Method and apparatus for diagnostic testing including a neural network for determining testing sufficiency

• US 5,130,936 A
• Filed: 09/14/1990
• Issued: 07/14/1992
• Est. Priority Date: 09/14/1990
• Status: Expired due to Term

• Current Assignee: ARINC Incorporated (Raytheon Technologies Corporation d/b/a RTX)
• Original Assignee: Arinc Research Corporation (Raytheon Technologies Corporation d/b/a RTX)
• Inventors: Simpson, William R., Sheppard, John W., Graham, Jerry L.
• Primary Examiner(s): Lall, Parshotam S.
• Assistant Examiner(s): AUCHTERLONIE, THOMAS
• Application Number:
  US07/582,383
• Time in Patent Office: 669 Days
• Field of Search: 364/513, 364/550, 364/551.01, 364/554, 364/580, 364/579, 371/15.1, 371/20.1, 371/25.1, 371/26, 324/73.1
• US Class Current: 702/123
• CPC Class Codes: G06F 11/2257 using expert systems
• CPC Class Codes: Y10S 128/925 Neural network
• CPC Class Codes: Y10S 706/911 Nonmedical diagnostics

Abstract

A diagnostic tester evaluates at least one inputted test signal corresponding to test data relating to at least one predetermined parameter of a system being tested, to produce first and second candidate signals corresponding respectively to first and second possible diagnoses of the condition of the system respectively having the first and second highest levels of certainty of being valid, and first and second certainty signals corresponding respectively to values of the first and second highest levels of certainty. The diagnostic tester further determines the sufficiency of the testing that has taken place responsive to the first and second certainty signals, and produces an output signal indicative of whether sufficient test data has been evaluated to declare a diagnosis. Preferably, an uncertainty signal corresponding to a measure of the uncertainty that the evaluated at least one test signal can be validly evaluated is also produced and used to produce the output signal.

Claims

1. An improved electronic diagnostic tester for evaluating a failure state of a system having a defined number of functional elements, said diagnostic tester comprising:

an electronic signal input interface for providing electronic test signals;

evaluating means, responsive to at least one inputted test signal corresponding to test data relating to at least one predetermined parameter of the system, for producing;

a set of possible diagnoses of the failure state of the system ranked according to a level of certainty associated with each diagnosis of being valid;

first and second candidate signals corresponding respectively to first and second ones of said possible diagnoses respectively having the first and second highest levels of certainty of being valid, and

first and second certainty signals corresponding respectively to values of said first and second highest levels of certainty; and

testing sufficiency means, responsive to said first and second certainty signals, for producing an output signal indicative of whether sufficient test data has been evaluated to declare a diagnosis.

2. The diagnostic tester of claim 1 wherein:

said evaluating means also produces an uncertainty signal corresponding to a measure of the uncertainty that the evaluated at least one test signal can be validly evaluated; and

said testing sufficiency means also evaluates said uncertainty signal to produce said output signal.

3. The diagnostic tester of claim 2 wherein:

said evaluating means also produces a signal S_N corresponding to a measure of the level of certainty that a diagnosis not corresponding to said first or second candidate signals or to said uncertainty signal is a valid diagnosis;

said testing sufficiency means also evaluates said signal S_N to produce said output signal.

4. The diagnostic tester of claim 2 wherein said evaluating means comprises:

means for storing a model of predetermined information flow relationships within the system under test;

means for associating said test signals with sets of possible failure state causes in accordance with said model, and for ascertaining the intersections of said sets of possible causes for combinations of test signals, said candidate signals being derived from said intersections; and

means for determining for each candidate signal a measure of the level of certainty that it represents a valid diagnosis.

5. The diagnostic tester of claim 4 wherein said evaluating means comprises UR determining means for determining the degree to which a conflict exists between at least said first candidate signal and said at least one test signal indicative of the level of certainty that an unanticipated result (UR) has been obtained, said uncertainty signal having a value corresponding to said UR level of certainty.

6. The diagnostic tester of claim 5 wherein said UR determining means comprises:

means for defining a plurality of conclusions corresponding to possible diagnoses of the system;

means for assigning a confidence value to each successively inputted test signal;

means for calculating the support and denial for each conclusion based on said confidence value of the currently inputted test signal;

means for determining if the currently inputted test signal conflicts with a then selected set of said conclusions corresponding to candidate signals;

means for obtaining a measure of internal conflict based on accumulated total support values and the current test signal support values for all of said conclusions;

means for computing the present value of said uncertainty signal based on the number of test signals evaluated, if the present test signal is not a conflict; and

based on the number of test signals evaluated and the number of conflicts, if the present test signal is a conflict;

means for computing a new total support value for each conclusion based on the old total support value, the present test signal support value and confidence value, an old uncertainty value, and said measure of internal conflict;

means for computing a new uncertainty value based on the old uncertainty value, the present test signal confidence value and said measure of internal conflict;

means for normalizing the total support value for each conclusion based on the old total support value, and the present uncertainty signal value;

means for computing a new total denial for each conclusion based on an old total denial and the present test signal denial value;

means for computing the present plausibility for each conclusion based on the present total denial and the number of test signals evaluated;

means for selecting the current set of candidate signals based on the present total support and the plausibility for each conclusion.

