Research On Incipient Fault Diagnosis Method Of Batch Process Based On Deep Feature Extraction

With the development of modern industry,batch process,as one of its components,plays an increasingly important role in modern industry.Accurate and effective fault diagnosis of batch process is very important to improve product quality and ensure production safety.Incipient fault has the characteristics of low amplitude,easy to be covered by noise,extremely difficult to be detected and diagnosed.It widely exists in the industrial production of batch process,and an effective method of incipient fault diagnosis of batch process is very important in modern times.In view of the strong nonlinearity,high dimensionality,small amplitude and easy to be covered by noise,this thesis proposes an intelligent fault diagnosis model based on the combination of depth feature extraction and machine learning.The main contents of this paper are as follows:(1)Incipient fault diagnosis of batch process based on machine learningAccording to the characteristics of strong nonlinearity and high dimensionality of batch process data,as well as the characteristics of low amplitude of micro fault and easy to be covered by noise,the shortcomings of traditional methods are analyzed,and the machine learning methods SVM,ELM and ANN,which are suitable for nonlinear problems,are introduced into the micro fault diagnosis of batch process,and a fault diagnosis model based on machine learning method is established,The penicillin fermentation data were used for experimental verification.And it paves the way for further research.(2)Incipient fault diagnosis of batch process based on KECA feature extraction and IGWO-KELMOn the premise that the single machine learning method is not effective,a compound fault diagnosis model based on intelligent feature extraction method and machine learning method is proposed.KECA algorithm is used to extract the features of the data and obtain the key features of the data.KELM is selected as the classifier,and the improved GWO algorithm is used to select the intelligent parameters in the training process of EKLM model.The optimal fault diagnosis model is established in multiple stages,and the validity of the method is verified by the data of penicillin fermentation process.(3)Incipient fault diagnosis of batch process based on deep time series feature extractionIn view of the fact that the batch process data is a kind of time series in essence,the paper improves the feature extraction method,selects the LSTM network with excellent processing ability for the time series as the feature extraction method,and improves the network with AE network,so that the network has the ability of data denoising and deeper feature extraction.Softmax is used as the classifier in the feature output of the network,and penicillin fermentation process data is used for experimental research,which shows the effectiveness and superiority of the deep-seated sequential feature extraction method,and paves the way for the subsequent model research based on the deep-seated learning method.(4)Incipient fault diagnosis of intermittent process based on space-time fusion featureIn order to further explore the application of deep learning method in the diagnosis of batch process incipient fault and obtain a better diagnosis model,considering that the single batch of intermittent process data has similar characteristics with the picture data,CNN network is introduced to extract its spatial dimension characteristics.Combined with LSTM network to extract features in time dimension,the two networks perform parallel operation and intelligent fusion at the output.Based on the time-space fusion feature,a model of intermittent process incipient fault diagnosis is proposed,which is a feasible and accurate method for intermittent process micro fault diagnosis.