Abstract
Understanding the Heat Transfer Coefficient (HTC) is essential for evaluating cooling media used in the immersion quenching of steels. This HTC characterizes the heat exchange between the immersed workpiece and the quenchant. Calculating the HTC involves solving an inverse heat transfer problem, which typically requires stochastic optimization algorithms. These algorithms use iterative processes and can be computationally demanding, often needing hundreds or thousands of iterations to find a solution.
To reduce this computational burden, this paper introduces an initialization technique that employs a non-iterative approach to solve the inverse heat transfer problem. The proposed method uses an artificial neural network (ANN), specifically a multi-layer feedforward neural network trained with the backpropagation algorithm. A synthetic database with 150,000 records of heat transfer coefficients, determined as a function of temperature, is created for training the network. Unconventionally, the Fourier transform of the cooling curve is used as input for the inference system. Additionally, the performance of the neural network is compared with other conventional learning algorithms. Results show that when combined with stochastic algorithms, the ANN achieves comparable solutions in a shorter amount of time.
