Explore
Communities in English
Advertise on Engormix

Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness

Published: May 10, 2013
By: Gang Sun (ASABE Member Engineer, Department of Agricultural and Biosystemns Engineering); Robert Fitzgerald (former Post-doctoral Research Associate, Department of Animal Science); Kenneth Stalder (Department of Animal Science); Locke Karriker (Associate Professor, Department of Veterinary Diagnostic and Production Animal Medicine), Anna Johnson (Assistant Professor, Department of Animal Science) and Steven Hoff (Professor, Department of Agricultural and Biosystems Engineering, Iowa State University)
Summary

Measuring sow weight distribution is vital for scientists to identify lame animals before clinical signs can be visually observed and help livestock producers decrease lameness incidence in their swine breeding herd. In this study, an embedded microcomputer-based force plate system was developed to measure vertical forces produced by each limb of the sow and evaluate data accuracy to the sow's known weight. It was found that all tested sows averaged more weight on their front legs than their hind legs and side-to-side weight differences had more variation than front-to-hind distribution. The deviation in front-to-hind weight distribution might be indicative of lameness in both hind or both front feet. To better illustrate the capabilities of the force plate, a 60-s data rolling average protocol was employed for the collected weight data which were recorded every second from each sow leg. The preliminary results indicate that the force plate system was able to identify sow lameness by separately measuring the weight of each leg. Future work will need to evaluate the magnitude of the difference in weight distribution between legs detected by the force plate system in order for producers to effectively determine lameness in sows. 

Keywords. Sow, Lameness, Force plate, Embedded microcomputer, Weight distribution.

INTRODUCTION
Lameness in swine, poultry, horses, and cattle have a large negative economical impact to livestock producers (Corr et al., 2003). Multiple definitions of lameness are used. Merriam-Webster (2008) defines lameness as “having a body part and especially a limb so disabled as to impair freedom of movement” while Wells (1984) defines lameness as ”impaired movement or deviation from normal gait.” Further, abnormal locomotion of pigs has been described as having a shortened stride length, stiff movements, and lowered ability to accelerate and change direction (Main et al., 2000). Locomotor disorders can be associated with neurological disorders, lesions of the hoof or limb, mechanical-structural problems, trauma, or metabolic and infectious disease (Wells, 1984; Smith, 1988). An evaluation of sows at two midwestern U.S. cull sow harvest facilities by Knauer (2006) found that 85% of sows evaluated at harvest had at least one lesion impacting at least one foot, and the authors further noted that lameness is a common reason why sows leave the breeding herd. 
There are numerous methodologies that can be employed to subjectively and objectively measure the degree of lameness an individual animal is expressing at a given point in time. Subjective lameness scoring systems are designed to categorize the degree of lameness expressed while the animal is walking and have been developed for dairy cows (Manson and Leaver, 1988), sheep (Welsh et al., 1993), broilers (Kestin et al., 1992), and other species. These scoring systems for livestock have been implemented so that caretakers can quickly and affordably quantify the prevalence of lameness in the herd on any particular day. However, there can be disagreement between the lameness score assigned to an individual animal (Flower and Weary, 2006); so a standardized objective method for assigning lameness scores to animals would likely be more accurate than subjective scoring measures and provide producers with a useful tool to assess lameness. One such method that shows promise is the force plate measurement system; this device quantifies the amount of force each limb applies to the surface of the assessment tool (Pastell et al., 2008). Force plate measurement systems can measure variables that have been associated with objectively classifying structural abnormalities into degrees of lameness. An animal will distribute less weight on the limb(s) that is painful or structurally unsound (Corr et al., 2003). The use of such equipment has been evaluated in other species such as dogs (Evans et al., 2005), chickens (Corr et al., 2003), dairy (Pastell and Kujala, 2007), and horses (Judy et al., 2001). Force plate measurements are typically obtained during locomotion for these species, which is acceptable because these animals spend a greater portion of their time moving. However, in current housing systems, sows spend the majority of their time in stalls. In this type of housing system, lameness identification more commonly occurs while sows are standing. When sows are housed in pen gestation systems, application of a force plate system could be made with electronic sow feeding systems so that daily evaluations of weight between limbs and total sow weight over time can be monitored. Available force plate measurement systems either do not allow for measurements over time (i.e. current force plates measure approximately 3-5 s) or the instruments are too large to obtain measures from sows in gestation stalls; the most commonly utilized gestation housing system throughout the United States.
