# US 2023/0264285 A1 · Welding system, welding method, welding support device, program, learning device, and method of generating trained model > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2023/0264285 A1 - Document ID: US20230264285A1 - Status: Published application - Kind: A1 - Application number: 18/040,121 - Filed: 2021-08-25 - Published: 2023-08-24 - Earliest claimed: 2020-10-16 - Assignees: Kabushiki Kaisha Kobe Seiko Sho (Kobe Steel, Ltd.) - Applicants: Kabushiki Kaisha Kobe Seiko Sho (Kobe Steel, Ltd.) - Inventors: Keita OZAKI; Akira OKAMOTO - Web page: https://patentgrep.com/patent/US20230264285A1 ## CPC - B23K 9/0953: {using computing means} - B23K 9/02: Seam welding; Backing means; Inserts - B23K 9/0956: {using sensing means, e.g. optical} - B23K 31/006: {relating to using of neural networks} - B23K 31/125: {Weld quality monitoring} - G01N 21/88: Investigating the presence of flaws or contamination - G06N 3/04: Architecture, e.g. interconnection topology - G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/09: Supervised learning - G06N 20/00: Machine learning - B23K: SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM - G01N: INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES (measuring or testing apparatus or processes other than immunoassay, involving enzymes or microorganisms C12M, C12Q) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract This welding system comprises: a welding device, various different types of a plurality of sensors which detect an event according to welding performed by a welding device; and an estimation unit which uses a trained model that is pre-generated by machine-learning by taking, as input data, a plurality of pieces of data for learning obtained by detecting events according to welding by means of the same types of sensors as the plurality of sensors, and, as training data, labels representing whether the welding is normal or abnormal, thereby estimating an abnormality of the welding performed by the welding device from a plurality of pieces of detection data generated by the plurality of sensors. ## Claims ### Claim 1 A welding system comprising: a welding device; a plurality of sensors in different types, which detect an event according to welding performed by the welding device; and an estimation unit that estimates an abnormality of the welding performed by the welding device from a plurality of pieces of detection data generated by the plurality of sensors, using a trained model that is pre-generated by machine-learning by taking, as input data, a plurality of pieces of data for learning obtained by detecting events according to welding by same types of sensors as the plurality of sensors, and, as training data, labels representing whether the welding is normal or abnormal. ### Claim 2 (depends on 1) The welding system according to claim 1, wherein the plurality of pieces of detection data include two or more of an image obtained by capturing a weld zone during welding, welding sound, a voltage of a welding power source, and a current of a welding power source, and include at least one of the image and the welding sound. ### Claim 3 (depends on 1) The welding system according to claim 1, wherein one of the plurality of pieces of detection data is welding sound, a voltage of a welding power source, or a current of a welding power source, a conversion unit is further provided configured to generate a spectrogram that represents the welding sound, the voltage of the welding power source, or the current of the welding power source in terms of three dimensions: time, frequency, and strength, and the estimation unit inputs the spectrogram to the trained model. ### Claim 4 (depends on 1) The welding system according to claim 1, wherein the trained model generates, as training data, a feature value of at least one of the plurality of pieces of data for learning, and the estimation unit further estimates a feature value of at least one of the plurality of pieces of detection data using the trained model. ### Claim 5 (depends on 4) The welding system according to claim 4, wherein one of the plurality of pieces of detection data is an image obtained by capturing a weld zone during welding, and the estimation unit estimates a feature point in the image as the feature value. ### Claim 6 (depends on 4) The welding system according to claim 4, wherein one of the plurality of pieces of detection data is welding sound, a voltage of a welding power source, or a current of a welding power source, and the estimation unit estimates, as the feature value, an interval of abnormal quality of the welding sound, the voltage of the welding power source, or the current of the welding power source. ### Claim 7 (depends on 4) The welding system according to claim 4, further comprising: a similarity calculation unit that calculates a similarity between the feature value extracted from the one piece of data for learning and the feature value estimated by the estimation unit; and a reliability determination unit that determines a reliability of an abnormality of welding estimated by the estimation unit based on the similarity. ### Claim 8 (depends on 1) The welding system according to claim 1, wherein the plurality of pieces of data for learning and the plurality of pieces of detection data are time series data. ### Claim 9 (depends on 8) The welding system according to claim 8, wherein the trained model includes a recurrent neural network. ### Claim 10 (depends on 8) The welding system according to claim 8, wherein the estimation unit estimates a feature value of at least one of the plurality of pieces of detection data using a second trained model that is pre-generated by machine-learning by taking, as input data, at least one of the plurality of pieces of data for learning, and as training data, a feature value in the at least one piece of data for learning, and inputs time series data of index based on the feature value to the trained model. ### Claim 11 (depends on 10) The welding system according to claim 10, wherein the time series data of index represents a shape of a molten pool in an image obtained by capturing a weld zone during welding. ### Claim 12 A welding method comprising: detecting an event according to welding performed by a welding device by a plurality of sensors in different types; and estimating an abnormality of the welding performed by the welding device from a plurality of pieces of detection data generated by the plurality of sensors, using a trained model that is pre-generated by machine-learning by taking, as input data, a plurality of pieces of data for learning obtained by detecting events according to welding by same types of sensors as the plurality of sensors, and, as training data, labels representing whether the welding is normal or abnormal. ### Claim 13 A welding support device comprising: an acquisition unit that acquires a plurality of pieces of detection data generated by a plurality of sensors in different types, which detect an event according to welding performed by a welding device; and an estimation unit that estimates an abnormality of the welding performed by the welding device from the plurality of pieces of detection data generated by the plurality of sensors, using a trained model that is pre-generated by machine-learning by taking, as input data, a plurality of pieces of data for learning obtained by detecting events according to welding by same types of sensors as the plurality of sensors, and, as training data, labels representing whether the welding is normal or abnormal. ### Claim 14 A program causing a computer to execute a process comprising: acquiring a plurality of pieces of detection data generated by a plurality of sensors in different types, which detect an event according to welding performed by a welding device; and estimating an abnormality of the welding performed by the welding device from the plurality of pieces of detection data generated by the plurality of sensors, using a trained model that is pre-generated by machine-learning by taking, as input data, a plurality of pieces of data for learning obtained by detecting events according to welding by same types of sensors as the plurality of sensors, and, as training data, labels representing whether the welding is normal or abnormal. ### Claim 15 A learning device comprising: an acquisition unit that acquires a data set for learning including a plurality of pieces of data for learning obtained by detecting events according to welding by a plurality of sensors in different types, and labels representing whether the welding is normal or abnormal; and a learning unit that generates a trained model to estimate an abnormality of welding from a plurality of pieces of detection data obtained by detecting events according to welding with same types of sensors as the plurality of sensors, by taking the plurality of pieces of data for learning as input data, and the labels as training data. ### Claim 16 (depends on 15) The learning device according to claim 15, wherein the learning unit generates the trained model to further estimate a feature value of at least one of the plurality of pieces of detection data by further taking, as training data, a feature value of at least one of the plurality of pieces of data for learning. ### Claim 17 A method of generating a trained model, the method comprising: acquiring a data set for learning including a plurality of pieces of data for learning obtained by detecting events according to welding by a plurality of sensors in different types, and labels representing whether the welding is normal or abnormal; and generating a trained model to estimate an abnormality of welding from a plurality of pieces of detection data obtained by detecting events according to welding with same types of sensors as the plurality of sensors, by taking the plurality of pieces of data for learning as input data, and the labels as training data.