US 20070106485 A9 Abstract The state or condition of a system may be evaluated by comparing a set of selected parameter values, converted into a trial vector, with a number of model or exemplar vectors, each of which was represents a particular state or condition of a sample system. Examples of such conditions may include “good”, “marginal”, “unacceptable”, “worn”, “defective”, or other general or specific conditions. Sets of parameter values from the system are converted into input vectors. Unprocessed vectors are then processed against the input vectors in an artificial neural network to generate the exemplar vectors. The exemplar vectors are stored in a memory of an operational system. During operation of the system, the trial vector is compared with the exemplar vectors. The exemplar vector which is closest to the trial vector represents a state which most closely represents the current state of the system. Thus, a high similarity between the trial vector and an exemplar vector which represent a “good” system is likely to have come from a “good” system.
Claims(36) 1. A method for detecting a state of a system, comprising:
generating a plurality J of n-tuple exemplar vectors representative of potential states of a system; operating the system; generating an n-tuple trial vector from n parameters representing an actual state of the system; of the J exemplar vectors, identifying an exemplar vector K which is closest to the trial vector, the distance from the trial vector to the identified exemplar vector K being an activation value; whereby, if the activation value is less than a first predetermined value, the actual state of the system is characterized by the exemplar vector K. 2. The method of initializing a plurality J of n-tuple unprocessed exemplar vectors with random values; globally conditioning the J unprocessed exemplar vectors; separating the conditioned exemplars; and consolidating the separated exemplar vectors. 3. The method of sequentially applying a plurality M of input vectors to the J unprocessed exemplar vectors; updating the J unprocessed exemplar vectors by a factor of α _{1}; and repeating the applying and updating steps over a predetermined number e _{1 }of epochs. 4. The method of sequentially applying the plurality M of input vectors to the J conditioned exemplar vectors; of the J conditioned exemplar vectors, selecting the conditioned exemplar vector which is closest to the currently applied input vector; updating the selected conditioned exemplar vector and a predetermined number of adjoining conditioned exemplar vectors by a factor of α _{2}; and repeating the applying, selecting and updating steps over a predetermined number e _{2 }of epochs. 5. The method of sequentially applying the plurality M of input vectors to the J separated exemplar vectors; of the J separated exemplar vectors, selecting the separated exemplar vector which is closest to the currently applied input vector; updating the selected conditioned exemplar vector by a factor of α _{3}; and repeating the applying, selecting and updating steps over a predetermined number e _{3 }of epochs. 6. The method of 7. The method of 8. The method of 9. The method of if the activation value is less than a second predetermined value, the second predetermined value being less than the first predetermined value, logging the activation value; if the activation value is greater than the second predetermined value, initiating a system calibration; and if the activation value is greater than a third predetermined value, the third predetermined value being greater than the second predetermined value and less than the first predetermined value, setting an alarm. 10. A method for detecting a state of a system, comprising:
operating the system; generating an n-tuple trial vector from n parameters representing an actual state of the system; retrieving a plurality of n-tuple exemplar vectors from a system memory, each exemplar vector representing a potential state of the system; and of the J exemplar vectors, identifying an exemplar vector K which is closest to the trial vector, the distance from the trial vector to the identified exemplar vector K being an activation value; whereby, if the activation value is less than a first predetermined value, the actual state of the system is characterized by the exemplar vector K. 11. The method of 12. The method of 13. The method of 14. The method of if the activation value is less than a second predetermined value, the second predetermined value being less than the first predetermined value, logging the activation value; if the activation value is greater than the second predetermined value, initiating a system calibration; and if the activation value is greater than a third predetermined value, the third predetermined value being greater than the second predetermined value and less than the first predetermined value, setting an alarm. 15. An apparatus for detecting a condition of a system, comprising:
a memory operable to store a plurality J of n-tuple exemplar vectors representative of potential states of a system; the memory further operable to store a first predetermined value; an input coupled to receive a plurality n of current parameter values from the system; means for generating an n-tuple trial vector from the n parameter values; a processor operable to determine a distance between the trial vector and each exemplar vector; the processor further operable to identify an exemplar vector K which is the least distance to the trial vector, the least distance comprising an activation value; and a comparator operable to compare the activation value with the first predetermined value, whereby if the activation value is less than the first predetermined value, the current state of the system is characterized by the exemplar vector K. 16. The apparatus of 17. The apparatus of 18. The apparatus of 19. The apparatus of compare the activation value with a second predetermined value, the second predetermined value being less than the first predetermined value, and if the activation value is less than the second predetermined value, log the activation value and if the activation value is greater than the second predetermined value, initiate a system calibration; and compare the activation value with a third predetermined value, the third predetermined value being greater than the second predetermined value and less than the first predetermined value, and if the activation value is greater than the third predetermined value, set an alarm. 20. A model for indicating a current state of a system, comprising:
