Fault diagnosis of the hybrid system composed of proton exchange membrane fuel cells and ammonia-hydrogen fueled internal combustion engines under adaptive power allocation strategies

Based on the complementary efficiency characteristic of PEMFCs and ammonia-hydrogen fueled internal combustion engines (AHICEs), an adaptive power allocation strategy is proposed by this paper to enhance the efficiency of hybrid systems in a wide load range from 0 to 500 kW. With the increased load from 0 to 500 kW, the fault diagnosis of hybrid systems is implemented by a robust diagnostic method for single-fault/hybrid-fault states, where the robust diagnostic method is composed of the multi-scale convolutional neural network (MCNN) and the bi-directional long short-term memory (BiLSTM) neural network. The diagnostic results show that the diagnosis accuracy is 97.5 % for single-fault states of AHICEs, 99.1 % for single-fault states of PEMFCs, 95.76 % for hybrid-fault states of hybrid systems respectively. Based on that fact, the diagnosis accuracy of MCNN-BiLSTM methods is higher than that of widely employed diagnosis methods, attributing to the enhanced capability of feature extraction and temporal processing. Here these employed methods consist of the support vector machine (SVM), gated recurrent unit (GRU), MCNN-least squares support vector machine (MCNN-LSSVM) and MCNN-long short-term memory neural network (MCNN-LSTM).