Mohsen Dehghanpour Abyaneh

In Laser Powder Bed Fusion, a layer-by-layer melting of metal powder takes place and is specifically suited to high-performance applications in advanced technologies using Inconel™ 718. Vickers hardness (HV20) of the samples was measured as an output variable of the machine learning model after the heat treatment and surface polishing. Particle swarm optimization and genetic algorithm were proposed to define the relationships between the input and output data. The prediction of hardness was then done using five regression models, such as support vector machine, Gaussian process regression (GPR), single-layer and deep-layer artificial neural networks (ANNs), and random tree (RT). A novel technique, rollback, for optimizing output data increased the accuracy metrics. GPR and ANNs performed best in terms of training results. The rollback process was also implemented on the test results. GPR and ANN displayed the best results, with the highest R ² , NSE, and KGE values between 0.97 and 0.99, and the lowest MAPE, MAE, and RMSE on the testing data, which proved them as the best solutions. The Kruskal–Wallis test and Taylor diagram were also used to evaluate model performance.

As mechanical engineering becomes more data-driven, accurate and explainable prediction models are increasingly required. This study applied benchmarks for data-driven strategies and conducted a comprehensive analysis, using an 84-run grinding dataset on aluminum alloy 6061, with the surface roughness as the output variable. The conventional fitting method was the first applied and tuned using a closed-form formula, motivating the use of machine learning algorithms. Gaussian process regression, Artificial neural network, and Extreme gradient boosting were the algorithms that were used in predicting the relationship between input variables and the output. The best single model performance was given by GPR with an accuracy of 97.50%, a MAPE of 2.49%, and an R ² of 0.99. There were also three models of stacking ensembles that were applied. The stacking ensemble using ANN and XGB as base learners and GPR as the meta-learner offered the best trade-off between its bias and variance and achieved an overall accuracy of 94.54, a MAPE of 5.45, and an R ² of 0.98. The sensitivity analysis was employed to assess the significance of input parameters. Shapley additive explanations were also used to give attribution to each case, to attribute the impact of individual input features to each prediction. Among the variables, grinding wheel type 89A180K6V111 and the specific removal rate were the most influential. The framework can be adaptable to other datasets and grinding machining scenarios. Graphical Abstract

In mechanical engineering, predictive tools are increasingly used to enable faster analysis. The study began with cold spraying of high-entropy alloy coatings, Al _0.1-0.5 CoCrCuFeNi and MnCoCrCuFeNi coatings at nitrogen gas temperatures of 650°C, 750°C, and 850°C. The surface roughness, Ra, as the output, was measured using profilometry and microscopy techniques. The experimental sample set consisted of 20 samples of different experiments. The stacking ensemble structure included three base learners: Linear Regression, Extreme Gradient Boosting, and Gaussian Process Regression with RidgeCV as the meta-learner. This ensemble was compared with Extreme Gradient Boosting, which is a powerful single machine learning approach. Results indicated that the stacking ensemble performed better than Extreme Gradient Boosting in all the regression metrics. Extreme Gradient Boosting reached RMSE of 0.22 µm, MAPE of 3.32%, and R ² of 0.83, whereas the stacking ensemble achieved RMSE of 0.17 µm, MAPE of 2.73%, and R ² of 0.89. A sensitivity analysis was used to qualitatively assess the influence of input variables. The results suggest that stacking ensembles can improve predictive performance even in scenarios of small data.

The combination of both Passive Thermography and machine learning in materials science and engineering allows rapid progress in advanced fatigue analysis. Focusing on mechanical aspects, the combination of these approaches is capable of interpolating the fatigue resistance in diverse conditions with minimal data, when compared to the classical solution, in which analyses are conducted using statistical processes such as the Staircase Method. Even though the thermal increment and thermal area are crucial parameters for the fatigue limit analysis, the implementation of machine-learning interpolation improves data consistency and reduces variability in the fatigue limit estimation through Type-A repeatability uncertainty reduction. This way, the two-layer artificial neural network does not have any predefined form of functions; second, it maintains the inherent non-linear features of the data. The validation of the proposed approach was conducted for a C45 steel, and two different experimental campaigns were conducted using a resonant machine. At the end, the analysis of the fatigue limit was conducted by means of an interpolation-assisted Two-Curve Method, starting from the classical thermal data evolution properly optimized with a machine-learning approach, achieving a more precise result in estimating the fatigue limit.

The digitalization of mechanical engineering requires reliable prediction tools that are interpretable and facilitate faster analysis. The study applied benchmarks for data-driven strategies and conducted a comprehensive analysis, using an 84-run grinding dataset on aluminum alloy 6061 that includes three different grinding wheel types, four different coolants, and seven specific removal rates with the surface roughness as the output variable. Particle Swarm Optimization was the first applied and tuned using a closed-form formula, but with unreliable accuracy metrics. Gaussian Process Regression (GPR), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGB) were the machine learning algorithms that were tested in predicting the relationship between input variables and the output. The best single model performance was given by GPR with an accuracy of 97.50%, a MAPE of 2.49% and an R ² of 0.99. There were also three models of stacking ensembles that were applied. The stacking ensemble using ANN and XGB as base learners and GPR as the meta-learner offered the best trade-off between its bias and variance and achieved an overall accuracy of 94.54, a MAPE of 5.45, and an R ² of 0.98. The sensitivity analysis was employed to assess the significance of input parameters. Shapley Additive Explanations (SHAP) was also used to give attribution to each case to attribute the impact of individual input features to each prediction. Among the variables, grinding wheel type 89A180K6V111 and the specific removal rate were the most influential. These novel stacking-ensemble and analysis methods can be broadened for future applicability.