Objectively measuring vertical forces produced by sows will potentially allow scientists and producers to identify lame individuals before lameness impairs sow productivity as well as aid producers in their efforts to decrease herd lameness. Further, quantifying the relationship between leg structure and lameness or quantifying ideal weight distribution based on least lameness prevalence would allow producers to accurately implement environmental or genetic programs to decrease the lameness incidence thereby increase sow longevity, improve individual sow welfare, and improve pork producers profitability. Therefore, the objective of this study was to initiate the process of measuring vertical forces produced by each leg of the sow and develop a system to study weight distributions for visually lame and non-lame sows.
MATERIALS AND METHODS
Force Plate Mechanical Design
The force plate was designed with a total dimension of 1524 × 565 × 106 mm (length × width × height), with 6.4-mm thick aluminum plating comprising the top and bottom plate. A semi-flexible epoxy (FlexCoat, Vanberg Specialized Coatings, Lenexa, Kans.) was mixed with sand at the manufacturer's recommendations and applied to the top plate at a desired thickness of 4.8 mm. This epoxy-sand composite provided concrete-like flooring and mimicked what the sow was used to standing on every day. The use of the flooring material attempted to reduce the variability that would have been a result of the sow standing on material that felt foreign. In this manner sows were more likely to express their normal posture and thereby allow for more accurate measurement and reduce slipping while evaluation was occurring. The force plate was designed so that the weight of each of the sow's four limbs would be independently measured. This was accomplished by dividing the top plate into quadrants, each having dimensions of 762 × 279 mm (length × width). A diagram of the force plate is shown in figure 1a. This diagram shows a middle bar that was designed to keep the sow from drifting from side-to-side. 
 Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 1
Figure 1. Diagram of the force plate (1524 × 565 × 106 mm) (a) and the inside of the force plate with four load cells (b)
Force Plate Electronic Design
An embedded microcomputer-based measuring system was designed and built for the force plate device. The whole force plate measuring system, illustrated in figure 2, can be separated into three parts: (1) load cell (load cells 2 and 3 are not shown in fig. 2), (2) signal conditioning, and (3) embedded microcomputer controlling.
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 2
 Figure 2. Diagram of the embedded microcomputer-based force plate measuring system
A single point load cell (VLC-A133, Virtual Measurements and Control, Santa Rosa, Calif.) was placed between the top and bottom plate of each quadrant (i.e. each force plate system needs four load cells, fig. 1b) and utilized to measure real-time force applied to maximum 750- × 750-mm top plate by each of the sow's four limbs. The load cell is an electronic transducer which is able to convert force into an analog electrical signal. The conversion can be implemented by the applied force deforming strain gages which are bonded onto the load cell beam and wired into a Wheatstone bridge configuration. These strain gages are connected into a Wheatstone bridge in order to convert the very small change in resistance of the gages into an electrical signal. The output signal from the VLC-A133 is 2mv/v ±10%, which is linearly proportional to the load applied (load reading). 
It should be noted that choosing the correct load cell capacity is vital and needs to consider several factors contributing to the weight load on the load cell, e.g., a sudden load change resulting from sow's kicking and other moving behaviors during a very short period of time could cause permanent damage to load cells. The capacity for each quadrant of the force plate system can be calculated and obtained from the following, which is based on the following equations (Vishay Revere Transducer Company, Malvern, Pa.).
