces, C M .; Krebs, L A . In Situ Corrosion Sensor for Coating Testing and Screening.Mater-. Perform .2000, 39, 46 -1.

24. Pachecc, M .; W ang, Z .; Skoglund, L.; Liu, Y .; M edina, A .; Ram an, A .; D ias, R .; Goyal, D .; Ramanathan, S. Advanced Fault Isxiaticnand Faiure Analysis Techniques for Future Package Technologies. Intel Technol. J. 2005, 9, 337 -52.

25. Sophiam, A .; Tian, G Y .; Zairi, S . Pulsd M agnetic Flux Leakage Techniques fcrCrack Detection and Characterization. Sensor. Actuator. A-Phys. 2006, 125, 186 -91.

26. Van Velsor, J.; Owens, S.; Rose, J. Guided W aves for Nondestructive Testing cf Pipes. US. PatentUS 2009/0150094 A1, June 11, 2009.

27. W an, K .; Leung, C . Fiber Optic Sensorf:>rlhe M cntcring cf M ixed M ode Cracks in Structures. Sensor. Actuator. A-Phys. 2007, 135, 370 -80.

28. Sot-iris,V.; Tse, P.; Pecht, M . Anomaly Detection through a Bayesian Support Vector M achine. J. Trans. 2010, 1, 1 -1.

29. Kumar, S.; DoHev, E .; Pompetzzki, M .; Pecht, M . A Residual Estimaton Based Approach for Haxating Faulty Param eters. In Proceedings cf IEEE Aerospace Conference, Big Sky, M T, USA , M arch 7 -4, 2009.

30. Bernstein, D. SerLscrPerfcrmance Specifi-catons. IEEE ContiDlSyst Mag. 2001, 21, 9 -8.

31. ISO /EC Guide 99:2007. International Vocabulary cf M etiology tBasic and General Concepts and Associated Terms (VIM ); ISO : Geneva, Switzerland, 2007.

32. W ilson, J. Seh^SDrTechhclCgy Handbook; EllsevierNewnes: Burlington, MA, USA, 2005.

33. W ester, CT. The M easurem ent, Instrum entatihandSehSDrs H andbock; CRC/EEE Press: Bcca Raton, FL, USA, 1999.

34. Fraden, J. AIP Handbook cfM odern Sensors. Physics, Designs and Applications; AIP: W oodbuny, NY, USA, 1997.

35. Gu,J.;B arker, D .; PechLt, M . U ncertainty A s^sot ent cf Prognostics cf Electrcnics Subject tc Random Vibration. In Proceedings cf AAAI Fall Symposium on Artificial Intelligence for Prognostics, ArUingtcn, VA,USA, November, 2007; pp. 50 -7.

36. Gu, J.; Vichare, N .; Tinsley, E .; Pecht, M . ComputerUsage Monitoring fOrDesign and Reliability Tests;. IEEE Trans. Compon. Packag. Technol. 2009, 32, 550 556.

37. Akyildiz, I.; Su, W .; Sankarasubram aniam , Y .; Cayirci, E. W ireless Sensor Networks: A Survey. Comput. Netw. 2002, 38, 393 -22.

38. Want, R. An Introduction to RFID Technology. IEEE Perv. Comp. 2006, 5, 25 -3.

39. RFID Journal. Available online: http//www rfdjomrnal com /faq/17 and http//www rfidjournal. com/EaqA8/ (accessed on 22 July 2009).

40. Chatschik, B. An Overview of the B luetoothW ireless Technology.lEEE Commun. M ag. 2001, 39,86 94.

41. SIG Bluetooth.. Bluetooth Specifications. Available online: http/bluetoothcom /Bluetooth/ Technology/Buil^ing/Specifica^tio^s/Defaullth^trn / accessed on 22 July 2009).

42. 2 1 +D UEPetrick-TA . , ( ( ( + D Q GER R N .EEE PiHEV2ai)4J Q H U 1 V

43. Baronti, P.; Pillai, P.; Chook, V .; Chessa, S .; Gotta, A .; Hu, Y . F. W ireless Sensor Networks: A Survey on the State of the Art and the 80215 4 and ZigBee Standards. Com put. Commun. 2007, 30,1655 9695.

44. Foerster, J; G reen, E .; Som ayazulu, S .; Leeper, D . U ltra-W ideband Technology for Short or M edium Range W ireless Communication. Intel Technol. J. 2001, Q2, 1-11.