plurality J of n-tuple exemplar vectors stored in a system memory and representative of potential states of a system; inputs for receiving n current parameter values from a system; and a routine stored in the memory and adapted to be implemented on a system processor to generate a trial vector from the n parameter values, compare the trial vector with each of the exemplar vectors and output an indication of a current state of the system based on the exemplar vector closest to the trial vector. 21. The model of 22. The model of 23. The model of 24. The model of log the activation value if the activation value is less than a second predetermined value, the second predetermined value being less than the first predetermined value; initiate a system calibration if the activation value is greater than the second predetermined value; and set an alarm if the activation value is greater than a third predetermined value, the third predetermined value being greater than the second predetermined value and less than the first predetermined value. 25. The model of 26. The model of 27. A method for creating exemplar vectors representative of potential states of a system, comprising:
initializing a plurality J of n-tuple unprocessed exemplar vectors with random values; globally conditioning the J unprocessed exemplar vectors; separating the conditioned exemplars; and consolidating the separated exemplar vectors. 28. The method of sequentially applying a plurality M of input vectors to the J unprocessed exemplar vectors; updating the J unprocessed exemplar vectors by a factor of α _{1}; and repeating the applying and updating steps over a predetermined number e _{1 }of epochs. 29. The method of sequentially applying the plurality M of input vectors to the J conditioned exemplar vectors; of the J conditioned exemplar vectors, selecting the conditioned exemplar vector which is closest to the currently applied input vector; updating the selected conditioned exemplar vector and a predetermined number of adjoining conditioned exemplar vectors by a factor of α _{2}; and repeating the applying, selecting and updating steps over a predetermined number e _{2 }of epochs. 30. The method of sequentially applying the plurality M of input vectors to the J separated exemplar vectors; of the J separated exemplar vectors, selecting the separated exemplar vector which is closest to the currently applied input vector; updating the selected conditioned exemplar vector by a factor of α _{3}; and repeating the applying, selecting and updating steps over a predetermined number e _{3 }of epochs. 31. The method of 32. A computer program product of a computer readable medium usable with a programmable computer, the computer program product having computer-readable code embodied therein for detecting a state of a system, the computer-readable code comprising instructions for:
operating the system; generating an n-tuple trial vector from n parameters representing an actual state of the system; retrieving a plurality of n-tuple exemplar vectors from a system memory, each exemplar vector representing a potential state of the system; and of the J exemplar vectors, identifying an exemplar vector K which is closest to the trial vector, the distance from the trial vector to the identified exemplar vector K being an activation value; whereby, if the activation value is less than a first predetermined value, the actual state of the system is characterized by the exemplar vector K. 33. The computer program product of 34. The computer program product of 35. The computer program product of 36. The computer program product of logging the activation value if the activation value is less than a second predetermined value, the second predetermined value being less than the first predetermined value; initiating a system calibration if the activation value is greater than the second predetermined value; and setting an alarm if the activation value is greater than a third predetermined value, the third predetermined value being greater than the second predetermined value and less than the first predetermined value. Description The present invention is related to commonly assigned and co-pending U.S. application Ser. No. 10/______ [TUC920040199US1], entitled IDENTIFYING A STATE OF A DATA STORAGE DRIVE USING AN ARTIFICIAL NEURAL NETWORK GENERATED MODEL, filed on the filing date hereof, which application is incorporated herein by reference in its entirety. The present invention relates generally to system diagnostics and, in particular, using an artificial neural network to generate a model representing one or more possible states of a system and comparing an actual state of the system to the model. Numerous types of systems include automated processes to generate one or more parameters which may be used to evaluate the current state of the system. The automated process may also be used to improve system performance or even repair certain defects or faults. For example, a data storage drive, such as a tape drive, may include an adaptive equalizer with many finite impulse response (FIR) taps whose input coefficients are automatically modified to optimize system performance. However, it may be difficult to assess the quality of the result of an automated process because of obscure relationships between the measured parameters and the system response. It will be appreciated that if the integrity of the parameter