Prediction Of Remaining Useful Life Of Rotating Machinery Based On Deep Learning

In the era of Industry 4.0,scientific and technological progress drives the development of machinery and equipment in the direction of scale and intelligence,which puts forward more stringent requirements for the reliability of mechanical components.Rotating machinery is a key component of large machinery and equipment,however,after long-term operation,rotating machinery will inevitably wear out,deteriorate performance,and eventually lead to failure.Once a failure occurs,it will threaten life safety and cause significant economic losses.Therefore,the prediction of the remaining service life of rotating machinery is of great significance for the safe production and maintenance of machinery and equipment.Network models based on deep learning have wide recognition in academic and technical fields because of their powerful feature recognition capabilities.However,due to the complex working conditions of rotating machinery and the accompanying noise interference,it will cause serious interference to the network model training process.Aiming at the problem that the original vibration signal features are difficult to extract,the parameter adaptive signal decomposition is proposed,which can effectively extract the weak features sensitive to early faults,extract the characteristic parameters from multiple angles in the time domain,frequency domain and time and frequency domain,construct health factors that can quantitatively reflect the degradation degree of rolling bearings,and then propose a residual life prediction framework based on multi-parameter fusion.The research results have certain reference value and guiding significance for improving the life prediction ability of rotating machinery.The main contents of this article are as follows:(1)Aiming at the problem that the modal number and penalty factor are difficult to determine in the traditional variational mode decomposition method,an improved variational modal decomposition method with parameter adaptive selection is proposed,and the weighted spectral peak ratio index is constructed based on the Gini index and spectral peak ratio index as the objective function,the sparrow algorithm is used to optimize the optimal combination of modal number and penalty factor,and the monotonicity and robustness index are used to screen out the sensitive features that can reflect the bearing degradation process.(2)In order to solve the problem that a single characteristic index cannot effectively reflect the degradation trend of rotating machinery,a method for extracting health factors based on Gaussian mixture model is proposed.The Gaussian mixture model was established by using the data of normal state after dimensionality reduction of t-distribution random near-neighbor embedding,and based on Bayesian inference of the distance between the test data and the Gaussian model of the healthy state,it was used as a quantitative index to reflect the degradation degree of rotating machinery,and the experimental results showed that the model had good performance in reflecting the bearing degradation trend.(3)Aiming at the problem that the remaining service life prediction accuracy of rotating machinery is insufficient,a data-driven remaining useful life prediction(RUL)model is proposed.First,the self-attention mechanism(SA)is used to improve the local feature extraction ability of Convolutional Neural Networks(CNN),and the Bi-directional Long-Short Term Memory(Bidirectional Long-Short Term Memory)is used by Particle Swarm Optimization(PSO).Bi LSTM),and finally input the features extracted by SA-CNN into the optimized Bi LSTM to construct a rotating machinery RUL prediction model,and the experimental results show that the prediction accuracy of SA-CNN-PSO-Bi LSTM is about 6.391%,21.268% and 35.888% higher than that of SA-CNN-Bi LSTM,Bi LSTM and LSTM,respectively.