·Home ·Table of Contents ·Methods and Instrumentation

The Research for the On-Line Metal Detector which Based on Genetic Algorithm Zhao Jiyin, Huang Ting, Gai Shuang, Zhao Jingrong, School of Information Science & Engineering, Jilin University of Technology,Changchun, China 130025 Contact

Abstract:

During the production of chemical product and food, some small metal objects always mix in them, which will damage to the product and instrument. To improve the quality of the product and reduce the economic lost, it is necessary to manufacture a metal detector, which has high sensitivity and reliability. This paper introduce a metal detector based on genetic algorithm of ANN(artificial neural network). The metal detector worked very well and the minimum detecting-accuracy nearly 3mm (test by the diameter of a ball).
Key words: metal detector , genetic algorithm , AC magneto-sensor, Fluxgate Magnetometers

I. Introduction

During the production of chemical product, food and electric power, some small metal objects always mix in them, so we must filter the raw materials and proof-test the finished product and semi-finished product. Only through this way we can exclude the metal impurity. It is necessary to develop a high-sensitivity metal detector. This paper introduce a new metal detector which includes AC magneto-sensor, signal processing unit, data processing unit based on single-chip and ANN based on the genetic algorithm soft. The metal detector exert important effect in the above domain.

II. Measuring Principle of the metal detector

We can use the emission loop and a pair of differential detector to find the metal impurity in the test sample. The main structure is showed as figure 1.

Fig 1: the main structure of metal detector

The samples pass the Fluxgate Magnetometers one by one, which can be regarded as the source of the magnetic field. This will effect circumstance-magnetic field all around. When some small metal objects mix in the samples, the take-over loop will produce remarkable change, we can estimate the result. The Fluxgate Magnetometers phenomenon is a kind of electromagnetism phenomenon [ 1 ] , and the principle obeys Faraday electromagnetism faradic law.

We consider this physical model as the research object. We wire emission loop and take over loop on a magnet core which is made by soft-magnetism material. The area of the cross section is S, the conductance of magnetism is m, the magnetic field intensity constituted by the emission loop is H, and the circle-count of the take-over-loop is W₂.

Supposing the parameter S, m , H and are not functions of t (time), we can get the faradic volt in the faradic coil based on the Faraday electromagnetism faradic law.



When parameter S and W₂ are invariable and the magnet core far from the status of saturation, the parameter mis approximately constant, the faradic volt will only relational with magnetizing field H. If the magnetic field intensity is:



Where H_m is the amplitude of the magnetizing field and f₁ is the frequency of the magnetizing power. The formula (1) will be:



This is the model of ideal transformer.
Because the magnetization-curve of the magnet core is non-linear, the variety of the magnetizing field will bring about the change of the conductance of magnetism (m). In such instance, we can get the real mathematics-model of the transformer as below:



The magnetizing field has two polarity and varies periodically, but the conductance of magnetism(m)only has one polarity and varies one-way. So m(t) is a even function. We can open out m(t) as Fourier-progression:



Where the m_om is the constant of m(t), them_2m, m_4m, m_6mis the amplitude of even syntonic wave.
We take the formula (5) into the formula(4), then we can get the formula(6):



This is mathematical model of transformer effect which magnet core conductance (m) changes following the magnetic field. Compare the formula (6) with formula (3), we can know that there will be odd times waves following the change of magnet core conductance(m)
Any transformer will always locate an environmental magnetic-field, so the addition magnetic field of magnet core includes magnetizing field and environmental magnetic field at least. When we think of the parameter H^// that the environmental field put on the magnet core axis in fact, the formula (4) will be:



When H^// is far smaller than the amplitude of magnet core saturation magnetic field intensity H_sand the amplitude of magnetizing coil magnetic field intensity H_m, its influence to the magnetic conductance can be ignored. Now the first two parts of formula (7) are equal to that of the formula (4), and the last part or formula (7) is the increment e(H₀^//) of faradic voltage e that come from H₀^//. So we can get e(H₀^//) by take the formula (5) to the formula(7).



Formula(8) certified that even syntonic wave in the faradic voltage would change with environment magnetic field if the magnet core conductance(m) changes with magnetizing field intensity.
e(H₀^//) will increase evidently when magnet core works in periodicity supersaturated state, which is named fluxgate magnetometers phenomena. It is evident that we can use the physics phenomena to measure the environment magnetic field. Thus we can detect metal objects mixed in the samples[2].
In the design of fluxgate magnetometers sensor, the magnet core must work in supersaturated state, and only in such status m can change from the max value to the min value periodically and we can receive more stronger signal. So the amplitude of the magnetizing field intensity (H_m) should far more than the amplitude of magnet core magnetic-field intensity (H_s)[3].

IV.The ANN(artificial neural network) design of genetic algorithm

To improve the sensitive degree and the anti-jamming ability of detecting, the metallic detector uses genetic algorithm based on ANN. ANN is a system that simulates the structure of nature nerve cell , which has the ability of self-adaptive and self-organizing and can satisfy the given demand by continually adjusting the joint weight through "studying" or movement. The ANN system in the detector, which adopts the improving multi-layer perceptivity model network, is used to detect and recognize the metal in the chemical product or food by "studying". The genetic algorithm is a self-adaptive algorithm that simulates the biologic evolution in nature, whose frame is:

Procedure genetic algorithm
Begin

t=0
initialize population(t)
evaluate population(t)
while (not termination-condition)do
Begin
t=t+1
select population(t) from population(t-1)
recombine population(t)
evaluate population(t)
End
End

During the t times circulation, every resolution X₁^t in the resolution group of the retentive resolutions in genetic algorithm X₁^t , X₁^t,...X₁^t (the number of the group N is fixed) is "evaluated" through calculating adaptive degree, and on the base of adaptive degree, the individual is selected to recombine and form the new group. In the gene recombination of our system, two operations - "hybridization" and "variation" are important. "Evaluation" shows the performance of the property, and because choosing the individual is based on the principle of superiors win and inferiors lost, the "better" individual will propagate and the "worse" individual will be eliminated. Then, through some evaluations the properties of the whole group will incline to improving, which can make the group incline to the best. In our system, the treat object is the joint weight of ANN whose value can be coded by binary character string figure 2 is the ANN testing and controlling frame of genetic algorithm.

Fig 2: the ANN testing and controlling frame of genetic algorithm

Our system, which uses genetic algorithm to optimize the initial joint weight of ANN and adjusts it by studying during running, detects and controls the detected metal and disturbances in real-time. The running on-the-spot shows that our system achieves high stable and high sensitive detecting and controlling.

III. The main structure of the metal detector

The hardware of the metal detector include AC magneto-sensor, signal processing unit, A/D interface unit, single-chip MCS-51and computer. The main structure is showed as figure 3.

Fig 3: The main structure of the metal detector

When the sample mixed some metal objects through the metal detector, we can display the equal-value of the metal object through the LED, then inform the worker to dispose.

V.Conclusion

This metal detector introduced in this paper which is based on the genetic algorithm soft overcome the yawp of the detector effectively by applying differential take-over coil. This paper uses genetic algorithm of ANN to analysis made this instrument has high sensitivity and reliability. The metal detector worked very well in the factory, the minimum detecting-accuracy nearly 3mm for the ferromagnet and the minimum detecting-accuracy nearly 4mm for the stainless steel matter (test by the diameter of a ball).

Reference

© AIPnD , created by NDT.net

|Home|    |Top|