7. The diagnostic tester of claim 5 wherein the values said certainty and uncertainty signals correspond to probability values.

8. The diagnostic tester of claim 7 wherein said training data includes first, second and third unambiguous combinations, said first unambiguous combination comprising said first certainty signal having a value corresponding to 100% probability and said second certainty signal and said uncertainty signal having zero values, said second unambiguous combination comprising said first certainty signal and said uncertainty signal each having a value corresponding to 100% probability and said second certainty signal having a zero value, and said third unambiguous combination comprising both of said certainty signals having values corresponding to uniform probabilities and said uncertainty signal having a zero value.

9. The diagnostic tester of claim 8 wherein said training data includes two data sets each having a predetermined number of ambiguous combinations wherein:

both data sets have certainty signal values within a closed interval [0,1] generated at random according to a lognormal distribution;

a first one of said data sets has uncertainty signal values generated at random according to a lognormal distribution;

a second one of said data sets has uncertainty signal values set to zero;

the values to generated are adjusted by normalizing the certainty signal values in each ambiguous combination to sum to (1-S_U), all normalized certainty signal values and the uncertainty signal values less than or equal to a predetermined value based on the number of elements in the system are set to zero; and

then the resultant certainty and uncertainty signal values in each of said ambiguous combinations are normalized to sum to the value one; and

the larger of the resultant normalized certainty signal values is designated as the first certainty signal value in each of said ambiguous combinations where the certainty signal values are unequal.

10. The diagnostic tester of claim 7 wherein:

the sum of the weighted outputs inputted to each of the intermediate nodes and the output mode is processed in the respective node with a sigmoidal activation function;

the weights of the first, second and third outputs of the first input node are (-9.640), 5.049, and (-1.624), respectively;

the weights of the first, second and third outputs of the second input node are 15.641, (-8.491), and 2.436, respectively;

the weights of the first, second and third outputs of the third input node are (-6.680), 3.627, and (-1.237), respectively;

the weights of the first, second and third outputs of the fourth input node are (-0.422), 0.442, and (-0.427), respectively;

the weight of the output of the first intermediate node is (-19.214);

the weight of the output of the second intermediate node is 11.457;

the weight of the output of the third intermediate node is (-2.483); and

the weight of the output of the fourth intermediate node is 4.081.

11. The diagnostic tester of claim 10 wherein said activation function is logistic.

12. The diagnostic tester of claim 11 wherein said activation function is 1/(1+e^-x), where x is the unweighted output of a given node.

13. The diagnostic tester of claim 2 wherein said uncertainty measure has a probability value.

14. The diagnostic tester of claim 2 wherein said testing sufficiency means comprises a neural network comprising:

a set of first, second, third and fourth input nodes each having first, second and third outputs weighted by a predetermined fixed weight, said first and second input nodes respectively receiving said first and second certainty signals as an input, said third input node receiving said uncertainty signal, and said fourth input node receiving a predetermined bias signal;

a set of first, second, third and fourth intermediate nodes each having an output weighted by a predetermined fixed weight, said first intermediate node receiving the sum of the first weighted outputs of said input nodes as an input, said second intermediate node receiving the sum of the second weighted outputs of said input nodes as an input, said third intermediate node receiving the sum of the third weighted outputs of said input nodes as an input, and said fourth intermediate node receiving said predetermined bias signal; and

an output node, responsive to the sum of said weighted outputs of said intermediate nodes, for producing said output signal.

15. The diagnostic tester of claim 14 wherein said output weights of said input and intermediate nodes correspond to a model of expert data correlating combinations of certainty signal values with at least one expert opinion regarding the ability to declare a diagnosis on the basis of each combination of certainty signal values.

16. The diagnostic tester of claim 14 wherein said output weights of said input and intermediate nodes correspond to a model of expert data correlating combinations of certainty and uncertainty signal values with at least one expert opinion regarding the ability to declare a diagnosis on the basis of each combination of certainty and uncertainty signal values.

17. The diagnostic tester of claim 14 wherein said output weights of said input and intermediate nodes have values obtained by training a backpropagation neural network having the same architecture as said neural network using empirical certainty and uncertainty signal training data obtained from testing of real systems, and conclusions derived from at least one expert regarding the sufficiency of the information represented by said combinations to declare a diagnosis for a specific system to be tested.