values is not verified, there is a risk that the automated process produces an undesirable system response. Thus, not only might the process fail to improve performance but, if the parameter values are undetectably invalid, may also cause the system to fail completely. For example, in an adaptive equalizer of a tape drive, FIR tap values are computed from information captured from the storage drive. If the information is corrupted, or if execution of the algorithm which is used to compute the tap values is corrupted, the FIR tap values will be invalid. More specifically, a media defect or servo error may corrupt the captured information. Similarly, an overflow or underflow may occur during the execution of the FIR tap algorithm, resulting in tap values which bear no relation to the proper results. Or, rather than the captured information being invalid, the information may result in the creation of tap values which are outside the range of values which can be handled by the equalizer. In each of these circumstances, the invalid or improper condition of the tap values may be undetected. Consequently, there remains a need for an automated process which provides an assessment of the quality of parameter values which are used to adjust the system. The present invention provides an assessment the state or condition of a system. Examples of such conditions may include “good”, “marginal”, “unacceptable”, “worn”, “defective”, or other general or specific conditions, depending on the specific system being evaluated and the desired specificity of the evaluation. Sets of n parameter values each from a model system are converted into n-tuple input vectors. Unprocessed n-tuple vectors are then processed against the input vectors in an artificial neural network (“ANN”) to generate a set of n-tuple exemplar vectors. The ANN preferably includes three stages, a global conditioning stage, a vector separation stage and a vector consolidation stage, to fine tune the creation of the exemplar vectors. Each exemplar vector will thereby represent a particular potential state or condition of the system. The exemplar vectors are stored in a memory of an operational system to be evaluated. During operation of the system, a set of n selected parameter values are converted into an n-tuple trial vector. The trial vector is compared with the exemplar vectors. The exemplar vector which is closest, measured by the distance between the two, to the trial vector represents a state which most closely represents the current or actual state of the system. Thus, a high similarity between the trial vector and an exemplar vector which represent a “good” system is likely to have come from a “good” system. Conversely, a high similarity between the trial vector and an exemplar vector which represent an “unacceptable” system is likely to have come from a “unacceptable” system. Optionally, the presence of predetermined states may be flagged or logged for follow-up attention. Additionally, if the distance between the trial vector and the closest exemplar vector is greater than a predetermined distance, the current state of the system may be a previously unknown state and the trial vector may subsequently be analyzed and used as another exemplar vector. Alternatively, such a situation may indicate that the parameters underlying the trial vector are invalid, thereby triggering a flag for immediate attention. Overview The first section The second section Exemplar Vector Creation Depending upon the particular system to be evaluated, not all of the available parameters may be necessary to obtain a satisfactory evaluation. For example, of the In some systems, there may be some ambiguity associated with the parameters, thereby obscuring the relationship between the parameter values and the physical “world” which is to be affected. For example, it may be difficult to determine what is happening within the equalizer merely by looking at FIR tap coefficient values because the tap values are a time domain representation of the frequency domain result. Thus, it may be desirable to convert the values from the time domain to the more recognizable state of the frequency domain, such as by using a Discrete Fourier Transform. Such a conversion, or other comparable operation, may be accomplished in an optional pre-processor The M sets of n parameter values are then processed by a vector creator In a further optional operation, the magnitudes of the vectors may be normalized to a common scaled magnitude (such as 1) in another unit In addition to the input vectors whose creation has just been described, unprocessed exemplar vectors are also created for input to the ANN Both the input vectors and the unprocessed exemplar vectors are input to the ANN In a first stage, global conditioning This adjusting step In a second stage, vector separation In a third, vector consolidation The finalized exemplar vectors processed through the ANN System Evaluation The trial vector and the exemplar vectors Even more specific evaluation may also be performed. For example, if the minimum distance or activation value is less than a second predetermined limit (which is less than the first limit), the activation value may be logged. If the minimum activation value is greater than the second predetermined limit, a system calibration or other comparable operation may be initiated. And, if the activation value is greater than a third predetermined limit (which is greater than the second limit), an alarm may be activated. As is known, the equalizer The controller The objects of the invention have been fully realized through the embodiments disclosed herein. Those skilled in the art will appreciate that the various aspects of the invention may be achieved through different embodiments without departing from the essential function of the invention. The particular embodiments are illustrative and not meant to limit the scope of the invention as set forth in the following claims. Referenced by
Classifications
Legal Events
Rotate |