Research And Application Of Deep Intelligent Prediction Model For Smart Industry

In the midst of ongoing advancements in sensor technology and cloud computing,deep learning has progressively emerged as a central driving force for the transformation and modernization of traditional industries towards intelligent industries.Owing to its outstanding adaptability,efficacious prediction capabilities,and relatively affordable costs,deep learning has found extensive application in the domain of equipment remaining useful life prediction.Nevertheless,during the practical deployment of deep learning across various industrial sectors,numerous problems and challenges persist.For instance,within the aerospace industry,engine data is replete with complex conditions and multiple sensors,leading to potential overfitting or underfitting issues when utilizing a single model.In contrast,within the mechanical processing industry,bearing data is characterized by information redundancy,substantial data volume per individual training sample,and heightened computational memory requirements.This paper concentrates on the exploration of remaining useful life prediction technology,and aims at the difficulties in the application of deep learning in different industrial fields to propose the different innovation prediction models.The primary research objectives of this paper are as follows:1)To address the challenges faced by engine data in the aerospace manufacturing sector,this study proposes a parallel engine remaining useful life prediction model,which synergistically employs a Dense Multiscale Convolutional Network(DMCN)and a Long Short-Term Memory Network Based on Temporal Pattern Attention(LSTM-TPA).The primary objective of this innovative approach is to solve the complexities inherent in multi-sensor variable prediction problems.The DMCN serves as a sophisticated feature extraction module,integrating multi-scale features and enhanced dense connectivity techniques.By fusing multiscale characteristics,the DMCN is capable of broadening the model’s perceptual field,achieving feature reuse,and ultimately reducing overall model complexity.The LSTM-TPA functions as an optimized LSTM feature extraction module,grounded in the temporal pattern attention mechanism.This approach can improve the model’s generalization capabilities when faced with complex datasets by meticulously determining the contributions of sensor variables through a weighting process.Moreover,this parallel structures effectively utilizes the spatial and temporal features extracted by DMCN and LSTM-TPA,hence can achieve a more accurate prediction of the remaining useful life.2)To address the challenges presented by bearing data in the mechanical processing industry,this study introduces a bearing remaining useful life prediction model founded upon Variational Modal Decomposition Reconstruction,an Improved Dense Temporal Convolutional Attention Network and Gated Recurrent Units(IDTCAN-GRU).The primary purpose of this innovative approach is to tackle the prediction problem associated with computationally complex samples.Firstly,a variational modal decomposition method is employed to eliminate redundant information present in the original signal and subsequently reconstruct the signal.Secondly,an attention mechanism featuring enhanced time step contribution is designed,leading to improvements in the basic structure of the temporal convolutional network.In addition,the IDTCAN module is designed to parallelize the computation of large receiver field timing data,effectively managing computational complexity.Finally,the GRU is employed to address the long-term dependencies inherent in time series data,ultimately enhancing the model’s nonlinear fitting capabilities and convergence speed.3)This study combines the two prediction models proposed above and designs an intelligent prediction software for remaining service life based on deep learning algorithms.The software encompasses a comprehensive range of features,including data pre-processing,model training,model prediction,visualization and analysis,as well as maintenance recommendations.This software aims to accurately predict the remaining life of equipment,subsequently providing users with precise maintenance recommendations and robust decision support.The software can provide companies advanced technological tools and cutting-edge capabilities and help them stay ahead of the competition in the market.