| Polyethylene of which PE is short is a kind of common thermoplastic resin and is made from the polymerization of vinyl monomer. Polyethylene was developed in 1922 and it is time of over 60 years from its productization to now. Wha’s more, it has been the first of the global five largest domestic resin. In our country, PE is the variety of the largest production and the largest import volume in the field of synthetic resin and gas-phase polythene technique is more than 60%. Although gas-phase process has numerous advantages, it, compared with solvent method, has had uneven mass and heat transfer phenomenon which directly results in some problems including material agglomeration, material level abnormal fluctuation and unstable flow pattern end so on. If these problems have not been solved promptly, they would bring economic loss and potential risk.In view of the gas-phase polythene technique characteristics and the existing detection technology, this research has designed and implemented the cold fluidized bed experimental device and a gas-phase PE reactor agglomeration monitoring system based on the information fusion. The whole process, at first, has used acoustic emission technology to detect and collect acoustic wave vibration signal form the reactor. The original signal collected, after pretreated, has need data feature extraction and analyses. And, then, the process has established single and multi class gas-phase PE agglomeration monitoring models based on a sensor by using SVDD and completed the recognition. The recognition results have been dealt by the membership function to realize softness. At last, multi-sensor recognition results have been fused by the D-S evidence theory at the decision level. The final result has achieved state recognition including normal material, micro agglomeration material and serious agglomeration material.The large experimental results have shown that the acoustic wave vibration signal could reflect the difference of material particle size in the reactor. MFCC and LPCC feature parameters of the original signal can effectively enhance the between-class distance and increase the data separability. The monitoring model built of many types of even particle material data can complete the recognition of particle mixing state. At last, many recognition results from the different sensors have been fused by D-S evidence theory to promote overall importance of state recognition and reduce uncertainty, which verify the effectiveness of the information fusion technology. The whole process is not complex to be carried out simply and has openness and engineering practical significance. |
