EP0965092A4

EP0965092A4 - Distributed diagnostic system

Info

Publication number: EP0965092A4
Application number: EP98908960A
Authority: EP
Inventors: Vojislav Divljakovic; Thomas Grudkowski; Joseph A Kline; Austin H Bonnet; George W Buckley; James P Lynch; Israel E Alguindigue; Nancy L Quist; Robert P Bauer; Roland I Hannula
Original assignee: Emerson Electric Co
Current assignee: Emerson Electric Co
Priority date: 1997-03-04
Filing date: 1998-03-04
Publication date: 2002-10-30
Also published as: AU6686298A; WO1998039718A1; EP0965092A1

Abstract

A distributed diagnostic system in which a plurality of local monitoring devices (12, 13) collect local information concerning various machines (11) and process that information, according to redefined diagnostic parameters, for diagnostic purposes. The local information collected by the plurality of local monitoring devices (12, 13) is provided to a global processor (15) that globally processes the collected information to provide updated diagnostic parameters to the local monitoring devices (12, 13).

Description

DISTRIBUTED DIAGNOSTIC SYSTEM

FIELD OF THE INVENTION

The present invention relates to systems and methods for diagnosing machines and, in

particular, to systems and methods for predicting the expected lifetime for and the failure of

rotating machines.

BACKGROUND OF THE INVENTION

Presently, many industrial process and operations depend on the proper and continued

operation of machines and, in particular, on the proper and continued operation of rotating

machines such as motors. The number of such rotating machines in operation today is

significant. For example, some have estimated that approximately 70% of all of the

electricity produced in this country goes to power rotating machines. Further, the proper

operation of such machines can have a significant economic impact on the operation of

industrial plants as the failure of a key machine, for even a short time period, can cause an

entire assembly line to come to a halt. In certain industries, for example the paper mill

industry, typical motor failure can result in costs in excess of $20,000 per hour when the

motor is down.

In an effort to ensure reliable and continued operation of such machines, and to avoid

unexpected failures, many have attempted to employ non-intrusive diagnostic or monitoring

methods or systems to locally monitor such machines in an effort to determine and, ideally,

predict machine failure. One goal of such systems and methods is to allow their users to

identify potential problems at an early stage and either take steps to avoid the potential

problem or replace the suspect machinery. Despite the widespread interest in diagnostic systems as described above, a practical,

reliable, low cost and convenient diagnostic system for machines in an industrial environment

has not yet been developed. This is especially true with respect to rotating machines such as

motors, where the absence of an effective, low-cost diagnostic system or method of

diagnosing motor health and life prediction is noticeable.

Prior art attempts to develop effective motor diagnostic systems and methods have

been limited. The vast majority of such systems simply locally monitor a specific machine

according to a fixed monitoring process to determine whether it is operating within a "fault

state" (i.e., a limited, predefined operating state) or a "no-fault state." These systems, while

providing some slight advance warning before a machine fails, do not provide information of

the type that may be readily used for preventative maintenance or for scheduled replacement

of certain machines. Also, the local monitoring processes used in such systems typically are

derived from laboratory tests on related machines and are, thus, not highly accurate in field

situations.

In addition to the above, many known motor diagnostic systems and methods require

the use of complicated, space-requiring, and expensive detectors and/or transducers for proper

operation. For example, when dealing with variable speed motors, one of the key parameters

often used in known systems is the rotational speed of the rotor. Often, the rotational speed is

determined through the use of an encoder or other similar device which includes a rotating

member coupled to the rotor of the motor and a stationary member that is coupled the stator

and that interacts with the rotating member to produce an output signal representative of the

rotational speed of the rotor. The components required by such encoders often require space that could otherwise be effectively used, result in increased motor costs, and are subject to

failure and/or breakage. Accordingly, many known diagnostic systems are necessarily

limited because of their dependence on such speed-sensing devices.

It is an object of the present invention to overcome these, and other limitations of the

prior art. Other objects of the present invention will be apparent to those of ordinary skill in

the art having the benefit of this disclosure.

SUMMARY OF THE INVENTION

In accordance with one exemplary embodiment of the present invention, a distributive

diagnostic system is provided for monitoring a plurality of machines where the system

includes a plurality of local monitoring devices, where each local monitoring device is

adapted to receive local data concerning at least one machine associated with the local

monitoring device, and where each local monitoring device further includes a data processor

adapted to: (i) communicate the local data concerning its associated machine and (ii) analyze

the local data concerning its associated machine using a set of provided parameters for local

diagnostics of the machine. The exemplary system also includes a global data processor

coupled to the plurality of local monitoring devices, where the global data processor is

adapted to receive from each local monitoring device the local data concerning its associated

machine and where, in response to the local data from the plurality of local monitoring

devices, the global data processor generates the set of provided parameters for each local

monitoring device. Other exemplary embodiments of the present invention and other features of the

present invention will be apparent to one of ordinary skill in the art having the benefit of this

disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 illustrates an exemplary distributed diagnostic control system constructed in

accordance with certain aspects of the present invention.

Figures 2A-2E illustrate in greater detail an exemplary machine and a local

monitoring device of the type illustrated in Figure 1.

Figure 3 generally illustrates a typical induction motor torque-speed, torque-slip

curve.

Figure 4 illustrates a novel circuit for determining the slip of an induction machine

using a flux sensor in accordance with certain aspects of the present invention.

Figure 5 generally illustrates the frequency spectrum that may be obtained through

appropriate processing of the digital signals corresponding to the output of a flux sensor in

accordance with certain aspects of the present invention.

Figure 6 illustrates an exemplary predictive routine in accordance with certain aspects

of the present invention.

Figure 7 generally illustrates the manner in which the input data for the exemplary

predictive routine of Figure 6 may be obtained.

Figure 8 illustrates the use of a local monitor device constructed according to various

aspects of the present invention with an external personal computer. Figure 9 provides a flow chart of the operation of a local monitor device constructed

according to various aspects of the present invention in a Birth Certificate mode.

Figure 10 illustrates a peak searching process that may be used by a local monitoring

device constructed according to certain teachings of the present invention.

Figures 11 A-l IC illustrate the types of loads often encountered by electric machines.

Figure 12 illustrates the operation of a local monitoring device constructed in

accordance with certain aspects of the present invention in a learning mode.

Figure 13 illustrates the operation of a local monitoring device constructed according

to various teachings of the present invention in a learning and diagnostics mode.

Figure 14 generally illustrates the use of time expansion factors in accordance with

one method of the present invention.

DETAILED DESCRIPTION OF THE INVENTION

Turning to the drawings and, in particular, to Figure 1 , an exemplary distributed

diagnostic system 10 constructed in accordance with certain aspects of the present invention

is illustrated. In general, the exemplary distributed diagnostic system 10 includes a plurality

of machines 11, where each machine is associated with and coupled to a local monitoring

device 12. In the exemplary system of Figure 1, each of the machines 11 is represented as a

conventional induction motor, although the present invention is applicable to other forms of

machines such as brushless DC machines, switched reluctance machines, and the like.

Each of the local monitoring devices 12 collects information concerning the

operational status of the machine 11 with which it is associated. For example, each local

monitoring device 12 may collect information concerning the vibrational characteristics of the machine 11, the temperature of the stator, windings and/or bearings of the machine 11,

and the flux established in the stator and rotor. This information may be stored in data

storage elements (not illustrated) positioned in the local monitoring devices 12.

The collected information concerning the various machines 11 is processed by each

local monitoring device 12 to produce a low-level indication reflecting the operational status

of its associated machine 11. This low-level indication may take the form of a visual

indicator of motor health that will provide a green indicator if the motor is operating properly,

a yellow indicator if the motor appears to be in a near-failing mode, and red indicator of the

motor is failing or has failed. The local monitoring devices 12 may also pre-process some or

all of the collected information for external communication and later processing as described

more fully below.

In one embodiment, each of the local monitoring devices includes a microcontroller or

microprocessor (not illustrated in Figure 1) that runs software establishing a local, low-level,

pattern recognition model that receives the collected information and, using the local model,

locally predicts the "health" of the motor. In this embodiment, the parameters that define the

local model may be downloaded to the local monitoring devices 12 as more fully described

below.

In the embodiment of Figure 1, each of the local monitoring devices 12 is adapted to

provide all or some of the collected information and all or some of the processed information

reflecting the operating condition of its associated machine 11 to a protocol translator 13. As

reflected in Figure 1 , various arrangements are possible wherein only a single local monitoring device 12' is coupled to a single protocol translator 13' or where multiple local

monitoring devices 12" are coupled to a single protocol translator 13".

Alternate versions of the protocol translator 13 may be used in the system of Figure 1.

In one embodiment, the protocol translators 13 simply receive information from the local

monitoring devices 12 using one communications protocol and converts the information such

that it can be transmitted to a site processor 14 using a second communications protocol. In a

second embodiment, the protocol translator 13 has some "intelligence" and periodically polls

and collects information from the local monitoring device 12 to which it is coupled. The

construction of either described protocol converter 13 will be well within the ability of one of

ordinary skill in the art having the benefit of this disclosure.

Referring to the example of Figure 1, all of the protocol translators 13 are coupled to

communicate with a site processor 14, which in the exemplary system is a personal computer.

The site processor 14 receives and processes the collected information from the local

monitoring devices 12 via the protocol translators 13. In embodiments where the site

processor 14 is capable of receiving information using the same communications protocol

utilized by the remote monitoring devices 12, the protocol translator 13 may be eliminated.

In one embodiment of the present invention, the site processor 14 is a computer

running a global diagnostic program that constantly receives the collected information

generated by the local monitoring devices and generates information concerning the likely

failure of the various machines. The site processor may use this information to provide an

indication of likely machine failure and/or to initiate machine shutdown or other corrective

action. In a further embodiment, the site processor 14 is a personal computer that is running a

global neural network program that receives as its inputs the information from the local

monitoring devices 12 and provides as outputs information representative of the operating

characteristics of the various machines 11. As explained more fully below, these inputs and

outputs may be used to derive the parameters used by the local monitoring devices 12 to

establish the local models used by the local monitoring devices 12 for diagnostic purposes.

The global diagnostic program running on the personal computer 14 may include a

self-correcting algorithm, such as a neural network, that receives information from the local

monitoring devices 12 via the protocol translators 13 and uses that information to develop an

updatable statistical model that can provide useful information concerning the operating

condition and failure potential of the various motors 1 1. Since the global program running on

the site processor is adaptive it can "learn" from the information provided to it from the local

monitoring devices 12 and can build one or more global neural networks that can predict

motor operating conditions and failure with greater precision as more and more information is

provided to the global neural network programs from the various local monitoring devices 12.

The use of the adaptive program running on the site processor 14 allows the program

to receive and analyze information concerning the operation and failure of the machines 11 at

a particular site. Accordingly, while the adaptive program may be initially "seeded" with

basic information concerning likely motor operation and failure for the machines 11 , it can

adapt that information, by adjusting the parameters of the global neural network, to the

particular operating conditions of the site. Further, the personal computer 14 can periodically

download some or all of the parameters of the neural network to the local monitoring devices 12 for the local monitoring devices to use the updated parameters to locally determine the

operating conditions of the machines 11 associated with the local monitoring devices.

In one embodiment, the site processor 14, in addition to maintaining the global,

adaptive, neural network described above, also performs "high" level processing of the

information provided by the local monitoring devices 12. Such high level processing may

provide specific information about individual motors such as: expected lifetime, expected

time to failure, and desired maintenance operations. The results of this high level processing

may be provided to a human user of the system through the screen of the personal computer

14 or through some form of "alarm" indicators which would likely draw the attention of a

human operator to a potential or actual motor failure. In addition the personal computer 14

may monitor the information from the protocol converts to provide "trend" information

concerning a specific machine 11 or a group of machines.

