A multi-variate rotating machinery fault optimization and recognition framework

Accurate diagnosis, prognosis, and forecasting of a machine's time-to-failure are crucial through the massive, profound monitoring parameters and expeditiously is essential to assist in the preservation of reliability and life remaining for the machine increase by recommending an appropriate decision to diminish the occurrence of ruinous malfunction and substantial financial losses proactively. Artificial intelligence is considered significantly crucial to specialists in diagnosing various faults. But all research, although have great accuracy, still till now have a lack on the number of considering parameters to achieve the diagnosis or prognosis.in addition to, it depends on repeat the processing for sensor reading in raw data which already the control system do it, so, This paper suggests an organizing principle is proposed for the malfunction diagnosis and is used to measure the variant parameters in monitoring values compared to the same conditions and restrictions. It depends on the unit control system converting all sensor readings from raw data forms to numerical values so it is easy and applicable to apply in industry. The suggested framework is a conventional ten-layer deep neural network (DNN): the first one is the input layer (71 inputs) which extracted from the control system directly, the next eight layers are hidden layers, respectively (60, 60, 50, 40, 30, 20, 10, 8), and finally the last layer is the output dataset with 1194 numerical reading extracted directly from the machine controller. The best recorded metrics are 99.9% accuracy after only 10 epochs using DNN.