Performance prediction methodology based on pattern recognition

Performance prediction is crucial to increasing drill availability and reliability as well as facilitating further proactive maintenance. This paper presents a novel method for pattern recognition based performance prediction. The features of vibration signals are extracted by wavelet packet transform after de-noising, and then band energy based feature measures are identified by ant colony clustering analysis to form an input vector for the performance assessment model. Prediction of performance degradation trend of drill is carried out using feature mapping based on the established performance assessment model. Experimental results have shown that the proposed method can analyze drill degradation quantitatively, and predict problems before they occur. Furthermore, ant colony clustering algorithm is improved to adjust comparison probability dynamically and detect outliers. Compared with other ant colony clustering algorithms, the algorithm has higher convergence speed to meet requirements of real-time analysis as well as further improvement of accuracy. Finally, effectiveness and feasibility of the proposed method are verified using vibration signals acquired from a drill test bed.

► Clustering analysis using improved ant colony algorithm in frequency domain.
► Extracting major features from pattern of frequency band perspective.
► Evaluating operating conditions of equipment using extracted major features.
► Predicting performance of equipment to facilitate further proactive maintenance.