A wide range of fault-bearing information is generated during equipment health monitoring tasks, but this information usually exists in the form of text in different stages, which leads to the inefficient use of this information. In addition, there are few studies on information extraction methods in the equipment failure field, and the entity nesting in the field of equipment failure is so severe that most existing named entity recognition (NER) methods don't apply. To address this problem, this paper proposes a dual neural network entity recognition model based on scan filtering for identifying entities in equipment fault texts. The method incorporates a domain dictionary to increase the granularity of text sequences in terms of words, uses sliding windows to extract sequence fragments, and filters sequence fragments by rules to reduce the proportion of negative samples. And feature encoding of sequence fragments using the pre-trained model of bi-directional encoder representations from transformers (Bert), fusion extraction of sequence features using bidirectional long-short-term memory (BiLSTM) network, and convolutional neural network (CNN) to predict the entity labels of sequence fragments. It has been experimentally demonstrated that the model outperforms the baseline model in the equipment faulty corpus.