Appearance inspection device
AssigneeFanuc Corporation
InventorKeisuke WATANABE
Life of this patent
Estimating the term…
Abstract
An appearance inspection device creates a combination of a first image which is a normal product image and a second image which is a product image as a comparison object, on the basis of a reference data set and a learning data set, and a machine learning device learns classification of a product corresponding to the second image as normal or not normal for the combination. The machine learning device observes the combination of the first image and the second image as a state variable representing a current state of an environment, acquires a label given to the second image as label data, and performs learning by associating the state variable with the label data.
Claims9
An appearance inspection device for inspecting data capable of being subjected to appearance inspection, the appearance inspection device comprising:
- a reference data set storing section capable of storing a set of pieces of data, as reference data set, which has been used for learning in the past and each of which is given a label indicating normality or abnormality;
- a learning data set storing section capable of storing a set of pieces of data, as a learning data set, which is used for additional learning and each of which is given a label indicating normality or abnormality;
- a data combination creating section that creates at least one combination composed of first data which is normal data and second data which is data as an object to be compared with the first data, on the basis of the reference data set and the learning data set;
- and
- a machine learning device that learns classification of the second data as normal or abnormal for the combination of the first data and the second data created by the data combination creating section, wherein
- the machine learning device includes
- a state observing section that observes the combination of the first data and the second data as a state variable representing a current state of an environment,
- a label data acquiring section that acquires a label given to the second data as label data, and
- a learning section that learns by associating the state variable with the label data, and wherein
- the data combination creating section creates each of the combinations such that at least one of the first data and the second data is a piece of data selected from the learning data set storing section.
An appearance inspection device for performing appearance inspection of a product as an inspection object, the appearance inspection device comprising:
- a reference data set storing section capable of storing, as reference data set, a data set of product images which has been used for learning in the past and each of which is given a label indicating a normal product or a defective product;
- a learning data set storing section capable of storing, as learning data set, a data set of product images which is used for additional learning and each of which is given a label indicating a normal product or a defective product;
- a data combination creating section that creates at least one combination composed of a first image which is a normal product image and a second image which is a product image as an object to be compared with the first image, on the basis of the reference data set and the learning data set;
- and
- a machine learning device that learns classification of a product corresponding to the second image as normal or defective for the combination of the first image and the second image created by the data combination creating section, wherein
- the machine learning device includes
- a state observing section that observes the combination of the first image and the second image as a state variable representing a current state of an environment,
- a label data acquiring section that acquires a label given to the second image as label data, and
- a learning section that learns by associating the state variable with the label data, and wherein
- the data combination creating section creates each of the combinations such that at least one of the first image and the second image is an image selected from the learning data set storing section.
An appearance inspection method for inspecting data capable of being subjected to appearance inspection, the appearance inspection method comprising:
- a step of creating at least one combination composed of first data which is normal data and second data which is data as an object to be compared with the first data, on the basis of a reference data set and a learning data set, wherein the reference data set is a set of pieces of data which has been used for learning in the past and each of which is given a label indicating normality or abnormality, and the learning data set is a set of pieces of data which is used for additional learning and each of which is given a label indicating normality or abnormality;
- a step of observing the combination of the first data and the second data as a state variable representing a current state of an environment;
- a step of acquiring a label given to the second data as label data;
- and
- a step of learning by associating the state variable with the label data, wherein
- the data combination creating step includes creating each of the combinations such that at least one of the first data and the second data is a piece of data selected from the learning data set.
An appearance inspection method for performing appearance inspection of a product as an inspection object, the appearance inspection method comprising:
- a step of creating at least one combination composed of a first image which is an image of the product that is normal and a second image which is an image of the product as an object to be compared with the first image, on the basis of a reference data set and a learning data set, wherein the reference data set is a data set of product images which has been used for learning in the past and each of which is given a label indicating a normal product or a defective product, and the learning data set is a data set of product images which is used for additional learning and each of which is given a label indicating a normal product or a defective product;
- a step of observing the combination of the first image and the second image as a state variable representing a current state of an environment;
- a step of acquiring a label given to the second image as label data;
- and
- a step of learning by associating the state variable with the label data, wherein
- the image combination creating step includes creating each of the combinations such that at least one of the first image and the second image is an image selected from the learning data set.
Dependencies are inferred from the claim text.