Rotating Machine Crack Estimation From Surface Deformation

The patent replaces the conventional least square method (a direct mathematical approach) with a neural network-based computational model. The neural network learns the complex nonlinear mapping between surface deformation and hidden crack characteristics through training on simulation data, thereby achieving more accurate and reliable crack estimation while avoiding the ill-posedness issues of traditional inverse analysis methods.

The patent changes the approach from using direct mathematical equations (least square method) to using learned parameters from neural network training. By training the neural network on a comprehensive dataset generated from structural analysis simulations, the system captures complex relationships between surface deformation and crack parameters, transforming the ill-posed inverse problem into a more robust parameter estimation problem.

The patent performs preliminary action by pre-training the neural network on comprehensive simulation data that covers various crack types, positions, and structural configurations. This pre-training phase creates a ready-to-use estimation model that can directly process surface deformation measurements without requiring complex real-time inverse analysis, thereby reducing operational complexity while maintaining high detectability.

The patent creates a computational copy of the complex inverse analysis problem by training a neural network on simulated data. Instead of directly solving the difficult inverse problem in real-time, the system uses a pre-trained neural network model that has learned the complex relationships during training, thereby simplifying the actual crack detection process while maintaining accuracy.

The patent replaces expensive and complex physical inspection methods (ultrasonic inspection, X-ray inspection) with a computational approach using neural networks and surface deformation measurement. This substitution eliminates the need for costly inspection equipment while achieving comparable or superior crack detection capability through intelligent algorithms that process data from simple surface sensors.

The patent creates a virtual copy of the inspection process by using neural network models trained on simulation data. Instead of physically penetrating or vibrating the structure to detect cracks, the system uses a computational model that learns from surface deformation patterns, thereby replacing expensive physical inspection devices with a cost-effective information processing system.