45. W ireless USB Specification Revision 1.0. Available online: http//www usb org/developers/ wusb/wuäb_2007_0214 zip/ accessed on 22 July 2009).

46. Karygiannis, T.; Eydt, B .; Barber, G .; Bunn, L.; Phillips, T. Guidelines for Securing Radio Frequency Identification (RFID) Systems; NIST Special Publication 800-98; National Institute of Standards and Technology: Gaithersburg, M D ,USA , 2007.

47. Shoreey, R. M obile, W ireless, and Sensor Networks: Technology, Applications, and Future Directions; W iley-IEEE Press: Hoboken, NJ,USA, 2006.

48. W alters, J; Liang, Z .; Shi, W .; Chaudhary, V . W ireeless Sensor Network Security: A Survey, m Security in Distributed, Grid, Mobile, and Pervasive Computing; Xiao, Y , Ed.; CRC Press: Boca Raton, FL, USA ,2007; Chapter 16.

49. SpaceAge Control. Sensor Total Cost of Ownership. Available online: http//www .spaceage controlcom /sa54ahtm / (accessed on 22 July 2009).

50. Ibarguengoytia, P.; Vadera, S.; Sucar, L. A Prbabilistic M odel for Information and Sensor Validation. Comput. J". 2006, 49, 113 -26.

51. M etras, H . Trends inW ireless Sensors: Potential Contributions of N anotechnologies and Sm art Systems Integration Issues. In Proceedings of 2006 IST Conference, Helsinki, Japan, November2006.

52. STM icroelectronics. STM icroelectronics Unveils UItra-Low-Power Technologiy Plattform for 8-Bitand 32-Bit M icrocontrollers, Enabling Next-GeertonPower-Saving Products. Available online: http//www stcom /stonline/stappl/cm s/oress/hew s/year2009/o2389 htm / (accessed on 20 July 2009).

53. Fry, N D .; H olcom b, E D .; M unro, K J; O akes, CL.;M aston, JM . Compact Portable Electric Power Sources; Oak Ridge N atonal Laboratory ReportORNL/TM -13360; Oak Ridge National Laboratory:Oak Ridge, TN ,USA , 1997.

54. Glynne-Jones, P.; W hie, M N . Self-Pcwered System s: A Review cf Energy Sources. Sens. Rev. 2001,21,91 -7.

55. Rcurdy, JS. Energy Scavenging for W ireless Sensor Nodes with a Focus on Vibration to Electricity Conversion; PhD . D-sedation, Departm ent cf M echanical Engineering, University cf California, Berkeley, CA, USA, 2003.

56. Qiwai, A M .; Thom as, P J; KeDogg, C J; Baucom , J. Energy H arvesting Concepts for Sm all Electric Unmanned Systems. Proc. SPIE 2004, 5387, 84 95.

57. Paradise, A J.; Starrer, T. Energy Scavenging for M cbile and W ireless Electronics. IEEE Pervasive Comput. 2005, 4, 18 -7.

58. M ateu, L.; Moll, F. Review cf Energy Harvesting Techniques and Applications for M iroelectrriics.Prcc. SPIE 2005, 5837, 359 -73.

59. Eschenauer, L.; G ligcr, V . A Key-M anagem ent Schem e for D istrribuited Sensor Networks. Ir Proceedings cf the 9th ACM Conference on Computer and Communications Security, W ashington,DC, USA, November18 -2, 2002; pp. 41 -7.

60. Beats,B. Smart Sensors. IEEE Spectrum Mag. 2006, 43, 50 -3.

61. Akyill^iz, I.; Su, W .; Sankarasubram aniam , Y .; Cayirci, E. W ireless SensorNetworks: A Survey. Computt Netw. 2002, 4, 393 -22.

62. Kwcn, D .; A zarian, M .; Pecht M . EamlyD etectorLof Interconnect D egradatonby Continuous M ohitcring of RF Impedance. IEEE Trans. Dev. M ater. Rel. 2009, 9, 296 -04.

© 2010 by the authors; licensee M DPI, Basel, Switzerland. This article is an Open Access article distrdbuted under the terms and conditions of the Creative Commons AtttrbuitiorL license (http //creatveccm m orLscrg/licenises/by/3 .0/).

Copyright of Sensors (14248220) is the property of MDPI Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.