Research And Application Of Fault Diagnosis Method Using Extension Neural Network

As the rapid development of science and technology level, process automation is applied more and more extensively in the modern chemical industry. And the system function is more perfect and the system structure is more complex, which result in a lot of security risks. Therefore, the safety of the chemical processes becomes the basis for enterprises to obtain the economic benefits. If some serious failure is occurred, it will inevitably lead to catastrophic accidents and hinder the economic development. From the maintenance of the personal safety and the environment protection, the reliability and security of the system must be effectively protected. Thus, designing an effective method for fault diagnosis and monitoring has attracted considerable attention in scholars and business.In order to overcome the difficulty of fault diagnosis for complex industrial process, a thorough study is made on the modern fault diagnosis technologies. Extreme learning machine, which has fast learning speed and stable generalization performance, is adopted to make fault diagnosis after analyzing and comparing the advantages of several different neural network models. A fuzzy C-means algorithm based extreme learning machine ensemble (FCM-ELME) method and an extension sample classification based extreme learning machine ensemble (ESC-ELME) method are proposed in this research. The main idea of the classification for fault types by fuzzy C-means algorithm is the optimization of the objective function. The extension sample classification algorithm is used to classify the fault types by adjusting the dual weights of extension neural network. For each fault type, a specific extreme learning machine (ELM) is established and trained independently. Then, all specific ELMs are integrated to determine which fault is happened by the majority voting method. Tennessee-Eastman (TE) process is an effectively simulation of complex industrial processes. The proposed FCM-ELME method and ESC-ELME method are used for the fault diagnosis of TE process. The diagnosis accuracy and response speed have been compared with the traditional ELM and a duty-oriented hierarchical artificial neural network. The results demonstrate that the ESC-ELME method provides higher diagnosis accuracy and faster response than other methods, which the extension sample classification method achieves more reasonable clustering result and the proposed ensemble method has higher practicability.