Recent research has indicated that Al_0.1–0.5CoCrCuFeNi and MnCoCrCuFeNi high-entropy alloys (HEAs) exhibit superior mechanical and thermal properties under extreme conditions. This chapter provides an account of the wear and surface characteristics of cold-sprayed HEA coatings at various temperatures. The inputs of surface roughness and volume variation are analyzed by analysis of variance (ANOVA), employing formulas optimized by genetic algorithms. Gaussian process regression, support vector regression (SVR), and artificial neural networks are machine learning (ML) methods that predict surface roughness and volume variation with high accuracy. For surface roughness, SVR achieves a coefficient of determination of 0.97, which is lower than that determined from the other models. Furthermore, all three models achieve a coefficient of determination of 0.99 for volume variation. The results indicate the ability of ML to generalize effectively across datasets, capturing nonlinear patterns with precision. These findings emphasize the potential of HEAs for high-wear applications and the reliability of predictive modeling.

Digitalisation of mechanical engineering necessitates the use of predictive tools that are explainable and allow for more rapid analysis. The study began with cold spraying (CS) of high entropy alloy (HEA) coatings, Al0.1-0.5CoCrCuFeNi and MnCoCrCuFeNi coatings at nitrogen gas temperatures of 650, 750, and 850 °C. The surface roughness, Ra, was measured using profilometry and microscopy techniques. The data set comprised 20 samples with varying experimental conditions and seven measured parameters (five input variables and surface roughness (Ra) as the output. The stacking ensemble structure included three base learners, Linear Regression (LR), Extreme Gradient Boosting (XGBoost), and Gaussian Process Regression (GPR) with RidgeCV as the meta-learner. LR was chosen due to its simplicity and interpretability, GPR was chosen due to its ability to capture nonlinear trends with small datasets, and RidgeCV was used to stabilize the model by shrinking coefficients and reducing variance. This ensemble was compared with XGBoost, which is a powerful single machine learning approach. Results indicated that the stacking ensemble performed better than XGBoost in all the regression metrics. XGBoost reached 96.79% accuracy, RMSE of 0.22 µm, MAPE of 3.32%, and R2 of 0.83, whereas the stacking ensemble achieved 97.26% accuracy, RMSE of 0.17 µm, MAPE of 2.73%, and R2 of 0.89. The results validate that stacking ensembles can improve predictive and robust performance even in scenarios of small data. These novel stacking-ensemble and analysis methods can be broadened for future applicability.

In Laser Powder Bed Fusion, a layer-by-layer melting of metal powder takes place and is specifically suited to high-performance applications in advanced technologies using Inconel™ 718. ANOVA and hardness measurements identified relationships that cannot be predicted using conventional statistical methods. Particle Swarm Optimization and Genetic Algorithm were proposed to define the relationships between the input and output data. Although the mean accuracy was high (97% for Particle Swarm Optimization and 91% for Genetic Algorithm), the drawbacks were calculation variation and sensitivity to parameter changes. The prediction of hardness was then done using five regression models, such as Support Vector Machine, Gaussian Process Regression (GPR), Single-Layer and Deep-Layer Artificial Neural Network (ANN), and Random Tree (RT). The low R² values were observed in the initial implementation with a mean accuracy of 90%. The linear method of optimizing output data increased R² values up to and beyond 0.9 with high average accuracy. GPR and single-layer ANN performed best in terms of training results. A rollback process was implemented on the test results. GPR and ANN displayed the best results with the highest R² (0.99) and MAPE (1.3%) values on the testing data, which proved them as the best solutions.

The estimation of the tooth root bending fatigue strength of gears is a topic of great interest in the field of mechanical engineering. The assessment of this mechanical property is generally conducted through the execution of a series of tests and, in many cases, a long-time experimental campaign is necessary for the bending fatigue strength evaluation. The present study aims at the estimation of the bending fatigue strength in gears by using the well-known Thermographic Method with integrated Machine Learning techniques implementing Gaussian process regression and artificial neural networks. This approach allows for the combination of a Non-Destructive, green technique with Artificial Intelligence algorithms, determining a rapid and reasonable estimation of the bending fatigue strength for gears. Among all methods, the statistical analyses confirm that all models have high accuracy. However, Gaussian process regression and deep neural networks may be superior in comparison with other methods, and their precision and reliability may be higher for advanced fatigue assessment. This tool could be helpful to cut down experimental workload with the help of Thermographic Method for the tooth root bending fatigue strength estimation, hence enabling very fast Non-Destructive evaluation of gear performance. Thermography approach combined with Machine Learning agrees sustainability by saving critical resource-intensive testing and leads to an advanced mechanical properties evaluation framework in gear systems, hence offering important alternative to the classical methods.