18. The diagnostic tester of claim 17 wherein said output weights of said input and intermediate nodes have generic values obtained by training a backpropagation neural network having the same architecture as said neural network using normalized randomly generated certainty and uncertainty signal training data, and conclusions derived from at least one expert regarding the sufficiency of the information represented by said combinations to declare a diagnosis for systems in general.

19. The diagnostic tester of claim 1 further comprising a controller responsive to said evaluating means and to said testing sufficiency means output signal for controlling said evaluating means and said signal input interface to control the test signals provided for evaluation and the continued evaluation of test signals.

20. The diagnostic tester of claim 19 wherein said signal input interface comprises operator actuated input means for entering test data relating to at least one predetermined aspect of the system, and for generating test signals corresponding to the test data.

21. The diagnostic tester of claim 19 wherein said signal input interface comprises an automatic tester responsive to a first control signal produced by said controller for performing selected tests on the system, and for generating test signals corresponding to the test results.

22. The diagnostic tester of claim 21

wherein said controller produces said first control signal to cause said automatic tester to perform at least one further selected test on the system if said output signal indicates that insufficient test data has been evaluated;

and produces a conclusion signal indicative of a diagnosis in accordance with said first candidate signal if said output signal indicates that sufficient test data has been evaluated.

23. The diagnostic tester of claim 22 wherein said controller produces said first control signal to select said at least one further selected test in accordance with at least one prior evaluation of test signals.

24. The diagnostic tester of claim 19 further comprising:

a display; and

said controller is responsive to said output signal for controlling said display to indicate the need for additional test data and for controlling said signal input interface to accept a further at least one test signal for evaluation by said evaluating means if said output signal indicates that insufficient test data has been evaluated; and

for controlling said display responsive to said first candidate signal to indicate a diagnosis if said output signal indicates that sufficient test data has been evaluated.

25. The diagnostic tester of claim 1 wherein the values of said first and second highest levels of certainty are probability values.

26. An automatic diagnostic tester for testing a system state according to a set of defined relationships between a plurality of sources for providing information about the system and a plurality of conclusions about the state of the system, said tester comprising:

an electronic signal input interface for providing electronic test signals corresponding to the information provided by the plurality of sources;

first control means for selecting the test signals provided by said signal input interface corresponding to successive ones of the information sources for evaluation according to the relative amount of information provided by each information source until the test signals for all of the information sources have been selected or the information which would be provided thereby has been determined;

second control means for further selecting the test signals provided by said signal input interface corresponding to successive ones of the information sources for evaluation according to which information source will increase the level of certainty that a selected candidate conclusion constitutes a valid diagnosis of the system state.

27. The diagnostic tester of claim 26 wherein said second control means comprises:

means for assigning a confidence value to the information provided by an information source;

means responsive to said confidence value and to the information provided by evaluated information sources for computing a level of certainty for each of the conclusions that the associated conclusion constitutes a valid diagnosis;

means for selecting candidate conclusions according to the relative values of the computed levels of certainty for each conclusion.

28. The diagnostic tester of claim 27 further comprising:

first means for identifying, if more than one candidate conclusion has been selected, which information sources, if any, depend on one of the two candidate conclusions with the then highest levels of certainty and not the other, and, if more than one such information source is identified, identifying which of such information sources furnishes the most information about the support for the candidate conclusion on which it depends.

29. The diagnostic tester of claim 28 further comprising:

second means, responsive to said first identifying means determining that no information source satisfies the identification criteria of said first identifying means, and responsive to said selecting means having then selected only one candidate conclusion, for identifying the information source, if any, that provides the most support for the candidate conclusion having the then highest level of certainty, if more than one candidate conclusion has then been selected, or for the candidate conclusion, if only one candidate conclusion has then been selected.

30. The diagnostic tester of claim 29 further comprising:

third means, responsive to said second identifying means determining that no information source satisfies the identification criteria of said second identifying means, for identifying the information source, if any, that supports the most conclusions other than the candidate conclusion having the then highest level of certainty.

31. An automatic controller for controlling the operation of a physical system, said controller comprising:

sensor means for sensing at least one predetermined system parameter indicative of an operating state of the system and producing output data signals indicative of the value of said at least one sensed parameter;

control means responsive to said sensor means output data signals for controlling a value of at least one operating parameter of the system to control the operating state of the system, said control means comprising;

evaluating means, responsive to said sensor means output data signals, for producing;

a set of possible diagnoses of the operating state of the system ranked according to a level of certainty associated with each diagnosis of being valid,

first and second candidate signals corresponding respectively to first and second ones of said possible diagnoses respectively having the first and second highest levels of certainty of being valid, and

first and second certainty signals corresponding respectively to values of said first and second highest levels of certainty; and

testing sufficiency means, responsive to said first and second certainty signals, for producing an output signal indicative of whether sufficient sensor output data signals have been evaluated to declare a diagnosis, said control means being responsive to said testing sufficiency means output signal.