Gas-solid Fluidized Bed Agglomeration Monitoring Based On Multi-sensor Voiceprint Feature Extraction

With the continuous development of the ethylene/polyethylene industry,gas-solid fluidized bed owing numerous advantages has become the mainstream industrial production plant of the polyethylene.However,in the actual production process,due to the low heat-transfer efficiency and electrostatic accumulation,the small polyethylene particles may easily aggregate to form agglomeration.If the problems are not been discovered and handled in time,it may lead to the "melting-bed" phenomenon,resulting in an emergency shutdown of the plant,that ultimately affects the safety and stable production of the device.Therefore,it is of great significance to research the on-line monitoring and early-warning technology of polymer agglomeration in fluidized bed,but also quite challenging.So far,there have been some agglomeration monitoring methods,and the acoustic emission(AE)method has attracted more and more attention due to its advantages of non-destructive equipment,convenient installation and real-time reliability.In this paper,the experimental platform based on cold-mold fluidized bed apparatus and the fluidized bed agglomerate monitoring system based on multi-sensor voiceprint feature extraction were established.The AE detection technology was applied to collect acoustic vibration signals generated by fluidized material particles acting on the inner-wall of fluidized bed,and the LP-MFCCs voiceprint feature of the acoustic vibration signals under different conditions(the normal,micro-agglomeration,and agglomeration condition)was extracted and analyzed;Then,the agglomeration monitoring model was established with the extreme learning machine(ELM),futher,the recognition results were expressed in the form of probability by soften dealt,and obtain the ELM network softening-output model.Finally,the fuzzy integration method was used to fuse the recognition results of the three sensors at the decision-making level,and the multi-sensor agglomeration monitoring model was established to achieve comprehensive-monitoring and recognition of the agglomeration status in fluidized bed.The experimental results has shown that there are certain differences in acoustic vibration signals under different particle size polyethylene materials(normal,micro-agglomeration and agglomeration),which can better reflect the changes of the material particle size,and the extracted LP-MFCCs feature of different acoustic signals have good stability and distinguishability;The ELM monitoring model can effectively and accurately recognise the different agglomeration condition of the system,and has a higher recognition rate for the corresponding condition.By fusing the multi-sensors recognition information,the single sensor's uncertainty has been effectively reduced to a certain extent,and the accuracy of the materials agglomeration condition recognition and the overall status monitoring have been also improved.