 C= ((Netload+Liveload*Fa) + Ft + Fw)* Fb      (1)
        Ft =(Liveload *ZS%)/100                     (2)
where
C             = capacity of load cell (kg)
Fa           = dynamic load factor
Ft            = combined effect of zero setting devices
ZS%        = the total zero setting value (2% for VLC-A133 load cells)
Fw           = effect of wind force
Fb            = load factor of safety
Netload    = net load of top force plate (kg)
Liveload   = average weight of each sow's limb (kg)
Sow body weight typically increases with an increase in parity (number of litters). Large sows can reach a body weight equal to about 330 kg (Fitzgerald et al., 2008). Thus, the approximate maximum weight of one sow limb could reach 83 kg and the corresponding combined effect of zero setting devices Ft was equal to 1.66 kg. In this study, the dynamic load factor and load factor of safety were 1.4 and 1.5, respectively. These factors were employed for enhancing the capacity of the load cell and extending its reliability. The net load for each quadrant was constant, about 5 kg. Integrating Ft into equation 1 results in the following load cell capacity: 
 C = ((5+83*1.4) +1.66+ 0)*1.5 = 184.3 kg     (3)
Since the standard capacity of the VLC-A133 load cells are 100, 250, 500, 635, and 1000 kg, the capacity of 250 kg was selected for this study.
An advanced signal conditioning circuit was developed to convert the output signal (2mV/V ± 10%) from the load cell into an electrical signal that met the requirements (0-5 V) of the microcomputer system. The core of the signal conditioning circuit, as shown in figure 3, was a PGA 204 which is a low-cost, programmable-gain instrumentation amplifier with excellent accuracy (PGA 204, Texas Instruments, Dallas, Tex.). The gain was selected by setting address lines A0 and A1 to ground and +5 V, respectively so that the load cell signal was amplified by 100 times for the analog-to- digital (A/D) converters in the data logger. The +5 V, +15 V, and -15 V voltages were provided by high efficiency multiple power supplies (MAP55 AC-DC Power supply, Power One, Campbell, Calif.). A 4.7-μf capacitance was connected to each supply voltage to filter line noise. 
The amplified load cell signal was fed into an embedded data logger (USB-1408FS, Measurement Computing Corporation, Norton, Mass.) with four 14-bit differential A/D converters and 16 input/output (I/O) interfaces. These A/D converters were used to convert the analog voltage from the signal conditioning circuit to the digital number proportional to the magnitude of that voltage. 
Furthermore, the force plate monitoring and recording software was developed based on Visual Basic 6.0 event-driven programming language and integrated development environment (VB6, Microsoft, Redmond, Wash.). The software provided real-time weight measurements for each quadrant and recorded quadrant data for each selected variable time interval. In this case, weights were recorded every second. Weight measurements were recorded for each of the four independent load cells every 1 s over the 30-min evaluation measurement period for each sow.
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 3
Figure 3. A signal conditioning circuit using PGA 204 (Pins 1, 6, 7, and 9: Not use)
Force Plate Validation Method and Procedure
Each force plate load cell and the platform scale used to obtain the reference weight were validated using procedures outlined in Handbook 44 of the National Institute of Standards and Technology (NIST, U.S. Department of Commerce, Gaithersburg, Md.). Four certified weights each weighing 22.68 kg (±0.113 g; Weights and Measures Bureau, Iowa Department of Agriculture and Land Stewardship, Ankeny, Iowa) were used in the validation procedures. The precision of each load cell was validated to ±0.45 kg. For validation, a 68.02-kg weight was placed on a quadrant. The correction factor for each quadrant was calculated by dividing 68.02 by the reading from the measurement system. Thus, values greater than 1 would increase the output weight compared to the original values and values <1 would reduce the output weight by the correction factor. Once correction factors were obtained for each quadrant, quadrants were validated as follows using procedures outlined in the Handbook 44 of the National Institute of Standards and Technology: 1) Each quadrant was warmed up by the operator standing on the quadrant for 30 s then stepping off for 30 s. This step was repeated five times; 2) For each quadrant, the increasing-decreasing procedure was performed for calibration and to estimate hysteresis effects. A 22.7-kg weight was placed on the center position E (fig. 4) of each quadrant; the output weight was read from the accompanied software and was confirmed to be ±1 unit (0.454 kg). This was repeated by incrementally adding 22.7-kg weights and confirming the total weight until a total of 68.02-kg static weight was placed on the quadrant. The total weight was incrementally decreased by 22.7 kg and confirmed until no weight was on the quadrant, and 3) A 45.4-kg weight was placed on the four positions (A, B, C, and D, shown in fig. 4) of the quadrant to test the ability to detect changes in hoof placement on the quadrant. Each reading must be within 0.45 kg. Each quadrant was validated after a pass status was achieved for each step. 
