Research and Development of Advanced Intelligent Health Management and Control System of Power-Driven Arrangement for Smart Manufacture

Industrial fans and rolled spindle are significant fundamental rotating machinery in manufacturing factories, a sudden shutdown of the critical arrangements is highly unfavorable. Implementation of big data and powerful machine learning techniques will be utilized in rotating arrangements. To detect early fault effectively and accurately, the first is to confirm the abnormal features of rotating machinery in preliminary damage. Because of the mechanical properties of general rotating machinery, the abnormal features of early fault would first appear fall within the medium and high frequency fields. Another emphasis will be to select the proper sensor according to the abnormal features and sensing characteristics for the early fault detection. The sensing resolution bandwidth features of acoustic emission and vibration sensors will be verified by the Karhunen-Loeve transform and Mel-frequency cepstral coefficient spectrum feature filtering. Acoustic emission sensors have an advantage of sensing resolution bandwidth especially within the medium and high frequency fields compared to vibration sensors. The experiment shall prove that acoustic emission signal and acoustic feature analysis method could effectively realize the real-time early fault detection and prediction on industrial fans and rolled spindle. This project will also apply the machine learning (ML) approaches for industrial fans and rolled spindle. The feature algorithm is implemented on industrial fans and rolled spindle using an acoustic filtering technique. It will be implemented on the platform of National Instruments embedded compact-RIO. The fault diagnostic process will be based on the machine learning algorithms by using acoustic feature information for dynamic arrangements. Acoustic signals will be acquired from array microphone of industry. Experimental investigations were carried out for the practical industrial fans and rolled spindle system. The effectiveness of the proposed system on the fault features will be classified by using the multiple layer perceptron (MLP) and support vector machine (SVM) methods. A smart fault prediction for industrial fans and rolled spindle was proposed that can distinguish the feature differences of normal and abnormal ones. For classification and identification of rotating arrangement, we will adopt the MLP and SVM method of ML algorithm. Implementation results on GPU processors will be discussed The proposed rotating system can be regulated by active control approach when they have early damage.

Project ID:PB10907-2905
External Project ID:MOST109-2221-E182-036