In a further embodiment of the present invention, one or all of the local monitoring

devices 12 is programmed to run a low level adaptive program similar to the higher level

adaptive program running on the site processor 14. In this embodiment, the local adaptive

program running in such local monitoring device 12 will receive measured information

concerning the machine 11 to which it attached and use that information to update the

parameters of its particular adaptive program. Some or all of the local parameters generated

by these local monitoring devices will then be communicated to the personal computer 14

which can use these local parameters to generate "site-wide" updated parameters for feedback

to the local monitoring devices 12. In this embodiment, each "intelligent" local monitoring device can learn from its own motor and receive information derived from an analysis of all

of the motors that communicate with the personal computer.

As the above indicates, the system of Figure 1 represents a distributed, multi-layered,

diagnostic system in which a global self-correcting predictive algorithm running on site

processor 14 operates on information locally acquired by the local monitoring devices 12 to

determine and predict machine operation and failure. In sum, site processor 14 provides a

self-correcting predictive algorithm based on the collection of a number of similar motors

working under similar environmental and load conditions.

The "site" system disclosed above may be expanded by allowing the site processor 14

to communicate data, via a modem line or a direct communications connection, to a

centralized database 15. The data may be transferred using disk or tape, if necessary.

In the example of Figure 1 , the centralized processor 15 represents a centralized

processor running a "super-global" adaptive program that, receives information from the site

processor 14, as well as information from similar site processors 14' and 14" operating in

sites different from that of site processor 14. For example, site processor 14 could be a

processor running in one portion of an industrial plant, while site processors 14' and 14"

operate in different portions of the same plant. Alternately, site processor 14 could operate in

one industrial plant, and site processors 14' and 14" operating in different industrial plants or

in different parts of a given country. In either embodiment, the centralized processor 15

receives the parameters from the global predictive algorithms running on the site processors

14, 14' and 14" and uses that information to generate updated parameters that reflect all of the

information received from all of the site processors 14, 14' and 14". These updated centralized parameters, may be stored in a centralized database and feed back to the various

site processors 14, 14' and 14", which in turn may provide the parameters to the appropriate

local monitoring devices 12 for use in determining and predicting the operation and failure of

the machines 12 associated with the local monitoring devices.

As will be apparent to those of ordinary skill in the art having the benefit of this

disclosure the system of Figure 1, including the centralized processor 15, provides a

distributed, multi-level diagnostic and monitoring system in which several self-collecting

predictive algorithms run based on information and parameters derived from the comparison

of locally acquired information with statistical data collected in centralized database

maintained by central processor. Thus, the system of Figure 1 provides a useful system

capable of providing information useful for the monitoring and preventative maintenance of

electric machines and, in particular, electric motors. Through the use of centralized

processor 15 information may shared plant or industry wide for more effective machine

diagnostics.

Those of ordinary skill in the art will appreciate that the general system of Figure 1 is

but one exemplary system utilizing certain aspects of the present invention. For example, the

number of machines 11, local monitoring devices 12 and site processors 14 feeding into the

centralized processor 15 may be changed without departing from the present teachings.

Further, while the various devices of Figure 1 communicate over hard- wired lines, it will be

appreciate that wireless communication devices and/or a combination of wireless and

hardwired communications may be used without departing from the present invention. Various details of the components of the system of Figure 1 , and alternate

embodiments of such components, are provided below.

Figures 2A-2E illustrate in greater detail an exemplary machine 1 1 and a local

monitoring device 12 of the type illustrated in Figure 1. In this example, the machine 1 1 is a

squirrel-cage induction machine of the type available from U.S. Electrical Motors or the

Emerson Electric Co.

Turning to figure 2A, the machine 1 1 includes a rotating member referred to as a rotor

and an outer stationary member referred to as a stator (not illustrated). Both the rotor and the

stator are contained in a motor housing 20. The machine 1 1 may be of conventional

construction.

Coupled to the motor housing 20 is a local monitoring device 12. In the example of

Figure 2A, the local monitoring device comprises one or more electronic boards (not

illustrated in Figure 2A) coupled to several sensors affixed to the machine 1 1. The electronic

boards are positioned within a device housing 22 that is coupled to the motor housing 20.

Alternate embodiments are envisioned wherein the local monitoring device 12 is not directly

connected to the machine 1 1 but is, instead, positioned at a different location. In

embodiments where the local monitoring device is coupled 1 1 is directly coupled to the

machine 1 1 , the device housing 22 should be capable of protecting its contents from the

expected thermal and environmental conditions in which the machine 1 1 will operate.

In the example of Figure 2A, the device housing 22 supports visual indicators 23. In

the illustrated embodiment, the visual indicators comprise three lights (red, yellow, and

green) which provide local visual indications of the operating condition of the machine 1 1 as described above. The controls for the visual indicators 23 are provided by the electronics on

the electronic control boards of the local monitoring device 12. Alternate embodiments are

envisioned wherein other forms of indicator (e.g., audible) of visual indicators with more or

less than three lights are used.

A communications link 24 extends from the local monitoring device 12 to allow the

local monitoring device 12 to communicate and receive information and data from outside

sources. The nature of the communication link 24 will vary depending on the communication

scheme employed by the local monitoring device. For example, the communication link 24

may comprise co-axial cable, twisted wire cable or optical fiber, depending on the

communication scheme utilized by the local monitoring device 12.

Figure 2B illustrates in greater detail the electronics control boards housed in the

device housing 22.

In general the electronics control boards housed in the device housing 22 include a

communications board 26, such as a CT Network Communications Board, that is adapted to

communicate (i.e., transmit and receive) information and data over an appropriate

communications link 24. In the example of Figure 2B, the communications board 26 is

coupled to the communications link 24 such that information can be communicated over the

information link. The communications board 26 should include appropriate hardware,

software and/or firmware to allow the communications board 26 to receive and transmit

information according to one or more appropriate protocols. For example, the

communications board 26 may be adapted to communicate using wireless communication

techniques, or using standard communication protocols such as the HART, CT Net, Modbus+, Fieldbus or other similar protocols. In the exemplary embodiment of Figure 2B,

the communications board 26 is also adapted to control the visual indicators 23.

The communications board 26 may be constructed and configured using known

devices and techniques and the appropriate construction of such a board will be apparent to

those of ordinary skill in the art having the benefit of this disclosure.

Coupled to the communications board 26 is a main control board 27 that, in the

exemplary embodiment of Figure 2A, includes a microprocessor or microcontroller 28 and a

first data storage device 29. In one embodiment, the microprocessor 28 is a Motorola

MC68LC302, HC11 or HC05 type processor and the data storage device 29 comprises flash

memory, such as a flash memory device contained within the microprocessor 28 or an

external flash memory device such as an AT29C256FLASH part. Other external memory

devices, such as EPROM and DRAM devices may be used in conjunction with the

microprocessor to implement the system described herein. The construction of the main

control board 27 and the selection of the appropriate external memory devices will be

apparent to one of ordinary skill in the art having the benefit of this disclosure.

A standard modem device 30, such as an RS-485 modem, is also coupled to the

microprocessor 28 such that the microprocessor can communicate over the modem device 30.

As discussed in more detail below, the microprocessor 28 may use the modem device 30 to

communicate with a number of different instruments including a hand-held data

logger/transmitter 31. In the exemplary embodiment of Figure 2B, additional communication

devices are provided to allow the microprocessor 28 to communicate data and information.

Specifically, a RF transceiver 32 is provided to allow for "wireless" communications and a HART ASIC 33 or other appropriate device (e.g., a FR 3244 transmitter) is provided to allow

for the microprocessor to communicate using the HART protocol. Those of ordinary skill in

the art will appreciate that the communication devices 30, 32 and 33 are only examples of the

types of communication devices that may be used with microprocessor 28 and that other

devices (and other combinations of devices) maybe used. Embodiments are also envisioned

wherein communication devices such as devices 30, 32 and 33, are eliminated and all

microprocessor communications are accomplished through the CT protocol board.

In certain embodiments a dual-port memory device 40 (e.g., a dual port RAM) may be

positioned between the microprocessor 28 and the various devices used by the

microprocessor for communications. Figure 2B illustrates the use of such a device 40 in the

communications link between the microprocessor 28 and the control board 26.

As reflected in Figure 2B, the microprocessor 28 is adapted to receive as inputs

information provided from a sensor set that is adapted to sense various operating parameters

of the machine 11. Figure 2B illustrates one such exemplary sensor set including seven

sensors: 34a-34e, 35 and 36.

Sensors 34a-34e are RTD transducers that are positioned appropriately with respect to

the machine 11. In one embodiment, two of the RTD transducers 34a-34e are positioned near

two bearing devices positioned within machine 11 and the other RTD transducers are

positioned to detect the temperature of the windings 11 of machine 11, the temperature of the

machine housing, and/or the temperature of the environment in which machine 11 is

operating. The precise placement and use of such RTD temperature sensors will vary from

application to application and those of ordinary skill in the art will appreciate that fewer or more than five RTD transducers may be used to implement the teachings contained herein.

Also, those of ordinary skill in the art will appreciate that temperature detection devices other

than RTD transducers may be used to detect and provide information concerning the

temperature of machine 11 , its bearings, housing and/or environment.

In the embodiment of Figure 2B, the microprocessor 28 includes a plurality of built-in

A/D converters and each of the RTD transducers 34a-34e comprises a RTD device and an

amplifier (not illustrated in Figure 2B) that receives the output of the RTD device and

conditions the signal such that the output of the amplifier is an analog signal within the range

acceptable by the appropriate A/D converter of the microprocessor 28. In embodiments

where a microprocessor 28 not having built-in A/D capabilities is utilized a separate A/D

converter, and possibly separate communications devices, may be coupled between the RTD

transducers 34a-34e and the microprocessor 28.

In addition to the five RTD transducers 34a-34e, the microprocessor 28 also receives

as an input the output signals from a vibration sensor 35 that, in the embodiment of Figure

2B, includes a vibration detector coupled to an amplifier for proper conditioning of the

vibration signal in a manner similar to the conditioning associated with the RTD transducers

discussed above. The vibration sensor 35 may be positioned with respect to machine 11 to

detect mechanical vibrations (or the absence of such vibrations) from the machine 11 that are

induced when the machine 11 is running and/or at rest. In one embodiment, the vibration

detector 35 comprises an accelerometer, such an automotive accelerometer available from

various manufactures including Motorola. In the embodiment of Figure 2B, the microprocessor 28 also receives as an input the

output signals from an electromagnetic flux sensor 36 that includes a flux sensing device and

a conditioning amplifier. In general the flux sensor 36 should be positioned appropriately

with respect to the associated machine 1 1 to detect the magnitude of the flux existing in the

stator of machine 11. As explained more fully below, the flux sensor allows for a

determination of, among other things, the rotor speed and the load of machine 11 for use in

the normalization of the temperature and vibration information provided by sensors 34a-34e

and sensor 35.

The selection, construction, and positioning of sensors 34a-34e, 35 and 36, and the

coupling of such sensors to microprocessor 28, will be apparent to those of ordinary skill in

the art having the benefit of this disclosure. In general, the number, type and positioning of

the sensors should provide enough information for reliable prediction of machine failure

when combined with the statistical life data of similar machines. Sensors other than those

discussed above may be used without departing from the teachings herein. For example,

other sensor types (e.g., current and/or voltage sensors) may be added or substituted as

required for reliability or cost optimization.

