No abstract

Smart remaining useful life (RUL) prognosis methods for condition-based maintenance (CBM) of engineering equipment are getting high popularity nowadays. Current RUL prediction models in the literature are developed with an ideal database, i.e., a combination of a huge “run to failure” and “run to prior failure” data. However, in real-world, run to failure data for rotary machines is difficult to exist since periodic maintenance is continuously practiced to the running machines in industry, to save any production downtime. In such a situation, the maintenance staff only have run to prior failure data of an in operation machine for implementing CBM. In this study, a unique strategy for the RUL prediction of two identical and in-process slurry pumps, having only real-time run to prior failure data, is proposed. The obtained vibration signals from slurry pumps were utilized for generating degradation trends while a hybrid nonlinear autoregressive (NAR)-LSTM-BiLSTM model was developed for RUL prediction. The core of the developed strategy was the usage of the NAR prediction results as the “path to be followed” for the designed LSTM-BiLSTM model. The proposed methodology was also applied on publically available NASA’s C-MAPSS dataset for validating its applicability, and in return, satisfactory results were achieved.

An efficient Remaining Useful Life (RUL) prediction method is one of the most important features of a condition-based maintenance system. A running machine’s RUL prognosis in its real-time is a challenging task, especially when there is no historic failure data available for that particular machine. In this paper, an online RUL of an in-operation industrial slurry pump having no historical failure data has been predicted. At first, the available raw vibration datasets were filtered out for valid datasets. The obtained valid datasets were utilized for constructing the Health Degradation Trends (HDTs) using principal component analysis and a moving average method. Then, a novel procedure for automatically selecting the HDT’s data points for initiating the iteration process of prediction was formulated. Afterward, a hybrid deep LSTM model embedded with a smart learning rate mechanism was developed for estimating the online RUL using the selected points of HDTs. The online RUL prediction results produced by the developed model were quite satisfactory when they were compared with other online RUL prediction methods.

Conclusion: In our patient tension pneumocephalus was complicated by diabetes insipidus and leading to cerebral sinus thrombosis after surgery.