If I understand your question correctly, your goal is essentially to simulate key output variables based on input variables?

For this, I would use a supervised learning approach. Probably start with an ANN if performance is not dependent upon previous timesteps, or an LSTM if performance is dependent on previous timesteps.

The trick is that you need enough high-quality data to train the models. If you don't have any data, then ML will not be a good approach. In that case you are either left with mechanistic modeling, or finding a way to collect the data.

Yes, machine learning can effectively simulate the thermal processes of a bioenergetic device by analyzing complex heat transfer patterns, optimizing system performance, and predicting thermal behavior under varying conditions. Data-driven models, such as neural networks and regression algorithms, can complement traditional physics-based simulations for improved accuracy and efficiency.