Incremental Cluster Research For Batch Drilling Process Quality Based On Weighted Characteristics Of Monitoring Signals

Aiming at problems of quality detection of batch drilling-holes with the same process parameters, a new approach had been proposed. An Acoustic Emission sensor and a Hell-power sensor were used to monitor the working process. And then the signals'characteristic vector was constructed from the signals'characteristics in Time-domain and frequency-domain. Then we assigned a weight to every characteristic according to its reflection level for working process. Finally, the incremental clustering algorithms had been put forward to analyze the distribution law and develop trend of the batch drilling process quality indirectly. Thus, we can provide a theoretical instruction for the manual sampling inspection. To illuminate the approach, we will discuss it from three main aspects as follow.1. Characteristic extraction. The quantity of data points of these monitoring signals was too huge to analyze directly. So the signals'characteristics in Time-domain and frequency-domain had been extracted. These characteristics were used to represent the signals. Then we constructed a characteristic vector for every group of signals. All the characteristic vectors composed a database for the following analysis.2. Weight assignment. The characteristics extracted from the monitoring signals can reflect the working process. However, the reflection level was different to every characteristic, and this is significant to the following incremental cluster analysis. So we assigned a weight to every kind of characteristic according to its reflection level for working process. In this paper, we used AHP as a tool for the weight assignment, and applied GAs to solve objectivity and logic problems in the Judge-matrix of AHP.3. Incremental cluster analysis. After the characteristic extraction and weight assignment, every group of monitoring signals corresponds to a weighted characteristic vector. Using the vectors as data objects, clustering method was applied to classify the drilling-holes according to their own drilling process quality. Because the weighted characteristic vector database in front of the cluster analysis was changing continuously, we researched three incremental clustering algorithms: InDBSCAN which is based on density, InGrid which is based on grid, SOM which is based on model. And we compared their incremental clustering effects using the cluster evaluating method of'accuracy rate'.Calculation and analysis results show that: the characteristic extraction reduces the difficulty of data analysis effectively. And the result of incremental cluster is more reasonable with the weight assignment. Besides, the improved density based incremental clustering method of InDBSCAN has the highest accuracy rate for batch drilling process quality detection. The rate is 84.90%.