PSI - Issue 56

B. Kalita et al. / Procedia Structural Integrity 56 (2024) 105–110 B. Kalita/ Structural Integrity Procedia 00 (2019) 000–000

108

4

aspect is that the training data should be properly labelled since machine learning based regression models need to comprehend the link between input and output variables. Regression analysis, whether used to support stock market predictions or forecast healthcare trends, can give the different organisations important information to help make decisions. Using a sort of regression approach known as polynomial regression, the relationship between the features and the outcome is modelled as an nth degree polynomial.

Fig. 1. : Linear regression curve [7]

2.5. Comparing ML Algorithms ML techniques vary among themselves in terms of complexity. For instance, while neural networks employ a real valued function of linear variable bounds, linear regression fits linear functions. More complicated models run the risk of overfitting but can produce more accurate results. While some models such as linear regression and neural network (NN) have minimal parameters, others require more decision-making to be optimized such as Support Vector Machines (SVMs). Models vary in how quickly they can fit the required parameters. The speed at which a model can predict a given query varies between the models. Some models are more understandable and straightforward than others (white box vs. black box models). If a model can take the inputs, and routinely get the same outputs, the model is interpretable. Interpretability poses no issue in low-risk scenarios. If a model is recommending movies to watch, that can be a low-risk task. Interpretability sometimes needs to be high in order to justify why one model is better than another. 3. Results and Discussion Relevant experimental Fatigue data of 17-4 PH Stainless Steel has been gathered in order to create the ML model from already published and pertinent literature. For the purpose of determining the FCGR properties, conventional Compact Tension (CT) samples were made from grade 5 17-4 PH SS spherical powder. Sizes of the powder particles ranged from 15 to 45 micrometers. FCGR tests were carried out in accordance with ASTM E647 guidelines. A pre crack of size 1mm was generated and manufactured prior to FCGR examination. V. Cain et al. observed three alternative build orientation XY, XZ, and ZX as well as post-processing methods for each orientation as-built, heat treated, and stress relieved were varied along with cyclic loading of 5 Hz and a 0.1 stress ratio under processing conditions. Following stress release, heat-treated samples were annealed at 890 °C for two hours after being soaked at 650 °C for a few hours. The data for the post-processing mentioned above were pulled from the literature while the process parameters were optimized. For the predictive study of the FCGR of 17-4 PH SS, data from the Paris law graphs are collected and used for each distinct processing and post-processing scenario. The dada points which are collected are plotted for visualization during the analysis and shown in Figure. 2.

3.1. Data and description