US 12,437,386 B2
Weld quality inspection with domain knowledge infused adaptive-network-based fuzzy inference system
Shashwat Shahi, Bangalore (IN); Shailesh Shankar Deshpande, Pune (IN); Gargi Uday Kulkarni, Bangalore (IN); Mahesh Kshirsagar, Mumbai (IN); and Sonam Sharma, Noida (IN)
Assigned to Tata Consultancy Services Limited, Mumbai (IN)
Filed by Tata Consultancy Services Limited, Mumbai (IN)
Filed on Jun. 9, 2023, as Appl. No. 18/207,812.
Claims priority of application No. 202221033807 (IN), filed on Jun. 13, 2022.
Prior Publication US 2023/0410281 A1, Dec. 21, 2023
Int. Cl. G06T 7/00 (2017.01); G06T 7/11 (2017.01)
CPC G06T 7/0004 (2013.01) [G06T 7/11 (2017.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30108 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A processor implemented method for weld quality inspection, the method comprising:
receiving, via one or ore hardware processors, a labeled set of a plurality of images with each of the plurality of images comprising a weld of a joint type from among a set of joint types of interest, wherein the plurality of images are labeled as a good quality weld and a bad quality weld;
configuring, via the one or more hardware processors, weightages to be assigned to each of a plurality of geometrical parameters for the weld of the one or more joint types, wherein the weightages are one of (i) default with equal weightages across the plurality of geometrical parameters, and (ii) customized with varying values of the weightages across the plurality of geometrical parameters;
preprocessing, via the one or more hardware processors, the plurality of images and marking a closed outline around the weld in each of the plurality of images;
segmenting, via the one or more hardware processors, each of the preprocessed plurality of images to determine a plurality of fractals of the weld, wherein pixel coordinates of the largest fractal among the plurality of fractals are identified;
extracting, via the one or more hardware processors, the plurality of geometrical features of the weld in accordance with the joint type using the pixel coordinates of the associated largest fractal and weighing the plurality of geometrical features of the weld in accordance with the configured weightages to introduce domain biases for joint types;
generating, via the one or more hardware processors, a quality index for each of the plurality of images based on the extracted plurality of geometrical features using domain knowledge in form of gold standard of a good quality weld of each joint type;
encoding, via the one or more hardware processors, class labels for generated the quality index for each of the plurality of images with ‘0’ indicating image with the bad weld quality and ‘1’ indicating image with the good weld quality, wherein the encoded class labels serve as a training data for training the weld inspection model;
computing, via the one or more hardware processors, domain threshold values for each of the plurality of geometrical features associated with images encoded with class label ‘1’ associated with good weld quality; and
training, a weld inspection model executed by the one or more hardware processors, for weld quality inspection to predict a confidence score of the weld of the joint type of each of the plurality of images comprising a Domain Knowledge Infused Adaptive-Network-Based Fuzzy Inference System (DKI-ANFIS), wherein the DKI-ANFIS comprises a plurality of network layers infused with domain knowledge, wherein a first layer comprises a Weld Good Membership Function and a Weld Bad Membership Function with trainable parameters which generate a set of rules for each of the plurality of geometrical parameters.