Method and apparatus for analyzing cause of product defect

1. An apparatus for analyzing a cause of a product defect, comprising:

a pre-processing unit configured to receive process data and perform pre-processing for analyzing a cause of a product defect;

a search unit configured to search for a primary defect cause-conditional sentence to represent a primary defect cause through solution encoding and decoding and solution fitness calculation for a plurality of candidate solutions in order to search for a conditional sentence using the pre-processed process data and to output the primary defect cause-conditional sentence; and

a post-processing unit configured to receive the primary defect cause-conditional sentence, remove a redundant conditional sentence, and output a final defect cause-conditional sentence

wherein the search unit

searches for the primary defect cause-conditional sentence based on a search setting value, comprising a maximum length of a conditional sentence that limits an attribute entering the conditional sentence, a number of evaluations for searching for the conditional sentence, and a number of conditional sentences to represent the primary defect cause-conditional sentence,

outputs the primary defect cause-conditional sentence through a genetic algorithm for constructing a population comprising the plurality of candidate solutions, calculating fitness of each candidate solution, and searching for an optimal solution while evolving the population through evolutionary computation,

converts each of the plurality of candidate solution, each encoded into a real number of 0 to 1, into an operator and a conditional expression using head, body, and tail tables,

generates a tree until the tree is sequentially available,

converts the tree into the primary defect cause-conditional sentence, and

evaluates a degree that the primary defect cause-conditional sentence indicates a cause of a defect through the solution fitness calculation.

2. The apparatus of claim 1 , wherein the pre-processing unit

receives the process data comprising an identifier of a product and sensor values collected according to a production flow of the product,

removes process data having an error, and

limits an attribute having a consecutive value or an attribute having a plurality of types of values to a predetermined number of types of values.

3. The apparatus of claim 1 , wherein the post-processing unit

receives the primary defect cause-conditional sentence,

removes a redundant portion from each of the primary defect cause-conditional sentences, and

removes a conditional sentence based on results of the solution fitness calculation calculated in the search unit.

4. A method of analyzing a cause of a product defect, comprising:

receiving, by a pre-processing unit, process data and performing pre-processing for analyzing a cause of a product defect;

searching for, by a search unit, a primary defect cause-conditional sentence to represent a primary defect cause through solution encoding and decoding and solution fitness calculation for a plurality of candidate solutions in order to search for a conditional sentence using the pre-processed process data and outputting the primary defect cause-conditional sentence; and

receiving, by a post-processing unit, the primary defect cause-conditional sentence, removing a redundant conditional sentence, and outputting a final defect cause-conditional sentence,

wherein searching for, by the search unit, the primary defect cause-conditional sentence to represent the primary defect cause through solution encoding and decoding and solution fitness calculation for the plurality of candidate solutions in order to search for the conditional sentence using the pre-processed process data and outputting the primary defect cause-conditional sentence comprises:

searching for the primary defect cause-conditional sentence based on a search setting value, comprising a maximum length of a conditional sentence that limits an attribute entering the conditional sentence, a number of evaluations for searching for the conditional sentence, and a number of conditional sentences to represent the primary defect cause-conditional sentence, and

outputting the primary defect cause-conditional sentence through a genetic algorithm for constructing a population comprising the plurality of candidate solutions, calculating fitness of each candidate solution, and searching for an optimal solution while evolving the population through evolutionary computation, and

the method further comprising:

converting each of the plurality of candidate solutions, each encoded into a real number of 0 to 1, into an operator and a conditional expression using head, body, and tail tables,

generating a tree until the tree is sequentially available,

converting the tree into the primary defect cause-conditional sentence, and

evaluating a degree that the primary defect cause-conditional sentence indicates a cause of a defect through the solution fitness calculation.

5. The method of claim 4 , wherein receiving, by the pre-processing unit, process data and performing pre-processing for analyzing the cause of the product defect comprises:

receiving the process data comprising an identifier of a product and sensor values collected according to a production flow of the product,

removing process data having an error, and

limiting an attribute having a consecutive value or an attribute having a plurality of types of values to a predetermined number of types of values.

6. The method of claim 4 , wherein receiving, by the post-processing unit, the primary defect cause-conditional sentence, removing the redundant conditional sentence, and outputting the final defect cause-conditional sentence comprises:

receiving the primary defect cause-conditional sentence,

removing a redundant portion from each of the primary defect cause-conditional sentences, and

removing a conditional sentence based on results of the solution fitness calculation calculated in the search unit.