Scroll Compressor Fault Diagnosis Based On Multiscale Permutation Entropy

Scroll compressor is a new type of positive displacement fluid machinery, which has the advantage of compact structure, reliable operation, energy conservation and environmental protection. It is widely used in refrigerating air-conditioner, medical equipment, transportation, food packaging and other fields. Thus regular condition monitoring and fault diagnosis are very important not only for the stability of system,but for the service life of the equipment. However, researches on the scroll compressor fault diagnosis at present still stay at the single-scale level, which cannot solve the multiscale coupling problem and reflect the dynamic characteristics of the system accurately. Aiming at this situation, a scroll compressor fault feature extraction method based on multiscale permutation was proposed in this paper. By analyzing the characteristics of typical faults, the dynamic characteristic of the fault system on different scales was obtained. The fault diagnosis based on Mahalanobis distance validate the validity and accuracy of this method. Concretely, it includes several aspects as follows:(1) Research the noise reduction method of singular value decomposition based on singular values difference spectrum, the simulation results show that choose valid singular values according to the maximal peak value of singular values difference spectrum can achieve a good result of noise reduction. But when a direct component is included in the signal, this method will cause a lot of useful information loss. At this time, we should choose the valid singular values according to the second largest peak of singular values difference spectrum.(2) In view of the limitation of signal analysis methods of single scale, research the feature extraction method based on multiscale permutation entropy(MPE), which was used to quantitatively describe the energy distribution of vibration signal on different status and different scales. The MPE characteristic curves of four kinds of scroll compressor faults show that, the PE value of bearing looseness fault is the largest, the PE value of mechanical assembly looseness fault is the second, and the PE value of rotor imbalance fault is the minimum, which is consistent with the degree of uncertainty of the system in four kinds of fault states.(3) In order to further illustrate the preponderance of multiscale permutation entropy, respectively taking PE and MPE as the characteristic parameters, the fault diagnosis for scroll compressors was carried out based on the weighted Mahalanobisdistance. The results show that the average diagnostic rate of PE is only 76.9%, while the average diagnostic rate of MPE is 92.5%, which demonstrate that the diagnosis method based on multiscale permutation entropy has obvious superiority.The research results in this paper show that, multiscale permutation entropy basically can reflect the essential characteristics of scroll compressor fault system,meanwhile, the feature weight reflect the distribution of useful information at different scales.