Method for simulation assisted data generation and deep learning intelligence creation in non-destructive evaluation systems

A system for detecting one or more anomalies in an object, the system comprising:

• at least one processor;
• and
• a memory storing instructions that, when executed by the at least one processor, cause the system to:
• receive experimental data of the object, wherein the experimental data is obtained while testing one or more anomalies in the object, and wherein the experimental data is associated with predefined anomaly types of the object;
• apply a probability density function (PDF), representative of variability in the experimental data arising from measurement conditions, upon one or more variables associated with the experimental data to determine corresponding one or more PDF estimates,
• wherein the one or more variables correspond to measurement-system parameters affecting acquisition of the experimental data and comprise at least one of instrument noise, transducer sensitivity variation, and expected defect morphology parameters, wherein the PDF application identifies variations and sources of variations in the experimental data attributable to the one or more variables,
• generate stochastic synthetic data corresponding to each of the one or more variables based on the corresponding one or more PDF estimates;
• provide the stochastic synthetic data to a simulator model to generate simulated data associated with the object based on the stochastic synthetic data and priori data associated with the testing of the object, wherein the simulated data is generated based on characteristics of the object indicated by the priori data, and wherein the simulated data comprises one or more statistically synthesized anomalies not present in the experimental data along with the one or more anomalies of the experimental data,
• train a learning model based on the one or more statistically synthesized anomalies and the one or more anomalies of the experimental data,
• validate the trained learning model using a portion of the experimental data of the object, and
• apply the trained learning model to detect anomalies in a new object related to the object,
• wherein the object is one of a part, a product, a weld, a system, an instrument, and a component,
• wherein the one or more anomalies include one of a size, shape, and orientation,
• wherein the learning model is retrained based on a result of the validation, and
• wherein the one or more variables are defined based on registered experimental data such that defect statistics are independent of positional variation in the experimental data.