Artificial Intelligence-based Data Analytical Models For Materials Failure Cases And Corrosion Prediction

Any material seriously suffers from failures and corrosion under the influence of materials-related and environment-related factors,such as atmosphere,temperature,pressure,flow rate,abrasives and contaminants.The potential for damage due to failure cases and corrosion is an essential concern in the design of instrumentation for most process control systems.In materials science,data analysis demand on different materials-related problems is more increasing.Many researchers in materials science are conducting data analysis works on the determination of the causes of failure cases and corrosion performance at the metal surface.Since material service safety is vital to be maintained,many researchers devote their time to experimental testing and data analyzing to discover the different factors influencing materials performance.Traditional analytical models used in materials science are still not perfect,limited to capture nonlinearities of data,time-consuming,dependent on man-labor and individual research.They have a low ability to handle a large amount of data,difficult to reuse,and so on.This thesis proposes different data analysis approaches to assist materials researchers in constructing an efficient reasoning model for materials failure cases and a better predictive model of materials corrosion with the help of artificial intelligence(AI)-based methods:rule-based reasoning,case-based reasoning,rule-case based reasoning,artificial neural networks(ANNs)and adaptive neuro-fuzzy inference system(ANFIS).It consists of separate data analysis works focusing on a large amount of atmospheric corrosion data of carbon steel(including reliable data from reports and published research papers)and data from materials failure cases.Firstly,materials failure problems are becoming a serious concern because there would be a variety of consequences,possibly affecting public safety.Therefore,materials failure analysis is needed to find out the reason for failure and avoid similar failures' reoccurring.However,materials failure cases mostly rely on manual analysis by experts with sufficient domain knowledge,leading to the situation of time-consuming,low efficiency,and difficult evaluation.This research proposes an Al-based method,rule-case-based hybrid reasoning approach for materials failure analysis with the aid of ontology.Hundreds of materials failure cases from different industries were collected and analyzed by rule-based reasoning,case-based reasoning,and the hybrid method.It is demonstrated that the rule-case-based hybrid reasoning method can provide better analysis results in comparison with rule-based reasoning and case-based reasoning alone.Secondly,the optimization of network topologies to retain the generalization ability by deciding when to stop overtraining an artificial neural network(ANN)is an existing vital challenge in ANN prediction works.The larger the dataset the ANN is trained with,the better generalization the prediction can give.In this work,a large dataset of atmospheric corrosion data of carbon steel compiled from several resources is used to train and test a multilayer backpropagation ANN model as well as two conventional corrosion prediction models(linear and Klinesmith models).Unlike previous related works,a grid search-based hyperparameter tuning is performed to develop multiple hyperparameter combinations(network topologies)to train multiple ANNs with a mini-batch stochastic gradient descent optimization algorithm to facilitate the training of a large dataset.After that,one selection strategy for the optimal hyperparameter combination is applied by an early stopping method to guarantee the generalization ability of the optimal network model.The correlation coefficients(R)of the ANN model can explain about 80%(more than 75%)of the variance of atmospheric corrosion of carbon steel,and the root mean square errors(RMSE)of three models show that the ANN model gives a better performance than the other two models with acceptable generalization.The influence of input parameters on the output is highlighted by using the fuzzy curve analysis method.The result reveals that TOW,Cl-and SO2 are the most crucial atmospheric chemical variables,which have a well-known nonlinear relationship with atmospheric corrosion.Lastly,the estimation of atmospheric corrosion of materials can improve materials5 design,prolong the durability of materials' structures and the lifetime of materials,and reduce overhead costs of corrosion.To predict the atmospheric corrosion of materials which includes a strong non-linear relationship between different environmental variables and the corrosion rate is becoming a challenging research trend in materials science.In this study,an attempt is made to build three artificial intelligence(AI)based predictive models,namely,deep feedforward neural network(DFF-NN),k-means based-radial basis function neural network(RBF-NN),and fuzzy logic neural network known as adaptive neuro-fuzzy inference system(ANFIS)trained with a large amount of data available in the literature to evaluate their performance in describing the phenomenon of atmospheric corrosion of carbon steel.The results exhibit that all AI-based models can perform well for the prediction of atmospheric corrosion of carbon steel with acceptable outcomes.DFF-NN gives the best prediction result with a good correlation coefficient(R)that explains about 90%of the variance of atmospheric corrosion of carbon steel and with mean absolute percentage error(MAPE)that shows satisfactory prediction accuracy(about 98%).Moreover,its root mean square error(RMSE=0.0688)also dedicates the best performance among three individual models with acceptable generalization.