Control Chart Pattern Recognition System: A Fuzzy Neural Network Approach

Control chart is recognized as a kind of effective tools for monitoring and fault diagnosis in statistical process control community, thereby, widely used in modern industrial productions. The main objective of control chart supervision is to detect the abnormal process situations timely and take corrective actions accordingly to avoid possible accidents or waste of materials.In recent years, the applications of intelligent technology in the field of industrial process have been increasingly extensive. In this sense, it becomes an inevitable trend to develop more efficient intelligent methods for control chart pattern recognition. Aiming at taking advantage of prior process knowledge in control chart pattern recognition, a novel fuzzy neural network based hybrid intelligent approach is explicitly introduced in the thesis, whose detailed coverages are presented as follows.1. A novel method to convert control chart discriminant criterions to symbolized fuzzy rules which help guide to model and initialize a specific fuzzy neural network is proposed. Therein, a fuzzy neural network able to combine prior knowledge is explicitly introduced, along with approaches to structure deign and learning algorithms.2. Two-order time delay process models are employed to generate more practical control charts consistent with characteristics of industrial processes. With in-depth analysis of impacts of generalized process model parameters on control charts, a generalized fuzzy neural network based control chart pattern recognition system is further established. Case studies consisting in penicillin fermentation are carried out to demonstrate the validity and effectiveness of the intelligent system.3. An application system concerning control chart pattern recognition using MFC programming (Visual C++) is developed, according to the utility requirements. The overall design of the application system together with its software development and implementation are elaborated. Additionally, screen captures of the friendly easy-access graphical user interface are presented to exemplify the contributions.