Integrating machine learning and microstructure analysis for the design of high-performance ceramic-reinforced cladding layers

The preparation of ceramic-reinforced, metal-based fusion cladding steel surfaces is an effective way to improve the wear resistance of cutter rings in shield machine. However, designing high-performance metal-ceramic powders is challenging due to their complex composition, often requiring extensive experimental work. To address this problem, this study utilizes experimental data to predict the wear resistance of ceramic-reinforced, iron-based plasma fusion cladding layers through machine learning. The goal is to support the design of high-performance metal-ceramic powders. First, four nonlinear regression models were established. The optimal model was selected after optimization and comparison to predict the wear resistance of the cladding layer. Experiments were then conducted to validate the reliability of the prediction model. Finally, model interpretation, microstructural analysis of the cladding layer, and thermodynamic calculations were performed to elucidate the relationship between powder composition, microstructure, and performances of the cladding layer. The results show that the random forest model (RF) has the best prediction accuracy, achieving an coefficient of determination (R²) of 0.84 after optimization. This model effectively predicts the trends in wear resistance of the cladding layer. Based on the predicted results, a metal-ceramic powder was designed and used to prepare plasma cladding layers with excellent wear resistance. The interaction between powder compositions significantly influences the microstructure and, consequently, the wear resistance of the cladding layer. In particular, coupling effects, such as Cr₃C₂ & NbC and Fe60 & NbC, showed a stronger impact on wear resistance than individual compositions. The excellent wear resistance of the optimized cladding layer is attributed to a multi-scale strengthening mechanism. The micron-szied primary M₇C₃ carbides are well bonded to the substrate. Their elongated shape and high hardness contribute to improved wear resistance. The matrix of the cladding layer consists of eutectic structure composed of micron-sized M₂₃C₆ hard phase and submicron-sized ductile α-Fe phase. The honeycomb-like structure helps prevent excessive wear of α-Fe and the detachment of M₂₃C₆ during abrasion, which is crucial for enhancing overall wear resistance. The nanoscale precipitates strengthen the α-Fe phase, further improving the wear resistance of the fused cladding.