Detection And Isolation Of Process Faults In Wastewater Treatment Plants

There is a serious shortage of water resources in China.Due to the uneven development of urbanization and industrialization in the early stage of reform and opening up,this problem has become more serious.Wastewater treatment is an effective means to realize water cycle and alleviate water shortages.In this process,wastewater treatment plants play an important role in it.The stable operation of the wastewater treatment plant is the guarantee for the effective treatment of wastewater.However,there are problems in the wastewater plant,such as complex biochemical reactions,harsh environment,strong external disturbance,erosion of control equipment and mechanical equipment,etc.,which easily lead to process faults such as sludge bulking,equipment failure.Once these faults occur and are not handled in time,it may result in water quality not meeting the discharge standards and even cause more serious safety problems in the wastewater treatment plant.Therefore,the research on process fault detection and isolation(fault diagnosis)is of great significance to the stable operation of wastewater treatment plants.In this thesis,the research on the process fault diagnosis of the wastewater treatment plant is carried out.Firstly,in the wastewater treatment plant,the relationship between several process faults and effluent variables is analyzed,and a fault signature matrix is established.Secondly,in order to obtain key water quality variable data that can reflect important information in real-time,a soft-sensor model based on Gaussian process regression(GPR)is designed.In the soft-sensor model,the input variables are selected to improve the model prediction accuracy.Subsequently,based on the interval prediction model and Bayesian reasoning,a process fault detection and isolation method is proposed.The prediction interval of the interval prediction model is used as the threshold,combining the variables data obtained by hardware-sensors or softsensors for fault detection.On this basis,fault isolation is also carried out by using the fault signature matrix and Bayesian reasoning.Finally,a set of software for process fault diagnosis of the wastewater treatment plant is developed,which can effectively monitor the system.The main research work of this thesis is as follows:(1)Analysis of the process fault mechanism of the wastewater treatment plant.The wastewater treatment benchmark simulation model 1(BSM1)is used as the research object,which can simulate process faults more conveniently and help understand the process of faults.Four process faults are simulated by changing the parameters in BSM1.By studying the biochemical reaction mechanism model and the kinetic model of the secondary settling tank in BSM1,the relationship between the process faults and the effluent variables is found to establish the fault signature matrix.Finally,a process fault simulation experiment is carried out to verify the correctness of the process fault analysis.(2)Research on soft-sensor model based on GPR.Aiming at the problem that some key water quality variables are difficult to obtain in real-time,a soft-sensor method is studied to replace the traditional measurement method.In the soft-sensor model,the selection of input variables will have a great impact on the quality of the model.In this thesis,an automatic relevance determination method is introduced in the training phase of GPR to select appropriate input variables.This method considers the nonlinear relationship between input variables and the output variable,it is more suitable for the wastewater treatment process.Afterward,the GPR is used to model the wastewater treatment plant,which has a simple structure and fewer parameters to be adjusted.Finally,the method is applied in an example of a wastewater treatment plant to demonstrate the effectiveness of the method.(3)Process fault detection and isolation for wastewater treatment plants using interval prediction model and Bayesian reasoning.Firstly,the radial basis function neural network is used to model the wastewater treatment plant,and the set membership identification algorithm,which is based on the error-bounded assumption,estimates the output weights of the neural network.Then the neural network outputs a confidence interval,which can ensure that the actual value of the effluent variable is in the range,so the upper and lower bounds of the interval are used as thresholds for fault detection to eliminate false alarms.After a fault is detected,the posterior probability of process faults can be calculated based on the interval prediction model,fault signature matrix and Bayesian reasoning.When the probability exceeds a certain threshold,the fault can be successfully isolated.The final experimental results verify the effectiveness of the method.(4)The wastewater treatment plant process fault diagnosis software.In order to monitor the system intuitively,a set of software for process fault diagnosis of the wastewater treatment plant was designed and developed.First,the requirements for the design of the software are put forward.Then,a specific development plan is formulated according to the requirements.The software mainly includes several functions such as user management,process fault monitoring,log and notification,and related information introduction.The software provides users with concise and intuitive information on the wastewater treatment plant.When a process fault occurs,the staff can respond quickly and minimize the loss.