Figure 2C illustrates in greater detail an alternate sensor set 200 that may be used to

provide information for reliable prediction of machine failure.

Referring to Figure 2C, a schematic for a sensor set 200 is provided. The illustrated

exemplary sensor set includes a number of various sensing elements that will be discussed in

greater detail below. In the illustrated example, the various sensor elements maybe

appropriately affixed to a 2-sided, 4-layered printed circuit board. The components used to construct the sensors may utilize surface mount technology, although through-hole

components may also be used.

The sensor set of Figure 2C includes four three-terminal temperature sensing devices

201, 202, 203, and 204. In the illustrated embodiment, each of the temperature sensors is an

AD22100 device that provides a variable analog output that varies with the ambient

temperature in the area of the sensing element. In the embodiment of Figure 2C, the

temperature sensor 201 is positioned to detect the ambient temperature of the electric circuit

board to which the sensor elements 201, 202, 203 and 204 are attached. Each of sensors 202

and 203 are positioned so as to detect the temperature near the front and rear endshield

bearings of the motor to which the sensor set is attached. Sensor 204 is positioned so as to

provide a temperature reflective of the temperature of the windings of the motor to which the

sensor set 200 is attached.

In one embodiment, the 202 and 203 sensors are coupled to the sensor circuit board by

suitable connectors and the temperature sensors themselves are embedded in the endshield or

other structure that holds the front and rear bearings. Figure 2D illustrates one such

embodiment. Referring to Figure 2D, an endshield 205 or other appropriate structure is

illustrated. The endshield 205 defines a angular bearing bracket or recess 206 adapted to

receive a suitable motor bearing. One or more pockets 207 is formed in the structure 205 and

the pockets 207 are sized to receive a temperature sensor of the type used for temperature

sensors 202 and 203. Thus, a temperature sensor may be placed in recess 207, and a bearing

may be placed in recess 206 such that the temperature sensor will provide an output signal

reflective of the temperature of the bearing. In this manner, the temperature sensor is held in close proximity to the appropriate bearing structure such that the bearing helps to maintain

the temperature sensor in its desired position and an accurate reading of the bearing

temperature may be obtained. In general, the depth of the recess 207 should be such that the

sensor B or C is positioned as closely as possible to the bearing.

Referring back to Figure 2C, the fourth temperature sensor, sensor 204, is positioned

to obtain an accurate reading of the temperature of the windings of the motor to which the

sensor set is coupled. In general, the sensor 204 should be positioned to obtain a temperature

reading that corresponds to the average temperature of the various motor windings.

Small filter capacitors 208 provide some limited filtering of the analog sensors 201-

204.

In addition to including temperature sensors 201-204, the sensor set 200 of Figure 2C

also includes a novel circuit for determining the flux associated with the machine to which

the sensor board is coupled. In general, the flux detecting circuit includes a magnetoresistive

microcircuit 209. The flux detector may be positioned to the machine housing of the

machine to which sensor set 200 is coupled. In general, the flux sensor 200 should be

positioned as far as possible from the electrical connections between the phase windings and

any drive devices coupled to the machine.

In the illustrated embodiment, the magnetoresistive microcircuit 209 comprises a

resistive circuit in the form of a Wheatstone bridge having three elements of a substantially

known resistance and a fourth resistive element whose resistance varies depending on the

strength of the magnetic field within which the magnetoresistive circuit is positioned. Two

terminals of the device are coupled to a known voltage supply and circuit ground, and the other two terminals are monitored to provide an indication of the strength of the magnetic

field within which the device is positioned (and thus an indication of the strength of the flux

associated with the machine). One magnetoresistive circuit suitable for the above-described

application is the HMC1001 one-axis magnetoresistive microcircuit available from

Honeywell.

Referring to Figure 2C, two terminals of the circuit 209 are coupled, respectively, to a

Vcc power supply and to a ground. The other two terminals from the device 209 are coupled

to the inputs of a differential amplifier 210. The differential amplifier is configured, via a

feedback capacitor and resistor, to provide an analog output signal that will vary according

the magnetic field near the circuit 209. Because this magnetic field will vary with the leakage

flux from the machine to which sensor set 200 is coupled, the analog output signal from

differential amplifier 210 will provide an indication of the leakage flux of the machine.

Certain magnetoresistive circuits, such as circuit 209, have a pre-set easy axis (a

preferred direction of the magnetic field to be detected) that is set along one axis of the

circuit. Under the influence of particularly strong magnetic fields, however, the preferred

axis can "flip," thus changing the electrical characteristics of the circuit. Certain such

circuits, such as circuit 20,9 have an on-chip current strap that allows for external re-flipping

of the axis in the event that the axis flips in the presence of a strong magnetic field. In the

illustrated schematic, a set/reset circuit is provided that will allow for resetting the circuit 209

in the event a high magnetic field is encountered.

In the illustrated embodiment, this resetting function is accomplished as follows: The

analog output from differential amplifier 210 is monitored by, for example, a microprocessor that converted the analog value to a digital value. If it is determined that the analog signal

has exceeded a preset maximum value corresponding to a high magnetic field, the

microprocessor or other monitoring device will general a flux circuit reset signal that is

provided to the set/reset circuit 211. The set/reset circuit 21 1 will, in response, generate a

reset signal that is applied to the circuit 209 so as to reset the circuit 209. Additional

information about alternate approaches for setting/resetting magnetoresistive circuits may be

found in the data sheets for the Honeywell HMC1001 and in Honeywell application note AN-

201 "Set/Reset Pulse Circuits for Magnetic Sensors" by Mike Caruso, both of which are

hereby incorporated by reference.

Referring back to Figure 2C, in addition to including the temperature and flux sensors

described above, the sensor set 200 also includes a novel sensor circuit for detecting

insulation failures in an electric machine. In general, the novel insulation failure detector

includes an insulation sensor 212 that has one insulation failure output for each phase of the

machine to which the sensor 212 is coupled. In the example of Figure 2C, the insulation

failure sensor 212 is coupled to a three-phase machine and there are, therefore, three output

leads from the insulation sensor 212. Each of the output leads from the insulation detector is

coupled to one terminal of a phase-specific current-limiting resistor 213. The remaining

terminals of the resistors 213 are coupled together at a common point 214. A cut-off device

215, such as an optically isolated transistor, is coupled between the common point 214 and a

detection node 216. Two current paths exist between the detection node 216 and ground. A

first path allows current to flow from ground, through a unidirectional current device 217 (e.g., a diode), to detection node 216. A second current path, through a light-emitting diode

and light detection circuit 218, allows current to flow from the detection node 218 to ground.

The insulation failure detection circuit 212 is constructed such that the current will

begin to flow through one of the two current paths between detection node 216 and ground

when the insulation of the machine fails. One exemplary embodiment of such an insulation

failure sensor for a single machine phase is illustrated in Figure 2D.

Referring to Figure 2E, an insulation failure sensor is illustrated for sensing the

failure of the insulation surrounding an insulated wire 219. In the specific illustrated

example, the insulated wire 219 is open of the wires that form the phase winding of the

machine to which the sensor set 200 is coupled. The insulated wire 219 is wound about a wire

220 that includes an uninsulated portion 221. Because the insulated wire is actually a portion

of the phase winding of the machine, it is subject to the same stresses as the phase windings

of the machine. As the phase winding is subjected to electrical stress and the insulation

begins to fail, the resistance of the electrical path between the insulated wire 219 and the

uninsulated wire 220 will begin to decrease. Eventually, an electrical path will be created

between the insulated wire 219 and the uninsulated wire 219 that can be detected by the

insulation failure detection circuit described above. Through use of the described circuitry,

insulation failures can be detected and monitored.

It should be noted that in an actual electrical machine, the degree of insulation

between any two wire segments of the phase winding will be approximately twice that of the

insulating layer separating the insulated wire 219 and the uninsulated wire 220 since each

wire segment will have one insulated coating and, thus, there will be two layers of insulation separating each segment of the phase winding. As such, the insulation detector described in

connection with Figure 2D may provide an indication of a potential insulation failure,

sufficiently prior to the occurrence of such a failure such that appropriate corrective action

may be taken.

Additional details concerning the insulation failure detection circuit may be found in a

related co-pending application, assigned to the assignee of the present application, Serial No.

08/972,579 entitled "Apparatus for and Method of monitoring the Status of the Insulation of

the Wire in a Winding by V. Divljakovic et al., filed on November 18, 1997 the disclosure of

which is hereby incorporated by reference.

Referring back to Figure 2C, the novel circuit set also includes an accelerometer

circuit 224 for detecting the acceleration/deacceleration of the electrical machine to which the

sensor set is connected. In the illustrated embodiment, the accelerometer circuit comprises a

piezioelectric device 225, that provides an analog voltage signal having a magnitude

corresponding to the degree of vibration to which the sensor 224 is subjected. In one

embodiment, the vibration detector may be an A5100 piezioelectric sensor, available from

Oceana. The sensor 225 should be positioned in a portion of the electric machine known to

vibrate when the machine is accelerated or deaccelerated. Because

acceleration/deacceleration of an electric machine results in the establishment of vibration

within the machine, the use of vibration detector 224 can provide information concerning the

acceleration/deacceleration of the machine.

The various outputs from the sensors comprising the sensor set of Figure 2C may be

provided to the main control board via suitable electrical connections. Depending on whether the microprocessor used to construct main control board 27 has a built-in analog-to-digital

(A D) converter, an external A/D converter may be used to transform the analog signals from

the sensor set to digital signals of the type appropriate for input to the microprocessor 28.

In general, the sensor set and the main control board together form a local monitoring

device 12. The specific physical structure of the local monitoring device may vary depending

on the particular application and on the electrical machine to be monitored. In general,

however, the local monitoring device 12 will consist of a number of appropriate sensors for

detecting physical parameters associated with the electrical machine to be monitored, analog-

to-digital converters for converting the sensed data into digital form, a microprocessor and

memory circuit for assessing and operating on the sensed data, and a communication circuit

for communicating with the microprocessor. A power supply for the referenced circuitry will

also typically be provided. A high-level block diagram of such a local monitoring device is

provided in Figure 2E.

As explained more fully below, a software routine running on the microprocessor 28

receives the infoπnation provided by the sensors described above, normalizes that

information, and uses that information ~ along with provided parameters — to perform local

diagnostics on the machine 11 with which the local monitoring device 12 containing the

microprocessor 28 is associated. The construction and assembly of the main control board 27

and any software or firm ware required to properly operate microprocessor 28, may be

conventional and will be within the ability of one of ordinary skill in the art having the

benefit of this disclosure. In one embodiment of the present invention, the microprocessor 28 comprises a

microcontroller (such as a Motorola HC1 1 microcontroller) in which is embedded a data

acquisition and local prediction program. This program may be embedded in software or

firmware (e.g., a EPROM or ROM) and may use several provided parameters to establish a

local model of the machine 1 1 and, in response to the information provided by the various

sensing devices, utilize the model to provide local diagnostic information concerning the

appropriate machine.

The data acquisition and local prediction program described above may comprise two

general routines: (i) a normalization routine which receives the raw information from the

sensors 34a-34e, 35 and 36 and normalizes the raw information to provide normalized

information about the state of the machine 11 that is not dependent on the machine load or the

environmental conditions in which the machine 11 is operating; and (ii) a predictive routing

(such as a neural network or clustering algorithm) that receives the normalized information

and, in response to such information, provides an output signal indicative of the remaining

life for the machine 11. In some embodiments the local prediction program may also include

or be combined with a routine that, in response to the raw or normalized information,

provides a recommendation concerning the operation of the local machine (e.g., decrease

load).

