Abstract
This paper proposes a decoupled fan control (DFC) for a multivariable server system that combines a proportional-integral-derivative (PID) control and fan weighting settings. To reduce labor and time, PID parameters are automatically tuned online by using a neural network, and the values of the fan weighting setting can be easily assigned to decouple the multiple fans. Furthermore, the PID auto-tuning achieves the optimal combination of control gains in terms of the integral square error criterion. To verify the proposed method, a mockup system is constructed to simulate a real server. The experimental results show that the temperature responses satisfy the cooling requirements of the server, namely the minimal temperature overshoot and low-amplitude oscillations. In addition, the DFC reduces power use by 16.8%.