Research On Weighing Method Based On Digital Filtering And BP Neural Network

Pomelo weight classification plays an important role in pomelo post-harvest treatment.However,current classification equipments were focused on general fruit weighting in factory environment so that most of the pomelo sorting is depends on human handling since almost all of the pomelo produced in middle or small orchards in China.Therefore,to develop a small modular designed weighting and sorting device for pomelo is meaningful for this industry.In this work,a small-scale modular pomelo weighting and sorting was proposed by designing a sorting line consisting of a fruit loading machine,a separation machine and a sorting machine,compiling data acquisition and analysis software based on Lab VIEW and MATLAB and designing a control program of the sorting machine based on PLC to realize automatic fruit loading,set 6 pomelo weight grades,automatic single row,electronic weighing and intelligent sorting.The theoretical sorting speed is 6000/hour.The user could choose whether to add other mechanisms to the core module of weighing and grading according their needs.In addition,aiming at the impact load and weighing signal interference of pomelo single fruit weight on the weighing sensor,the research on the weighing algorithm of signal analysis interval mean weight prediction and the research on the weighing algorithm based on BP neural network were carried out.The main works including:1.A small modular pomelo weight sorting machine was designed.The fruit loading machine of the sorting machine was composed of a belt conveyor with baffles.A fruit box was arranged at the bottom of the fruit loading machine to collect the dumped pomelo.The pomelo was discharged into the sorting machine orderly after being separated from the machine.The sorting machine carries out electronic weighing and automatic grading on the pomelo.2.The control circuit and communication circuit of sorting system were designed and debuged.A single-point cantilever beam sensor was used on that hardware of the sorting machine to generate a weight signal.The signal was amplified by a voltage transmitter and sent to a data acquisition card.The acquisition card was connected with a PC in a serial port communication mode.T he collected voltage signal was transmited to the PC in a digital quantity mode.T he PC obtains the weight after being analyzed and processed by Lab VIEW and MATLAB software.T he weight was transmitted to the PLC through the PC.T he PLC carries out grade judgment on pomelo according to a programmed grading program.T he photoelectric switch is used as a pomelo arrival triggering device.T he triggering signal is transmitted to the PLC,and when the current grade is met.T he PLC outputs a sorting execution signal to control the electromagnet to realize pomelo sorting.3.The electronic weighing software of sorting system was developed,the filtering algorithm and weight prediction model based on Lab VIEW and MATLAB were designed and the grading program and upper computer interface conforming to the grading standard were designed.4.To measure the natural frequency of the dynamic weighing device,unit pulse mode was used to act on the weighing device.T he output signal was analysis for observing signal frequency after Fourier transform.Test shown that the weighing device had a strong formant frequency of 39 Hz in addition to the frequency domain diagram and a weak formant frequency of 54 Hz.In the no-load test of weighing device,the no-load signal is tested at three speeds.The results shown that the main frequencies of the signal varied at different speeds.The main frequency was 82 Hz at 0.33 m/s,85 Hz at 0.4 m/s and 90 Hz at 0.5 m/s,while the peak value of the main frequency increases with the speed increasing.In dynamic weighing signal analysis,5 kinds of digital filters were used for signal filtering and the prediction results of the filtered signal weight were analyzed.The results shown that IIR filter,FIR filter and wavelet denoising have good effects among which IIR filter had the smallest error while adaptive filter had the largest error.For the digital filter algorithm execution time analysis,IIR program running time was 0.9 ms,while 2.78 ms for FIR program,3.085 ms for wavelet denoising program,12.4 ms for NLMS program,and 39.6 ms RLS program running time is.The BP neural network prediction model for pomelo weighting was established based on selected characteristic variables of dynamic weighing signal,the maximum average error and the maximum error in the experiment both occur at a speed of 0.5m/s.The BP neural network algorithm error is smaller than the mean method,and the maximum average error is 2.17%,the maximum error is 4.56%,the sorting accuracy is 92.3%.