Correction of the Dynamic Effect in Weight Measurement using the Load Cell

Transcription

1 Correction of the Dynamic Effect in Weight Measurement using the Load Cell Nabil Mohamad Usamah School of Mechanical Engineering, Universiti Sains Malaysia, Penang, MALAYSIA Mohamad Izudin Alisah School of Mechanical Engineering, Universiti Sains Malaysia, Penang, MALAYSIA Zaidi Mohd Ripin School of Mechanical Engineering, Universiti Sains Malaysia, Penang, MALAYSIA Abstract: Weight measurement is an important task in metering of food using load cell which usually mounted on a structure connected to other packaging mechanism. The metering of the food is usually done using vibration conveyor or auger conveyor. In both cases, there is a vertical separation between the outlet of the conveyor and also the weighing plate. The vertical drop of the food onto the weighing plate together with the delay of the response of the system usually resulting in overweight or overfill because of the impact. In this study, a load cell is mounted on a vibrating base to simulate the vibrating structure and grains are poured on the weighing plate supported by a load cell. The dynamic response of the system is measured to determine the settling time. Correction of the dynamic effect can be achieved by using filtering techniques. Filtering is applied by using the exponential smoothing filter to the measured data to get the corrected data. The filtered value is compared with the static measurement to determine the error of the approach. In this approach by using the exponential smoothing filter, the error due to the dynamic impact of the grain on the weighing system is reduced to less than two percent. The results showed that the corrected value produces error of less than two percent of the set value at faster time rate. Keywords: Load cell measurement; Dynamic response; Signal correction; Automation Weighing; Environmental noise. 1 INTRODUCTION Load cells are sensors widely employed in automatic weighing instruments such as checkweighers in the food industry. As productivity is highly dependent on the time taken to measure the weight it is crucial to minimize the duration of the weighing process. This entails that the dynamic effect from the weighing will needs to be compensated or corrected if the value of the measurand is to be determined before stable equilibrium (Jafaripanah et al., 2005; Recommendation et al., 2006). Some techniques have been proposed in literature in order to overcome the limitations related to the use of traditional low-pass linear filters to load cell response correction. Filtering schemes based on model (Jafaripanah et al., 2005; Shi et al., 1993; Piskorowski et al., 2008; Halimic et al, 1995) and without modeling (Yasin et al., 1999; Shu & W. Q., 1993; Hernandez & W., 2006; Pietrzak & P., 2009) have been shown to be of value using singledegree-of-freedom spring-mass-system model (Jafaripanah et al., 2005; Shi et al., 1993; Piskorowski et al., 2008; Halimic et al., 1995). Artificial neural network (Yasin et al., 1999) and recursive least square procedures (Shu & W. Q., 1993; Hernandez & W., 2006) have also been implemented and a discrete nonstationary filter has been proposed in (Pietrzak & P., 2009). These were made by assuming that load cells are fitted on rigid and clamped housings and therefore they just address the problem of compensating load cell under-damped vibrational dynamics which may not suitable for low frequency disturbances. This problem is overcome by the approach proposed in (Boschetti et al., 2013) where a network of supplementary accelerometers and the multibody model of the system comprising the housing and the load cell and apply a compensation technique which estimated the effects of environmental vibrations on load cell signals. These effects can then be easily removed from the signals and hence the cell response can be corrected. Signal compensation techniques based on sensor networks have been proposed (A. Deraemaeker & A. Preumont., 2006) where the outputs of a large network of sensors are combined into a single output by means of a linear combiner acting as a modal filter, i.e., to mimic the behavior of a single degree of freedom system for vibration-based damage detection. The identification of the base vibration is important and one of the way is the use of accelerometers to estimate 81

