Research And Application Of Industrial Time-series Analysis Method Based On Wavelet Transform And Neural Network

The Internet of Things connects things through networks,and then creates new value through data collection,analysis and automation.The Industrial Internet of Things refers to products,sensors,production equipment and so on connected through wireless or wired networks.Since the introduction of the Internet of things,manufacturing,logistics,oil,natural gas,transportation and other industries have begun to invest their eyes.Time series are the main data in industry.The analysis of industrial time series can explore data,so as to improve product quality,avoid accidents and reduce environmental impact.More importantly,the analysis results provide actionable insights that enable engineers to save time when making smarter,more data-driven decisions.Aiming at the problems in different industrial fields,according to the characteristics of large volume and wide sources of industrial time series,this paper studies them under the framework of wavelet transform and neural network model.Firstly,according to the characteristics of industrial time series and the current demand for overall timeliness and application of industrial Internet of things,after studying and comparing different time series databases,this paper selects Influx DB as the management tool of industrial time series.Influx DB has the characteristics of fast writing and multi interface query,which is helpful for developers to manage industrial time series such as sensor data with Internet of things equipment,so as to facilitate subsequent data processing and analysis.Then,aiming at the problem of industrial time series analysis in the scene of tool wear condition monitoring,a wavelet transform and neural network model based on Adam optimization algorithm and Savitzky-Golay smoothing algorithm are proposed.Wavelet transform is universal for extracting frequency features from time series,so using such features in condition monitoring has a good effect.Due to the correlation between the front and back states of the data,the final state detection can be more accurate by using the Savitzky-Golay filter with convolution property for data smoothing.The experimental results also show that the model considering wavelet characteristics,Adam optimization algorithm,neural network and Savitzky-Golay smoothing algorithm is better than the traditional model in the tool wear condition monitoring scene.At the same time,aiming at the problem of industrial time series analysis of motor fault diagnosis scene,a wavelet feature selection method based on Parseval’s energy theory is proposed.Starting from the characteristics of fault occurrence,the signal of fault occurrence can be obtained after the selected features are entered into the neural network,and the robust decision-making in the presence and absence of fault is realized.Finally,aiming at the problem of industrial time series analysis of turbine engine remaining useful life prediction scene,a wavelet feature selection method based on Variance Threshold is proposed.When there are many data and representative features need to be extracted,the results obtained by this wavelet selection method combined with neural network can better predict the remaining useful life of turbine engine.Through the industrial time series analysis of a variety of industrial scenes,using the universality of wavelet transform and neural network algorithm,the model is optimized according to the data characteristics of different scenes.It can not only carry out condition monitoring,but also carry out fault diagnosis and life prediction.The application of time series database in industrial scenes makes the whole theory reliable on the basis of application.