Bayesian belief networks for industrial processes

Details Bayesian belief networks for industrial processes Bayesian belief networks for industrial processes

1999-08-13

2002-07-02

Chaki, Kakali (Department: 2122)

Data processing: artificial intelligence

Knowledge processing system

Knowledge representation and reasoning technique

C706S024000, C702S084000, C702S183000

Reexamination Certificate

active

06415276

ABSTRACT:

BACKGROUND OF THE INVENTION
1. Field of the Invention (Technical Field)
The present invention relates to sensor and process analysis and diagnosis, especially through use of Bayesian Belief Networks (BBNs).
2. Background Art
The following patents discuss control systems:
U.S. Pat. No. 5,726,915, entitled “Automated System for Testing an Imaging Sensor,” to Prager, et al., issued Mar. 10, 1998. This patent discloses univariate techniques fortesting imaging sensors. This patent assumes a known signal. The method also uses a frequency-domain analysis step that forces collection of large amounts of data and shifting to a frequency domain. The method is a test suitable to determine if a sensor is suspect, which is likely to involve taking the sensor off-line and then carrying out this procedure.
U.S. Pat. No. 5,629,872, entitled, “System for Monitoring an Industrial Process and Determining Sensor Status, to Gross, et al., issued May 13, 1997 with a terminal disclaimer to U.S. Pat. No. 5,223,207. This patent discloses univariate techniques for system monitoring. The disclosed method also uses a frequency-domain analysis step that forces collection of large amounts of data and shifting to frequency domain. The method is a test suitable to determine if a sensor is suspect, which is likely to involve taking the sensor off-line and then carring out this procedure.
U.S. Pat. No. 5,661,666, entitled “Constant False Probability Data Fusion System,” to Pawlak, issued Aug. 26, 1997. The thrust of this patent's disclosed method is to compare input sensor data against the values in a lookup table. The threshold lookup table provides two thresholds. A “data fusion” processor uses the sensor “decisions” to generate a log-likelihood ratio, which is used as a test existence metric. The values of the threshold lookup table appear as weighted sums, which are normalized to represent in the probability domain. The invention provides sensor fault reconciliation only, which is based on a weighted sum and table lookup method. The method does not detect and isolate specific sensor faults.
U.S. Pat. No. 5,223,207, entitled “Expert System for Online Surveillance of Nuclear Reactor Coolant Pumps,” to Gross, et al., issued Jun. 29, 1993. This patent disclosed expert system technology for online surveillance of nuclear reactor coolant pumps through use of an artificial intelligence inference engine (an expert system) for early detection of pump or sensor degradation. The degradation is based on a sequential probability ratio test (SPRT). The invention provides sensor fault detection only and serves only as an early alert system to allow an “orderly shutdown of the pump” to avert serious damage to it. The method does not isolate the specific sensor fault. Further, it does not reconcile sensor data.
U.S. Pat. No. 5,548,378, entitled “Image Operating Apparatus Providing Image Stabilization Control,” to Ogata, et al., issued Aug. 20, 1996. This patent discloses a simple principle of comparing sensor-input data with certain standard values stored in the sensor controller. With this simple comparative method, a fault is detected as a disagreement between the sensor data and the stored standard values. The method is developed for a specific application, namely, use of photoelectric sensor for conveyor belt control (i.e., spacing of bottles on a conveyor belt).
U.S. Pat. No. 5,267,277, entitled “Indicator System for Advanced Nuclear Plant Control Complex,” to Scarola, et al., issued Nov. 30, 1993. This patent discloses a collection of tools to centralized signal display in a central location for a nuclear power plant, having concise information processing and display, reliable architecture and hardware, and easily maintainable components, while eliminating operator information overload. This method provides for a rapid response to changes in plant parameters and component control system. The “complex” includes six major systems: (1) the control center panels, (2) the data processing system, (3) the discrete indication and alarm system, (4) the component control system consisting of the engineered safeguard function component controls, (5) the plant protection system, and (6) the power control system. The six systems collect data from the plant, “efficiently” present the required information to the operator, perform all automatic functions and provide for direct manual control of the plant components.
U.S. Pat. No. 5,680,409, entitled “Method and Apparatus for Detecting and Identifying Faulty Sensors in a Process,” to Qin, et al., issued Oct. 21, 1997. This patent discloses a principal component analysis methods to detect a faulty sensor from the signals provided to it through a set of sensors. It also develops a validity index, which is a ratio of two residuals, to isolate the faulty sensor. A residual is calculated from the difference between the sensor output and the average of all sensor output signals. The principal component analysis is a transformation technique that converts a set of correlated sensor measurements into a set of uncorrelated variables. The effect of this transformation is to rotate the coordinate system in a way that results in the alignment of information represented by the sensor measurement on a fewer number of axes than the original coordinate system. This transformation results in a comparison of the variables by allowing those variables that are highly correlated with one another to be treated as a single variable.
U.S. Pat. No. 5,130,936, entitled “Method and Apparatus for Diagnostic Testing Including a Neural Network for Determining Testing Sufficiency,” to Sheppard, et al., issued Jul. 14, 1992. This patent discloses a combination of evidence theory and neural networks for improved diagnostic testing and determining the sufficiency of testing in diagnostic testing. The method receives input corresponding to at least one predetermined parameter of the system corresponding to its condition and produces a ranked set of diagnostic signals. Further, the system determines the sufficiency of the signal to ensure the validity of its diagnosis.
U.S. Pat. No. 5,715,178, entitled “Method of Validating Measurement Data of a Process Parameter from a Plurality of Individual Sensor Inputs,” to Scarola, et al., issued Feb. 3, 1998. This patent discloses a method to reduce information overload on a plant operator by providing means to display information in a concise, reliable, and easily maintainable manner. This patent's disclosure is aimed at improving overall effectiveness of a control room complex by providing novel designs for the alarm indicators, alarm processors, displays and the like. The patent also discloses a knowledge-based heuristic algorithm based on the explicit calculation of residuals among redundant sensors that measure a same variable.
U.S. Pat. No. 5,237,518, entitled “Optimization Method for Adaptive Sensor Reading Scheduling and Delayed Alarm Evaluation in Real-Time Diagnostic Systems,” to Sztipanovits, et al., issued Aug. 17, 1993. This patent discloses an optimization algorithm for use in an automated fault diagnostic system. It aims at scheduling an optimal sequence of evaluations of alarms that may be triggered by the diagnostic system.
None of the preceding patents disclose use of Bayesian belief networks for sensor diagnosis.
SUMMARY OF THE INVENTION (DISCLOSURE OF THE INVENTION)
The present invention comprises a method for diagnosis of sensors comprising: providing at least one sensor-status node wherein each sensor-status node comprises a known probability table; providing at least one process-variable node wherein each process-variable node comprises a known probability table; providing at least one sensor-reading node wherein each sensor-reading node comprises a probability table conditional on at least two known probability tables; providing the at least one sensor-reading node with at least one sensor reading; inferring a status of at least one sensor. The method of the present invention, more specifically, comprises at least one Bayesian belief network. Th

Affiliated with

Also associated with

Rating

Bayesian belief networks for industrial processes does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Bayesian belief networks for industrial processes, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Bayesian belief networks for industrial processes will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2860606