Digital hydraulic systems are composed of parallel fast switching valves (FSVs) and have unique fault tolerance characteristics, while fault identification and localization are the premise of fault tolerance. However, due to the similar fault features, it is difficult to accurately diagnose the faulty valve in an equal-coded digital hydraulic system (EDHS) without its additional sensors. Aiming at solving the problem by relying on the self-contained sensor information only, the identification and localization method for the faulty valve in EDHS based on system pressure information is proposed in this paper, which is the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) with the multi-attribute time series data. Firstly, the flowrate mathematical models of the FSV, the digital flow control unit (DFCU) and its EDHS system are established under typical faults. On this basis, the system performance for typical faults in DFCU on the pump and tank side under different operating conditions are simulated and analyzed. Furthermore, the system pressures under normal and fault conditions, input control signals and the system pressures are used as the database, that is the multi-attribute time series data. And then, the time–space network fault diagnosis model with the combination of CNN and LSTM is developed, and the model reliability and independence are verified by indexes of precision, recall rate, and F1. The results show that the average precision of the hybrid CNN-LSTM model is up to 98.68%, achieving efficient diagnosis performance for faulty valves, which could contribute to fault tolerance of digital hydraulic systems.

中文翻译：

---

基于混合 CNN-LSTM 模型的等编码数字液压系统中快速切换阀的故障识别和定位


数字液压系统由并联快速切换阀 （FSV） 组成，具有独特的容错特性，而故障识别和定位是容错的前提。然而，由于具有相似的故障特征，在没有额外传感器的情况下，很难在等编码数字液压系统 （EDHS） 中准确诊断故障阀。针对仅依靠自包含传感器信息的问题，该文提出一种基于系统压力信息的EDHS故障阀门识别定位方法，该方法将卷积神经网络（CNN）和长短期记忆网络（LSTM）与多属性时间序列数据相结合。首先，在典型故障下建立 FSV、数字流量控制单元 （DFCU） 及其 EDHS 系统的流量数学模型;在此基础上，对不同工况下泵和罐侧 DFCU 典型故障的系统性能进行了仿真分析。此外，正常和故障条件下的系统压力、输入控制信号和系统压力被用作数据库，即多属性时间序列数据。然后，建立了 CNN 和 LSTM 相结合的时空网络故障诊断模型，并通过精度、召回率和 F1 指标验证了模型的可靠性和独立性。结果表明，混合 CNN-LSTM 模型的平均精度高达 98.68%，实现了对故障阀门的高效诊断性能，有助于数字液压系统的容错能力。