Research On Intelligent Method Of Bearing Fault State Assessment And Prediction For Unbalanced Data

Rolling bearings are widely used in many fields and are a key component of rotating machinery.In general,bearings operate in harsh and complex environments with long running times and high loads,factors that make them highly susceptible to failure during operation,resulting in significant economic losses.Bearing fault condition data is more difficult to obtain in the actual operating environment than normal operating condition data,which results in an unbalanced bearing sample data set,which in turn affects the performance of the bearing fault condition assessment and prediction model,resulting in problems such as overfitting.In addition,most of the traditional bearing condition assessment and prediction algorithms and models rely on complex feature extraction techniques,which have problems such as high expert experience requirements,long time consumption,and poor model generalisation.To address the above problems,the main research elements of this paper are as follows:(1)To address the problems in data imbalance bearing fault state assessment and prediction tasks based on CNN and LSTM techniques,a two-layer CNN-LSTM weighted deep learning model with a sampling strategy(w TLCL-U)is proposed in Chapter 3 of this paper,which uses a sliding time window strategy to transform the fault state assessment and prediction problem into a classification problem and incorporates the sampling strategy and weighted cost loss strategy.The model structure consists of a two-layer CNN and an LN.The model structure is internally and externally stacked by a two-layer CNN and LSTM,which avoids the complex data preprocessing process and better learns the temporal and spatial characteristics of the fault samples.Experimental results show that w TLCL-U has higher stability and robustness than other conventional models for fault prediction tasks,improving the prediction accuracy of bearing faults under unbalanced conditions.(2)In Chapter 4 of this paper,an adaptive boundary weighted oversampling algorithm(ABWSMO)for unbalanced data sets is proposed to address the problems of inner-class and between-class imbalance in unbalanced data learning and applied to bearing fault condition assessment..ABWSMO improves the SMOTE algorithm based on the distribution of the underlying data,calculates the sample space cluster density based on the K-Means clustering algorithm,and incorporates local and global weighting strategies to Generating data mechanisms that enhance the learning of important samples at the boundaries of the dataset overcomes inner and between class imbalances and avoids the traditional oversampling algorithm noise problem.The effectiveness of the algorithm for synthesising data is verified by experimentally comparing five traditional oversampling algorithms on 13 unbalanced ratio datasets and four classifiers in the UCI database.Meanwhile,experimental results on the bearing dataset show that the oversampling algorithm improves the learning of fault sample features by the model and increases the accuracy of fault state assessment.