Data processing: artificial intelligence – Neural network
Reexamination Certificate
1999-09-28
2004-06-29
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Neural network
C706S005000, C706S904000, C706S905000, C706S906000, C706S907000
Reexamination Certificate
active
06757665
ABSTRACT:
TECHNICAL FIELD
The present invention generally relates to a system and method for obtaining and using synthesized fault data for pump diagnosis and failure prediction.
BACKGROUND OF THE INVENTION
Motors, pumps and bearings require frequent maintenance attention in typical commercial systems and industrial plants. With conventional maintenance strategies such as exception-based and periodic-checking, faults developed in critical equipment (e.g. pumps) have to be detected by human experts through physical examination and other off-line tests (e.g., metal wear analysis) during a routine maintenance in order for corrective action to be taken. Faults that go undetected during a regular maintenance check-up may lead to catastrophic failure and un-scheduled shut-down of the plant. The probability of an un-scheduled shut-down increases as the time period between successive maintenance inspections increases. The frequency of performing maintenance, however, is limited by availability of man-power and financial resources and hence is not easily increased. Some maintenance inspections, such as impeller inspection may require stopping the process or even disassembling machinery. The lost production time may cost ten times more than the labor cost involved. There is also a possibility that the reassembled machine may fail due to an assembly error or high start up stresses for example. Finally, periodically replacing components (via routine preventive maintenance) such as bearings, seals, or impellers is costly since the service life of good components may unnecessarily be cut short.
Cavitation, blockage and impeller damage are common problems/faults encountered with pumps. Cavitation can cause accelerated wear, and mechanical damage to pump components, couplings, gear trains, and drive motors. Cavitation is the formation of vapor bubbles in the inlet flow regime or the suction zone of the pump. This condition occurs when local pressure drops to below the vapor pressure of the liquid being pumped. These vapor bubbles collapse or implode when they enter a high pressure zone (e.g. at the discharge section or a higher pressure area near the impeller) of the pump causing erosion of impeller casings as well as other pump components. If a pump runs for an extended period under cavitation conditions, permanent damage may occur to the pump structure and accelerated wear and deterioration of pump internal surfaces and seals may occur. Detection of such conditions before they become severe or prolonged can help to avoid cavitation-induced damage to the pump and facilitate extended plant up time. Such detection also can avoid accelerated pump wear and unexpected failures and further enable a well planned and cost-effective maintenance routine. Depending on the type of pump, other problems can occur such as inlet or outlet blockage, leakage of air into the system due to faulty pump seals, or the impeller or impeller parts impacting the pump casing.
Prior efforts in pump diagnostics have included vibration analysis and acoustic analysis techniques. For example, a modern chemical plant may require a service engineer to physically go to hundreds or even thousands of critical pumps periodically (e.g., monthly) to record vibration data from the pump. The data is then subsequently analyzed using vibration analysis algorithms to detect pump problems such as broken impeller vanes or out of balance conditions. Other research efforts have looked at performing pump diagnostics using process instrumentation such as flow meters and pressure transducers. Some efforts have looked at the relationship between inlet and outlet pressures and flow rate with pump speed to determine if a pump problem exists. Others have performed trending on these parameters over time.
Still other techniques have focused on signal analysis of unconditioned process sensors, such as flow sensors and pressure sensors. Flow sensors such as orifice-plate differential pressure, vortex, turbine, time-domain pressure techniques are invasive sensors and must be installed in-line within the process framework or pump system. Other flow sensors such as corriolis flow meters, are extremely costly and must be installed in-line with process piping.
In view of the above, there is a strong need in the art for a system and/or method for condition monitoring which mitigates some of the above-noted problems associated with conventional pump monitoring systems and/or methods.
SUMMARY OF THE INVENTION
The present invention provides for a system and method for condition monitoring of synthesized fault data and determining an operating condition of a pump. It has been found that fault data relating to the operating condition of a pump is encoded in variations in current of a motor driving the pump. These features present in the stator frequency spectrum of the motor stator current are caused by load effects of the pump on the motor rather than changes in the motor itself. The present invention provides a system and method for extracting (e.g., synthesizing) the fault data directly from the instantaneous motor current data. This data relates not only to pump machinery conditions, but also pump process conditions. Thus, by employing current signature analysis of the instantaneous current of the motor driving the pump, problems with the pump and/or process line can be detected without using invasive and expensive pressure and flow meters. Instead, a lower cost current sensor may be used and this sensor may be located in a motor control center or other suitable location remote from the motor and pump.
More particularly, in a preferred embodiment, the present invention includes utilizing an artificial neural network (ANN) to analyze the current signature data of the motor that relates to pump faults. Although, multi-iterative, supervised leaning algorithms could be used, which could be trained and used only when a fully-labeled data set corresponding to all possible operating conditions, the application of unsupervised ANN techniques that can learn on-line (even in a single iteration) are preferred. The present invention will be described with respect to an AC induction motor and both centrifugal and positive displacement pumps. However, it is to be appreciated that the present invention has applicability to substantially any type of pump and motor combination where current signature analysis can be performed to determine the operating state (e.g., health) of the pump. It should also be appreciated that the current signature analysis can be performed both on the pump to determine the operating state of the pump, and on the motor driving the pump to determine the operating state of the motor, simultaneously.
The present invention also provides for preprocessing of the fault signature data before it is being used to train an ANN or design a decision module based on ANN paradigms. The preprocessing eliminates outliers and performs scaling and bifurcation of the data into training and testing sets. Furthermore, it is desired to further post process the output generated by unsupervised ANN based decision modules for condition monitoring applications. This is because unsupervised ANN based decision modules when presented with a new operating condition can only signal the formation of a new output entry indicating that a possible new condition has occurred, but is not necessarily able to provide particular fault information. Post processing is carried out by utilizing the domain knowledge of a human expert to develop an expert system, or by correctly classifying this new operating state and encoding this information in a fuzzy expert system for future reference.
Furthermore, the present invention provides an intelligent, stand alone decision module which can identify the operating condition of the pump/plant without significant human supervision. The stand alone decision module employs adaptive preprocessing and intelligent post processing in conjunction with a one shot unsupervised ANN algorithm. Preferably, the ANN algorithm is either an adaptive resonance theory (ART-2)
Babu Vetcha Sarat
Discenzo Frederick M.
Unsworth Peter J.
Amin & Turocy LLP
Davis George B.
Gerasimow Alexander M.
Rockwell Automation Technologies Inc.
Speroff R. Scott
LandOfFree
Detection of pump cavitation/blockage and seal failure via... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Detection of pump cavitation/blockage and seal failure via..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Detection of pump cavitation/blockage and seal failure via... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3299992