Data processing: measuring – calibrating – or testing – Measurement system – History logging or time stamping
Reexamination Certificate
2000-10-27
2002-11-05
Hoff, Marc S. (Department: 2857)
Data processing: measuring, calibrating, or testing
Measurement system
History logging or time stamping
C702S186000, C702S183000, C702S196000, C702S199000
Reexamination Certificate
active
06477485
ABSTRACT:
TECHNICAL FIELD
This invention relates to monitoring machines or systems, such as elevators or chillers, using multiple sensor data, treated as a random sequence, to construct a stochastic process model of the machine or system, comprising an empirical distribution of the sequence of discrete data; comparison of empirical distribution of data acquired daily on-line with the base information of the stochastic process model provides quantitative and qualitative indicators of system health; use of a cumulative distribution norm, for each machine or system, allows relative comparison to other machines or systems.
BACKGROUND ART
Conventional methods for detecting abnormal machine or system behavior typically use models to reconstruct the behavior from sensor data. Then a combination of schemes, based on knowledge of experts along with elementary statistical methods, are used to focus on what are deemed to be relevant features, looking for known troublesome patterns in the data. Such methodology suffers from an abundance of human influence: the relevant features are selected in accordance with not only what an expert knows about the system, but also what an expert believes about system behavior. Such approaches are typically troubled by too much data, by extracting only signals deemed to be relevant, and treating the remaining numerous records as noise.
DISCLOSURE OF INVENTION
Objects of the invention include determining abnormality in system behavior without the use of domain knowledge (knowledge and beliefs of experts about the machine or system domain); detecting abnormal system behavior well before there is any human-perceptible change; providing quantitative measure of severity as a function of deviation from normalcy; providing qualitative classification of abnormality, indicative of a problem to be fixed; monitoring system behavior without use of expert domain knowledge until after abnormality has been detected; monitoring behavior of systems in a manner that allows comparing relative system health of one system to that of a different system; and system health monitoring methodology that can be applied universally to different systems, without tailoring.
This invention is predicated on the concept that treating time sequences of system sensor data as a stochastic process model of the system allows a separate stochastic process model to be built for each different system requiring no specific domain knowledge as a form of self-learning step. The invention is further predicated on use of the cumulative distribution norm to allow comparison of relative system health between one system and another similar, or even dissimilar system. This invention is further predicated on the discovery that, although a vector represented by tens of bits has several million possible rearrangements, as little as on the order of 100 different states may be of value in monitoring a system.
According to the present invention, a data stream, such as periodic samplings of a plurality of sensors, representing the values of relevant parameters of the system, are first utilized to build an empirical distribution of the process, which represents a stochastic process model of the system. In one embodiment, bootstrapping methodology is used to build an empirical distribution, using a five-dimensional Markov chain model. Thereafter, the stream of system data is monitored, such as reading the sensors each five milliseconds or whenever any event in a system occurs, e.g., car call, floor switch, compressor turn-on, and abnormality is determined by comparison of the current (e.g., daily) information against the empirical distribution of the process. According to the invention, confidence intervals (deviations) from normal behavior are identified to provide a quantitative measure of system malfunction or other abnormality (if any). According further to the invention, selective processing of the data stream (by eliminating data from one or more sensors in each iteration of processing) qualitatively identifies abnormal behavior, by indicating one or more sensors with which the abnormality is associated.
In accordance with the invention still further, the cumulative distribution norm of deviation from normal behavior is utilized to permit comparing the relative health of one system with the relative health of another system, whether the systems be alike, similar or dissimilar, such as to determine which system should receive service first.
The present invention is totally data driven, and does not require use of human intervention until the source of a system abnormality is identified. The invention permits qualitative and quantitative identification of abnormalities in a given system, and permits comparing relative system health between different, even dissimilar systems.
Other objects, features and advantages of the present invention will become more apparent in the light of the following detailed description of exemplary embodiments thereof, as illustrated in the accompanying drawing.
REFERENCES:
patent: 5761090 (1998-06-01), Gross et al.
patent: 6202038 (2001-03-01), Wegerrich et al.
LaBarre Robert E.
Radulovic Dragan
Hoff Marc S.
Otis Elevator Company
Suarez Felix E.
LandOfFree
Monitoring system behavior using empirical distributions and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Monitoring system behavior using empirical distributions and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Monitoring system behavior using empirical distributions and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2970716