Classification of draglines failure types by K-NearestNeighbor algorithm

• Main Authors: Taghızadeh, Amir, Demirel, Nuray
• Other Authors: Ghodrati, Behzad, Kumar, Uday, Schunnesson, Håkan
• Format: Conference Object
• Language: unknown
• Published: 2017
• Subjects: Machine learning; K-nearest neighbor; Dragline; Reliability; Maintenance; Luleå
• Online Access: https://hdl.handle.net/11511/83657; https://docmh.com/classification-of-draglines-failure-types-by-k-nearest-neighbor-algorithm

Availability of mining machines plays a significant role in mine production. Dragline’s reliability has a great impact on sustaining economic feasibility of open cast coal mining projects. In that sense, reliability of draglines and optimizing its preventive maintenance are key issues to be addressed. The objective of this study is to apply machine learning methodologies for classifying failure types of a dragline based on a real data. The mean time between failure data was acquired from an operating open cast coal mine in Turkey. Three modified form of K-Nearest Neighbors algorithms have been used as a predictor for failure classification. An approximation function has been generated based on the time to failure and break-down type. In case of parameter tuning, cross validation method has been utilized. This caused more reliable evaluation of the test sample, so average testing performance has been used for test data estimation. The basic model was for parameter tuning; Moreover, for achieving more efficient parameter Grid Search method was utilized. Since, usage of the algorithm is computationally expensive, so Randomized Search method has been carried out in order to figure out the functionality of modeled function in the high dimension datasets. The results of the study revealed that the application of K-Nearest Neighbors method reached the Regression Analysis of 73 percent. Thus, the higher accuracy of prediction of failure type can be helpful in prognostic of dragline’s procedure. The main novelty of this study is utilization of machine learning approach for dragline maintenance for the first time.