Data Collected Using Computer-Vision Screening System Helps Plant Improve Process Control on Poultry Kill Line

Georgia Tech’s computer-vision screening system is currently on-line at Gold Kist’s poultry processing plant in Carrollton, Ga. Installed on two of the plant’s kill lines, the system is collecting data that is being analyzed to identify trends that can be used to better control processing operations.

Recent advancements in computer-vision technologies now make it possible to collect a wealth of information pertaining to process performance in poultry processing plants. This type of data collection can provide poultry plants with information that enables personnel to better understand how product quality is varying and to implement control measures to reduce process conditions that may be contributing to that variability. As a result, productivity and profitability can be significantly improved throughout the plant.

In a study nearing completion at Gold Kist’s Carrollton, Ga., poultry processing plant, Georgia Tech researchers and plant management are discovering that kill line product quality screening data can be a valuable tool in identifying trends that can be used to better control processing operations.

The Georgia Tech/Gold Kist partnership began several years ago when Georgia Tech researchers sought a plant in which to test an on-line imaging system they had developed. Initially, a single imaging cell was installed on one of the plant’s two kill lines, primarily to screen for systemically defective birds (septicemia/toxemia, unbled birds, and severe overscalding). As the system began to prove its reliability, a second cell was placed on the other line, and new features were added that included the ability to track the occurrence rates for broken wings and bruising. These new data sets soon proved to be extremely valuable to plant management.

“The wealth of data produced by the system is valuable in its own right,” comments John Stewart, Georgia Tech’s lead investigator on the project. “Alarms can be set to be triggered when defect rates exceed pre-set levels, quickly alerting managers to problems on the line as they are occurring so corrective actions can be taken as quickly as possible to reduce re-work and downgrades.” Post analysis of the data, he explains, also allows managers to compare the performance of different processing shifts and growers. These comparisons can help identify more subtle problems needing correction.

However, to be effective, more information is needed relative to understanding the relationships between defect levels measured by the inspection system and the factors contributing to them. Ongoing study in this area is being conducted using funds from Georgia’s Traditional Industries Program for Food Processing in coordination with the Food Processing Advisory Council (FoodPAC).

“Kill room operations are the one part of the process that can be dynamically controlled to influence the defect levels being observed by the screening system. As these settings change, their impact on quality is immediately seen by the screening system. Therefore, relationships between screening system measurements and kill room settings offer a strong basis for automated supervisory control,” says Stewart.

Studying those relationships, nonetheless, called for a more in-depth tracking system for kill line operations than currently exists. This was achieved by instrumenting one line at the plant with additional sensors to monitor scalder temperatures, picking machine positions and motor current, stunner settings, and environmental conditions. BOC-Thinkage, an industrial partner on the project, donated the necessary measurement instrumentation.

Figure 1. Histogram showing a typical breast width distribution of birds being processed during the study.

As the study has progressed, the team has begun to observe that bird size plays an important role in broken wing rates. After experimenting with several size measurement options, including total projected area, breast width, and total length, researchers have found that the strongest relationship exists between breast width and broken wings. Analysis of production data where the pickers were set for optimum feather removal shows that birds with wider breast widths had more than twice the rate of broken wings compared to smaller birds. A typical distribution of breast widths is shown in Figure 1.

While these findings are confirming suspicions that picker settings have the highest potential impact on the rates, further analysis is still ongoing. One of the main challenges to be faced is balancing the need for effective feather removal with the need to minimize wing breakage. Thus far the team has clearly shown that high breakage rates can be decreased on a consistent basis by reducing the aggressiveness of the picker settings. However, less aggressive picker settings can result in poor feather removal efficiency. In addition, Stewart says, the team is evaluating the impacts of stunner settings and scalder temperatures on broken wing rates and discoloration.

Regardless of the outcome, the study is already proving that the imaging cell can provide feedback to operators who are adjusting the pickers, enabling them to immediately understand the repercussions of those adjustments on wing breakage. This is an important step forward in striking a better balance between feather removal efficiency and the rate of broken wings.

The team is also working with plant personnel in the hopes of devising control schemes that might eventually allow kill room machinery to be changed automatically to minimize the defect rate measured by the vision system.

“The imaging system will be a valuable tool in the near future. It will enable production to track flock size, percent of bruising, percent of wing damage, along with other quality issues. This will give production time to react to the quality of product being sent to second processing,” adds Thomas Bradford, unit manager 1st processing at Gold Kist-Carrollton.

The Georgia Tech imaging system is the target of ongoing commercialization efforts that include a partnership with Gainco, Inc., which is currently marketing the design under an exclusive manufacturing license. In addition, BOC-Thinkage holds a use license on the design and is currently offering it as part of its in-plant optimization service.