Self-optimizing machine and method

• US 4,368,509 A
• Filed: 06/22/1981
• Issued: 01/11/1983
• Est. Priority Date: 08/24/1979
• Status: Expired due to Fees

• Current Assignee: Chou H. Li
• Original Assignee: Chou H. Li
• Inventors: Li, Chou H.
• Primary Examiner(s): Ruggiero, Joseph F.
• Application Number:
  US06/275,758
• Time in Patent Office: 568 Days
• Field of Search: 364/105, 364/106, 364/111, 364/112, 364/431, 364/500, 364/501, 364/502, 364/148-166, 364/178, 364/179, 364/442, 364/497, 364/479
• US Class Current: 700/53
• CPC Class Codes: F02B 2075/027 four
• CPC Class Codes: G05B 13/0205 not using a model or a simulator of the controlled system
• CPC Class Codes: G06Q 10/04 Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" market predictions or forecasting for commercial activities G06Q30/0202

Abstract

The invention discloses method for self-optiming, in real time, a machine relative to a specific performance characteristic in response to variations on a prescribed number m of variables. This method comprises selecting a statistical design; repeatedly and at n time instants adjusting conditions of the m variables and simultaneously recording the dependent values of the performance characteristic; analyzing the n sets of data to determine the functional relationship between the m variables and the specific performance characteristic; determining the combination of the m variables that gives optimum value of the performance characteristic; and setting the conditions of the m variables at the thus-determined combination. A machine in the form of a programmed microprocessor embodying this self-optimizing method is also disclosed.

Claims

1. A method for self-optimizing, in real time, a machine relative to a specific performance characteristic in response to variations on a prescribed number m of variables comprising:

selecting a statistical design which prescribes the number n of tests and the exact design matrix in which the test level of each of the m variables is specified for all the n tests;

according to the design matrix simultaneously sensing and adjusting to their respectively specified test levels all said m variables at a specific time instant;

recording and storing the value of the specific performance characteristic in response to all said m variables at their respective test levels for this specific time instant;

sequentially and repeatedly sensing and adjusting the conditions of all said m variables according to the design matrix at n-1 substantially consecutive time instants within minutes subsequent to said specific time instant;

sequentially and repeatedly recording and storing the n-1 values of the specific performance characteristic in response to all said m variables at their respective levels for the n-1 subsequent, consecutive time instants;

analyzing the n sets of recorded values and combinations of levels of said variables to determine the functional relationship between said variables and said specific performance characteristic;

from the functional relationship determining the singular combination of the conditions of said m variables that gives optimum value for the specific performance characteristic for the particular time instants and condition, environment, and service requirements of the machine; and

setting the conditions or levels of said m variables at their respective, thus-determined conditions.

2. A method as in claim 1 wherein said statistical design is an orthogonal design so that the effect of each of said m variables can be determined independently of the presence, and without being affected by the simultaneous variation and testing, of the other m-1 variables.

3. A method as in claim 2 wherein m is at least 7 and said design matrix is a factorial design matrix.

4. A method as in claim 2 wherein m is at least 31 and said design matrix is a fractionally replicated, factorial design matrix.

5. A method as in claim 1 wherein said statistical design is a symmetrical design so that each condition or level of every variable is tested exactly the same number of times in the n tests.

6. A method as in claim 5 wherein m is at least 250, and said statistical design is a sequential screening design.

7. A method as in claim 1 wherein said selecting, sensing and adjusting, recording and storing, sequential and repeated sensing and adjusting and recording and storing, analyzing, determining, and setting steps are done under the control of a microprocessor having calculating speeds better than several nonoseconds.

24. The method as in claim 1 including periodically reevaluating said performance characteristic at time instants within minutes subsequent to said n successive time instants to collect additional data on said m variables and performance characteristic;

redetermining the new, real-time functional relationships and optimum combinations; and

substantially instantly resetting said m variables at the new optimum combinations thereby dynamically maintaining the system optimized at all subsequent time instants.

