System and method for additive manufacturing process monitoring

Abstract

A computer-implemented method for predicting material properties in an Additive Manufacturing (AM) process is provided. The method comprises: receiving sensor data during the build of a metallic component using the AM process wherein the sensor data includes time-series temperature data of a surface of the metallic component recorded by a photodiode and time-series temperature data of a surface of the metallic component recorded by a pyrometer; receiving ICME (Integrated Computational Materials Engineering) model output data for building the component wherein the ICME model output data includes predicted melt pool dimensions time-series data, predicted melt temperature time-series data, and predicted defects forming as a result of melt pool evolution and movement; and estimating using the received sensor data and the received ICME model output data one or more material properties associated with the metallic component using a material property prediction module configured to predict one or more of the material properties.

What is claimed is: 
     
         1 . A computer-implemented method for predicting an anomaly in an Additive Manufacturing (AM) process, the method comprising:
 receiving sensor data during the build of a metallic component using the AM process;   receiving ICME (Integrated Computational Materials Engineering) model output data for building the component;   estimating one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component using a material property prediction model trained using machine learning techniques using sensor data from the build of a second component, ICME model output data for the build of the second component, and post-build inspection data of the second component to estimate the one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component; and   predicting an anomaly during the AM process based on the estimated one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of the at least portion of the metallic component.   
     
     
         2 . The method of  claim 1 , wherein the sensor data comprises one or more of time-series temperature data of a surface of the metallic component recorded by a photodiode, time-series temperature data of a surface of the metallic component recorded by a pyrometer, time-series laser power data, time-series build chamber oxygen data, time-series build chamber humidity data, and other time-series build chamber environmental data. 
     
     
         3 . The method of  claim 1 , wherein the material property prediction model comprises a Deep Neural Network (DNN). 
     
     
         4 . The method of  claim 3 , further comprising training the DNN wherein training the DNN comprises:
 collecting training data comprising sensor data from the build of a second component, ICME model output data for the build of the second component, and post-build inspection data of the second component; and   computing and correcting a neural network model based on an error estimator metric to determine a final weight for each node in the neural network model.   
     
     
         5 . The method of  claim 1 , wherein the predicting an anomaly during the AM build process comprises predicting the anomaly during the AM build process using a Deep Neural Network (DNN). 
     
     
         6 . The method of  claim 1 , wherein the estimating one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component comprises estimating the one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component using a plurality of Deep Neural Networks (DNNs) wherein each DNN of the plurality of DNNs is configured to estimate using the received sensor data and the received ICME model output data a unique one of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component. 
     
     
         7 . The method of  claim 1 , wherein the ICME model output data is predicted from an ICME model that includes as inputs laser power, laser speed, hatch spacing, and laser motion on top of a powder bed. 
     
     
         8 . The method of  claim 1 , wherein predicting the anomaly comprises predicting the anomaly during the AM build process of the component using an anomaly detection model that is configured to predict the anomaly based on a departure from nominal of the estimated one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component output from the material property prediction model. 
     
     
         9 . A computer-implemented system configured to predict an anomaly in an Additive Manufacturing (AM) process, the system comprising:
 a photodiode configured to provide time-series temperature data of a surface of a metallic component during the build of the metallic component during the AM process;   a pyrometer configured to provide time-series temperature data of a surface of the metallic component during the build of the metallic component during the AM process;   an ICME (Integrated Computational Materials Engineering) module configured to predict ICME model output data for building the component, the ICME model output data including predicted melt pool dimensions time-series data, predicted melt temperature time-series data, or predicted defects forming as a result of melt pool evolution and movement; and   a material property prediction model trained to estimate, in real-time, during the build of the metallic component, using the time-series temperature data from the photodiode, the time-series temperature data from the pyrometer, and the ICME model output data, one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component;   wherein the system is configured to predict an anomaly during the AM process based on the estimated one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of the at least portion of the metallic component.   
     
     
         10 . The system of  claim 9 , wherein the material property prediction model has been trained using sensor data from the build of a second component, ICME model output data for the build of the second component, and post-build inspection data of the second component to estimate the one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component 
     
     
         11 . The system of  claim 9 , wherein the material property prediction model comprises a Deep Neural Network (DNN) that has been trained using sensor data from the build of a second component, ICME model output data for the build of the second component, and post-build inspection data of the second component to estimate the one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component. 
     
     
         12 . The system of  claim 11 , wherein the DNN is trained through computing and correcting a neural network model based on an error estimator metric to determine a final weight for each node in the neural network model. 
     
     
         13 . The system of  claim 11 , wherein the DNN is further trained to predict the anomaly during the AM build process of the component. 
     
     
         14 . The system of  claim 9 , comprising a plurality of Deep Neural Networks (DNNs) wherein each DNN of the plurality of DNNs is configured to estimate, using the time-series temperature data from the photodiode, the time-series temperature data from the pyrometer, and the ICME model output data, a unique one of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component. 
     
     
         15 . The system of  claim 9 , wherein the ICME module is configured to predict the ICME model output data from inputs comprising laser power, laser speed, hatch spacing, and laser motion on top of a powder bed. 
     
     
         16 . A material property prediction module configured to predict an anomaly in an Additive Manufacturing (AM) process, the material property prediction module comprising one or more processors configured by programming instructions in computer readable media, the material property prediction module configured to:
 receive sensor data during the build of a metallic component using the AM process;   receive ICME (Integrated Computational Materials Engineering) model output data for building the component;   estimate one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component using a material property prediction model trained using machine learning techniques using sensor data from the build of a second component, ICME model output data for the build of the second component, and post-build inspection data of the second component to estimate the one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component; and   predict an anomaly during the AM process based on the estimated one or more of porosity size, porosity distribution, surface roughness, residual stresses, or strength of the at least portion of the metallic component.   
     
     
         17 . The material property prediction module of  claim 16 , wherein the ICME model output data is predicted from an ICME model that includes as inputs laser power, laser speed, hatch spacing, and laser motion on top of a powder bed. 
     
     
         18 . The material property prediction module of  claim 16 , wherein the material property prediction model comprises a Deep Neural Network (DNN). 
     
     
         19 . The material property prediction module of  claim 18 , wherein the DNN is further trained to predict the anomaly during the AM build process of the component. 
     
     
         20 . The material property prediction module of  claim 16 , wherein the material property prediction model comprises a plurality of Deep Neural Networks (DNNs), wherein each DNN of the plurality of DNNs is configured to estimate using the received sensor data and the received ICME model output data a unique one of porosity size, porosity distribution, surface roughness, residual stresses, or strength of at least a portion of the metallic component.