Fault Diagnosis for Energy Transportation-Oriented Intelligent Automation System via Mixed Neural Networks

A leak fault is a safety risk to interrupt the operation of the transportation process for intelligent automation system. To judge the system failure, a fault diagnosis method based on mixed neural networks is proposed in this paper. First, the multi-attribute extraction module is proposed to provide the local and global data changes caused by the leak event for the following neural networks, which reduces the noise influence from the complex environment. Second, a feature-sharing neural network is proposed to capture the multi-scale features of data changes and further implement leak fault detection and location through the sequence information and the corresponding sequential dependencies. Third, the task-dependency loss function is proposed to update the network parameters of the joint learning of different network outputs in the whole training process. Based on the proposed network structure, the proposed method could achieve two different targets of fault detection and location simultaneously for long-time series data. Finally, different case studies of the collected acoustic data are studied, and the analysis results show the effectiveness of the proposed method for leak fault detection. Note to Practitioners-A leak event is considered a sudden system fault in the transportation-oriented intelligent automation system. In order to reduce and minimize the system damage influence, leak fault diagnosis is the key point to identify operation risks based on the analysis result of the collected data changes. Thus, a mixed neural networks-based method is proposed to achieve the detection and location of leak events in this paper. With the combination of different neural network structures, multi-scale data changes could enhance the analysis capability of the proposed method for different fault diagnosis targets in complex operation scenarios. The case results show that the proposed method is an implementation way to ensure system safety and is better than other detection methods through the comparisons of different evaluation metrics.