US 12,432,895 B2
Solder printing inspection device
Kazuyoshi Kikuchi, Aichi (JP); Tsuyoshi Ohyama, Aichi (JP); and Norihiko Sakaida, Aichi (JP)
Assigned to CKD Corporation, Aichi (JP)
Filed by CKD CORPORATION, Aichi (JP)
Filed on Mar. 28, 2023, as Appl. No. 18/190,996.
Application 18/190,996 is a continuation of application No. PCT/JP2021/027300, filed on Jul. 21, 2021.
Claims priority of application No. 2020-169513 (JP), filed on Oct. 7, 2020.
Prior Publication US 2023/0232603 A1, Jul. 20, 2023
Int. Cl. H05K 13/08 (2006.01); G06T 7/00 (2017.01); G06T 7/50 (2017.01); H05K 13/04 (2006.01)
CPC H05K 13/0817 (2018.08) [G06T 7/001 (2013.01); G06T 7/50 (2017.01); G06T 2207/30141 (2013.01); G06T 2207/30152 (2013.01); H05K 13/0465 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A solder printing inspection device configured to perform a pre-reflow inspection for a printing state of a solder paste printed on a printed circuit board, the solder printing inspection device comprising:
an illumination device that irradiates, with a predetermined light, the printed circuit board on which the solder paste is printed;
an imaging device that takes an image of the printed circuit board irradiated with the predetermined light and obtains image data; and
a control device that:
based on the image data, obtains three-dimensional measurement data of the solder paste printed on the printed circuit board,
based on the three-dimensional measurement data, extracts upper portion shape data of an upper portion of the solder paste, the upper portion having a height equal to or higher than a predetermined height, and
compares the upper portion shape data with a predetermined criterion and determines whether a quality of a three-dimensional shape of the upper portion of the solder paste is good or poor, wherein
the solder printing inspection device further comprises:
a storage that stores a neural network and a model, wherein the neural network includes an encoding portion extracting a characteristic amount from input shape data and a decoding portion reconstructing shape data from the characteristic amount, and the model is generated by learning of the neural network using, as learning data, only upper portion shape data with regard to a non-defective solder paste, and
the control device:
inputs, as original upper portion shape data, the extracted upper portion shape data into the model, and obtains, as reconstructed upper portion shape data, the extracted upper portion shape data reconstructed by the model,
compares the original upper portion shape data with the reconstructed upper portion shape data, and
based on a comparison result, determines whether the quality of the three-dimensional shape is good or poor.