Statistical Analysis
Eight sows were classified as having visual lameness in at least one hoof (n = 4 sows) or normal (n = 4 sows). For each second in the approximate 30-min measurement period, front-to-hind and left-to-right distributions were calculated by adding both the respective quadrants and dividing by the total weight. This equaled to the percentage of weight on the front, left, right, or hind sections during that second. Rolling averages were calculated by averaging the current second with the previous 9 (Lag 10), 19 (Lag 20), …, 59 s (Lag 60). 
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 4
Figure 4. Four positions (A, B, C, and D) of each quadrant for the shift test and one center position (E) for the repeatability and increasing/decreasing test
RESULTS AND DISCUSSION
Validation Results
Validation results for one randomly selected quadrant in the force plate system are shown in table 1. Note that all the quadrants had the same validation results. It can be seen that all the readings were equal to the applied test load on the quadrant, which indicates that the force plate system could accurately measure the known static weights.  
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 5
Data Management
Force plate measurement recording began before each sow stepped on the force plate. Recording ended after the 30-min measurement period. Thus, zero values were recorded at the beginning of the measurement period with very few if any at the end of the period. Additionally, it was common to measure zero values dispersed throughout the data. Data were deleted if the sum of the quadrants for each measurement second were less than 90.7 kg (an arbitrary minimum total weight for which all sows measured weighed more than the cutoff), both front quadrants were less than 4.5 kg (e.g. the sow had no weight on the front quadrants), or both hind quadrants were less than 4.5 kg (e.g. the sow place no weight on the rear quadrants).
The measurement period was pre-determined to last for 30 min per sow to confirm that enough usable data were collected per evaluation. Approximately 1800 measurements were recorded per sow during this period. Future research projects should be performed to determine the optimal measurement frequency and time length.
Preliminary Testing Results
Over 13,843 recorded measurements, the average difference between the total weight calculated from the force plate and the total weight recorded for each sow using the reference scale equaled -2.0 kg, with a 5.22% coefficient of variation. Coefficients of variation for each sow ranged from 1.54% to 13.11%. Table 2 summarizes the average front-to-hind and left-to-right distribution recorded from normal and lame sows. Overall, sows averaged 56.5% [(LF + RF) / (LF + RF + LH + RH)] and 43.5% [(LH + RH) / (LF + RF + LH + RH)] of their total body weight on their front and hind legs, respectively, where LF, RF, LH, and RH indicate the left front, right front, left hind, and right hind legs, respectively. Average front weight distributions ranged from 54.2% to 58.3% for the eight test sows. Over the 30-min measurement period, variations in front-to-hind weights were as large as 2.2% of the weight on the front legs (97.8% of the total body weight on the rear legs) to 95.3% of the sow's total weight on the front legs. For example, Sow 1 placed 2.2% of her weight on her front legs and 97.8% of the total weight on her rear legs for one observation, possibly because sows consumed feed from a trough during the measurement period and they may have reduced the weight on their front legs by resting their lower jaw on the feed trough. Another possibility of why such a large difference in front-to-hind weight distribution would be that the sow shifted her most of her weight to the rear legs for just one second, and that happened to be the second the measurement was taken. Left-to-right weight distribution was sow dependent, in that six of the sows averaged more weight on their right side while two sows averaged more weight on their left side. All sows in this study averaged more weight on their front legs than their hind legs, regardless of lameness status. Based on these results, it appears that sows carry more weight on their front legs than rear legs. Key signatures that would classify sows as lame, especially in their front leg(s), may include when sows average more weight on their rear legs than their front legs. Side-to-side weight differences had more variation than front-to-hind distribution, in that sows shifted their weight from side-to- side more than from front-to-hind. Any large deviation in front-to-hind weight distribution may be indicative of lameness in both hind or both front feet. 
