This paper focusses on the identification of the blockage fault in the inlet pipe of centrifugal pump over a range of speed using data obtained from different types of sensors. Acceleration, pressure and motor line current signatures taken using accelerometer, pressure transducer and current probe, respectively, and are used for identification of the pipe blockage level. Methodology is given for the blockage detection based on the multiclass classification using the deep learning algorithm at different blockage levels and speed of rotation of the pump. Importance of the multi-source data collection is emphasized based on the obtained results. Effect of the motor speed is also discussed when considered as an input feature to the classifier. It is observed that the use of combinations of different types of sensors help to identify the blockage level with better accuracy (close to 100% for many combinations). Blockage level prediction at each speed separately, is also given, which can be used for the fault diagnosis of single speed pumps. Finally, the performance of the classifier is tested using some unknown data (different from the data at training speed) to check the blockage prediction accuracy of the classifier.

本文重点介绍了使用从不同类型的传感器获得的数据，在一定速度范围内识别离心泵进口管中的堵塞故障。分别使用加速度计、压力传感器和电流探头获取的加速度、压力和电机线路电流特征，用于识别管道堵塞水平。给出了基于多类分类的堵塞检测方法，该方法使用深度学习算法在不同的堵塞级别和泵的转速下进行。根据获得的结果强调多源数据收集的重要性。当考虑作为分类器的输入特征时，还讨论了电机速度的影响。据观察，使用不同类型传感器的组合有助于以更好的准确度识别堵塞水平（许多组合接近 100%）。还分别给出了各速度下的堵塞程度预测，可用于单速泵的故障诊断。最后，使用一些未知数据（不同于训练速度的数据）来测试分类器的性能，以检查分类器的阻塞预测精度。