In this work, a novel deep learning computational framework is developed to determine and identify the damage load conditions of different types of structures, including cantilever beams of inelastic materials, elasto-plastic shell structures of inelastic materials and crashed cars subjected to mechanical forcing actions.
There are a variety of methods to measure engineering responses based on the corresponding load conditions. The aim of this work is to establish reverse analysis algorithms. This artificial intelligence framework offers a practical solution to solve the inverse problem of engineering failure analysis based on final material and structure damage states and fields. More precisely, the machine learning inverse problem solver may be a practical solution to characterize failure load parameters and conditions based on the permanent plastic deformation distribution or the residual displacement condition of beam, shell structures and cars.
The study presents the detailed machine learning algorithm, data acquisition and learning processes, and validation and verification examples. Neural Network Modeling offers a cohesive approach to the computational mechanical problems based on TensorFlow. Different activation functions and loss functions are compared theoretically and numerically during implementing neural network.
Feature selection is used in model construction for simplification of models to make them easier to interpret and lead to shorter training times. It is demonstrated that the developed machine learning algorithm can accurately identify a practically unique prior static loading as well as impact loading state for different structures, in an inverse manner, using the permanent plastic deformation or the residual displacement as the forensic signatures.
The data-driven based method developed in this work, employs Artificial Neural Networks to provide a powerful tool for forensically diagnosing, determining, and identifying damage loading conditions for engineering structures in accidental failure events, such as car crashes and infrastructure or building structure collapses. The machine learning inverse problem solver developed here may have potential broader impacts on general forensic material and structure analysis using permanent plastic deformations.