Data processing: artificial intelligence – Neural network – Learning task
Reexamination Certificate
1999-02-12
2002-07-02
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Neural network
Learning task
C706S023000, C706S046000, C706S903000
Reexamination Certificate
active
06415273
ABSTRACT:
BACKGROUND OF THE INVENTION
This invention relates to a method of feed-forward control of a machine using a control logic model (or map) capable of learning, and particularly to a feed-forward control method based on anticipated or estimated deviation of the machine from the control logic model (or map), which method allows rapid learning and highly responsive control.
Heretofore, a feed-forward control system has been employed in various technological areas for controlling a subject, wherein behavior of the subject in response to input is estimated using a model of the subject.
By using the feed-forward control, it is possible to improve responsive characteristics of the control in response to change in input and to operate the control subject in optimum conditions. However, the subject itself is likely to be influenced by environmental changes, and cause deterioration with time. As a result, deviation occurs sometimes between the actual subject and the model constituted in advance. Thus, it is necessary to adjust or compensate for the deviation as necessary.
A learning method is known as a method for compensating for the deviation of the machine, comprising the steps of feeding actual output from the subject back to the control system, compensating for the deviation based on the feedback value, formulating teacher data based on a compensation obtained when the feedback is stabilized, and undergoing learning using the teacher data.
According to the above learning method based on the actual output from the subject, the teacher data is obtained from the actual output. Thus, deviation between the subject and the model can be detected accurately, thereby leading to accurate learning. However, first, because it takes time until the feedback is stabilized, a long period of time is required for obtaining teacher data. Second, because teacher data is obtained through feedback control, control of the subject must be continued when obtaining teacher data, regardless of its using conditions or using circumstances. Further, the above control must be continued until the feedback is stabilized. Thus, it is extremely difficult to obtain satisfactory teacher data without influencing behavior of the subject.
An objective of the present invention is to resolve the above problems in the above conventional learning method, by providing a method for performing learning by rapidly obtaining teacher data wherein deviation of a machine from a model in feedback control is estimated using a control logic capable of learning which allows timely modification of the model.
SUMMARY OF THE INVENTION
One important aspect of the present invention attaining the above objective is to provide a machine control system for controlling performance of a machine which is operable by a causative signal and the performance of which is indicatable by an indicative signal. The control system comprises: (a) a model-based control unit for outputting an estimated value of a causative signal when receiving an indicative signal, wherein the input-output relationship of the model-based control unit is the inverse of that of the machine, which model-based control unit receives a target value of the indicative signal and is provided with a learning function which modifies output from the model-based control unit based on teacher data; (b) a feedback control unit for providing first teacher data to the model-control unit, which feedback control unit receives feedback information of an actual value of the indicative signal from the machine, and provides the feedback information to the model-based control unit as first teacher data, wherein the model-based control unit undergoes learning using the first teacher data to modify its output of the causative signal (the feedback control unit is inactivated if the running condition of the machine deviates from a stable running condition); and (c) an anticipatory control unit for providing second teacher data to the model-based control unit. The anticipatory control unit comprises: (i) a deviation detection program programmed to determine whether the running condition of the machine deviates from a predetermined stable running condition, output from which program is provided to the feedback control unit; and (ii) a compensation formulation program programmed to formulate anticipatory compensations to compensate for deviation of the machine from the predetermined stable running condition based on feedback information received from the machine, output from which program is provided to the model-based control unit as second teacher data, wherein the model-based control unit undergoes learning using the second teacher data to modify output of the causative signal.
According to the present invention, the anticipatory learning method in feed-forward control using control logic (or computer simulation or map) capable of learning allows quick and accurate learning of a feed-forward model of a machine, even if a stable running condition of the machine does not continue long enough for completing feedback control to compensate for deviation of the machine from the model. That is, even if a stable running condition of the machine is very short, or while the machine is actually in operation, feed-forward control based on the model can be performed by timely compensation for deviation of the machine from the model. Unlike conventional model-based control, it is not necessary to wait until compensation values by feedback control are stabilized to obtain teacher data. According to the present invention, the machine can be controlled without deviation by quickly coping with deviation of the machine.
Another important aspect of the present invention, which exhibits the above advantageous effects, is to provide a method for controlling an unstable running condition of a machine which is operable by a causative signal and the performance of which is indicatable by an indicative signal. The method is achieved using a control system comprising: (i) a mode-based control unit for outputting an estimated value of a causative signal when receiving an indicative signal, wherein the input-output relationship of the model-based control unit is the inverse of that of the machine, which model-based control unit receives a target value of the indicative signal and is provided with a learning function which modifies output from the model-based control unit based on teacher data; and (ii) a feedback control unit for providing first teacher data to the model-control unit, which feedback control unit receives feedback information of an actual value of the indicative signal from the machine, and provides the feedback information to the model-based control unit as first teacher data. The method comprises the steps of: (a) providing the feedback information as first teacher data to the model-control unit from the feedback control unit, wherein the model-based control unit undergoes learning using the first teacher data to modify its output of the causative signal; (b) determining whether the running condition of the machine deviates from a predetermined stable running condition; (c) inactivating the feedback control unit when the running condition of the machine deviates from a stable running condition; (d) formulating anticipatory compensations to compensate for deviation of the machine from the predetermined stable running condition based on feedback information received from the machine; and (e) providing the anticipatory compensations to the model-based control unit as second teacher data, wherein the model-based control unit undergoes learning using the second teacher data to modify output of the causative signal.
Preferably, in the above, the feedback information received in step (d) is previous feedback control patterns. The feedback information received in step (d) can be information from the feedback control unit when the feedback control unit is inactivated. Further, the model-based control unit may comprise: a forward model or map defining and simulating the input-output relationship of the machine, wherein the m
Davis George B.
Knobbe Martens Olson and Bear LLP
Yamaha Hatsudoki Kabushiki Kaisha
LandOfFree
Method of feed-forward control using control logic with... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of feed-forward control using control logic with..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of feed-forward control using control logic with... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2858806