Data Analytics And High-Efficiency Monitoring Methods For Nonstationary Complex Industrial Processes

With the development of technology and expansion of market demands,modern industries have experienced an increase in the scale and complexity.As a consequence,the requirements for process safety,product quality and economic benefits are growing significantly.Process monitoring can be used to timely detect the abnormal events,and provide diagnosis information to effectively eliminate the detrimental effects of the process anomalies.Therefore,process monitoring has attracted increasing attention in both academic research and industrial applications.By translating historical data into process information,data driven methods detect and diagnose process upset resort to massive data measured in industrial processes rather than first-principle knowledge.In this way,some complex industrial processes whose physics and mathematics models cannot be effectively developed can also be operated safely,efficiently and economically.In real industrial applications,the complexities of the process data bring a great challenge to the existing data-driven based fault detection and diagnosis methods.As for fault detection,the process changes over time because of the complexity nonstationary characteristics,and thus the process anomalies cannot be effectively detected using a fixed model.To be specific,continuous processes exhibit nonstationary characteristics due to the normal slow changes and normal shifts of operation conditions;batch processes generally include multiple operation conditions and thus are also nonstationary.As for fault diagnosis,many factors should be taken into consideration,including the smearing effect of process anomalies,the problem of insufficient data,the extraction of fault tendency,the limit performance of single model,ect.In this dissertation,a systematic set of data-driven based fault detection and diagnosis methods are proposed to solve the aforementioned problems for complex nonstationary industrial processes In the first chapter,the research background and current situation of fault detection and diagnosis are introduced.After that,the process monitoring methods for continuous processes and batch processes are provided in the second and third chapters,respectively.After detecting the process anomaly,its faulty variables are isolated to infer its location in the process system Based on the result of Chapter four,a diagnosis model is developed to identify the category of the detected anomaly,so that the fault cause can be further inferred based on historical data.Due to the limited performance of single model,an ensemble learning strategy is proposed to integrate the diagnosis results of multiple methods in a probabilistic manner.The detailed information of the dissertation is provided as follows.(1)A recursive exponential slow feature analysis(RESFA)algorithm is developed for fine-scale adaptive monitoring to solve the false model updating problem in conventional adaptive methods.First,an exponential slow feature analysis(ESFA)method is proposed to nonlinearly extract slow features,so that the general trend of the process variations can be better captured.On the basis of ESFA model,a fine-scale adaptive monitoring scheme is developed to accurately identify the normal changes of industrial processes,including normal slow varying and normal shift of operation conditions.In this way,the normal slow varying can be effectively distinguished from incipient faults with unusual dynamic behaviors to avoid falsely adapting for fault case,and the monitoring model can be correctly updated for new operation condition after distinguishing real process anomalies from normal shifts of operation conditions.A simulation process and two real industrial processes are adopted to validate the performance of the proposed adaptive monitoring method.Experimental results show that the proposed method can effectively identify different operation statuses to decide whether to update the monitoring model or to raise an alarm.(2)A stationary subspace analysis(SSA)based hierarchical monitoring model is developed to divide batch process in the low dimension space and detect incipient faults for batch processes.The proposed method extracts the global stationary features from the historical process data,and establishes a global monitoring model for the time-invariance information throughout the whole batch process.Based on the remaining nonstationary global features,a phase partition method is developed to divide the process using dynamic information in the low dimension space.According to the partition result,local monitoring models are constructed for each operation phase using equilibrium relationship and dynamic information.The operation status of the newly collected sample is codetermined by both the global and local models,and a physical interpretation is provided for better process understanding.The proposed method is illustrated using a simulated process and a real industrial process.Experimental results show that the proposed method can extract key features to accurately divide the batch process into different operation phases,and effectively detect the incipient fault so that immediate and corrective actions can be taken.(3)A sparse exponential discriminant analysis(SEDA)algorithm is proposed for seeking the discriminant directions simultaneously with variable isolation.The sparse discriminant model is developed by introducing the penalty of lasso or elastic net into the exponential discriminant analysis(EDA)algorithm,so that the key variables responsible for the fault can be automatically selected.Since the formulated model is non-convex,it is recast as an iterative convex optimization problem using the monorization-maximization(MM)algorithm.After that,a feasible gradient direction method is developed to solve the optimization problem effectively.The sparse solutions indicate the key faulty information to improve classification performance and thus distinguish different faults more accurately.A simulation process and a real industrial process are used to test the performance of the proposed method,and the experimental results show that the SEDA algorithm can isolate the faulty variables and simplify the discriminant model by discarding variables with little significance.(4)A broad convolutional neural network(BCNN)is designed with incremental learning capability to extract fault tendency for solving the fault diagnosis problem.The proposed method combines several consecutive samples as a data matrix,and then extracts both fault tendency and nonlinear structure from the obtained data matrix using convolutional operation.After that,weights in fully connected layer can be trained based on the obtained features and their corresponding fault labels.Because of the architecture of this network,the diagnosis performance of BCNN model can be improved by adding newly generated additional features.Finally,the incremental learning capability of the proposed method is also designed,so that BCNN model can update itself to include new coming abnormal samples and fault classes.The proposed method is applied to a simulated process and a real industrial process.Experimental results illustrate that it can better capture the characteristics of the fault process,and effectively update diagnosis model to include new coming abnormal samples and fault classes.(5)A probabilistic ensemble learning strategy is proposed based on Bayesian network(PEL-BN)to integrate the diagnosis results of multiple models in a probabilistic manner.First,an ensemble index is proposed to evaluate the candidate diagnosis models in a probabilistic manner,so that the diagnosis models with better diagnosis performance can be selected.Then,based on the selected classifiers,the architecture of the Bayesian network can be constructed using the proposed three types of basic topologies.Finally,the advantages of different diagnosis models are integrated using the developed Bayesian network,and thus the fault causes of the observable anomalies can be accurately inferred.A simulation process and a real industrial process are adopted to verify the performance of the proposed method,and the experimental results illustrate that the proposed PEL-BN strategy improves the diagnosis performance of single faults.