Temperature-dependent grain size prediction model for Nimonic 80A superalloy based on multi-feature information of laser ultrasonic and Bayesian neural network

Nimonic 80A superalloy is an important material for key components in aerospace and other fields. Its grain size directly affects high-temperature performance, and achieving grain size detection at different temperatures is crucial for ensuring product quality. Laser ultrasonic technology is characterized by its non-contact nature, making it highly suitable for microstructural characterization of hot metal materials. Bayesian neural networks (BNNs) can handle complex nonlinear relationships and provide quantified prediction results with confidence levels. This study combines laser ultrasonic technology with BNNs. Specifically, the research was conducted as follows: Nimonic 80A superalloy samples with different grain sizes were prepared through rolling deformation and solution treatment. In situ laser ultrasonic detection experiments were conducted on heated samples to obtain signal data from superalloy materials with varying grain sizes ranging from room temperature to 1000 °C. These experiments revealed the effects of temperature and grain size variations on ultrasonic energy attenuation. Ultrasonic attenuation coefficients were calculated based on the amplitudes of laser ultrasonic time-domain and frequency-domain signals. Using temperature and attenuation coefficients as inputs, a grain size prediction method based on BNN was proposed. For comparison, Gaussian Process Regression (GPR) and Quantile Random Forest (QRF) models were also constructed. The three types of models were trained and tested using signal samples, and the performance of these models was compared, demonstrating the applicability and superiority of the BNN model. The results showed that within the sample grain size range of 45.75 μm to 141.38 μm, the maximum prediction error of the BNN model is +7.29 μm (compared with metallographic measurements), and its prediction accuracy is superior to that of the GPR and QRF models. In addition, the predicted values of Nimonic 80A alloy grain size grades were provided in accordance with the industrial standard for grain size grades.