Apparatus, systems, and methods for diagnosing anomalous mass flow controller operation

• US 5,410,495 A
• Filed: 07/20/1993
• Issued: 04/25/1995
• Est. Priority Date: 07/20/1993
• Status: Expired due to Term

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Ramamurthi, Krishnamoorthy
• Primary Examiner(s): Voeltz, Emanuel T.
• Assistant Examiner(s): MILLER, CRAIG S
• Application Number:
  US08/095,236
• Time in Patent Office: 644 Days
• Field of Search: 395/900, 395/175, 395/61, 395/912, 395/10, 395/50, 73/1 R, 73/3, 73/861, 364/510, 364/916.2, 364/221.3, 364/571.04, 364/571.05, 204/296.03
• US Class Current: 702/100
• CPC Class Codes: G05B 9/02 electric
• CPC Class Codes: G05D 7/06 characterised by the use of electric means G05D7/005 takes precedence
• CPC Class Codes: Y10S 706/90 Fuzzy logic

Abstract

A processing system (10) includes a mass flow controller (22) operating in response to a plurality of control signals, and a data processor system (40). Data processor system (40) collects and stores data characterizing the plurality of control signals from the mass flow controller (22). Features are extracted from the data which characterizes failure modes of the mass flow controller (22), such as miscalibration and malfunction. Data processor system (40) combines there features represented by membership functions using fuzzy rules to diagnosis anomalous operation of the mass flow controller (22). If the mass flow controller 22 is miscalibrated, the data processor system 40 automatically recalibrates it by the rate-of-nie calibration procedure. In case of malfunction, a warning is raised to the operator. Processing system (10) employs self-learning techniques to initialize and refine the knowledge base.

Claims

1. A method of diagnosing drift in an in-situ mass flow controller comprising the steps of:

collecting and storing data characterizing a valve voltage, control set point voltage, and feedback signal of the in-situ mass flow controller;

computing an actual valve voltage difference as a voltage difference between the valve voltage of the in-situ mass flow controller and a representation of valve voltage characterizing a calibrated mass flow controller at the same control set point voltage;

computing an actual feedback error as a voltage difference between the control set point voltage and the feedback signal;

selecting for the actual valve voltage difference at least one membership function in which the actual valve voltage difference has a membership from a knowledge base comprising a plurality of membership functions, the selected at least one membership function corresponding to a possible state of the actual valve voltage difference;

selecting for the actual feedback error at least one membership function in which the actual feedback error has a membership from a knowledge base comprising a plurality of membership functions, the selected membership function corresponding to a possible state of the actual feedback error;

combining said membership functions selected for said actual valve voltage difference and said feedback error features using fuzzy rules to diagnose anomalous operation of said mass flow controller and;

recalibrating the mass flow controller when a drift exceeding a preselected level is detected during said step of diagnosing.

2. The method of claim 1 and further comprising the step of generating the knowledge base, said step of generating the knowledge base comprising the substeps of:

collecting and storing data characterizing valve voltage and feedback signals at selected control set points for a calibrated mass flow controller and an uncalibrated mass flow controller;

computing valve voltage difference and feedback error values from the stored data corresponding the calibrated and uncalibrated mass flow controllers;

generating a first pair of clusters from the valve voltage difference data, each cluster of the first pair corresponding to one of the calibrated and uncalibrated controllers respectively;

generating a second pair of clusters from the feedback error values, each cluster of the second pair corresponding to one of the calibrated and uncalibrated controllers respectively;

computing a mean for each of the clusters of the first pair and a corresponding decision boundary;

computing a mean for each of the clusters of the second pair and a corresponding decision boundary;

generating a membership function for each of a preselected number of possible states of the valve voltage difference from the means and decision boundary computed for the first pair of clusters; and

generating a membership function for each of a preselected number of possible states of the feedback error from the means and the decision boundary computed for the second pair of clusters.

3. The method of claim 2 and further comprising the steps of:

identifying any misclassifications existing in said knowledge base; and

modifying said knowledge base using heuristic rules to eliminate any misclassifications.

4. The method of claim 1 wherein said step of combining comprises the substeps of:

combining the membership function of each selected possible states of valve voltage difference with the membership function of the selected possible state of feedback error;

applying the MIN function to a result from each combination of membership functions;

applying the MAX function to the results from the application of the MIN function to generate a composite membership function; and

diagnosing drift in the in-situ mass flow controller using the composite membership function.

5. The method of claim 1 wherein said step of recalibrating comprises a step of recalibrating based on a rate of rise criterion.

6. A method of detecting anomalous operation in a mass flow controller operating in response to a plurality of control signals comprising the steps of:

collecting and storing data characterizing the plurality of control signals;

extracting features characterizing selected failure modes of the mass flow controller from the data;

selecting for each feature at least one membership function in which the feature has a membership from a knowledge base comprising a plurality of membership functions, each membership function corresponding to a possible state of the feature;

combining the membership functions selected for the features using fuzzy rules to diagnose anomalous operation of the mass flow controller and;

recalibrating the mass flow controller when the diagnosed measurement drift exceeds a preselected amount.

7. The method of claim 6 wherein:

said step of collecting and storing comprises a step of collecting and storing data characterizing valve voltage, control set point voltage and control feedback of the mass flow controller;

said step of extracting comprises a step of computing a valve voltage difference and a feedback error from the stored data;

said step of selecting comprises a step of selecting for each of the valve voltage difference and feedback error at least one membership function from a knowledge base comprising a plurality of membership functions; and

said step of combining comprises a step of combining the membership functions selected for the valve voltage difference and the feedback error using fuzzy rules to diagnose measurement drift in the mass flow controller.

8. The method of claim 6 wherein:

said step of collecting and storing comprises a step of collecting and storing data characterizing rise time, control set point voltage and control feedback of the mass flow controller;

said step of extracting comprises a step of computing feedback error from the stored data;

said step of selecting comprises a step of selecting for each of the rise time and feedback error at least one membership function from a knowledge base comprising a plurality of membership functions; and

said step of combining comprises a step of combining the membership functions selected for the rise time and the feedback error using fuzzy rules to diagnose malfunction in the mass flow controller.