Fault Monitoring And Diagnosis For Papermaking Wastewatertreatment Process Based On Statistical Methods

Fault diagnosis, mainly contains fault detection, identification and separation, namelydetects whether the fault occurred, positioning the region and the types of failure, determinethe size and time of the fault and so on. It is rather meaningful to improve the productionefficiency, product quality and the safety of equipments as well.In this research, four Fault Diagnosis Models (FDM) based on multivariate statisticsanalysis method for two different scales of papermaking wastewater treatment process, labscale and plant scale, are developed separately. Based on the further research on principalcomponent analysis (PCA), a series of fault diagnosis methods such as multi-way principalcomponent analysis (MPCA), improved MPCA and sub-PCA are applied. The PC number ofmodel is selected when accumulative contribution rate is85%and the model used as faultdiagnosis under the condition that the confidence coefficient α are95%and99%. CombiningPC loading and PC scores or just contribution chart is applied to fault identification.For the lab scale papermaking wastewater treatment process,4process variables (ORP,pH, DO, LL) of wastewater are the inputs of PCA model, MPCA model and improved MPCAmodel. Data sample190and265are used as PCA modeling and model verificationrespectively. The study results show that when the number of PC is2, confidence coefficient αis95%, the FDM dovetails the physical fault condition closely, in addition, fault location ofsensors in the wastewater treatment process is further achieved through the combination ofthe scores with the loadings of PCs. Based on the above research,50normal batches areselected to develop MPCA and improved MPCA models, then,55batches including7abnormal batches are used to test FDM. The study results show that PC numbers of the twomodels are14and3. The FDM dovetails the physical fault condition closely while confidencecoefficient α is95%. By comparing the study results, improved MPCA with time-varyingcharacteristics is superior to traditional MPCAin terms of fault detection rate and sensitivity.For plant scale wastewater treatment system, phase transition behaviors in SBRwastewater treatment which contains five phases, therefore phase-based sub-PCA was usedand then compared with improved MPCA. Selecting four variables like fan motor ampere, fanvalve opening, DO and level, which associated with the devices and sensors.37normalbatches of operating data are used as modeling data, one normal and two abnormal batchesare used to modeling validation. Under the influence of various factors, the whole batch isdivided into5phases, and thus develops5PCA models whose PC numbers are2or3. On theother hand, the PC number of improved MPCA is3. By comparing the study results, phase-based sub-PCAcan more detailedly reveal the process operation status and the changesof underlying characteristics over different phases, which will help to improve the onlinemonitoring performance and be superior to improved MPCA.