The normalization of the raw data from the sensors 34a-34e, 35 and 36 may be

performed locally within each local monitoring device 12 by a routine running on the

microprocessor 28. Such normalization is necessary because, the local machine model

established by the program running on microprocessor 28 will generally not be specifically directed to particular load or environmental conditions. As such, to conform the raw sensor

data — which is affected by load and environmental conditions — to data acceptable for use

in the model, normalization is required. For example, if the machine load changes from a

relative low load condition to a relatively high load condition the temperature of the machine

will typically rise. For diagnostic purposes, this rise in temperature should be attributed to

the change in the machine load and not to a change in the properties of the machine 11.

Equation 1 (below) provides one example of how the raw data from the temperature

sensors 34a-34e, 35 and 36 may be normalized to account for load and environmental

variations. Specifically, Equation 1 provides an exemplary normalization equation for

normalizing temperature data (from the bearings, windings or stator) to account for load and

environmental variations.

Equation 1 : T_N = (T_sensor - T_ambient)/L

In Equation 1 , T_N represents the normalized temperature information; T_sensor

represents the raw temperature reading from the appropriate sensor; T_ambient represents the

ambient temperature of the environment; and L represents the machine load. The information

for T_sensorand T_ambient may be obtained from appropriate sensors 34a-34e. The information L,

representing the machine load, may be obtained through conventional load sensors or load

measuring techniques. Alternately, for induction machines, the output of flux sensor 36 may

be used to generate the load information L according to a novel method in accordance with

certain aspects of the present invention.

For induction machines it is generally known that the rotational frequency of the rotor

f(r) is related to the synchronous speed of the stator field f(s) by a parameter referred to as the "slip" S of the machine. Generally, the slip S is expressed as a fraction of the synchronous

speed where S = (f(s) -f(r))/f(s) (Equation 2) and where f(r) mάfis) are in RPMs or equivalent

units. With this definition of slip, the slip S will vary from a value of 1 at start-up to a value

approaching zero at full speed.

In the typical operating range of most induction motors, there is a clear relationship

between the slip S of the machine, the rotational speed of the rotor fir) as a percent of the

synchronous stator speed/(s), and the torque output of the machine as a percent of rated

torque. Figure 3 generally illustrates a typical induction motor torque-speed, torque-slip

curve.

From Figure 3 it will be apparent to those of ordinary skill in the art having the benefit

of this disclosure that, knowing the slip of an induction machine S, it is possible to

approximately determine the torque output of the machine and, thus, the load L of the

machine and the rotational speed of the rotor.

Figure 4 illustrates a novel circuit for determining the slip S of an induction machine

using the flux sensor 36.

Referring to Figure 4 the output of flux sensor 36 is passed through a low pass filter

41 to produce a filtered version of the flux sensor output. The filtered output is applied to one

input of a two input digital comparator 42. The voltage across a bias resistor 43 is provided

to the other input of digital comparator 42. The digital comparator 42 will compare the

filtered output of the flux sensor with the voltage across resistor 43 and produce a signal

having a value of logic 1 when the filtered flux signal is greater than the voltage across

resistor 43 and a value of logic 0 when the converse is true. During normal operation of the machine, the output of flux sensor 36 will vary in an approximately sinusoidal fashion and,

thus, the value of the filtered flux signal will periodically vary above and below the voltage

across resistor 43. Thus, the output of comparator 42 will be a series of digital pulses.

The present inventor has recognized that, in general, the frequency associated with the

digital pulses at the output of comparator 42 will correspond to the rotational frequency of the

rotor fir). Thus, by monitoring the frequency of the digital pulse train at the output of

comparator 42 it is possible to obtain an indication of fir), which will provide an indication of

the speed of the rotor. The selection of the appropriate low pass filter 41 and the appropriate

voltage across resistor 43 will be apparent to those of ordinary skill in the art having the

benefit of this disclosure.

In Figure 4, the low pass filter, comparator 42 and resistor 43 are all individual

components and the digital pulse train from comparator 42 is provided as input to

microprocessor 28 which monitors the pulse train according to known techniques to derive a

digital signal corresponding to fir). Alternate embodiments are envisioned wherein the raw

analog output from sensor 36 is converted to a digital value and the low pass filtering and

comparison associated with comparator 42 are accomplished through appropriate software.

In either embodiment the frequency of the pulse train produced by comparator 42 is used to

generate a digital signal corresponding to fir).

Referring to Figure 4, it may be noted that the raw output from flux sensor 36 is also

passed through a band-pass filter 44 which will pass only signals within a selected frequency

range. For most induction machines, the band-pass filter should be constructed to pass

frequencies near the expected synchronous stator frequency/^ which will typically be around 60 Hz. In the embodiment of Figure 4 the output form the bandpass filter 44 is

applied to an A-D converter (which may be built-in to microprocessor 28) and a Fast Fourier

Transform ("FFT") is performed on the digital signal at block 46 to determine the major

frequency component of the signal. This major frequency component will be a digital signal

and will correspond to the synchronous stator frequency /fs .

Using the digital signals corresponding Xofir) and fis) Equation 2 may be used to

determine the slip S of machine 1 1 and, using a look-up table or algorithm coπesponding to

the slip-torque curve of Figure 3, the output torque or load L of the machine 1 1 may be

determined. This load value L then be used for normalization purposes using Equation 1 ,

above.

In accordance with another embodiment of the present invention the load information

L for an induction motor may be derived through a routine running on the microprocessor 28.

According to this embodiment, the output from the flux sensor 36 is applied to an A/D

converter on-board the microprocessor 28. The digital signals corresponding to the flux

sensor output are processed, through the use of a digital low pass filter and FFT or other

appropriate techniques to, and the peak frequency component below a first predetermined

frequency is identified. For most applications the first predetermined frequency will be

approximately 50 Hz. This peak frequency component below the first predetermined

frequency corresponds to the rotor frequency fir).

In addition to analyzing the digital signals representing the output of the flux sensor

36, the routing may also use a digital high pass filter and FFT or other appropriate techniques

to determine the peak frequency component above a second predetermined frequency. For most applications, the second predetermined frequency will be just below the first

predetermined frequency. For example, if the first predetermined frequency is 50 Hz., the

second predetermined frequency may be 49 Hz. The peak frequency above the second

predetermined frequency will generally corresponds to the synchronous stator frequency or

f(s).

Using the values ϊoxfir) and fis) it is possible to determine the slip S, the torque

output or load L, and the rotor speed, using the methods previously described.

One benefit of the digital approach for determining the load information L is that it is

possible to confirm that the fir), fis) and S values are accurate. Appropriate confirmation

techniques are generally illustrated by Figure 5. Figure 5 generally illustrates the frequency

spectrum that may be obtained through appropriate processing of the digital signals

corresponding to the output of the flux sensor 36. Specifically, Figure 5 illustrates the peak

frequency below the first predetermined frequency fir) and the peak frequency above the

second predetermined value fis). As explained above these fir) ana fis) values may be used to

derive the slip value S.

In most induction motor applications, the synchronous stator frequency fis) will

correspond to the fundamental frequency of the power supply. Accordingly, a frequency peak

will typically be found at frequencies corresponding to three times and seven times this

fundamental frequency. Thus, once fis) is derived using the techniques described above, the

routine can look for frequency peaks near or at 3 *f(s) and 7*fis). The presence of peaks at

these frequencies (as reflected in Figure 5) will confirm that the calculated/fo) is the appropriate/fs). If appropriate peaks are not found, the routine can continue to calculate fis)

until a confirmed, valid/fo) is obtained.

In addition to providing for easy conformation of the validity of the calculated/^ the

digital technique described above allows for easy confirmation of the slip value S. For an

operating machine, a frequency peak will be expected at the frequency corresponding to

S*f(s). Thus, once S is calculated and a confirmed valid fis) is obtained, the routine can look

for a frequency peak near S*fis). The presence of such a peak (illustrated in Figure 5) will

confirm the validity of the S value.

The above example demonstrated how normalization techniques in accordance with

the present invention may be used to normalize data from temperature sensors. Similar

techniques may be used to normalize the data obtained from the vibration sensors. Moreover,

for the vibration sensor 35, techniques may be used to normalize the vibrational data to filter

out vibrations not attributable to the machine but rather to the mounting configuration of the

machine.

From the normalized infoπnation corresponding to the temperature sensors, flux

sensor and vibration sensor, various normalized information groups may be collected and

maintained in memory by a routing running on the microprocessor 28. For example, for a

given machine 11 and local monitoring device 12 the microprocessor 28 may collect and

store data corresponding to:

(i) the difference between the measured normalized bearing temperature and the

environmental temperature for various points in time; (ii) the difference between the normalized machine winding temperature and the

ambient environmental temperature;

(iii) the rotational speed of the machine as determined from the output of the flux

sensor;

(iv) the harmonics of the output of the flux sensor (for use in detecting broken

rotor bars);

(v) the spectral lines from the vibration sensor for the appropriate rotational

frequencies and their harmonics (for use in determining the frequency of

bearing failures);

(vi) the normalized aggregate time spent by the machine at certain temperatures;

(vii) the normalized aggregate time spent by the machine at certain rotational

speeds; and

(viii) the number of times the machine is started and stopped (which may be derived

from the output of the vibration sensor),

The identified data, collected and stored by the microprocessor 28, may be used

locally by the microprocessor 28 for diagnostic purposes or communicated by the

microprocessor 28 to the site processor 14 or to an appropriate protocol converter 13 for other

uses. The external communication of the collected and stored information may be initiated

locally by a routine running on the microprocessor 28 or in response to a polling signal from

the site computer 14, a protocol converter 13, or other device. Those of ordinary skill in the art will appreciate that the identified categories of data

to be collected and stored by the microprocessor are exemplary only and that other categories

of data may be included and that some of the identified categories may be omitted.

For example, in one embodiment the local monitoring device 12 may be configured to

store and update a number of different operating parameters relating to the electric machine

coupled to the local monitoring device. Specifically, the local monitoring device 12 may be

configured to include a unique identifier, such as a serial number, which may be used to

uniquely identify the electrical machine coupled to the device 12.

The local monitoring device 12 may also be configured to be stored in a memory

location counter data coπesponding to the number of motor starts. In general, this counter

may be incremented every time the electric machine coupled to the local monitoring device

12 is powered up. The counter may be temporarily stored in RAM memory associated with

microprocessor 28 and transferred to the flash memory on a daily basis such that the flash

memory in the local monitoring device includes information (updated daily) relating to the

number of times the electrical machine has been started.

Another important operating parameter that may be monitored by the local monitoring

device 12 is the total elapsed running time of the electric machine. This data may be

maintained by the microprocessor and/or written to the flash memory in the local monitoring

device on a periodic basis (e.g., once an hour).

As explained above, once the normalized and pre-processed data is obtained and

stored by the normalization routine running on microprocessor 28, one or more local predictive routines may use that data to provide diagnostic information concerning the

appropriate machine 11.

One exemplary predictive routine is generally illustrated in Figure 6.

The predictive routine illustrated in Figure 6 may be used to receive information

concerning the normalized temperature of the bearings of machine 11 and, based on that

information, provide local diagnostic information concerning the expected life of the

machine. The exemplary illustrated routine utilizes a local neural network, such as a

Kohonen network that receives as inputs appropriate normalized bearing temperature

information and provides as outputs an indication of the expected life of the motor bearings.

