Sequential fuzzy clustering based dynamic fuzzy neural network for fault diagnosis and prognosis

In recent years, increasing demands for more efficient high speed milling (HSM) processes have propelled the application and development of more effective modeling methods and intelligent machining approaches. Hence, a desired reference model has to incorporate more efficient and sophisticated feature extraction and artificial intelligence (AI) techniques aiming for more repeatability and generalizability.In our work, wavelet analysis is applied for feature extraction to provide a better insight into time–frequency changes of the process. Considering the high dimension of extracted wavelet features, dimension reduction methods, such as clustering techniques are inevitable. They are applied as an interpretation layer between the feature extraction and artificial intelligence subsystems to form a new model structure for HSM with generalizability and sequential learning capability.We introduce a new architecture that incorporates the advantages of fuzzy clustering into well known dynamic fuzzy neural networks (DFNN) to form an online condition monitoring system which is tolerant to slight drifts in process dynamics and adaptable to variations in parameters and device. Sequential Fuzzy Clustering Dynamic based Fuzzy Neural Networks (SFCDFNN) is developed and successfully applied for monitoring of an HSM process. It is able to sequentially learn the model and adapt itself to variations and also provide an estimation or prediction on the status of the process. It facilitates for nonintrusive fault diagnosis and prognosis. Lastly, its performance on modeling of experimental data is practically illustrated and compared to its other counterparts.