Digital ear for the determination of ball mill load.

Rusu-Anghel, Stela ; Tirian, Gelu Ovidiu ; Rusu-Anghel, Nicolae 等

1. INTRODUCTION

Ball mills use about 2/3 of the total energy consumed in a cement factory and are driven by high power electric motors: 3000 / 5000 [kW], their role being to grind the clinker and the additives up to a prescribed grain size, in order to obtain cement. Their principle of functioning is simple: a rotary cylinder lifts the balls by means of the centrifugal force, and due to a special construction. After reaching a certain height, they fall over the material that needs to be crushed. The efficiency is low as the energy consumed to lift the balls does not depend on the degree of loading the mill. If the mill is too full, the efficiency diminishes, as the balls fall on a bed of material that is too thick. In order to use energy efficiently, one has to find an optimal degree of loading the mill. This target is achieved by means of a complex system of mill control (Martin et al., 2004), shown in fig. 1.

One can notice from fig. 1 that the real load of the mill needs to be very precisely measured.

Usually, the load level is measured by use of special microphones. The resulting signal (in terms of frequency and intensity) depends on the quantity of material in the mill. Example: when the mill is empty, the sound is produced by the steel balls, hitting one another; when the mill is full, the predominant sound is produced by the impact between the balls and the material.

[FIGURE 1 OMITTED]

There are simple systems using the intensity of the sound in order to determine the charge of the mill (FLS Automation, 2006), (Hecht & Derik, 1996). A complex filtering system allows the calculation of the mill charge. In practice, the calibration of the machine is difficult. The method is not too accurate, as the microphone receives other sounds too, coming from the neighboring mills, whereas for the two-chamber mills, separation is highly difficult.

The more complex systems, such as Smart Fill Level Control by KIMA (Kalkert, 2005) are based on a non-linear algorithm of sound analysis, which eliminates the disadvantages mentioned before, and ensures an acceptable measuring accuracy.

This paper introduces a new system of mill load determination, based on the frequency analysis of the sound it produces. The algorithm used ensures high measurement precision, and the system itself is very cheap.

2. THE METHOD SUGGESTED

In order to measure the noise produced in one of the chambers of the ball mill, we used an acoustic transducer, mounted directly on its walls, phonically isolated from the exterior. In this way, ambient noises do not influence the accuracy of the measurement. Also, the measuring can be done separately for both chambers of the mill, without influencing each other. The signal is sent over to a PC, via wireless, and the transducer is fed by a small electric generator driven by the rotation of the mill.

The signal thus obtained is decomposed into harmonics, by means of Fast Fourier Transform (Brigham, 2002) and the program packages "Signal Math" and "Signal View" (State College, Pennsylvania USA, 1998). An original scanning program determines in real time the frequency for which the amplitude of the signal grows with the load, and the frequency for which the amplitude of the signal diminishes with the load of the mill, with a minimal error in both cases. We used two significant frequencies, in order to grant a high accuracy measurement of the load level. After identifying the frequencies mentioned before, we determined the functions linking the load and the amplitude of the signal, considering both frequencies.

3. INDUSTRIAL APPLICATION

In order to implement the method, we performed trials, using a 2-chamber ball mill, with a maximum capacity of 80 [t/h]. For space reasons, we are going to show only some of the results we obtained.

In fig. 2 and 3 we gave the frequency spectrum for two extreme loads of the mill: 0 [t/h] and 80 [t/h].

Using the scanning program, we identified two characteristic frequencies:

--at 390 [Hz], the amplitude of the noise signal ([y.sub.1]) grows with the mill load (x) according to equation:

[y.sub.1] = -2,1707 x [10.sup.-6] [x.sup.2] + 3,7467 x [10.sup.-4] x + 3,0825 x [10.sup.-3] (1)

--at 878 [Hz] the amplitude (y2) decreases with the load (x) according to equation:

[y.sub.2] = 1,9641 x [10.sup.-6] [x.sup.2] - 3,5498 x [10.sup.-4] x + 1,7053 + [10.sup.-2]. (2)

We created a new function:

[y.sub.1] + 51 x [10.sup.-6] / y = [y.sub.2] / 2, (3)

y = 3,2354 x [10.sup.-6] [x.sup.2] + 4,7154 x [10.sup.-5] x + 4,3369 x [10.sup.-3]. (4)

This function establishes a connection between amplitude and the mill load (in [t/h]), but we considered the load=f(amplitude) to be more useful in practice, for which, at 390 [Hz], results:

[y.sub.p1] = 9,2752 x [10.sup.-4] [x.sup.2] + 2,1123 x [10.sup.3] x - 6,4423, (5)

at 878 [Hz]:

[y.sub.p2] = 1,2746 x [10.sup.5] [x.sup.2] - 6,8032 x [10.sup.3] x + 79,35, (6)

which, combined according to algorithm:

[y.sub.p] = 61,6 - [y.sub.p2] + [y.sub.p1]/2, (7)

give equation:

[y.sub.p] = -1,7355 x [10.sup.4] [x.sup.2] + 4,45 78 x [10.sup.3] x - 12,099. (8)

As can be noticed from fig. 4, the dependency load-amplitude is practically linear in the usual functioning zone of the mill. The calibration of the measuring system can be done starting from a mill with no material (after a capital repair, eventually), by introducing through the feeding system, well determined quantities of material. At the same time, we determined the measuring error, which did not exceed 4,5%, a very good value for this kind of systems.

After the mill starts working under normal conditions, the calculation of the load is done on-line. Later recalibration of the measuring system is possible (whenever necessary). Calculation time is irrelevant, the grinding process being a slow one.

The curve [y.sub.p] determined by relations (8) is given in fig. 4.

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

4. CONCLUSIONS

The paper introduces a highly accurate system of on-line determination of a cement mill load, based on an original algorithm. The information obtained is important for the control of the grinding process, which needs to determine its optimal functioning conditions, at a minimum cost of energy. The researches have been implemented in an industrial cement mill, for both its chambers. As compared to the traditional systems, this one has a series of advantages, such as: high measuring accuracy, simplicity and reliability, low cost (paid back in a few months of functioning), insensibility to turbulences. Proposed measurement system is based on software analysis of data and is simpler, more accurate and cheaper than those on market. Can be integrated directly into a digital control system.

5. REFERENCES

Brigham, E.O. (2002). The Fast Fourier Transform, Prentice Hall, New York

Hecht, H.M. (1996). A low cost automatic mill load level Control Strategy, Proceedings of Cement Industry Technical Conference, pp 348-354, Los Angeles, USA

Kalkert, P. (2005). A new disturbance free, precise level sensor for ball mills, World of Mining, no.4 2005, pp 281-284, ISSN 1613-2408

Martin et al. (2004). Method and Apparatus for Controlling a Non-Liniar Mill, US Patent, no/ US 6.735.483 B2

***(2006). http://www.flsautomation.com, Ball Mill Control by Electric Ear, Accessed: 2008-05-03