Refeπing to Figure 6, a two-layer neural network 60 is illustrated. As illustrated the

neural network includes three input nodes 61, 62 and 63 and six output nodes 64a-64e. As

those of ordinary skill in the art will appreciate, each of the output nodes receives as inputs

some or all of the outputs from each of the input nodes. In accordance with conventional

neural network techniques the outputs from the input nodes are appropriately "weighted"

such that the value of each output node will correspond generally to the sum of its weighted

inputs. In one embodiment the neural network is a "winner-take-all" network in which the

output of the network is determined by the output node with the highest value.

In the embodiment illustrated in Figure 6 the inputs to the three input nodes

coπespond to: (i) the normalized temperature Tnorm of the appropriate bearing at a time t₀

(for node 61); (ii) the gradient ATI between the normalized temperature Tnorm for the

bearing at time t₀ and the normalized temperature Tnorm for the bearing at an earlier time ti

(for node 62); and (iii) the difference ΔT2 between the normalized temperature Tnorm for the bearing at time t₀ and the normalized temperature Tnorm for the bearing at a time t.₂ , where t.

₂ coπesponds to a time one time period before t.,. Figure 7 generally illustrates the measured

value of T_normover time and the manner in which ΔT1 and ΔT2 may be obtained.

Using the input information applied to the input nodes 61, 62 and 63 and the weights

assigned to the various outputs of the input nodes, the neural network will yield one output

node with a higher value that the other output nodes. In the example of Figure 6, each output

node 64a-64e corresponds to an particular value of expected bearing life. For example, node

64a represents an expected bearing life of 1 year, while node 64e represents an expected

bearing life of 7 years. Thus, by processing the information provided to the input nodes 61,

62 and 63, the neural network 60 will select one output node as the "winner" and provide an

indication of the expected life of the bearing being analyzed.

The information concerning the expected bearing life derived from the neural network

60 may be stored by microprocessor 28 for use in determining the overall health of the motor

(for selection of the appropriate red, yellow or green indicator) and/or for external

communication purposes as described more fully below.

In one embodiment of the present invention the parameters of the neural network 60

that define the weights of the various outputs for the input nodes 61, 62 and 63 are stored in

the data storage device 29 associated with microprocessor 28 and are accessed by the routine

running on microprocessor 28 that establishes the neural network 60. These parameters are

referred to herein as the "weighting parameters."

The weighting parameters may be provided to the various microprocessors 28

associated with the various local monitoring devices 12 in a number of ways. In accordance with one embodiment of the present invention the weighting parameters are developed by

first establishing a global neural network similar to neural network 60 and then "training" the

neural network through known training techniques using data obtained through accelerated

aging tests. In this embodiment, accelerated aging data (e.g., data coπesponding to the t._n t.

_l, to_, ti t_n points) is obtained and converted into real time intervals by properly annotating the

outputs of the neural network to be trained. Once this global neural network is trained with

the accelerated aging data, the resulting weighting parameters can be downloaded into the

data storage devices 29 of each of the local monitoring devices through the communication

systems generally illustrated in Figure 1.

In the above embodiment the global neural network that is trained may be any type of

appropriate neural network or predictive algorithm, including a back propagation network, a

general recession network, a self-organized map, or a feed-forward network.. The

accelerated aging data used to train the global network may include accelerated data relating

to the thermal aging of the machine insulation, the thermal aging of the machine bearings,

and the electrical aging of the machine bearings.

While the previously-described approach to establishing the weighting parameters for

the various local monitoring devices 12 is acceptable for many applications, it is limited in

that the data used to establish the weighting parameters through training of the global neural

network was obtained from laboratory tests. As those of ordinary skill in the art will

appreciate, laboratory tests, while highly accurate, often cannot exactly replicate all scenarios

actually encountered in the field. Moreover, the data used to train the global neural network

in the previously-described approach is obtained from a limited number of motors. Each of these motors will have been constructed according to a particular manufacturing process and

from a certain group of materials. Accordingly, while the weight parameters obtained when

the laboratory data is used to train the global neural network may be valid for the laboratory

tested motors, they may not be as valid for motors manufactured using a different

manufacturing process or formed from different materials or for motors operating in different

environments.

As a further enhancement of the previously-described approach to machine

diagnostics, the present invention contemplates the use of a distributed diagnostic system in

which data about a plurality of machines is regularly collected in the field by local monitoring

devices, in the manner previously described. This field-collected data is then provided to a

centralized data processor running one or more global neural networks. Each of these global

neural networks will use the field gathered data as training data to develop updated weighting

parameters that will, in turn, be provided back to the local monitoring devices for local

diagnostic purposes.

Figure 1, previously described, illustrates certain aspects of this distributed diagnostic

system. When the system of Figure 1 is used to implement a distributed diagnostic system as

described above, each of the local monitoring devices 12 will include a microprocessor

running a local predictive neural network, such as neural network 60 as described above.

Initially, each local predictive neural network will be established using weighting parameters

provided the appropriate local monitoring device. Typically, these initial weighting

parameters will be derived from accelerated aging data as described above. In the described system, each local monitoring device 12 will collect, pre-process and

normalize data about the machine 11 to which it is attached. At certain intervals, the local

monitoring devices 12 will provide this collected data (and data indicating when a machine

11 fails) to the site processor 14 via the protocol converters 13, along with data identify the

machine from which such data was obtained.

The site processor 14 will include a data processor running one or more global neural

networks for, e.g., predicting the expected life of machine 11. Each such neural network will

initially operate according to weighting parameters established from accelerated test data but

will also be adapted to receive the field-collected data from the local monitoring devices 13

and use such field collected data to update the weighting parameters. For example, whenever

a machine 11 fails, the collected data coπesponding to that machine may be used by such a

global neural network as a known data set for training purposes. Thus, each global neural

network running on site processor 14 will have weighting parameters that are initially

determined from accelerated testing data but that are refined, over time, in response to actual

field collected data. These globally updated weighting parameters may then be downloaded

to the local monitoring devices at various intervals to further enhance the local monitoring

devices ability to predict the lifetime of the machine to which it is attached.

As a still further enhancement of the describe system, the field collected data provided

to site processor 14 may be collected and forwarded, along with other information, to a

centralized database 15 that receives such information from other site processors 14', 14"

This centralized database 15 may include one or more "super-global" neural networks that receive the relevant field-collected data and develop updated weighting parameter data for

transmission to the various site processors 14.

The use of global or super-global neural networks as described above allows for

increased diagnostic capabilities. For example, a global or super-global neural network may

be able to analyze field collected data from a variety of machines manufactured at different

times or from different materials and determine that the failure modes or expected lifetimes

for machines manufactured at one time (or with a certain type of material) are different from

the failure modes and expected lifetimes for other machines. The processor running the global

or super-global neural network may then be able to take this information, develop specific

weighting parameters for such machines, and provide the updated, manufacturing or material

specific weighting parameters to the appropriate local monitoring devices. Similarly, a global

or super-global neural network may be able to develop weighting parameters that are specific

to a particular environmental or load condition.

According to one embodiment of the present invention the training of the global or

super-global neural networks may be based on the Weibull law. The Weibull law has been

found useful in determining the mean time to failure and mean time between failures for

various machines. Generally, the law holds that — for machines sharing some common

characteristic ~ the probability of a machine failure will be high at the inception of the

machine's life, will level down during the normal expected life, and will rise again as the end

of the expected machine life approaches. For a given group of machines, the data for this

Weibull characteristic will be initially unknown. It would be beneficial, however, to begin to

train the global and super-global neural networks using data reflecting the totality of the Weibull characteristic, rather than just the data reflecting the inception of the machines life.

Accordingly, for this embodiment, a Weibull factor is used in the training of the global and

super-global neural networks, such that training is initially disabled or minimized until

sufficient data covering the entire Weibull characteristic is obtained. In this way, the training

of the global and super-global neural networks will be enhanced.

In the example described above the local neural network 60 and the global land super-

global neural networks received information concerning the bearing temperature and

provided output information representative of the expected lifetime of the machine bearings.

Alternate local, global, and super-global neural networks are envisioned. For example, a set

of neural networks may receive as inputs data indicating the time spent by the machine at

various temperatures and the number of starts and stops. The outputs of such a neural

network may indicate the expected lifetime of the machine's insulation system.

Alternately, a neural network can receive data reflecting the past and present vibration

experience of the machine. Such data can, like the bearing temperature data, be used to

predict the expected lifetime of the machine's bearings. Still further, a neural network may

receive inputs reflecting the rotor flux frequency and provide an output indicative of rotor

failures.

Those of ordinary skill in the art will appreciate that, while the exemplary illustrated

neural network 60 comprised a two layer network, that other more or less complicated neural

networks may be used to practice the present invention. For example, neural networks having

three, four or more layers and a number of inputs and outputs different from that of network

60 may be used without departing from the scope of the present invention. The local monitoring device 12 described herein may be advantageously used in a

variety of applications including initial testing and quality control. For example, an electric

machine/local monitoring device pair may be operated in a "birth certificate" mode in which

the initial quality of the machine is assessed and the base operating parameter of the machine

are determined. The device pair may also be operated in a "confirmation" mode to ensure

proper installation of a machine and in a "monitoring" mode where the machine is monitored

over time. Each of the various operating modes will be discussed in greater detail below.

In each of the various operating modes of the local monitoring device/electric

machine pair, the local monitoring device will perform a number of "tasks" and may respond

to various machine events. For example, in the "birth certificate" mode, the local monitoring

device/electric machine pair, may communicate with an appropriately programmed personal

computer to both initialize the local monitoring device and to perform initial quality tests.

Referring to Figure 8, a local monitoring device/machine pair 80 may be placed on a

motor test pad 81 in compliance with NEMA vibration testing specifications. In the

illustrated example, the local monitoring device includes the sensor set illustrated in Figure

2C. The communications port of the local monitoring device is coupled via an appropriate

communications link to a personal computer 82. The personal computer 82 is coupled to a

drive device 83 that is capable of energizing the motor at various voltage and current levels

and at various operating frequencies. The drive 83 (which may be an converter, inductor,

PWM drive or other appropriate drive) has an output coupled to the phase windings of the

electric machine. A load or shaft drive device 85 may be coupled to the shaft output of the

machine. Figure 9 generally illustrates a flow chart of tasks that may be implemented by the

microprocessor contained in the local monitoring device during "birth certificate" mode

operation.

In the illustrated example, the flash memory for the local monitoring device/machine

pair includes a serial model number register that is initially set to zero. Accordingly, upon the

initiation of the birth certificate mode, the PC will provide at step 91 a data signal to the local

monitoring device assigning the local monitoring device/machine pair a specific serial

number and a model number coπesponding the electrical machine. The local monitoring

device will then store the serial number and model number at the appropriate location in its

Flash memory and may return the serial number and model number to the PC for

confirmation. Once the serial number and model number data is written into the flash

memory, initial data acquisition may begin at Task 1.

In Task 1 in the Birth Certificate mode, the local monitoring device 80 will acquire

data from the accelerometer circuit of Figure 2D as the electric machine is operated over a

variety of operating frequencies (e.g., from 0 Hz. to 1200 Hz.) in Step 92. The program

running in the local monitoring device may first mask the collected data with a Hanning

window to avoid problems with asynchronous acquisition of periodic data and then call

various statistical functions to operate on the collected data to provide useful baseline

vibration data concerning the operation of the electrical machine. In one example, the local

monitoring device will determine and store in flash memory: (i) the vibration sensor mean;

(ii) the vibration sensor variance or standard deviation; (iii) the vibration sensor range (e.g.,

the difference between the maximum and the minimum vibration data points); (iv) overall vibration characteristics of the machine; and (v) characteristic vibration spectrum frequencies

of the machine (e.g., rotational ball failure, inner and outer race).