Measurements recorded every second can vary from 0.25 to 2 times the average over the entire period (fig. 5). Hence, it is difficult to identify tendencies in weight distribution when measurements recorded each second from each leg are plotted over time. However, 10- to 60-s rolling averages remove the extreme variation compared to LH Lag 0 (fig. 5) and provide more visually appealing ways to view average weights for individual legs. Although 60-s rolling averages were used in this manuscript to illustrate the capabilities of the force plate, there has been no comparison between the rolling averages and lameness prediction. Future research should evaluate which time length best predicts the severity and location of lameness in sows.
Sixty second rolling averages for each leg are illustrated for a normal (fig. 6) and lame (fig. 7) sow. It is clear to see in figure 7 that the right hind (RH) leg response varied around 30 kg, which was much lower than the other three legs (around 65 kg). Obviously, this sow had lameness in her right hind leg. The observed visual evidence supported this judgment. These two graphs provide proof of concept that this force plate can measure each individual hoof and what happens to individual hoof weights when lameness occurs (in one particular case). Although each sow's lameness status can be visually observed and decided, this may or may not have resulted in the lame hoof to have a lower weight average than the other three hooves (e.g. when sows are lame on both rear legs, both right side legs, etc). Future research projects are aimed at determining what thresholds should be used to discern lame versus normal in a wide variety of cases.
The test sows were weighed on a validated scale prior to being measured on the force platform. Over the combined 2,000 measurements shown in figures 6 and 7 (1,000 measurements each), the average difference in weight between the sum of the four quadrants and the validated scale (not shown) was -0.82 kg with a standard deviation of 9.05 kg. Weight differences ranged from -44.0 to 45.85 kg. Although the range in weight differences was relatively large for these two ranges of measurements, average differences were within 2.5 kg of weights measured on different validated scales. Future improvements to the software may include methods to edit the output values or calculate rolling averages to remove these undesirable extreme weight differences. Furthermore, a study to compare the benefits of removing this extreme variation to the potential loss of capturing weight shifts would be beneficial.
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 9 Figure 5. Left hind (LH) leg weights from a sow observed to have no lameness that were recorded over a 30-min measurement period and illustrated using raw data (Lag 0), or rolling averages of 20 (Lag 20), 30 (Lag 30), or 60 s (Lag 60). Using the force plate system measuring individual weight from sow legs
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 10
 Figure 6. Weight distribution over a 1000 second measurement period of a sow with no observable lameness (LF: Left Front, RF: Right Front, LH:Left Hind, RH: Right Hind; Lag 60 = a 60 s rolling average of the current weight and the previous 59 s)
SUMMARY AND CONCLUSIONS
Calibration results obtained during the validation phase of this study lend evidence to the accuracy of this force plate. Weight data for each quadrant recorded every second, when plotted over time, appeared to contain large weight variation and may visually be difficult to discern a lame animal for which specific limb may be lame; however this variation was reduced when 10- to 60-s rolling averages were calculated for each quadrant. Future work will need to evaluate the magnitude of the difference in weight distribution between legs in order to predict lameness before visual evidence is obvious to observers, and before the degree of lameness impairs productivity. Specifically, this work mainly involves two parts: (1) determining the relationships between the distribution of measured leg weights and different degrees of lameness using the lameness scoring system, and (2) developing lameness diagnosis algorithms to distinguish normal and lame sows and validate prediction algorithms for detecting early stage of lameness. 
Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness - Image 11
Figure 7. Weight distribution over a 1000-s measurement period of a sow with observable lameness in the right hind leg
ACKNOWLEDGEMENTS
The authors graciously thank Andrew Blackburn and Ivan Hankins of the Iowa Weights and Measures Bureau for providing the calibrated weights and training, Marc Lott of Iowa State University for providing assistance during fabrication of the force plate, and Dan Johnson and employees at the Iowa State Swine Nutrition Farm. This paper of the Iowa Agriculture and Home Economics Experiment Station, Ames, Iowa, Project No. 3801, was supported by Hatch Act and State of Iowa Funds.
REFERENCES
Merriam-Webster. 2008. Merriam-Webster online dictionary. Available at: http://www.merriam-webster.com/.
Corr, S. A., C. C. McCorquodale, R. E. McGovern, M. J. Gentle, and D. Bennett. 2003. Evaluation of ground reaction forces produced by chickens walking on a force plate. American J. Vet. Res. 64(1): 76-82. 
Evans, R., C. Horstman, and M. Conzemius. 2005. Accuracy and optimization of force platform gait analysis in Labradors with cranial cruciate disease evaluated at a walking gait. Vet. Surg. 34(5): 445-449. 
Fitzgerald, R. F., K. J. Stalder, C. D. Johnson, L. L. Layman, and L. A. Karriker. 2008. An economic analyis of feeding cull sows. Prof. Anim. Sci. 24(4): 355-362. 
Flower, F. C., and D. M. Weary. 2006. Effect of hoof pathologies on subjective assessments of dairy cow gait. J. Dairy Sci. 89(1): 139-146.
Judy, C. E., L. D. Galuppo, J. R. Snyder, and N. H. Willits. 2001. Evaluation of an in-shoe pressure measurement system in horses. American J. Vet. Res. 62(1): 23-28. 
Kestin, S. C., T. G. Knowles, A. E. Tinch, and N. G. Gregory. 1992. Prevalence of leg weakness in broiler chickens and its relationship with genotype. Vet. Rec. 131(9): 190-194. 
Knauer, M. T. 2006. Factors influencing sow longevity. MS thesis. Ames, Iowa: Iowa State Univ. Main, D. C. J., J. Clegg, A. Spatz, and L. E. Green. 2000. Repeatability of a lameness scoring system for finishing pigs. Vet. Rec. 147(20): 574-576.
Manson, F. J., and J. D. Leaver. 1988. The influence of concentrate amount on locomotion and clinical lameness in dairy cattle. Animal Production 47(2): 185-190. 
Pastell, M. E., M. Hautala, V. Poikalainen, J. Praks, I. Veermae, M. Kujals, and J. Ahokas. 2008. Automatic observation of cow leg health using load sensors. Computers and Electronics in Agric. 62(1): 48-53.
Pastell, M. E., and M. Kujala. 2007. A probabilistic neural network model for lameness detection. J. Dairy Sci. 90(5): 2283-2292. 
Smith, B. 1988. Lameness in pigs associated with foot and limb disorders. In Pract. 10(3): 113-117. 
Wells, G. A. H. 1984. Locomotor disorders of the pig. In Pract. 6(2): 43-53. 
Welsh, E. M., M. Gettinby, and A. M. Nolan. 1993. Comparison of a visual analogue scale and a numerical rating scale for assessment of lameness, using sheep as a model. American J. Vet. Res. 54(6): 976-983.
Related topics
Authors:
Steven Hoff
Iowa State University
Follow
Join to be able to comment.
Once you join Engormix, you will be able to participate in all content and forums.
* Required information
Would you like to discuss another topic? Create a new post to engage with experts in the community.
Create a post
Join Engormix and be part of the largest agribusiness social network in the world.
LoginRegister