Self-optimizing method and machine

A method for self-optimizing, without prespecified control model or variable setpoints, an object relative to a specific criterion or activity during the operation of said object and in response to variations on a prescribed number m of variables, comprising:

• self-planning a set of n statistically designed tests around a specified experimental point in a m-dimensional experimental region for testing said m variables in said n tests;
• performing said n tests on said very object in actual operation;
• determining from said n tests the combination of said m variables which optimizes said specific criterion or activity during said actual operation;
• setting the conditions of said m variables at the thus-determined optimal combination before said combination significantly changes;
• repeating or recycling the self-planning, performing, determining, and setting steps according to the sequence given;
• and
• during each self-optimizing cycle except for the first employing said optimal combination as the experimental point in the immediately following recycle.

A method for self-optimizing, without prespecified control model or variables setpoints, a computer software relative to a specific criterion in response to variations on a prescribed number m of software-related variables, comprising:

• self-planning a set of n statistically designed tests around a specified experimental point in a m-dimensional experimental region for testing the m variables in said n tests;
• performing said n designed tests on said software;
• determining from said n tests the combination of said m variables which optimizes said specific criterion;
• setting the conditions of said m variables at the thus-determined optimal combination before said combination significantly changes;
• repeating or recycling the self-planning, performing, determining, and setting steps in the order given;
• and
• during each self-optimizing cycle except for the first employing said optimal combination as the specified experimental point in the immediately following recycle.

A method of self-optimizing relative to a prespecified optimizing criterion with a group of m preselected control variables, comprising:

• automatically planning and executing a statistically designed R&D experiment on said m variables;
• determining from said experiment the optimal variables combination;
• setting said m variables at the thus-determined optimal combination;
• repeating the planning and executing, determining, and setting steps according to the sequence given;
• and
• feeding back in any given R&D experiment except for the first information at least from the immediately preceding experiment.