Refeπing to Figure 9, in Task 1 the local monitoring device will first acquire data, in

Step 92, from the accelerometer of Figure 2D as the machine is operated over a desired range

of operating frequencies. The local monitoring device (or the PC which may receive the

accelerometer data from the communications circuit of the local monitoring device) may then

mask the collected data with a Harming window at step 93 and call a conventional statistical

analysis package to calculate the mean value of the detected vibration data, the standard

deviation of that data, the minimum and maximum data points and the range of the data

points at step 94. Those values may then be stored in the flash memory associated with the

local monitoring device at step 95.

After calculating the basic statistical data, the microprocessor in the local monitoring

device, or the PC, may perform a fast Forieur transform (FFT) on the acquired data at step 96

and perform a peak searching process to identify peaks in the collected data at step 97. The

peak searching process of the present invention is illustrated in greater detail in Figure 3, 10A

and IOB. In general, the process begins by analyzing the first peak of the FFT data by

"climbing" to the top of the peak and summing up the area of the peak as the climb rises from

the bottom of the peak to the top of the peak at Step 102. Once the top of the peak is

detected, the peak frequency value of the peak and the area under the peak is stored in

temporary storage at Step 103. The area for that peak is then updated until a flat part in the

FFT spectrum or the beginning of the next peak is detected at Step 104. The next peak is analyzed in the same fashion and the peak frequency and area values for the various peaks are

then aπanged in temporary storage in ascending order at Step 105.

In one embodiment, only the top twenty peaks are stored in temporary memory. Once

twenty peaks are obtained, the data for each subsequent peak is either lost (if the peaks are

less than the smallest stored peak) or are positioned within the stored data at the appropriate

location. Once the review of the FFT spectrum is completed, the data for the top twenty

peaks is stored in the flash memory.

Returning to Figure 9, once the peak searching process is completed and the data for

the top twenty peaks is stored in the flash memory, the overall vibration level in a desired

frequency range of interest is calculated at step 98 by summing up the areas for all of the

peaks that fall within the desired frequency range. The desired frequency range will vary

from application to application and machine to machine but, in general, will coπespond to a

range slightly greater than the frequency range that corresponds to the normal operation of the

machine.

Once the accelerometer data is obtained and processed in Task 1, the routine in the

local monitoring device or in the PC will obtain and analyze data from the flux sensor of

Figure 2D as the machine is operated over a desired frequency range at Step 99. In general,

the processing of the data from the flux sensor is handled like the data from the

accelerometer, in that, the data is first masked with a Hamming winder, analyzed using

conventional statistical techniques to produce mean, standard deviation, min max and range

data. A FFT is performed and the FFT spectrum is processed using the techniques described above to provide data concerning the top twenty flux peaks and the overall flux level for a

desired frequency range.

After completing Task 2 the appropriate routing will implement a Task 3 in which the

data from the voltage sensor is received and analyzed at Step 100. In Task 3, the voltage data

is collected, passed through a Hamming winder and analyzed to provide produce mean,

standard deviation, min/max and range data. In the illustrated embodiment, no FFT

processing of the voltage data is performed.

Following the processing of the voltage data in Task 3, the processor in the illustrated

embodiments will implement Tasks 4, 5 and 6 at Step 101. In each step, the temperature data

of either the first bearing temperature sensor (Task 4), the second bearing temperature sensor

(Task 5), or the winding temperature sensor (Task 6) is taken over a range of operating

frequencies and stored in temporary memory. The temperature data is then statistically

analyzed to produce appropriate mean, standard deviation, min/max and range data that is

stored in the flash memory associated with the local monitoring device.

In Task 7, the processor will collect temperature data from the ambient temperature

sensor and will calculate mean, standard deviation, min/max and range data that is stored in

the flash memory associated with the local monitoring device at Step 102. The processor

will then use the ambient temperature values determined in Task 7 to normalize the values

calculated in Tasks 4, 5 and 6, using the normalization methods and processes described

above, at Step 103. The normalized temperature values will then be stored in the flash

memory associated with the local monitoring device. Upon the completion of Task 7, the local monitoring device, or the PC via

communications with the local monitoring device, will take a number (e.g., 10 consecutive

readings of the output of the insulation failure sensor of Figure 2D) at Step 104. An eπor flag

will be set if an insulation failure is detected, otherwise, the process will pass Task 8 and

proceed to a final analysis step.

In the final analysis step, all of the initial data collected and stored in the flash

memory of the local monitoring device will be transmitted to the PC in response to an

appropriate data request at Step 105. The PC will then compare that data with statistical data

corresponding to other motors of the same model as the motor involved in the birth certificate

testing. That data from the motor under testing will be compared by the PC with the

statistical data for all other motors of the same model and, if the data for the motor differs

from the stored statistical average data for that model by more than a given amount (e.g.,

5%), the motor under testing will be rejected as falling outside established quality guidelines.

If the motor is within the quality guidelines, then the motor will pass the initial birth

certificate testing, and the motor data may be used to update the appropriate statistical data in

the PC at Step 106.

Once a local monitoring device and electric machine pair has received its "birth

certificate" data through the process described above, it may be operated in a "Confirmation

Mode" where the motor is analyzed that it has not been damaged in transient or after a

particular electrical or environmental act (e.g., a plant shutdown, a serious storm, etc.). In the

confirmation mode, the local monitoring device and machine pair may be placed on a test pad

of the type used in the birth certificate processing described above and the birth certificate tasks may be performed with the data being stored in a temporary location. If the data taken

during these confirmation mode operations differs from the data obtained in the birth

certificate mode by a given amount (e.g., more than 10%), a motor eπor or fault signal may

be provided indicating that the motor should be inspected for possible damage or serious

deterioration.

While the electric machine data obtained and stored by the local monitoring device in

the birth certificate mode may be useful in monitoring the corresponding electric machine

during its operation to determine the anticipate life of the machine, the degree to which such

data may be used depends, in large part, on the specific application for which the machine is

used. Knowing the typical load of the machine can allow for more efficient use of the birth

certificate data. Accordingly, the local monitoring devices of the present invention may be

used in a "learning" mode to determine the typical load for its associated machine. The

inclusion of such a learning mode is beneficial, in that, it allows for a single local monitoring

device configuration to be used on machines having a number of different applications.

As illustrated in Figures 11A-1 IC, there are three basic load characteristics for electric

machine operation. In the first application, illustrated in Figure 11 A, the electric machine

runs at a substantially constant speed and is subjected to a substantially constant load. This

type of load is often encountered when a motor is used to drive a conveyor belt on a

continuously operating assembly line. In the second typical application, illustrated in Figure

1 IB, the electric machine operates in an OF/OFF manner, where the machine is either ON

and running at a given speed and with a given load or OFF. This ON/OFF application may

be associated with the operation of a fan that is on only during part of the day or when room temperatures exceeds a desired value. In the third application, illustrated in Figure 11 C, the

electric machine is subjected to erratic load and speed changes. The local monitoring device

of the present invention my include routines that will allow it to properly assess the load

characteristics of the motor to which it is attached.

There are various reasons for determining the load characteristics of a motor. First, by

monitoring the load conditions of a machine, it is possible to identify specific load conditions

at which data may be taken to accurately diagnose the condition and anticipated future life of

the machine. In particular, to properly diagnose the operation of an electrical machine, data

readings from the sensor set of Figure 2D should be taken at the typical load of the machine.

Determining the load characteristics of the machine allows for the determination of this

typical load. Second, the load characteristics of a machine can provide insight into the

electrical consumption of the machine for energy management purposes and may allow for

analysis of an unknown load.

Figure 12 generally illustrates the operation of the local monitoring device in the

"Learning Mode." Once the local monitoring device is placed into the learning mode, it will

first attempt to determine the inertia of the load coupled to the electrical machine associated

with the local monitoring device. Once obtained, this load inertia data may be used to

diagnose the operation of the electrical machine.

The load inertia is obtained at step 120 upon initial start-up of the machine. In that

step, the local monitoring device collects data from the various temperature sensors on a

periodic basis (e.g., once every 1/100th of a second) and stores the collected temperature data

into an array after first passing it through a moving average filter (e.g., Temp(k)=(Temp(k-l)+ temp (K)+temp(k+l))/3). The data from the temperature aπay

is then analyzed after 10 points are collected and a curve fitting processed is used to produce

a linear equation that describes the collected temperature date over the ten points in terms of

Temperature^ M_x*time + b. Where M_x coπesponds to the gradient over the ten collected

points at Step 121. Once ten points are obtained, the gradient over the past ten points is

constantly calculated for each new point. An average gradient value is obtained as the

running average of all gradients.

While the temperature values are being calculated and the temperature gradients are

being calculated, the total time since energization is being monitored at Step 122. It is

assumed that, as the machine accelerates the load, the gradient of the temperature change in

the machine will continuously increase. Once the load is accelerated to the desired rotational

speed, it is assumed that the temperature gradient will drop. Once the average gradient value

drops, the initial lower value of the lower temperature gradient and the total time spent to

accelerate the load to this value will be stored in the temporary memory of the local

monitoring device at Step 123. If the gradient does not change or if the absolute temperature

exceeds a predetermined threshold (e.g., 135°C) the local monitoring device may assume that

the motor is in a locked rotor condition and set an appropriate alarm flag at Step 124.

Once the monitoring of the temperature gradient indicates that the machine has

appropriately accelerated the load, the local monitoring device will then collect a significant

number of samples (e.g., 4096) from the flux sensor at Step 125. This collected flux data

may then be passed through a Harming window and the resultant data may be subjected to a

FFT. The largest peak in the FFT spectrum between 0 and 120 Hz. may be identified and stored in the memory of the local monitoring device ~ as this value will correspond to the

frequency of the power line used to power the electrical machine at Step 126.

Once the initial load inertia data is obtained and stored by the local monitoring device,

the local monitoring device will proceed to Step 127 wherein it will determine the load

profile of the electric machine. As discussed above, for typical electrical machines, the load

profile will either correspond to a steady load, an ON/OFF load or an eπatic load. The ability

of the monitoring device to accurately determine the load profile of an electrical machine is

one of the key unique features of the device.

In the steps above, the local monitoring device will collect select data from the sensor

set over an extended period of time. In one method, the local monitoring device will generate

a temperature array that is updated every ten minutes for the first 100 hours of machine

operation. Each time the temperature array is updated, the local monitoring device will

monitor the winding and bearing temperatures by obtaining the output values of the

appropriate temperature sensors at a rate of, e.g., 10 samples per second. A temperature

reading will be obtained at a running average of three samples as described above. The

temperature readings will then be normalized using ambient temperature readings as

described above. The first normalized temperature reading will be stored in temporary

memory. Data from the flux sensor will be taken upon determination of a temperature

reading and the peak power will be obtained through an analysis of the flux data. Data from

the vibration sensor will be detected and, using the techniques described above, the rotational

speed of the machine will be determined. The data couplet of the time of the reading (k), the

temperature reading (which is a running average at time k), and the rotational speed will be written into one point of the aπay. This process will be repeated until the array is populated

with a statistically large number of data couplets (e.g., 1,000 couplets).

Once the array is populated with the appropriate number of

time/temperature/rotational speed couplets, the array may be analyzed by the local monitoring

device to determine the load profile of the electric machine. The method for determining the

load profile is illustrated in Figure 12, Step 127.

Referring to Step 127A, the local monitoring device will first attempt to whether the

load profile of the machine is a steady load as that is the most easily identifiable load profile.