25. The method as in claim 1 wherein said functional relationship is a linear functional relationship.

26. The method as in claim 1 wherein said functional relationship is a non-linear functional relationship.

27. The method as in claim 1 wherein n is a minor fraction of m.

28. The method as in claim 1 wherein m exceeds 1,000.

8. A machine for self-optimizing, in real time, a specific performance characteristic in response to variations on a prescribed number m of variables comprising:

means for selecting a statistical design which prescribes the number n of tests and the exact design matrix in which the test level of each of the m variables is specified for all the n tests;

means for simultaneously sensing and adjusting to their respectively prespecified test levels all said m variables according to the design matrix at a specific time instant;

means for storing the value of the specific performance characteristic in response to all said m variables at their respective test levels for this specific time instant;

means for sequentially and repeatedly adjusting the conditions of all said m variables according to the design matrix at n-1 substantially consecutive time instants within minutes subsequent to said specific time instant and also storing the n-1 values of the specific performance characteristic in response to all said m variables at their respective levels;

means for analyzing the n sets of stored values and combinations of levels of said variables to determine the functional relationship between said variables and said specific performance characteristic;

means for determining from the functional relationship the singular combination of conditions of said m variables that gives optimum value for the specific performance characteristic; and

means for setting the conditions or levels of said m variables at their respective, thus-determined conditions.

9. A machine as in claim 8 wherein said analyzing means includes means for computing the partial derivatives of the specific performance characteristic relative to every variable so as to determine the path of steepest ascent;

and including means for performing a series of additional optimizing tests by varying said variables along the determined path of steepest ascent.

10. A machine as in claim 9 wherein said computing means is a programmed microprocessor having calculating speeds better than several nanoseconds.

11. A machine as in claim 8 including means for periodically checking at pre-specified time intervals the constancy of the functional relationship;

and means for adjusting said variables to the new unique combination when said function relationship significantly changes with time.

12. A machine as in claim 8 including means for reducing the ranges of the test levels for said m variables and means for repeating prescribed steps of the optimizing procedure.

13. A machine as in claim 8 which is in the form of an industrial processing equipment, and wherein said specific performance characteristic is the product cost, and said variables include at least several of those selected from the group consisting of:

type and amount of raw materials, chemical elemental concentrations at designated locations, processing temperature, applied pressures, and processing times.

14. A machine as in claim 8 which is in the form of a car;

and wherein said specific performance characteristic is fuel economy, and said variables include at least several of those selected from the group consisting of;

car design and style, engine speed, temperature, or settings, type and amount of fuel and oil, air-fuel ratio, pressures in combustion chambers, fuel ignition conditions, loading on car, exhaust design, age and condition of car, wind type direction, and velocity, rain or snow presence, driver habit and condition, road age, slope and condition, and traffic type and density.

15. A machine as in claim 8 which is in the form of combination drug dispenser for a sick patient and embodying a programmed, nanosecond microprocessor, and wherein said variables are m drug components while said performance characteristic is a prescribed medical effect and wherein said setting means comprises means for dispensing said m drug components in optimum combination of dosages or levels.

16. The machine as in claim 8 wherein said consecutive time instants are no later than said specific time instant by several seconds.

17. The machine as in claim 8 in the form of an integrated and self-contained, industrial processing equipment and wherein said specific performance characteristic to be dynamically self-optimized is the unit processing cost and said m variables are processing variables.

18. The machine as in claim 8 in the form of a transportation vehicle and wherein said specific performance characteristic relates to a cost factor in the operation of the vehicle and said m variables include those selected from the group consisting of vehicle, weather, road, and vehicle loading, operator, and operating conditions.

19. The machine as in claim 8 in the form of a drug-dispensing machine self-optimized to dispense the optimum combination of drugs into a sick patient and wherein said specific performance characteristic is the medical condition of the patient including the function of a vital organ of the patient and said m variables are a multitude of chemicals or drugs to be administered to the patient.

20. The machine as in claim 19 wherein the drugs in optimum combinations are administered to the patient in a substantially continuous manner.

21. The machine as in claim 8 in the form of a mineral drilling machine and wherein said performance characteristic is a unit cost of drilling and said m variables include drilling, personal, and environmental variables.

22. The machine as in claim 15 having a total volume of less than 0.1 cubic meter.

23. The machine as in claim 8 wherein said analyzing means is sufficiently light, small, and power-thrifty to be installed on the machine, and yet has sufficiently large memory and also fast calculating speeds to handle billions of calculating steps for the required, real time optimization.