Studies On Fault Diagnosis Based On Wavelet Analysis And SVDD For Construction Machinery

At present, the service model of construction machinery industry is still a“fire-fighting” corrective maintenance. It is a problem that how to forecasting, earlywarning and avoid sudden shutdown. Artificial intelligence has become the mainmethod to solve fault diagnosis problem to complex equipment, for example, expertsystem, neural network-based method, support vector machine-based method. Thesemethods, however, have an "adaptive" problem: when new equipment failure happens,established classifiers need re-training, so it would be a waste of time and not a gooduse of the samples obtained. Support vector data description is a single classificationbased on support vector machine theory. It builds hyper sphere in the feature space todescribe target sample as a whole. In multiple classifiers fault identification, supportvector data description method can train independently to each state sample, andextend the new hyper sphere without the need for re-training.During the operation of the construction machinery, the signal is the informationcarriers of reflecting the equipment running status. It is an important condition andpremises to diagnose the state of equipment that getting real, fully signal reflectingthe status of the device. However, the construction machinery works in the field, theacquired signal is inevitably subject to a variety of noise pollution, and the signal andnoise are non-stationary. Wavelet analysis is a time-scale analysis method, verysuitable for the processing and analysis of non-stationary signals. The signal can beobserved both in time and frequency domain.In this paper, wavelet analysis and support vector data description method areapplied to fault identification of the concrete pump. The modulus maxima algorithmwas used to denoising noise in the signal, then the wavelet packet energy spectrumwas extracted as characteristic parameters. The traditional multi-classification methodcannot effectively deal with the problem of the new category samples. The articleproposed sphere boundary offset criterion, then the diagnostic model can identifyunknown samples, and through unsupervised clustering learning, the model wasupdated. Piston pump of concrete pump truck was studied and the results show thatthe effectiveness of the proposed method.