The local monitoring device will first identify all members of the array that have temperature

values within a given amount (e.g., two degrees) of one another at Step 127A-1. The

identified members are then candidates for a steady load profile. If the number of array

members in the identified category is larger or equal to three, it will be necessary to determine

whether the load is steady or erratic. This determination is made by inspecting the other

parameter of each time/temperature/speed couplet at Step 127A-2. Assume that there are a

select number of samples (Tk, Tk+1 , Tk+2. . .) that have temperature readings within the

predefined range (e.g., +/- 2C). If the speeds for these select samples are substantially

constant, then the identified members of the array correspond to a steady load condition. This

can be determined by comparing the speed values for the identified samples. If the speed

values are all within a given amount (e.g., 5 RPMs), then a potential candidate for a steady

load has been identified. The remaining members of the aπay are compared to the average of

the select samples. If more than a given number (e.g. 30% or 300) have temperature data

within 2 degrees of the average temperature of the select samples and speed data within 5 RPMs of the average speed of the select samples, then the load may be characterized as

steady. A baseline load profile may then be created and stored in the flash memory

associated with the local monitoring device by storing the time/speed/temperature couplets

that have temperature data within 2 degrees of the average temperature of the select samples

and speed data within 5 RPMs of the average speed of the select samples at Step 127A-3. For

electric machines having steady loads, the average temperature and average speed data may

be used as baseline data for later diagnosis of the machine. Such baseline data may be written

into the flash memory of the local monitoring device.

If the local monitoring device determines that fewer than 30% of the aπay couplets

have temperature data within 2 degrees of the average temperature of the select samples and

speed data within 5 RPMs of the average speed of the select samples, then the load profile for

the electrical machine is either erratic or ON/OFF. To determine whether the load is ON/OFF

or eπatic, the array couplets are first grouped by the local monitoring device into groupings

having substantially the same temperature reading at Step 127B-1. For example, for one set

of aπay data temperature groupings of tk, tm and tn may exist, the local monitoring device

will then select one of the array couplets within a temperature grouping that was taken at a

time k and compare it to the array readings that were taken at times immediately preceding

the selected couplet (e.g., k-1, k-2, k-3) at Step 127B-2. If for more than 30% of the couplets

in the aπay the following rule can be established, then the load can be characterized as an

ON/OFF load: speed at k-3= speed at k-2 = speed at k-1 = speed at k and temp at k-3< temp

at k-2 < temp at k-1 < temp at k. If the load is neither steady nor ON/OFF, then the load is

erratic. When the load is ON/OFF, the baseline speed and temperature data for the machine

may be identified by first arranging all of the couplets in the array into groupings having

substantially the same temperate. The temperature grouping with the largest number of

couplets may then be selected and, within that group, the couplets may be analyzed to

determine whether a series exists where the couplets share substantially the same

temperatures and substantially the same speeds. If more than one series is identified, then the

longest series may be selected. The series may be used as the baseline data for the machine

and written into the flash memory of the local monitoring device's flash memory at Step

127B-3.

To identify baseline data for an electric machine having an erratic load profile, the

time/speed/temperature couplets in the array are first organized into groups falling within a

predetermined temperature range (e.g., groups of the same temperature or groups +/- 2

degrees of a given temperature) Step 127C-1. The temperature group with the largest number

of candidates should be selected and, within the group, the approach discussed above in

connection with the ON/OFF load may be used to identify an appropriate series for storage as

the machine's baseline data.

In addition to having the ability to identify the type of load associated with an electric

machine, the local monitoring device of the present invention may also be used to perform

life prediction and diagnostics on a machine. Such life prediction and diagnostics may take

advantage of baseline data stored while the local monitoring device is in the learning mode.

Figure 13 illustrates the operation of the local monitoring device in the life prediction and diagnostics mode. In this mode, the local monitoring device provides full diagnosis features

for the prediction of life.

Figure 13 illustrates the operation of the local monitoring device in the life prediction

and diagnosis mode for an electric machine whose load profile has already been either

provided to the local monitoring device or determined by the local monitoring device in

accordance with the methods described above. In Step 130, upon start-up, the local

monitoring device will recognize from the data stored in its flash memory the operational

mode of the machine (steady, ON/OFF, erratic) and the load profile.

Once this data is obtained, the local monitoring device will, in step 131, attempt to

determine whether the motor is operating in its "stable" load condition that corresponds to the

baseline load information written into the local monitoring device's flash memory in the

"learning" mode. This is because, for diagnostic and life prediction purposes, it is beneficial

to compare collected data and baseline data taken under approximately the same load

conditions. In step 131, the local monitoring device identifies the stable load of the machine

by attempting to recognize stable load conditions that include a temperature that is stable

within 2 degrees of the baseline temperature data written into the flash memory and a speed

that is within 5 RPMs of the baseline speed stored in the flash memory.

Once a stable and recognized load condition is identified, the local monitoring device

will collect the data previously collected in the birth certificate mode in Step 132. Reference

should be made to the description of the birth certificate mode for a description of such data.

In one embodiment, the birth certificate data is collected a number of times by the local

monitoring device (e.g., 5), averaged, and recorded in the memory of the local monitoring device for later analysis of provision to an external device (e.g., a PC) via the

communications board for the local monitoring device.

Following the collection of the birth certificate data in step 132, the local monitoring

device will collect other data useful for diagnostics and life prediction. In particular, in Step

132, the local monitoring device may collect a variety of data for diagnostics including data

relating to the rotational speed of the machine and data from the flux detector. As described

above and in more detail below, the data from the flux detector may be particularly useful for

life prediction purposes. As such, the local monitoring device may record the readings from

the flux detector and process such data as set forth in step 132. First, the local monitoring

device will perform an FFT on the flux data. From the FFT spectrum, the power supply

frequency may be identified (as the highest peak) and an appropriate digital filter may be

selected to filter out flux variations induced by the power supply. Once that is completed, the

local monitoring device can collect and store: (i) 3x, 4x, 5x, 6x, 7x, 8x, 9x and lOx from the

original spectrum, (ii) lx and 2x rotational data from the original spectrum; and (iii) lx and

2x data for the power supply frequency.

In addition to collecting flux data in step 132, the local monitoring device may collect,

process, and store data from the vibration sensor as follows: (i) a decimation of the raw

vibration data; (ii) data relating to vibration power (for, e.g., 300-400 Hz., 400-500 Hz., 600-

700 Hz., 800-900 Hz, 900-1000 Hz, 1000-1100 Hz., 1100-1200 Hz. and the ball, inner race

and outer race frequencies). From this data and the flux data, the following data is generated

by the local monitoring device and stored in the memory of the device: IX, 2X, 3X, 4X, 5X, 6X, 7X, 8X, 9X and frequency band data and side band peaks based on the rotational speed

of the machine.

In step 133, the temperature data collected in step 132 is normalized using the general

equation: Tnorm = Tcalc - (Tmeasured - Tambient). The measured and ambient temperature

readings are obtained from the temperature sensors coupled to the local monitoring device.

The calculated temperature is determined as follows. First, the speed of the machine is

determined using a speed sensor or the techniques described above. Second, the voltage

applied to the machine is determined using a voltage sensor. Then, using a polynomial that

described the torque as a function of the voltage and measured speed, the torque of the

machine may be determined. The selection of such a polynomial will vary from machine to

machine, but will be within the ability of one of ordinary skill in the art having the benefit of

this disclosure. From the calculated torque, the losses of the machine may be determined

and, based on the calculated losses, the temperature Tcalc may be determined.

In step 134, the data collected in step 132 and normalized in step 133 is stored in the

local monitoring device for either: (i) processing by a diagnostic routine running on the local

monitoring device or (ii) submission to a personal computer or other higher level device for

processing. The data that will be used for diagnostic and life prediction evaluation will be a

series of points collected in the time frame that are temperature dependent. The data may be

segregated into two categories. One set of data may be used for predicting the end life of the

bearings and the other set may be used to predict the end life of the insulation. For both types

of failure analysis, a "time expansion factor" may be determined by the device performing the

diagnostic operation. In step 135, the life expectancy of or aging of the bearings is calculated either by the

local monitoring device or a data processor that receives data from the local monitoring

device. The bearing aging is determined using the difference between the sensed bearing

temperature and the ambient temperature because such temperatures are relatively robust to

factors external to motor operation. The value of bearing temperature is a function of

ambient temperature, heat generation within the bearings and the effect of heating from the

windings. By considering ambient temperature the effect of local bearing heating becomes the

dominant factor affecting the feature value. The result of the life calculation may be used to

set an alarm or change the status of an indicator reflecting the state of the machine as

described above.

According to one embodiment of the present invention, before calculating the

anticipated bearing life, the local monitoring device — or the other processor analyzing the

collected machine data ~ will calculate a "time expansion factor" determined from past tests

on similar bearings. For example, assume that in accelerated aging tests for a particular

bearing show that the bearing lubricant lasted 600 hours at a temperature of 140°C. If the

average temperature of the machine under test is different than 140°C, e.g., 60°C, it is

possible through the use of a "time expansion factor" to determine not only how long the

lubricant will last at 60°C, but also to determine how often the bearing temperature should be

sampled and how to analyze the sensed data. This approach does not rely on any complicated

conversion factors, but instead relies on controlling the sampling of the machine's

temperature data. Figure 14 generally illustrates one approach for utilizing time factor expansion as

described above. Refeπing to Figure 14, a neural network for predicting bearing failure is

first trained using known techniques and the methods described above and bearing failure

data taken at a set sampling rate and temperature (e.g., at 140°C and one sample per hour), in

Step 140. In the illustrated example, it is assumed that the bearings failed after 10 aging

cycles, with each cycle taking four days. It is further assumed that readings from the sensors

are taken every hour. Through calculation, experimentation, or through the use of another

neural network system, it is possible to determine that one hour of aging at 140°C is

equivalent to EF hours of aging at 60°C, where EF is the time expansion factor. This is

performed at Step 141. Through the use of this expansion factor EF, it is possible to control

the sampling of data from the electric machine at 60°C such that the data can be used with the

neural network without the use of complex conversion formulas.

To determine how often data must be sampled at 60°C to be equivalent to data

sampled hourly at 140°C, it is first necessary to determine the total number of hours for each

aging cycle used to train the neural network. In the illustrated example, each aging cycle is

equal to six days of 6*24 - 144 hours. Thus, converting this to hours in the reference

temperature frame of the electric machine, each cycle will be EF * 144 hours.

When the bearing failure neural network is trained, it may be trained using

linear/piecewise interpolation between two aging/testing cycles and the number of training

sets inserted between each adjacent training set will be a given number (in this example 5).

Thus, when trained, neural network will need data from at least five aging cycles to properly

access bearing aging. To obtain data for five such cycles at 60°C, the bearing temperature should be sampled every EF* 144/5 hours at Step 142. The sampled data is provided to the

bearing failure neural network that was trained at Step 143 at 140°C and the output of the

network is used to access bearing aging. Once the appropriate number of samples are acquire,

the bearing failure neural network will accurately provide bearing failure information for the

machine at 60°C.

It should be noted that the above method uses the same neural network that was

trained with 160°C data to perform bearing failure without using any conversion factors to

convert the detected temperature data. Instead, the rate at which the data is sampled and

provided to the neural network is controlled using a time expansion factor, EF. This novel

method allows the use of neural networks trained at one temperature to accurately diagnose a

machine operating at a different temperature without changing the parameters of the neural

network or performing complicated conversion operations on the detected data.

