The Design System And Algorithm Research Of Monitoring Agglomeration In Gas-Solid Fluidized Bed

Gas-solid fluidized bed process is very simple in industrial, itsrequirement to the equipment is not very high, and its economic benefit is veryimpressive, and its no harm to the ecosystem and environment. But in theprocess of production, it has very high requirements about raw materials,auxiliary materials and catalyst. If staff use wrong way to operate it, fluidizedbed will generate agglomeration and it will affect the quality of the products,more serious is that serious accident happens at sometime. When clearing theagglomeration in the gas-solid fluidized bed must result fluidized bed to stopproduction and lead to losses in economic. Clearing-up agglomeration influidized bed is not easy. So fluidized bed agglomeration monitoring inindustrial is very important. Alarms will occur, when agglomeration appear inthe fluidized bed. Alert ringtones remind the staff to change the parameters ofthe fluidized bed and lead to product under normal production conditions.Therefore, the agglomeration monitoring system of fluidized bed has a verybroad application prospects in the industry.The agglomeration monitoring system in a fluidized bed detect soundsignals of fluidized bed and let sound signals do the analog to digital conversion, let digital signal is transported to the computer via the serialcommunication interface, then using computer analyze and process digitalsignal, the finally, it will form a set of agglomeration monitoring system offluidized bed. When designing the monitoring system, the system is dividedinto two parts of hardware and software: the function of hardware part is ADCand transmission; the function of software part is the signal analysis andprocessing. The hardware section contains using sensors to collect soundsignals, the circuit of the analog to digital conversion, peripheral circuit designof FPGA and transmission of data, using the Verilog HDL language toachieving the function of designing. The software section contain using theMexican hat wavelet function to wavelet packet decompose the signal, usingMel Frequency Cepstrum Coefficient to extract the voiceprint signal, and thenusing PCA method to reduce the dimension of the signal, and finally using theBP neural network to train the signal.