# US 2025/0094785 A1 · Waveform signal processing system, structure evaluation system, and waveform signal processing method > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2025/0094785 A1 - Document ID: US20250094785A1 - Status: Published application - Kind: A1 - Application number: 18/761,806 - Filed: 2024-07-02 - Published: 2025-03-20 - Earliest claimed: 2023-09-19 - Assignees: Kabushiki Kaisha Toshiba - Applicants: Kabushiki Kaisha Toshiba - Inventors: Takashi USUI; Hiroshi OHNO - Web page: https://patentgrep.com/patent/US20250094785A1 ## CPC - G06N 3/0475: Generative networks - G06N 3/045: Combinations of networks - G06N 3/08: Learning methods - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract According to one embodiment, a waveform signal processing system according to an embodiment includes a neural network, a learner, and an extractor. The neural network is configured to generate at least first time-series data on noise and second time-series data on a signal other than noise on the basis of input random noise. The learner is configured to update parameters of the neural network on the basis of a loss function including a main limitation term of which a value becomes lower as synthetic time-series data obtained by adding the first time-series data and the second time-series data generated by the neural network and an observed time-series waveform including noise become more similar to each other. The extractor is configured to extract at least one of the first time-series data and the second time-series data generated by the neural network as a target signal on the basis of the parameters updated by the learner. ## Claims ### Claim 1 A waveform signal processing system comprising: a neural network configured to generate at least first time-series data on noise and second time-series data on a signal other than noise on the basis of input random noise; a learner configured to update parameters of the neural network on the basis of a loss function including a main limitation term of which a value becomes lower as synthetic time-series data obtained by adding the first time-series data and the second time-series data generated by the neural network and an observed time-series waveform including noise become more similar to each other; and an extractor configured to extract at least one of the first time-series data and the second time-series data generated by the neural network as a target signal on the basis of the parameters updated by the learner. ### Claim 2 (depends on 1) The waveform signal processing system according to claim 1, wherein the learner updates the parameters of the neural network on the basis of the loss function additionally including a noise limitation term which is a limitation condition for expressing features of noise. ### Claim 3 (depends on 2) The waveform signal processing system according to claim 2, further comprising a noise extractor configured to extract a noise area from the observed time-series waveform, wherein the learner uses a term of which a value becomes lower as the first time-series data becomes more similar to the noise area extracted by the noise extractor as the noise limitation term. ### Claim 4 (depends on 3) The waveform signal processing system according to claim 3, wherein the noise extractor extracts at least a part of time-series data in a pre-trigger period which has been recorded before the amplitude of the observed time-series waveform exceeds a predetermined threshold value as time-series data of the noise area. ### Claim 5 (depends on 2) The waveform signal processing system according to claim 2, wherein the noise limitation term additionally includes a term of which a value becomes lower as an average or a variance in a first range of the first time-series data becomes more similar to an average or a variance in a second range different from the first range. ### Claim 6 (depends on 5) The waveform signal processing system according to claim 5, wherein the learner sets at least one of a first range and a second range to another range whenever the parameters are updated. ### Claim 7 (depends on 1) The waveform signal processing system according to claim 1, wherein the learner updates the parameters of the neural network on the basis of the loss function additionally including a signal limitation term which is a limitation condition for expressing features of a signal other than the noise. ### Claim 8 (depends on 7) The waveform signal processing system according to claim 7, wherein the learner uses a term which is proportional to a sum of magnitudes of a first-order differential or a second-order differential of the second time-series data as the signal limitation term. ### Claim 9 (depends on 1) The waveform signal processing system according to claim 1, wherein the learner performs updating of the parameters at least 1000 times. ### Claim 10 (depends on 1) The waveform signal processing system according to claim 1, wherein the observed time-series waveform is a signal waveform that is obtained as a result of observation of an elastic wave generated due to damage in a structure or industrial equipment using a sensor. ### Claim 11 (depends on 1) A structure evaluation system comprising: one or more sensors configured to detect an elastic wave generated in a structure or industrial equipment; the waveform signal processing system according to claim 1 that extracts second time-series data on a signal other than noise as a target signal on the basis of the elastic wave detected by the one or more sensors; and an evaluator configured to evaluate a deterioration state of the structure or the industrial equipment on the basis of the target signal extracted by the waveform signal processing system. ### Claim 12 A waveform signal processing method comprising: generating at least first time-series data on noise and second time-series data on a signal other than noise on the basis of input random noise using a neural network; updating parameters of the neural network on the basis of a loss function including a main limitation term of which a value becomes lower as synthetic time-series data obtained by adding the first time-series data and the second time-series data generated by the neural network and an observed time-series waveform including noise become more similar to each other; and extracting at least one of the first time-series data and the second time-series data generated by the neural network as a target signal on the basis of the updated parameters.