While the example illustrated in Figure 14 illustrates the use of a time expansion

factor and a neural network to monitor bearing aging, the same approach may be used to

monitor and obtain life estimates for the stator. If such an approach is used, the average stator

winding temperature may first be determined. Second, an appropriate EF factor may be

calculated based on the difference between the average winding temperature and the

temperature at which the stator failure neural network was trained. Third, using the process

described above, the sampling rate may be calculated. Finally, the data may be sampled

according to the calculated sampling rate and provided to the neural network for analysis.

Referring back to Figure 13, once the data has been collected from the machine and

analyzed by the appropriate neural networks or stored for later analysis, the local monitoring device may set one or more flags depending on the status of the machine. This "alarm

annunciation" is accomplished at Step 136. While the exact nature of the alarms to be

identified will vary from machine to machine, typical alarm conditions include: (i) winding

temperature too high (which winding temperature exceeds predetermined maximum value

(e.g., 130°C); (ii) bearing temperature too high (which winding temperature exceeds

predetermined maximum value (e.g., 130°C); (iii) excess vibration (when the vibration

exceeds the baseline vibration determined in the birth certificate mode by more than a given

amount (e.g., 30%)); and (iv) excess flux (when the vibration exceeds the baseline vibration

determined in the birth certificate mode by more than a given amount (e.g., 30%)).

While the above example discusses the use of a temperature sensor data to predict

bearing or stator failure, it has been found that reliance upon data from a single source or

sensor can result in potentially eπoneous conclusions. In accordance with one embodiment of

the present invention, the neural networks are used to perform machine diagnostics and such

networks are trained and operate on inputs from a plurality sensor of the type discussed above

in connection with Figure 2C.

Although some success has been achieved in developing residual life predictors using

data just from one source, the vibration feature, the most robust results are achieved by fusing

thermal and vibration data with the prediction process. In this approach, the aging neural

network is presented with vector inputs which are not just time delayed vibration-based

feature values but also time delayed temperature-based features. Under this approach, the

neural networks described above will not utilize temperature data only but will also operate

on vibration data from the vibration sensor. In accordance with another embodiment of the present invention, the data from

multiple sensors is used to monitor machine aging in a formalized manner using a Hubert

space formulation. Effectively, this approach detects resonance between appropriate peaks in

the Fourier transform of data collected from different sensors, e.g., the flux and accelerometer

sensors. The technique is described in the following seven steps, assuming the case of two

sensors embedded in the motor:

1. Collected data from motor flux and accelerometer sensors, x, = [x, (l),x, (2), ...1 and

x₂ = [x₂(l),x₂(2),...] .

2. Apply Fourier transform (3) to x, and x₂ to obtain X, and X₂ correspondingly.

3. At each calculate frequency the correlation matrix between the two spectra,

^X,⁺(ω)X₁(ω) X,⁺(ω)X₂(ω)^

A{ω) , where ⁺ is the transpose of the complex vX₂ ⁺(ω)X,(ω) X₂ ⁺(ω)X₂(ω)

conjugate of a .

4. Calculate the eigenvectors e , (ω ) and e ₂ (ω ) of _4(ω ) , a Hermitian matrix, for all ω .

5. Calculate the signal vector, ξ(ω) = [X,(ω),X₂(ω)l, at each ω .

6. Multiplying the complex conjugate of an eigenvector with the observation vector and the

eigenvector. There will be one output for each sensor.

7. The complex number output of step six is used in creating a Nyquist plot, the axis being

the real and imaginary components, respectively.

The resultant plots are open to classical analysis using Kennedy-Pancu modal analysis

methods. Such plots can also be used in conjunction with neural networks to predict residual life. A typical plot indicating motor failure, midway during the accelerated aging test, derived

from flux and accelerometer readings is presented in Figure 15. If used in conjunction with

neural networks, it would be the pulsing and motion of the modal circles which would be the

feature tracked. Typically, a multiple feature detection algorithm such as the Hough

Transform would be used to extract data about the circles which, in turn, is fed into the neural

network, such as circle center co-ordinates and radii.

The above description of several embodiments is made by way of example and not for

purposes of limitation. The present invention is intended to be limited only by the spirit and

scope of the following claims.

Claims

1. A distributive diagnostic system for monitoring a plurality of machines, the system

comprising:

a plurality of local monitoring devices, each local monitoring device being adapted to

receive local data concerning at least one machine associated with the local monitoring

device, each local monitoring device further including a data processor adapted to

communicate the local data concerning its associated machine and further adapted to analyze

diagnostics of the machine; and

a global data processor coupled to the plurality of local monitoring devices, the global

data processor being adapted to receive from each local monitoring device the local data

concerning its associated machine;

wherein, in response to the local data from the plurality of local monitoring devices,

the global data processor generates the set of provided parameters for each local monitoring

device.

2. Apparatus for locally monitoring a rotating electric machine for diagnostic purposes,

the apparatus comprising:

a programmed processor; and

a set of sensors having outputs coupled to the programmed processor, each sensor

being positioned with respect to the rotating electric machine to provide information at its

output concerning the operation of the rotating electric machine; wherein the programmed processor: (i) receives the output information from the set of

sensors and normalizes the information to provide normalized information that is not

dependent on the machine load or the environmental conditions in which the machine

operates; and (ii)processes the normalized information to provide an output signal indicative

of the diagnostic condition of the rotating electric machine.

3. The apparatus of claim 2 further comprising a communication link, and a

communication board electrically coupled to the programmed processor and to the

communication link, the communication board being adapted to communication information

and data over the communication link.

4. The apparatus of claim 2 further including a visual indicator, electrically coupled to

receive electrical signal generated by the processor, for providing a visual indication of the

diagnostic condition of the rotating electric machine.

5. The apparatus of claim 2 wherein the rotating electric machine includes at least one

bearing structure and a phase winding and wherein the set of sensors includes:

at least one sensor positioned to provide an electrical signal coπesponding to the

temperature of the bearing structure;

at least one sensor positioned to provide an electrical signal corresponding to the

temperature of the phase winding; and

temperature of the environment in which the rotating electric machine is operating.

6. The apparatus of claim 5 wherein the set of sensors includes at least one vibration

sensor positioned to provide an electrical signal coπesponding to mechanical vibrations form

the rotating electric machine.

7. The apparatus of claim 6 wherein the set of sensors includes at least one flux sensor

that is positioned to provide an electrical signal corresponding to the magnitude of the electric

flux existing in the rotating electric machine.

8. The apparatus of claim 5 wherein the rotating electric machine includes at least one

endshield structure for holding the at least one bearing assembly and wherein the at least one

temperature sensor is embedded in the endshield.

9. The apparatus of claim 6 wherein the at least one flux sensor includes a

magnetoresistive circuit.

10. The apparatus of claim 2 wherein the rotating electric machine includes an insulated

phase winding and the set of sensors includes an insulation failure sensor, the insulation

sensor comprising:

a conductive element having a first end and an uninsulated conductive portion that is

would about a portion of the insulated phase winding; and

a detection circuit coupled to the first end of the conductive element for detecting the

presence of an electrical connection between the uninsulated conductive portion and the

portion of the insulated phase winding and, therefore, the existence of an insulation failure.

11. The apparatus of claim 5 wherein the programmed processor normalizes the bearing

structure and phase winding temperature information to provide normalized information that is not dependent on the machine load or the environmental conditions in which the machine

operates in accordance with the following equation:

^J- N ~ v ^J- sensor ~ J- ambient/' J-" where, T_N represents the normalized temperature information; T_sensor represents the raw

temperature reading from the appropriate sensor; T_ambient represents the ambient temperature

of the environment; and L represents the machine load.

12. Apparatus for producing an electrical signal indicative of the rotational speed

frequency of the rotor of an induction machine comprising:

a flux sensor positioned to provide an output signal that corresponds to the flux in the

induction machine;

a comparator having an output, a first input and a second input, the first input

receiving a reference voltage signal and the second input receiving the output of the flux

sensor, the comparitor producing at its output a signal having a first logic state whenever the

voltage at the first input is greater than the voltage at the second input and a second logic state

whenever the voltage at the second input is greater than the voltage at the first input; and

means coupled to the output of the comparator for producing an electrical signal

corresponding to the frequency at which the output of the comparator changes states, wherein

the electrical signal is indicative of the rotational frequency of the rotor.

13. A method of determining the slip of an induction machine having a rotor and a stator,

the rotor defining a rotational frequency and the stator defining a synchronous frequency, the

method comprising the steps of: monitoring the flux passing through a portion of the induction machine and providing

an electrical signal corresponding to the flux;

deriving a first digital signal coπesponding to the frequency at which the electric

signal coπesponding to the flux varies from above a predetermined value to below the

predetermined value, the first digital signal corresponding to the rotational frequency of the

rotor;

determining the major frequency component of the electrical signal coπesponding to

the flux and deriving a second digital signal corresponding to the major frequency

component, the second digital signal coπesponding to the synchronous stator frequency; and

determining the slip S of the induction machine in accordance with the following

equation:

S = (f(s) -f(r))/f(s) where fir)is the rotational frequency of the rotor, and fis) is the synchronous stator frequency.

14. A monitoring device for use in monitoring an electric machine, the monitoring device

comprising:

a programmed processor; and

being positioned with respect to the electric machine to provide information at its output

concerning the operation of the electric machine;

means operatively associated with the programmed processor for operating the

processor in a birth certificate mode wherein the outputs of the sensors are processed by the

programmed processor and stored as baseline operational parameters; means associated with the programmed processor for operating the device in a

monitoring mode, after the programmed processor has been operated in the birth certificate

mode, wherein the programmed processor processes the outputs from the sensors, compares

the processed outputs to the baseline operational parameters, and provides an indication of the

diagnostic condition of the electric machine based on the comparison.

15. The device of claim 14 wherein the sensor set includes a vibration sensor and the

baseline operational parameters include: (i) the vibration sensor mean; (ii) the vibration

sensor variance; (iii) the vibration sensor range ; (iv) the overall vibration characteristics of

the machine; and (v) predefined characteristic vibration spectrum frequencies of the machine.

16. The device of claim 14 wherein the sensor set includes an accelerometer for

measuring the acceleration of the electric machine, a voltage sensor for sensing the electric

voltage applied to the electric machine, and a temperature for sensing the temperature of at

least a portion of the machine.

17. The device of claim 14 further including a flash memory device coupled to the

programmed processor for storing the baseline operating parameters.

18. The device of claim 14 further including means associated with the programmed

processor for operating the device in a learning to determine the typical load associated with

the electrical machine and wherein the means for operating the device in the monitoring mode

provides an indication of the diagnostic condition of the electric machine based at least

partially on the typical load.

19. A method of determining the load characteristics of an electric machine, the method

comprising the steps of: using an electrical sensor set to periodically monitoring the temperature of at least a

portion of the electric machine and the rotational speed of the electric machine at given points

in time over first extended period and, for each point in time, storing in a digital memory a

data couplet containing information concerning the temperature, the rotational speed, and the

point in time;

using a digital processor to identify couplets having temperature values within a

predetermined temperature range; and

providing an indication of a steady state load if the temperature readings for at least a

first predetermined number of couplets are within a first predetermined temperature range and

the speed readings for the at least a first predetermined number of couplets are within a first

predetermined speed range; or

providing an indication of an ON/OFF load if, for at least a second number of

couplets, the following condition is true:

speed at k-3 = speed at k-2 = speed at k-1 = speed at k and temp at k-3 < temp at k-2 <

temp at k-1 < temp at k, where k coπesponds to the time of the couplet, and k-x corresponds

to the time of the xth preceding couplet.