Improving the thermo-electro-mechanical responses of MEMS resonant accelerometers via a novel multi-layer perceptron neural network

Microelectromechanical systems (MEMS) accelerometers have been considerably developed since their wide use in current industrial applications. MEMS resonant accelerometers have attracted considerable attention in recent years by introducing some advantages over analog ones in terms of supporting a large range of dynamic movements, providing good sensitivity, considerably reducing the effects of interferences, etc. However, since silicon-based designs have shown some temperature-related disadvantages, there are always significant errors in such sensors due to the variable temperature. In this paper, we propose a novel multi-layer perceptron (MLP) neural network to improve the accuracy temperature compensation model for MEMS resonant accelerometers. To do so, we propose a novel metaheuristic algorithm by improving the searching ability of an artificial bee colony (ABC), named global search artificial bee colony (GSABC) to optimally train the MLP for gaining a better temperature compensation model. We then first evaluate GSABC on engineering and some well-known benchmark datasets in comparison with some widely-used and novel metaheuristic algorithms. After that, we perform various experiments to train the proposed MLP. Our experiments show that the proposed GSABC-derived MLP outperforms the state-of-the-art in providing higher accuracy for temperature compensation of MEMS resonant accelerometers. It is shown that within the calibration measuring experiment of the MEMS system, the environmental temperature was varied from 0 °C to 90 °C with a temperature interval of 10 °C.