Research On Fault Diagnosis Method That Based On DSP

Along with modern industry system there is a rapid development toward to large-scale, integration and automation direction, the complexity of the system is also rising, occurring system the probability of failure is also growing. Real-time effective condition monitoring and fault diagnosis are an important means of safety and reliability under the guarantee of modern industrial system. In recent years the condition monitoring and fault diagnosis technology have a rapid development, effectively ensure the stable operation of the industrial system and prescient maintenance plays an important role. At the same time, along with the digital signal processing technology and embedded technology of continuous improvement, but also for the condition monitoring and fault diagnosis technology research injected fresh blood.This thesis mainly based on DSP for the fault diagnosis system of fault diagnosis methods research. The first to presentation respectively is widely used in the wavelet analysis method and the clustering analysis method to the theoretical analysis and research, again with DSP as the core to make a diagnosis system design and the transplantation of algorithm. Because the traditional wavelet analysis method and the clustering analysis method in consumption memory space are huge, and the operation time is long, the diagnostic efficiency is low, against the defects of hardware implementation, hindered the DSP operation function of the effective play, and then put forward the corresponding improvement methods.An ensemble real time fault diagnosis method based on lifting wavelet (LW) and recursive incremental clustering (RICLUSTER), called LW-RICLUSTER, is proposed to realize real-time monitoring for complex industrial processes. Firstly, data are denoised by lifting wavelet transform in real-time, then recursive incremental clustering is used for real-time monitoring. With RICLUSTER algorithm, storage space is saved and computing time is shortened, while the adaptability of diagnostic model is increased. Experiment results show that the LW-RICLUSTER algorithm can monitor time-varying process. LW-RICLUSTER is superior to CLUSTER in diagnosis precision, rate and adaptability.