Predicting and optimizing the wear behavior of centrifugal cast β-TCP/Mg–2Zn–1Mn alloys: A hybrid artificial neural network—Grey relational analysis approach

Abstract

This study investigates the wear behavior of temporary implant such as plates and screws made by Mg–2Zn–1Mn alloy reinforced with β-tricalcium phosphate (β-TCP) under both dry and wet sliding conditions. Grey relational analysis (GRA) and an artificial neural network (ANN) model were utilized to predict wear and optimize performance, considering process parameters such as β-TCP content (0%, 2.5%, 5%), load (20, 40, 60 N), sliding speeds (2, 4, 6 m/s), and sliding distances (800, 1600, 2400 m). The wear rate (W_s) and coefficient of friction (COF) were evaluated using an orthogonal array L27 to optimize both dry and wet conditions. The optimal process parameters obtained through GRA for dry conditions are 5% β-TCP, 20 N, 4 m/s, and 1600 m, while for wet conditions, they are 0% β-TCP, 20 N, 2 m/s, and 2400 m. The experiment was conducted to calculate W_s and COF using the above process parameters. The values of W_s and COF are 0.010149 mm³/m and 0.109 for dry conditions, and 0.001143 mm³/m and 0.128 for wet conditions. The ANN predicted values for dry conditions are 0.010471 mm³/m for W_s and 0.11045 for COF, with error percentages of 3.17% and 1.33%, respectively. Similarly, for wet conditions, the predicted values are 0.001107 mm³/m for W_s and 0.125 for COF, with error percentages of 3.13% and 2.77%. The ANOVA results for the Grey relational grade (GRG) show that load was the most significant factor influencing wear performance, contributing 51.91% in dry conditions and 34.69% in wet conditions. The close alignment of predictions from both models, with overall error percentages below 3.17%, demonstrates their reliability and effectiveness. Field emission scanning electron microscope (FESEM) analysis of worn surfaces under optimal conditions provided insights into the wear mechanisms. This study confirms that GRA and ANN models are highly effective for predicting and optimize the wear behavior.