2 the forces induced by vibrations with the aim of compensating the load cell measurements and estimating the actual force generated (Beebe & J. C., 2004; Nakamura et al., 2000). Recently, signal compensation techniques based on sensor networks have also been applied to weighing machines. In (Cuadrado et al., 2010) dummy load cells are employed to estimate environmental vibrations and to perform a signal compensation of the cells adopted for weighing. Since load cell dynamics is affected by the mass suspended to it, the dummy load cells are loaded with a mass similar to the one of the buckets connected to the load cells whose signal is compensated. It is clear that the above work is relatively complicated and costly to implement and a simpler approach is required. In this work compensation of the dynamic effect from the impact of the food items on the weighing system using a strain-gauge based load cell is evaluated and compensated for the dynamic effect. The efficacy of the approach is presented here. 2 APPLICATION OF EXPONENTIAL SMOOTHING FILTER 2.1 Known Mass Drop Test on Load Cell A drop test was conducted to measure the dynamic response of load cell under exponential smoothing filters. This is to determine the best exponential smoothing filters coefficient to speed the dynamic response of the load cell into steady state value (+2% and -0% of target value). The experiment also measures the effect of different release height of the mass on both filtered and non-filtered dynamic response of the load cell. Figure 3: Drop test experimental set up Figure 2: Drop Test Area (a) (b) Figure 3: Programmed measurement VI NI-LabVIEW & Implementation of exponential smoothing filter Figure 4: Raw and filtered dynamic response of the load cell under drop test. A drop test was performed using 200 grams steel dead weight on the magnetic lock mechanism at different height as shown in figure 2. Laser measurement sensor was used to set the drop test height. At first, 200-gram mass is attach/stick on the magnetic lock mechanism at height of 5mm and filter coefficient of to is selected in NI LabVIEW software with the increment of every The VI of LabVIEW programming is shown in Figure 3. The magnetic lock is manually opened (turn clockwise) in order to drop the mass on the weighing base which is placed on the load cell. A 3Kg Mettler Toledo Load Cell and HBM Load Cell controller (MVD2510) are used to measure the dynamic response of dropped mass. Finally, the analogue data of load cell dynamic response measured by the HBM load cell controller are recorded using an analogue input 9234 of NI compact DAQ. The experiment is repeated for the drop height of 10mm, 15mm, 20mm and 25mm. 82

3 2.2 Controlled Environmental Noise Test on Load Cell This experiment is to determine the suitable exponential smoothing filters coefficient in reducing the effect of the environmental noise on the load cell dynamic response. Figure 5: Environmental noise test experimental set up Figure 6: Environmental noise test area The experiment is done by installing vibrational actuator system on the main base to simulate the environmental noise on the weighing plate as shown in figure 5 and 6. A voice coil actuator is mounted on the main base to excite the voice coil actuator. Four springs are installed under the main base to increase the flexibility of the system. NI 9234 compact DAQ is used as a signal generator to control the amplifier. Two small accelerometers A and B are placed on the weighing plate and voice coil actuator accordingly. The accelerometers output are connected to two channel of analogue input NI 9234 compact DAQ where the acceleration of the base and actuator are obtained. The same load cell and controller are used and the analogue data is measured by the NI 9234 compact DAQ channel. The measurement runs on the GUI VI of LabVIEW software as shown in Figure 3. At first, the GUI VI is turned on to activate the NI DAQ measurement system. The exponential smoothing fiter coefficient is selected from to 0.1 with an exponential increment due to the lack effectiveness differentiation of the filters at higher coefficient. The voice coil actuator is excited at 10Hz with constant amplitude gain. The cell response which is affected by the environmental noise is masured and filtered. The effectiveness of the exponential smoothing filters in reducing the noise on the cell response is based on the reduction percentage of RMS between filtered and non-filtered cell response. The experiment is repeated for the environmental noise of 20Hz and 30Hz. 3 RESULTS & DISCUSSIONS Figure 7 shows the dynamic response of the load cell for a drop test from release height of 5mm and the response when subjected to filter. It can be seen from the figure that the unfiltered data curve gives a higher value than the filtered data which highlighted the problem of unfiltered load cell data where convergence to the static value requires a much longer time than the filtered data. This is on top of the computational effort and the resulting lag from this process. This is one of the main motivation why an alternative approach to filtering is needed. Referring to figure 8, the settling time of raw cell response on variable release height at all testing points are plotted. Its shows that the height does not affecting the settling time of the cell responses. The fluctuation of the raw settling time is caused by the uncertainty of the impact type between perfectly plastic and perfectly elastic impact. However, the total average of all testing points trend line is calculated to measure the effectiveness of the exponential smoothing filters in reducing the settling time of the cell response. Next, the settling time of the filtered cell responses are plotted in figure 9 with total average settling time of raw cell responses. It is shown that, all release height shows a same trends of settling time as the filter coefficient increases. Comparing to the settling of the raw cell responses, the settling of filtered cell responses lags behind in a lower filter coefficient and as the coefficient increase, the filtered cell response produces the fastest settling time for that particular release height as shown in figure 7. The settling time will gradually increase and getting slower than raw cell response s settling time as the filter coefficient increases. Focusing on the release height, the filter coefficient that produce fastest settling time shift toward higher filter coefficient as the release height increases. This is because the increase of release height caused the load cell response to have a faster rising time and higher overshoot peak while the raw cell response settling time keep constant, a higher exponential filter coefficient is required to reduce the rising time and remove the overshoot of the cell response. 83

