Evaluation of Downhole Bit Wear Status Based on Improved CNN-SVM
The existing bit wear evaluation methods have the problems such as the inability of fully extracting the dynamic feature of signals needed by correct classification in the manual feature extraction process and the need for massive calculation of various statistics.Therefore,a new bit wear rate evaluation algorithm based on improved CNN-SVM was proposed in the paper.This algorithm imported the collected near-bit raw vibration data into the CNN-Softmax model,extracted the main feature parameters from the near-bit data through the trained CNN model,input the extracted sparse feature vectors into SVM for fault classification,used genetic algorithm to achieve optimi-zation selection of SVM parameters,and finally used t-distribution stochastic neighborhood method to conduct near neighbor embedding to visualize the fault feature learning process and evaluate its feature extraction ability.In addi-tion,this algorithm was used to analyze the field test data of bit wear.The analysis results show that the accuracy of the downhole bit wear status evaluation algorithm based on CNN-SVM is as high as 98.33%.The conclusions provide theoretical support for realizing further monitoring of bit wear status.
bit wear status evaluationconvolutional neural networks(CNN)support vector machine(SVM)feature extraction visualizationmean pooling sampling
NETL
NSTL
万方数据