4 Figure 7: Example of Dynamic Cell Response under drop test. (mass = 0.2kg, height = 5mm, filter coefficient = 0.038) Figure 8: Drop Test Settling time of Raw Cell Response for different release height. Figure 10 shows the result of environmental noise reduction on cell response using exponential smoothing filter. For all type of environmental noise frequency, the trends of noise reduction percentage as filter coefficient increases are the same, where the reduction percentage was negative exponentially increases as the filter coefficient increases. Focusing on the noise frequencies, a higher noise frequency has a greater noise reduction percentage gradient and reduces as the frequencies decreases. Figure 11(a) shows the response of the load cell when subjected to the base excitation. The larger amplitude shows the load cell reading before filtering and the smaller amplitude curve represents the filtered reading. The acceleration excitation of the base is shown in figure 11(b) represented by the smaller curve and the transmitted acceleration of the weighing plate is shown as the larger curve, the acceleration transmissibility of the base plate-load cell system is less than unity to this frequency (20Hz). The frequency spectrum of the load cell response is shown in figure 12(a). It is clear that the dominant frequency of the base vibration is 20Hz with a harmonic at 40Hz. Similar acceleration spectrum of the weighing plate can also be overserved in figure 12(b) where the dominant frequency is 20Hz and a harmonic at 40Hz. 84

5 Figure 9: Drop Test Settling time versus Filter coefficient of filtered cell response at variable release height. Figure 10: Environmental noise reduction versus filter coefficient on load cell responses at variable noise frequencies. 85

6 Shi, W. J., White, N. M., and Brignell, J. E. (1993). Adaptive filters in load cell response correction. Sensors and Actuators A: Physical, 37, Piskorowski, J., and Barcinski, T. (2008). Dynamic compensation of load cell response: A time-varying approach. Mechanical Systems and Signal Processing, 22(7), (a) (b) Figure 11: (a) Raw(white) and filtered(red) load cell response under environmental noise (b) Main base(white) and weighing plate(red) accelerations (a) (b) Figure 12: (a) Frequency spectrum of load cell response (b) Frequency spectrum of acceleration on weighing plate 4 CONCLUSIONS Based on the drop test, the focused area of filter coefficient is between to 0.075, where it produces the faster settling time for the 5mm to 25mm release height. For the environmental noise test, it shows that a higher filter coefficient produces greater noise compensation effect. In an industrial weighing system, the conditions of load cell in weighing process are the impact of product free fall and environmental noise. A parametric study of filter coefficient is required to identify the filter for both conditions with the fastest settling time. Application of a simple exponential smoothing filter is implemented in a weighing system. ACKNOWLEDGMENTS The authors would like to thank the Ministry of Higher Education and USM for providing the facilities and the research opportunity. REFERENCES Jafaripanah, M., Al-Hashimi, B. M., and White, N. M. (2005). Application of analog adaptive filters for dynamic sensor compensation. IEEE Transactions on Instrumentation and Measurement, 54(1), Halimic, M., and Balachandran, W. (1995). Kalman filter for dynamic weighing system. In Industrial Electronics, ISIE'95., Proceedings of the IEEE International Symposium on (Vol. 2, pp ). IEEE. Yasin, S. A., and White, N. M. (1999). Application of artificial neural networks to intelligent weighing systems. IEE proceedings-science, Measurement and Technology, 146(6), Shu, W. Q. (1993). Dynamic weighing under nonzero initial conditions. IEEE transactions on instrumentation and measurement, 42(4), Hernandez, W. (2006). Improving the response of a load cell by using optimal filtering. Sensors, 6(7), Pietrzak, P. (2009). Fast filtration method for static automatic catchweighing instruments using a nonstationary filter. Metrology and Measurement Systems, 16(4), Boschetti, G., Caracciolo, R., Richiedei, D., and Trevisani, A. (2013). Model-based dynamic compensation of load cell response in weighing machines affected by environmental vibrations. Mechanical Systems and Signal Processing, 34(1), A. Deraemaeker, A. Preumont, Vibration based damage detection using large array sensors and spatial filters, Mech. Syst. Sig. Process. 20 (2006) Beebe, J. C. (2004). U.S. Patent No. 6,834,559. Washington, DC: U.S. Patent and Trademark Office. Nakamura, Y., Wakasa, Y., Naito, K., Tokumaru, H., and Tajiri, S. (2000). U.S. Patent No. 6,034,334. Washington, DC: U.S. Patent and Trademark Office. Cuadrado, J., Dopico, D., Naya, M. A., and Pastorino, R. (2010, May). Automotive observers based on multibody models and the extended Kalman filter. In Proceedings of the 1st Joint International Conference on Multibody System Dynamics (IMSD 2010) (pp ). Recommendation, O. I. M. L. (2006). R 76-1: Nonautomatic weighing instruments, Part 1: Metro logical and technical requirements Tests. OIML, Paris. 86